[llvm] 94620e4 - [CostModel][X86] Add CostKinds handling for vector shift by generic/non-uniform shift amounts
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Thu Sep 15 09:02:36 PDT 2022
Author: Simon Pilgrim
Date: 2022-09-15T16:51:58+01:00
New Revision: 94620e4fc340dc392e863af1c86441a20511ebd7
URL: https://github.com/llvm/llvm-project/commit/94620e4fc340dc392e863af1c86441a20511ebd7
DIFF: https://github.com/llvm/llvm-project/commit/94620e4fc340dc392e863af1c86441a20511ebd7.diff
LOG: [CostModel][X86] Add CostKinds handling for vector shift by generic/non-uniform shift amounts
These are the worst case generic vector shift costs, where nothing is known about the shift amounts - in particular this should stop us using the default sizelatency cost of 1 for so many pre-AVX2 vector shifts that can often actually expand during lowering to +20 uops, just for 128-bit vectors, resulting in some horrible inline/unroll decisions.
This was achieved with an updated version of the 'cost-tables vs llvm-mca' script D103695 (I'll update the patch soon for reference)
Added:
Modified:
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
llvm/test/Analysis/CostModel/X86/div-codesize.ll
llvm/test/Analysis/CostModel/X86/div-latency.ll
llvm/test/Analysis/CostModel/X86/div-sizelatency.ll
llvm/test/Analysis/CostModel/X86/div.ll
llvm/test/Analysis/CostModel/X86/fshl.ll
llvm/test/Analysis/CostModel/X86/fshr.ll
llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
llvm/test/Analysis/CostModel/X86/mul.ll
llvm/test/Analysis/CostModel/X86/rem-codesize.ll
llvm/test/Analysis/CostModel/X86/rem-latency.ll
llvm/test/Analysis/CostModel/X86/rem-sizelatency.ll
llvm/test/Analysis/CostModel/X86/rem.ll
llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
llvm/test/Analysis/CostModel/X86/vshift-ashr-cost-inseltpoison.ll
llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
llvm/test/Analysis/CostModel/X86/vshift-lshr-cost-inseltpoison.ll
llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
llvm/test/Analysis/CostModel/X86/vshift-shl-cost-inseltpoison.ll
llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
llvm/test/Transforms/SLPVectorizer/X86/alternate-int-inseltpoison.ll
llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 8e2c801559f1a..f2fdc541217f9 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -776,25 +776,25 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
return LT.first * KindCost.value();
static const CostKindTblEntry AVX512BWCostTable[] = {
- { ISD::SHL, MVT::v16i8, { 4 } }, // extend/vpsllvw/pack sequence.
- { ISD::SRL, MVT::v16i8, { 4 } }, // extend/vpsrlvw/pack sequence.
- { ISD::SRA, MVT::v16i8, { 4 } }, // extend/vpsravw/pack sequence.
- { ISD::SHL, MVT::v32i8, { 4 } }, // extend/vpsllvw/pack sequence.
- { ISD::SRL, MVT::v32i8, { 4 } }, // extend/vpsrlvw/pack sequence.
- { ISD::SRA, MVT::v32i8, { 6 } }, // extend/vpsravw/pack sequence.
- { ISD::SHL, MVT::v64i8, { 6 } }, // extend/vpsllvw/pack sequence.
- { ISD::SRL, MVT::v64i8, { 7 } }, // extend/vpsrlvw/pack sequence.
- { ISD::SRA, MVT::v64i8, { 15 } }, // extend/vpsravw/pack sequence.
-
- { ISD::SHL, MVT::v8i16, { 1 } }, // vpsllvw
- { ISD::SRL, MVT::v8i16, { 1 } }, // vpsrlvw
- { ISD::SRA, MVT::v8i16, { 1 } }, // vpsravw
- { ISD::SHL, MVT::v16i16, { 1 } }, // vpsllvw
- { ISD::SRL, MVT::v16i16, { 1 } }, // vpsrlvw
- { ISD::SRA, MVT::v16i16, { 1 } }, // vpsravw
- { ISD::SHL, MVT::v32i16, { 1 } }, // vpsllvw
- { ISD::SRL, MVT::v32i16, { 1 } }, // vpsrlvw
- { ISD::SRA, MVT::v32i16, { 1 } }, // vpsravw
+ { ISD::SHL, MVT::v16i8, { 4, 8, 4, 5 } }, // extend/vpsllvw/pack sequence.
+ { ISD::SRL, MVT::v16i8, { 4, 8, 4, 5 } }, // extend/vpsrlvw/pack sequence.
+ { ISD::SRA, MVT::v16i8, { 4, 8, 4, 5 } }, // extend/vpsravw/pack sequence.
+ { ISD::SHL, MVT::v32i8, { 4, 23,11,16 } }, // extend/vpsllvw/pack sequence.
+ { ISD::SRL, MVT::v32i8, { 4, 30,12,18 } }, // extend/vpsrlvw/pack sequence.
+ { ISD::SRA, MVT::v32i8, { 6, 13,24,30 } }, // extend/vpsravw/pack sequence.
+ { ISD::SHL, MVT::v64i8, { 6, 19,13,15 } }, // extend/vpsllvw/pack sequence.
+ { ISD::SRL, MVT::v64i8, { 7, 27,15,18 } }, // extend/vpsrlvw/pack sequence.
+ { ISD::SRA, MVT::v64i8, { 15, 15,30,30 } }, // extend/vpsravw/pack sequence.
+
+ { ISD::SHL, MVT::v8i16, { 1, 1, 1, 1 } }, // vpsllvw
+ { ISD::SRL, MVT::v8i16, { 1, 1, 1, 1 } }, // vpsrlvw
+ { ISD::SRA, MVT::v8i16, { 1, 1, 1, 1 } }, // vpsravw
+ { ISD::SHL, MVT::v16i16, { 1, 1, 1, 1 } }, // vpsllvw
+ { ISD::SRL, MVT::v16i16, { 1, 1, 1, 1 } }, // vpsrlvw
+ { ISD::SRA, MVT::v16i16, { 1, 1, 1, 1 } }, // vpsravw
+ { ISD::SHL, MVT::v32i16, { 1, 1, 1, 1 } }, // vpsllvw
+ { ISD::SRL, MVT::v32i16, { 1, 1, 1, 1 } }, // vpsrlvw
+ { ISD::SRA, MVT::v32i16, { 1, 1, 1, 1 } }, // vpsravw
{ ISD::ADD, MVT::v64i8, { 1, 1, 1, 1 } }, // paddb
{ ISD::ADD, MVT::v32i16, { 1, 1, 1, 1 } }, // paddw
@@ -822,26 +822,33 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
return LT.first * KindCost.value();
static const CostKindTblEntry AVX512CostTable[] = {
- { ISD::SHL, MVT::v4i32, { 1 } },
- { ISD::SRL, MVT::v4i32, { 1 } },
- { ISD::SRA, MVT::v4i32, { 1 } },
- { ISD::SHL, MVT::v8i32, { 1 } },
- { ISD::SRL, MVT::v8i32, { 1 } },
- { ISD::SRA, MVT::v8i32, { 1 } },
- { ISD::SHL, MVT::v16i32, { 1 } },
- { ISD::SRL, MVT::v16i32, { 1 } },
- { ISD::SRA, MVT::v16i32, { 1 } },
-
- { ISD::SHL, MVT::v2i64, { 1 } },
- { ISD::SRL, MVT::v2i64, { 1 } },
- { ISD::SHL, MVT::v4i64, { 1 } },
- { ISD::SRL, MVT::v4i64, { 1 } },
- { ISD::SHL, MVT::v8i64, { 1 } },
- { ISD::SRL, MVT::v8i64, { 1 } },
-
- { ISD::SRA, MVT::v2i64, { 1 } },
- { ISD::SRA, MVT::v4i64, { 1 } },
- { ISD::SRA, MVT::v8i64, { 1 } },
+ { ISD::SHL, MVT::v64i8, { 15, 19,27,33 } }, // vpblendv+split sequence.
+ { ISD::SRL, MVT::v64i8, { 15, 19,30,36 } }, // vpblendv+split sequence.
+ { ISD::SRA, MVT::v64i8, { 37, 37,51,63 } }, // vpblendv+split sequence.
+
+ { ISD::SHL, MVT::v32i16, { 11, 16,11,15 } }, // 2*extend/vpsrlvd/pack sequence.
+ { ISD::SRL, MVT::v32i16, { 11, 16,11,15 } }, // 2*extend/vpsrlvd/pack sequence.
+ { ISD::SRA, MVT::v32i16, { 11, 16,11,15 } }, // 2*extend/vpsravd/pack sequence.
+
+ { ISD::SHL, MVT::v4i32, { 1, 1, 1, 1 } },
+ { ISD::SRL, MVT::v4i32, { 1, 1, 1, 1 } },
+ { ISD::SRA, MVT::v4i32, { 1, 1, 1, 1 } },
+ { ISD::SHL, MVT::v8i32, { 1, 1, 1, 1 } },
+ { ISD::SRL, MVT::v8i32, { 1, 1, 1, 1 } },
+ { ISD::SRA, MVT::v8i32, { 1, 1, 1, 1 } },
+ { ISD::SHL, MVT::v16i32, { 1, 1, 1, 1 } },
+ { ISD::SRL, MVT::v16i32, { 1, 1, 1, 1 } },
+ { ISD::SRA, MVT::v16i32, { 1, 1, 1, 1 } },
+
+ { ISD::SHL, MVT::v2i64, { 1, 1, 1, 1 } },
+ { ISD::SRL, MVT::v2i64, { 1, 1, 1, 1 } },
+ { ISD::SRA, MVT::v2i64, { 1, 1, 1, 1 } },
+ { ISD::SHL, MVT::v4i64, { 1, 1, 1, 1 } },
+ { ISD::SRL, MVT::v4i64, { 1, 1, 1, 1 } },
+ { ISD::SRA, MVT::v4i64, { 1, 1, 1, 1 } },
+ { ISD::SHL, MVT::v8i64, { 1, 1, 1, 1 } },
+ { ISD::SRL, MVT::v8i64, { 1, 1, 1, 1 } },
+ { ISD::SRA, MVT::v8i64, { 1, 1, 1, 1 } },
{ ISD::ADD, MVT::v64i8, { 3, 7, 5, 5 } }, // 2*paddb + split
{ ISD::ADD, MVT::v32i16, { 3, 7, 5, 5 } }, // 2*paddw + split
@@ -909,16 +916,16 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
static const CostKindTblEntry AVX2ShiftCostTable[] = {
// Shifts on vXi64/vXi32 on AVX2 is legal even though we declare to
// customize them to detect the cases where shift amount is a scalar one.
- { ISD::SHL, MVT::v4i32, { 2 } }, // vpsllvd (Haswell from agner.org)
- { ISD::SRL, MVT::v4i32, { 2 } }, // vpsrlvd (Haswell from agner.org)
- { ISD::SRA, MVT::v4i32, { 2 } }, // vpsravd (Haswell from agner.org)
- { ISD::SHL, MVT::v8i32, { 2 } }, // vpsllvd (Haswell from agner.org)
- { ISD::SRL, MVT::v8i32, { 2 } }, // vpsrlvd (Haswell from agner.org)
- { ISD::SRA, MVT::v8i32, { 2 } }, // vpsravd (Haswell from agner.org)
- { ISD::SHL, MVT::v2i64, { 1 } }, // vpsllvq (Haswell from agner.org)
- { ISD::SRL, MVT::v2i64, { 1 } }, // vpsrlvq (Haswell from agner.org)
- { ISD::SHL, MVT::v4i64, { 1 } }, // vpsllvq (Haswell from agner.org)
- { ISD::SRL, MVT::v4i64, { 1 } }, // vpsrlvq (Haswell from agner.org)
+ { ISD::SHL, MVT::v4i32, { 2, 3, 1, 3 } }, // vpsllvd (Haswell from agner.org)
+ { ISD::SRL, MVT::v4i32, { 2, 3, 1, 3 } }, // vpsrlvd (Haswell from agner.org)
+ { ISD::SRA, MVT::v4i32, { 2, 3, 1, 3 } }, // vpsravd (Haswell from agner.org)
+ { ISD::SHL, MVT::v8i32, { 4, 4, 1, 3 } }, // vpsllvd (Haswell from agner.org)
+ { ISD::SRL, MVT::v8i32, { 4, 4, 1, 3 } }, // vpsrlvd (Haswell from agner.org)
+ { ISD::SRA, MVT::v8i32, { 4, 4, 1, 3 } }, // vpsravd (Haswell from agner.org)
+ { ISD::SHL, MVT::v2i64, { 2, 3, 1, 1 } }, // vpsllvq (Haswell from agner.org)
+ { ISD::SRL, MVT::v2i64, { 2, 3, 1, 1 } }, // vpsrlvq (Haswell from agner.org)
+ { ISD::SHL, MVT::v4i64, { 4, 4, 1, 2 } }, // vpsllvq (Haswell from agner.org)
+ { ISD::SRL, MVT::v4i64, { 4, 4, 1, 2 } }, // vpsrlvq (Haswell from agner.org)
};
if (ST->hasAVX512()) {
@@ -945,31 +952,31 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
static const CostKindTblEntry XOPShiftCostTable[] = {
// 128bit shifts take 1cy, but right shifts require negation beforehand.
- { ISD::SHL, MVT::v16i8, { 1 } },
- { ISD::SRL, MVT::v16i8, { 2 } },
- { ISD::SRA, MVT::v16i8, { 2 } },
- { ISD::SHL, MVT::v8i16, { 1 } },
- { ISD::SRL, MVT::v8i16, { 2 } },
- { ISD::SRA, MVT::v8i16, { 2 } },
- { ISD::SHL, MVT::v4i32, { 1 } },
- { ISD::SRL, MVT::v4i32, { 2 } },
- { ISD::SRA, MVT::v4i32, { 2 } },
- { ISD::SHL, MVT::v2i64, { 1 } },
- { ISD::SRL, MVT::v2i64, { 2 } },
- { ISD::SRA, MVT::v2i64, { 2 } },
+ { ISD::SHL, MVT::v16i8, { 1, 3, 1, 1 } },
+ { ISD::SRL, MVT::v16i8, { 2, 3, 1, 1 } },
+ { ISD::SRA, MVT::v16i8, { 2, 3, 1, 1 } },
+ { ISD::SHL, MVT::v8i16, { 1, 3, 1, 1 } },
+ { ISD::SRL, MVT::v8i16, { 2, 3, 1, 1 } },
+ { ISD::SRA, MVT::v8i16, { 2, 3, 1, 1 } },
+ { ISD::SHL, MVT::v4i32, { 1, 3, 1, 1 } },
+ { ISD::SRL, MVT::v4i32, { 2, 3, 1, 1 } },
+ { ISD::SRA, MVT::v4i32, { 2, 3, 1, 1 } },
+ { ISD::SHL, MVT::v2i64, { 1, 3, 1, 1 } },
+ { ISD::SRL, MVT::v2i64, { 2, 3, 1, 1 } },
+ { ISD::SRA, MVT::v2i64, { 2, 3, 1, 1 } },
// 256bit shifts require splitting if AVX2 didn't catch them above.
- { ISD::SHL, MVT::v32i8, { 2+2 } },
- { ISD::SRL, MVT::v32i8, { 4+2 } },
- { ISD::SRA, MVT::v32i8, { 4+2 } },
- { ISD::SHL, MVT::v16i16, { 2+2 } },
- { ISD::SRL, MVT::v16i16, { 4+2 } },
- { ISD::SRA, MVT::v16i16, { 4+2 } },
- { ISD::SHL, MVT::v8i32, { 2+2 } },
- { ISD::SRL, MVT::v8i32, { 4+2 } },
- { ISD::SRA, MVT::v8i32, { 4+2 } },
- { ISD::SHL, MVT::v4i64, { 2+2 } },
- { ISD::SRL, MVT::v4i64, { 4+2 } },
- { ISD::SRA, MVT::v4i64, { 4+2 } },
+ { ISD::SHL, MVT::v32i8, { 4, 7, 5, 6 } },
+ { ISD::SRL, MVT::v32i8, { 6, 7, 5, 6 } },
+ { ISD::SRA, MVT::v32i8, { 6, 7, 5, 6 } },
+ { ISD::SHL, MVT::v16i16, { 4, 7, 5, 6 } },
+ { ISD::SRL, MVT::v16i16, { 6, 7, 5, 6 } },
+ { ISD::SRA, MVT::v16i16, { 6, 7, 5, 6 } },
+ { ISD::SHL, MVT::v8i32, { 4, 7, 5, 6 } },
+ { ISD::SRL, MVT::v8i32, { 6, 7, 5, 6 } },
+ { ISD::SRA, MVT::v8i32, { 6, 7, 5, 6 } },
+ { ISD::SHL, MVT::v4i64, { 4, 7, 5, 6 } },
+ { ISD::SRL, MVT::v4i64, { 6, 7, 5, 6 } },
+ { ISD::SRA, MVT::v4i64, { 6, 7, 5, 6 } },
};
// Look for XOP lowering tricks.
@@ -985,28 +992,6 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
return LT.first * KindCost.value();
}
- static const CostKindTblEntry SSE2UniformShiftCostTable[] = {
- // Uniform splats are cheaper for the following instructions.
- { ISD::SHL, MVT::v16i16, { 2+2 } }, // 2*psllw + split.
- { ISD::SHL, MVT::v8i32, { 2+2 } }, // 2*pslld + split.
- { ISD::SHL, MVT::v4i64, { 2+2 } }, // 2*psllq + split.
-
- { ISD::SRL, MVT::v16i16, { 2+2 } }, // 2*psrlw + split.
- { ISD::SRL, MVT::v8i32, { 2+2 } }, // 2*psrld + split.
- { ISD::SRL, MVT::v4i64, { 2+2 } }, // 2*psrlq + split.
-
- { ISD::SRA, MVT::v16i16, { 2+2 } }, // 2*psraw + split.
- { ISD::SRA, MVT::v8i32, { 2+2 } }, // 2*psrad + split.
- { ISD::SRA, MVT::v2i64, { 4 } }, // 2*psrad + shuffle.
- { ISD::SRA, MVT::v4i64, { 8+2 } }, // 2*(2*psrad + shuffle) + split.
- };
-
- if (ST->hasSSE2() && Op2Info.isUniform())
- if (const auto *Entry =
- CostTableLookup(SSE2UniformShiftCostTable, ISD, LT.second))
- if (auto KindCost = Entry->Cost[CostKind])
- return LT.first * KindCost.value();
-
if (ISD == ISD::SHL && !Op2Info.isUniform() && Op2Info.isConstant()) {
MVT VT = LT.second;
// Vector shift left by non uniform constant can be lowered
@@ -1058,28 +1043,22 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
return LT.first * KindCost.value();
static const CostKindTblEntry AVX2CostTable[] = {
- { ISD::SHL, MVT::v16i8, { 6 } }, // vpblendvb sequence.
- { ISD::SHL, MVT::v32i8, { 6 } }, // vpblendvb sequence.
- { ISD::SHL, MVT::v64i8, { 12 } }, // 2*vpblendvb sequence.
- { ISD::SHL, MVT::v8i16, { 5 } }, // extend/vpsrlvd/pack sequence.
- { ISD::SHL, MVT::v16i16, { 7 } }, // extend/vpsrlvd/pack sequence.
- { ISD::SHL, MVT::v32i16, { 14 } }, // 2*extend/vpsrlvd/pack sequence.
-
- { ISD::SRL, MVT::v16i8, { 6 } }, // vpblendvb sequence.
- { ISD::SRL, MVT::v32i8, { 6 } }, // vpblendvb sequence.
- { ISD::SRL, MVT::v64i8, { 12 } }, // 2*vpblendvb sequence.
- { ISD::SRL, MVT::v8i16, { 5 } }, // extend/vpsrlvd/pack sequence.
- { ISD::SRL, MVT::v16i16, { 7 } }, // extend/vpsrlvd/pack sequence.
- { ISD::SRL, MVT::v32i16, { 14 } }, // 2*extend/vpsrlvd/pack sequence.
-
- { ISD::SRA, MVT::v16i8, { 17 } }, // vpblendvb sequence.
- { ISD::SRA, MVT::v32i8, { 17 } }, // vpblendvb sequence.
- { ISD::SRA, MVT::v64i8, { 34 } }, // 2*vpblendvb sequence.
- { ISD::SRA, MVT::v8i16, { 5 } }, // extend/vpsravd/pack sequence.
- { ISD::SRA, MVT::v16i16, { 7 } }, // extend/vpsravd/pack sequence.
- { ISD::SRA, MVT::v32i16, { 14 } }, // 2*extend/vpsravd/pack sequence.
- { ISD::SRA, MVT::v2i64, { 2 } }, // srl/xor/sub sequence.
- { ISD::SRA, MVT::v4i64, { 2 } }, // srl/xor/sub sequence.
+ { ISD::SHL, MVT::v16i8, { 6, 21,11,16 } }, // vpblendvb sequence.
+ { ISD::SHL, MVT::v32i8, { 6, 23,11,22 } }, // vpblendvb sequence.
+ { ISD::SHL, MVT::v8i16, { 5, 18, 5,10 } }, // extend/vpsrlvd/pack sequence.
+ { ISD::SHL, MVT::v16i16, { 8, 10,10,14 } }, // extend/vpsrlvd/pack sequence.
+
+ { ISD::SRL, MVT::v16i8, { 6, 27,12,18 } }, // vpblendvb sequence.
+ { ISD::SRL, MVT::v32i8, { 8, 30,12,24 } }, // vpblendvb sequence.
+ { ISD::SRL, MVT::v8i16, { 5, 11, 5,10 } }, // extend/vpsrlvd/pack sequence.
+ { ISD::SRL, MVT::v16i16, { 8, 10,10,14 } }, // extend/vpsrlvd/pack sequence.
+
+ { ISD::SRA, MVT::v16i8, { 17, 17,24,30 } }, // vpblendvb sequence.
+ { ISD::SRA, MVT::v32i8, { 18, 20,24,43 } }, // vpblendvb sequence.
+ { ISD::SRA, MVT::v8i16, { 5, 11, 5,10 } }, // extend/vpsravd/pack sequence.
+ { ISD::SRA, MVT::v16i16, { 8, 10,10,14 } }, // extend/vpsravd/pack sequence.
+ { ISD::SRA, MVT::v2i64, { 4, 5, 5, 5 } }, // srl/xor/sub sequence.
+ { ISD::SRA, MVT::v4i64, { 8, 8, 5, 9 } }, // srl/xor/sub sequence.
{ ISD::SUB, MVT::v32i8, { 1, 1, 1, 2 } }, // psubb
{ ISD::ADD, MVT::v32i8, { 1, 1, 1, 2 } }, // paddb
@@ -1169,27 +1148,32 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
{ ISD::SUB, MVT::v2i64, { 1, 1, 1, 1 } }, // psubq
{ ISD::ADD, MVT::v2i64, { 1, 1, 1, 1 } }, // paddq
- { ISD::SHL, MVT::v32i8, { 22 } }, // pblendvb sequence + split.
- { ISD::SHL, MVT::v8i16, { 6 } }, // pblendvb sequence.
- { ISD::SHL, MVT::v16i16, { 13 } }, // pblendvb sequence + split.
- { ISD::SHL, MVT::v4i32, { 3 } }, // pslld/paddd/cvttps2dq/pmulld
- { ISD::SHL, MVT::v8i32, { 9 } }, // pslld/paddd/cvttps2dq/pmulld + split
- { ISD::SHL, MVT::v2i64, { 2 } }, // Shift each lane + blend.
- { ISD::SHL, MVT::v4i64, { 6 } }, // Shift each lane + blend + split.
-
- { ISD::SRL, MVT::v32i8, { 23 } }, // pblendvb sequence + split.
- { ISD::SRL, MVT::v16i16, { 28 } }, // pblendvb sequence + split.
- { ISD::SRL, MVT::v4i32, { 6 } }, // Shift each lane + blend.
- { ISD::SRL, MVT::v8i32, { 14 } }, // Shift each lane + blend + split.
- { ISD::SRL, MVT::v2i64, { 2 } }, // Shift each lane + blend.
- { ISD::SRL, MVT::v4i64, { 6 } }, // Shift each lane + blend + split.
-
- { ISD::SRA, MVT::v32i8, { 44 } }, // pblendvb sequence + split.
- { ISD::SRA, MVT::v16i16, { 28 } }, // pblendvb sequence + split.
- { ISD::SRA, MVT::v4i32, { 6 } }, // Shift each lane + blend.
- { ISD::SRA, MVT::v8i32, { 14 } }, // Shift each lane + blend + split.
- { ISD::SRA, MVT::v2i64, { 5 } }, // Shift each lane + blend.
- { ISD::SRA, MVT::v4i64, { 12 } }, // Shift each lane + blend + split.
+ { ISD::SHL, MVT::v16i8, { 10, 21,11,17 } }, // pblendvb sequence.
+ { ISD::SHL, MVT::v32i8, { 22, 22,27,40 } }, // pblendvb sequence + split.
+ { ISD::SHL, MVT::v8i16, { 6, 9,11,11 } }, // pblendvb sequence.
+ { ISD::SHL, MVT::v16i16, { 13, 16,24,25 } }, // pblendvb sequence + split.
+ { ISD::SHL, MVT::v4i32, { 3, 11, 4, 6 } }, // pslld/paddd/cvttps2dq/pmulld
+ { ISD::SHL, MVT::v8i32, { 9, 11,12,17 } }, // pslld/paddd/cvttps2dq/pmulld + split
+ { ISD::SHL, MVT::v2i64, { 2, 4, 4, 6 } }, // Shift each lane + blend.
+ { ISD::SHL, MVT::v4i64, { 6, 7,11,15 } }, // Shift each lane + blend + split.
+
+ { ISD::SRL, MVT::v16i8, { 11, 27,12,18 } }, // pblendvb sequence.
+ { ISD::SRL, MVT::v32i8, { 23, 23,30,43 } }, // pblendvb sequence + split.
+ { ISD::SRL, MVT::v8i16, { 13, 16,14,22 } }, // pblendvb sequence.
+ { ISD::SRL, MVT::v16i16, { 28, 30,31,48 } }, // pblendvb sequence + split.
+ { ISD::SRL, MVT::v4i32, { 6, 7,12,16 } }, // Shift each lane + blend.
+ { ISD::SRL, MVT::v8i32, { 14, 14,26,34 } }, // Shift each lane + blend + split.
+ { ISD::SRL, MVT::v2i64, { 2, 4, 4, 6 } }, // Shift each lane + blend.
+ { ISD::SRL, MVT::v4i64, { 6, 7,11,15 } }, // Shift each lane + blend + split.
+
+ { ISD::SRA, MVT::v16i8, { 21, 22,24,36 } }, // pblendvb sequence.
+ { ISD::SRA, MVT::v32i8, { 44, 45,51,76 } }, // pblendvb sequence + split.
+ { ISD::SRA, MVT::v8i16, { 13, 16,14,22 } }, // pblendvb sequence.
+ { ISD::SRA, MVT::v16i16, { 28, 30,31,48 } }, // pblendvb sequence + split.
+ { ISD::SRA, MVT::v4i32, { 6, 7,12,16 } }, // Shift each lane + blend.
+ { ISD::SRA, MVT::v8i32, { 14, 14,26,34 } }, // Shift each lane + blend + split.
+ { ISD::SRA, MVT::v2i64, { 5, 6,10,14 } }, // Shift each lane + blend.
+ { ISD::SRA, MVT::v4i64, { 12, 12,22,30 } }, // Shift each lane + blend + split.
{ ISD::FNEG, MVT::v4f64, { 2, 2, 1, 2 } }, // BTVER2 from http://www.agner.org/
{ ISD::FNEG, MVT::v8f32, { 2, 2, 1, 2 } }, // BTVER2 from http://www.agner.org/
@@ -1259,16 +1243,19 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
static const CostKindTblEntry SSE41CostTable[] = {
- { ISD::SHL, MVT::v16i8, { 10 } }, // pblendvb sequence.
- { ISD::SHL, MVT::v8i16, { 11 } }, // pblendvb sequence.
- { ISD::SHL, MVT::v4i32, { 4 } }, // pslld/paddd/cvttps2dq/pmulld
+ { ISD::SHL, MVT::v16i8, { 15, 24,17,22 } }, // pblendvb sequence.
+ { ISD::SHL, MVT::v8i16, { 11, 14,11,11 } }, // pblendvb sequence.
+ { ISD::SHL, MVT::v4i32, { 14, 20, 4,10 } }, // pslld/paddd/cvttps2dq/pmulld
- { ISD::SRL, MVT::v16i8, { 11 } }, // pblendvb sequence.
- { ISD::SRL, MVT::v8i16, { 13 } }, // pblendvb sequence.
- { ISD::SRL, MVT::v4i32, { 16 } }, // Shift each lane + blend.
+ { ISD::SRL, MVT::v16i8, { 16, 27,18,24 } }, // pblendvb sequence.
+ { ISD::SRL, MVT::v8i16, { 22, 26,23,27 } }, // pblendvb sequence.
+ { ISD::SRL, MVT::v4i32, { 16, 17,15,19 } }, // Shift each lane + blend.
+ { ISD::SRL, MVT::v2i64, { 4, 6, 5, 7 } }, // splat+shuffle sequence.
- { ISD::SRA, MVT::v16i8, { 21 } }, // pblendvb sequence.
- { ISD::SRA, MVT::v8i16, { 13 } }, // pblendvb sequence.
+ { ISD::SRA, MVT::v16i8, { 38, 41,30,36 } }, // pblendvb sequence.
+ { ISD::SRA, MVT::v8i16, { 22, 26,23,27 } }, // pblendvb sequence.
+ { ISD::SRA, MVT::v4i32, { 16, 17,15,19 } }, // Shift each lane + blend.
+ { ISD::SRA, MVT::v2i64, { 8, 17, 5, 7 } }, // splat+shuffle sequence.
{ ISD::MUL, MVT::v4i32, { 2, 11, 1, 1 } } // pmulld (Nehalem from agner.org)
};
@@ -1281,20 +1268,20 @@ InstructionCost X86TTIImpl::getArithmeticInstrCost(
static const CostKindTblEntry SSE2CostTable[] = {
// We don't correctly identify costs of casts because they are marked as
// custom.
- { ISD::SHL, MVT::v16i8, { 13 } }, // cmpgtb sequence.
- { ISD::SHL, MVT::v8i16, { 25 } }, // cmpgtw sequence.
- { ISD::SHL, MVT::v4i32, { 16 } }, // pslld/paddd/cvttps2dq/pmuludq.
- { ISD::SHL, MVT::v2i64, { 4 } }, // splat+shuffle sequence.
-
- { ISD::SRL, MVT::v16i8, { 14 } }, // cmpgtb sequence.
- { ISD::SRL, MVT::v8i16, { 16 } }, // cmpgtw sequence.
- { ISD::SRL, MVT::v4i32, { 12 } }, // Shift each lane + blend.
- { ISD::SRL, MVT::v2i64, { 4 } }, // splat+shuffle sequence.
-
- { ISD::SRA, MVT::v16i8, { 27 } }, // unpacked cmpgtb sequence.
- { ISD::SRA, MVT::v8i16, { 16 } }, // cmpgtw sequence.
- { ISD::SRA, MVT::v4i32, { 12 } }, // Shift each lane + blend.
- { ISD::SRA, MVT::v2i64, { 8 } }, // srl/xor/sub splat+shuffle sequence.
+ { ISD::SHL, MVT::v16i8, { 13, 21,26,28 } }, // cmpgtb sequence.
+ { ISD::SHL, MVT::v8i16, { 24, 27,16,20 } }, // cmpgtw sequence.
+ { ISD::SHL, MVT::v4i32, { 17, 19,10,12 } }, // pslld/paddd/cvttps2dq/pmuludq.
+ { ISD::SHL, MVT::v2i64, { 4, 6, 5, 7 } }, // splat+shuffle sequence.
+
+ { ISD::SRL, MVT::v16i8, { 14, 28,27,30 } }, // cmpgtb sequence.
+ { ISD::SRL, MVT::v8i16, { 16, 19,31,31 } }, // cmpgtw sequence.
+ { ISD::SRL, MVT::v4i32, { 12, 12,15,19 } }, // Shift each lane + blend.
+ { ISD::SRL, MVT::v2i64, { 4, 6, 5, 7 } }, // splat+shuffle sequence.
+
+ { ISD::SRA, MVT::v16i8, { 27, 30,54,54 } }, // unpacked cmpgtb sequence.
+ { ISD::SRA, MVT::v8i16, { 16, 19,31,31 } }, // cmpgtw sequence.
+ { ISD::SRA, MVT::v4i32, { 12, 12,15,19 } }, // Shift each lane + blend.
+ { ISD::SRA, MVT::v2i64, { 8, 11,12,16 } }, // srl/xor/sub splat+shuffle sequence.
{ ISD::AND, MVT::v16i8, { 1, 1, 1, 1 } }, // pand
{ ISD::AND, MVT::v8i16, { 1, 1, 1, 1 } }, // pand
diff --git a/llvm/test/Analysis/CostModel/X86/div-codesize.ll b/llvm/test/Analysis/CostModel/X86/div-codesize.ll
index 0966b5238bcbb..801b5f4e9ec06 100644
--- a/llvm/test/Analysis/CostModel/X86/div-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/div-codesize.ll
@@ -1,14 +1,14 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512F
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512BW
;
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX1
define i32 @sdiv() {
@@ -270,61 +270,80 @@ define i32 @udiv_uniformconst() {
}
define i32 @sdiv_constpow2() {
-; SSE-LABEL: 'sdiv_constpow2'
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LABEL: 'sdiv_constpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 91 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 181 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 187 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 373 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 271 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 541 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sdiv_constpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 91 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 181 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 139 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 277 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 157 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 313 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX1-LABEL: 'sdiv_constpow2'
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 83 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 166 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 98 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 196 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 137 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 274 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX2-LABEL: 'sdiv_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 122 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'sdiv_constpow2'
@@ -337,13 +356,13 @@ define i32 @sdiv_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 137 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'sdiv_constpow2'
@@ -360,9 +379,9 @@ define i32 @sdiv_constpow2() {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = sdiv i64 undef, 16
@@ -389,24 +408,119 @@ define i32 @sdiv_constpow2() {
}
define i32 @udiv_constpow2() {
-; CHECK-LABEL: 'udiv_constpow2'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LABEL: 'udiv_constpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'udiv_constpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'udiv_constpow2'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'udiv_constpow2'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'udiv_constpow2'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'udiv_constpow2'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = udiv i64 undef, 16
%V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
diff --git a/llvm/test/Analysis/CostModel/X86/div-latency.ll b/llvm/test/Analysis/CostModel/X86/div-latency.ll
index f891c7c4dc578..6afb9c790e352 100644
--- a/llvm/test/Analysis/CostModel/X86/div-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/div-latency.ll
@@ -1,14 +1,14 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512F
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512BW
;
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SLM
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX1
define i32 @sdiv() {
@@ -270,61 +270,80 @@ define i32 @udiv_uniformconst() {
}
define i32 @sdiv_constpow2() {
-; SSE-LABEL: 'sdiv_constpow2'
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LABEL: 'sdiv_constpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 73 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 145 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 115 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 229 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 89 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 177 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 353 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sdiv_constpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 103 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 205 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 157 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 313 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 110 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 219 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 437 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX1-LABEL: 'sdiv_constpow2'
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 49 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 184 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 115 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 230 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX2-LABEL: 'sdiv_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 71 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 142 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'sdiv_constpow2'
@@ -337,13 +356,13 @@ define i32 @sdiv_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 71 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'sdiv_constpow2'
@@ -360,28 +379,28 @@ define i32 @sdiv_constpow2() {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 57 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SLM-LABEL: 'sdiv_constpow2'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 103 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 205 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 157 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 313 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 110 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 219 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 437 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = sdiv i64 undef, 16
@@ -408,24 +427,138 @@ define i32 @sdiv_constpow2() {
}
define i32 @udiv_constpow2() {
-; CHECK-LABEL: 'udiv_constpow2'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LABEL: 'udiv_constpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'udiv_constpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'udiv_constpow2'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'udiv_constpow2'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'udiv_constpow2'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'udiv_constpow2'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SLM-LABEL: 'udiv_constpow2'
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SLM-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = udiv i64 undef, 16
%V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
diff --git a/llvm/test/Analysis/CostModel/X86/div-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/div-sizelatency.ll
index 2eaff0fa55602..d54dae0a21a06 100644
--- a/llvm/test/Analysis/CostModel/X86/div-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/div-sizelatency.ll
@@ -1,14 +1,14 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+ssse3 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512F
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512BW
;
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE
-; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.8.0 -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-latency -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX1
define i32 @sdiv() {
@@ -270,61 +270,80 @@ define i32 @udiv_uniformconst() {
}
define i32 @sdiv_constpow2() {
-; SSE-LABEL: 'sdiv_constpow2'
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LABEL: 'sdiv_constpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 115 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 229 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 187 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 373 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 139 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 277 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 553 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'sdiv_constpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 115 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 229 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 163 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 325 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 97 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 193 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 385 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX1-LABEL: 'sdiv_constpow2'
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 81 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 162 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 49 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 67 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 150 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 300 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 91 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 201 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 402 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX2-LABEL: 'sdiv_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 224 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'sdiv_constpow2'
@@ -337,13 +356,13 @@ define i32 @sdiv_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 167 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'sdiv_constpow2'
@@ -360,9 +379,9 @@ define i32 @sdiv_constpow2() {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 79 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = sdiv i64 undef, 16
@@ -389,24 +408,119 @@ define i32 @sdiv_constpow2() {
}
define i32 @udiv_constpow2() {
-; CHECK-LABEL: 'udiv_constpow2'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+; SSE2-LABEL: 'udiv_constpow2'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; SSE42-LABEL: 'udiv_constpow2'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX1-LABEL: 'udiv_constpow2'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX2-LABEL: 'udiv_constpow2'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512F-LABEL: 'udiv_constpow2'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
+;
+; AVX512BW-LABEL: 'udiv_constpow2'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = udiv i64 undef, 16
%V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
diff --git a/llvm/test/Analysis/CostModel/X86/div.ll b/llvm/test/Analysis/CostModel/X86/div.ll
index 7c8084cf87213..4bbd70e87a1a9 100644
--- a/llvm/test/Analysis/CostModel/X86/div.ll
+++ b/llvm/test/Analysis/CostModel/X86/div.ll
@@ -637,17 +637,17 @@ define i32 @sdiv_constpow2() {
; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 49 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 98 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 196 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 67 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 134 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 268 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 93 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 186 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 372 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX1-LABEL: 'sdiv_constpow2'
@@ -671,21 +671,21 @@ define i32 @sdiv_constpow2() {
;
; AVX2-LABEL: 'sdiv_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = sdiv i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V2i64 = sdiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 90 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'sdiv_constpow2'
@@ -699,12 +699,12 @@ define i32 @sdiv_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 83 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'sdiv_constpow2'
@@ -732,17 +732,17 @@ define i32 @sdiv_constpow2() {
; SLM-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V4i64 = sdiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V8i64 = sdiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = sdiv i32 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 49 for instruction: %V4i32 = sdiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 98 for instruction: %V8i32 = sdiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 196 for instruction: %V16i32 = sdiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = sdiv i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 67 for instruction: %V8i16 = sdiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 134 for instruction: %V16i16 = sdiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 268 for instruction: %V32i16 = sdiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = sdiv i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 93 for instruction: %V16i8 = sdiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 186 for instruction: %V32i8 = sdiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 372 for instruction: %V64i8 = sdiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%I64 = sdiv i64 undef, 16
@@ -798,13 +798,13 @@ define i32 @udiv_constpow2() {
; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX1-LABEL: 'udiv_constpow2'
@@ -828,21 +828,21 @@ define i32 @udiv_constpow2() {
;
; AVX2-LABEL: 'udiv_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = udiv i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2i64 = udiv <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i64 = udiv <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i64 = udiv <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = udiv i32 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4i32 = udiv <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'udiv_constpow2'
@@ -856,12 +856,12 @@ define i32 @udiv_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'udiv_constpow2'
@@ -893,13 +893,13 @@ define i32 @udiv_constpow2() {
; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8i32 = udiv <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i32 = udiv <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = udiv i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8i16 = udiv <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V16i16 = udiv <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V32i16 = udiv <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = udiv i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16i8 = udiv <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V32i8 = udiv <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V64i8 = udiv <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%I64 = udiv i64 undef, 16
diff --git a/llvm/test/Analysis/CostModel/X86/fshl.ll b/llvm/test/Analysis/CostModel/X86/fshl.ll
index eae376c170261..68320887960c3 100644
--- a/llvm/test/Analysis/CostModel/X86/fshl.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshl.ll
@@ -43,9 +43,9 @@ define void @var_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
;
; AVX2-LABEL: 'var_funnel_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 %c64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_funnel_i64'
@@ -86,16 +86,16 @@ define void @var_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %b32, <4 x i32> %b128, <8 x i32> %b256, <16 x i32> %b512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
; SSSE3-LABEL: 'var_funnel_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 140 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_funnel_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_funnel_i32'
@@ -108,8 +108,8 @@ define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
; AVX2-LABEL: 'var_funnel_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_funnel_i32'
@@ -121,16 +121,16 @@ define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
;
; SLM-LABEL: 'var_funnel_i32'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_funnel_i32'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 %c32)
-; GLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_funnel_i32'
@@ -150,16 +150,16 @@ define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %b16, <8 x i16> %b128, <16 x i16> %b256, <32 x i16> %b512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
; SSSE3-LABEL: 'var_funnel_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 188 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 184 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_funnel_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_funnel_i16'
@@ -172,15 +172,15 @@ define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX2-LABEL: 'var_funnel_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_funnel_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_funnel_i16'
@@ -193,22 +193,22 @@ define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX512DQ-LABEL: 'var_funnel_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_funnel_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_funnel_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 %c16)
-; GLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_funnel_i16'
@@ -235,9 +235,9 @@ define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
;
; SSE42-LABEL: 'var_funnel_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 74 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 148 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_funnel_i8'
@@ -250,15 +250,15 @@ define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX2-LABEL: 'var_funnel_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_funnel_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_funnel_i8'
@@ -271,22 +271,22 @@ define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX512DQ-LABEL: 'var_funnel_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_funnel_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; SLM-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 116 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_funnel_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 %c8)
-; GLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 74 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 148 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_funnel_i8'
@@ -587,7 +587,7 @@ define void @splatvar_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %u128)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %u256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'splatvar_funnel_i8'
@@ -605,7 +605,7 @@ define void @splatvar_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %u128)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %u256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'splatvar_funnel_i8'
@@ -672,9 +672,9 @@ define void @constant_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
;
; AVX2-LABEL: 'constant_funnel_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %b64, i64 7)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> <i64 1, i64 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> <i64 1, i64 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_funnel_i64'
@@ -730,8 +730,8 @@ define void @constant_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
; AVX2-LABEL: 'constant_funnel_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %b32, i32 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 4, i32 5, i32 6, i32 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_funnel_i32'
@@ -779,9 +779,9 @@ define void @constant_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
;
; SSE42-LABEL: 'constant_funnel_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_funnel_i16'
@@ -794,15 +794,15 @@ define void @constant_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX2-LABEL: 'constant_funnel_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_funnel_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_funnel_i16'
@@ -815,22 +815,22 @@ define void @constant_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX512DQ-LABEL: 'constant_funnel_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_funnel_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_funnel_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %b16, i16 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_funnel_i16'
@@ -857,9 +857,9 @@ define void @constant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
;
; SSE42-LABEL: 'constant_funnel_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_funnel_i8'
@@ -872,15 +872,15 @@ define void @constant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX2-LABEL: 'constant_funnel_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_funnel_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_funnel_i8'
@@ -893,22 +893,22 @@ define void @constant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-LABEL: 'constant_funnel_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_funnel_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_funnel_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %b8, i8 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_funnel_i8'
@@ -1213,9 +1213,9 @@ define void @var_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
;
; AVX2-LABEL: 'var_rotate_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 %c64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_rotate_i64'
@@ -1256,16 +1256,16 @@ define void @var_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
; SSSE3-LABEL: 'var_rotate_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_rotate_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_rotate_i32'
@@ -1278,8 +1278,8 @@ define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
; AVX2-LABEL: 'var_rotate_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_rotate_i32'
@@ -1291,16 +1291,16 @@ define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
;
; SLM-LABEL: 'var_rotate_i32'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_rotate_i32'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 %c32)
-; GLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_rotate_i32'
@@ -1320,16 +1320,16 @@ define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
; SSSE3-LABEL: 'var_rotate_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 176 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 172 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_rotate_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_rotate_i16'
@@ -1342,15 +1342,15 @@ define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX2-LABEL: 'var_rotate_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_rotate_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_rotate_i16'
@@ -1363,22 +1363,22 @@ define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX512DQ-LABEL: 'var_rotate_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_rotate_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_rotate_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 %c16)
-; GLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_rotate_i16'
@@ -1405,9 +1405,9 @@ define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
;
; SSE42-LABEL: 'var_rotate_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_rotate_i8'
@@ -1420,15 +1420,15 @@ define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX2-LABEL: 'var_rotate_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_rotate_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_rotate_i8'
@@ -1441,22 +1441,22 @@ define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX512DQ-LABEL: 'var_rotate_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_rotate_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; SLM-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_rotate_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 %c8)
-; GLM-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_rotate_i8'
@@ -1748,7 +1748,7 @@ define void @splatvar_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %u128)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %u256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'splatvar_rotate_i8'
@@ -1766,7 +1766,7 @@ define void @splatvar_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %u128)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %u256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'splatvar_rotate_i8'
@@ -1826,9 +1826,9 @@ define void @constant_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
;
; AVX2-LABEL: 'constant_rotate_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshl.i64(i64 %a64, i64 %a64, i64 7)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> <i64 1, i64 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.fshl.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> <i64 1, i64 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.fshl.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_rotate_i64'
@@ -1884,8 +1884,8 @@ define void @constant_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
; AVX2-LABEL: 'constant_rotate_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshl.i32(i32 %a32, i32 %a32, i32 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I32 = call <4 x i32> @llvm.fshl.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> <i32 4, i32 5, i32 6, i32 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.fshl.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_rotate_i32'
@@ -1933,9 +1933,9 @@ define void @constant_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
;
; SSE42-LABEL: 'constant_rotate_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_rotate_i16'
@@ -1948,15 +1948,15 @@ define void @constant_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX2-LABEL: 'constant_rotate_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_rotate_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_rotate_i16'
@@ -1969,22 +1969,22 @@ define void @constant_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX512DQ-LABEL: 'constant_rotate_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_rotate_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_rotate_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshl.i16(i16 %a16, i16 %a16, i16 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V8I16 = call <8 x i16> @llvm.fshl.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.fshl.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_rotate_i16'
@@ -2011,9 +2011,9 @@ define void @constant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
;
; SSE42-LABEL: 'constant_rotate_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_rotate_i8'
@@ -2026,15 +2026,15 @@ define void @constant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX2-LABEL: 'constant_rotate_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_rotate_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_rotate_i8'
@@ -2047,22 +2047,22 @@ define void @constant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-LABEL: 'constant_rotate_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_rotate_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_rotate_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshl.i8(i8 %a8, i8 %a8, i8 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16I8 = call <16 x i8> @llvm.fshl.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32I8 = call <32 x i8> @llvm.fshl.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V64I8 = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_rotate_i8'
diff --git a/llvm/test/Analysis/CostModel/X86/fshr.ll b/llvm/test/Analysis/CostModel/X86/fshr.ll
index 0d6f4d49886ad..dbd28ce3dfc79 100644
--- a/llvm/test/Analysis/CostModel/X86/fshr.ll
+++ b/llvm/test/Analysis/CostModel/X86/fshr.ll
@@ -43,9 +43,9 @@ define void @var_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
;
; AVX2-LABEL: 'var_funnel_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 %c64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> %c128)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_funnel_i64'
@@ -86,16 +86,16 @@ define void @var_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %b32, <4 x i32> %b128, <8 x i32> %b256, <16 x i32> %b512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
; SSSE3-LABEL: 'var_funnel_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 140 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_funnel_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_funnel_i32'
@@ -108,8 +108,8 @@ define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
; AVX2-LABEL: 'var_funnel_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_funnel_i32'
@@ -121,16 +121,16 @@ define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
;
; SLM-LABEL: 'var_funnel_i32'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_funnel_i32'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 %c32)
-; GLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_funnel_i32'
@@ -150,16 +150,16 @@ define void @var_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %b16, <8 x i16> %b128, <16 x i16> %b256, <32 x i16> %b512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
; SSSE3-LABEL: 'var_funnel_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 188 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 184 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_funnel_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_funnel_i16'
@@ -172,15 +172,15 @@ define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX2-LABEL: 'var_funnel_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_funnel_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_funnel_i16'
@@ -193,22 +193,22 @@ define void @var_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX512DQ-LABEL: 'var_funnel_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_funnel_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_funnel_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 %c16)
-; GLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_funnel_i16'
@@ -235,9 +235,9 @@ define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
;
; SSE42-LABEL: 'var_funnel_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 74 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 148 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_funnel_i8'
@@ -250,15 +250,15 @@ define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX2-LABEL: 'var_funnel_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_funnel_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_funnel_i8'
@@ -271,22 +271,22 @@ define void @var_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX512DQ-LABEL: 'var_funnel_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_funnel_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; SLM-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 58 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 116 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_funnel_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 %c8)
-; GLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 74 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 148 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_funnel_i8'
@@ -587,7 +587,7 @@ define void @splatvar_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %u128)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %u256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'splatvar_funnel_i8'
@@ -605,7 +605,7 @@ define void @splatvar_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> %u128)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> %u256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> %u512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'splatvar_funnel_i8'
@@ -672,9 +672,9 @@ define void @constant_funnel_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
;
; AVX2-LABEL: 'constant_funnel_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %b64, i64 7)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> <i64 1, i64 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %b128, <2 x i64> <i64 1, i64 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %b256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %b512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_funnel_i64'
@@ -730,8 +730,8 @@ define void @constant_funnel_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
; AVX2-LABEL: 'constant_funnel_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %b32, i32 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %b128, <4 x i32> <i32 4, i32 5, i32 6, i32 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %b256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %b512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_funnel_i32'
@@ -779,9 +779,9 @@ define void @constant_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
;
; SSE42-LABEL: 'constant_funnel_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_funnel_i16'
@@ -794,15 +794,15 @@ define void @constant_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX2-LABEL: 'constant_funnel_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_funnel_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_funnel_i16'
@@ -815,22 +815,22 @@ define void @constant_funnel_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX512DQ-LABEL: 'constant_funnel_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_funnel_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_funnel_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %b16, i16 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %b128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %b256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %b512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_funnel_i16'
@@ -857,9 +857,9 @@ define void @constant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
;
; SSE42-LABEL: 'constant_funnel_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_funnel_i8'
@@ -872,15 +872,15 @@ define void @constant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX2-LABEL: 'constant_funnel_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_funnel_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_funnel_i8'
@@ -893,22 +893,22 @@ define void @constant_funnel_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-LABEL: 'constant_funnel_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_funnel_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_funnel_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %b8, i8 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %b128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %b256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %b512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_funnel_i8'
@@ -1214,9 +1214,9 @@ define void @var_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
;
; AVX2-LABEL: 'var_rotate_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 %c64)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> %c128)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_rotate_i64'
@@ -1257,16 +1257,16 @@ define void @var_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512, i32 %c32, <4 x i32> %c128, <8 x i32> %c256, <16 x i32> %c512) {
; SSSE3-LABEL: 'var_rotate_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_rotate_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_rotate_i32'
@@ -1279,8 +1279,8 @@ define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
; AVX2-LABEL: 'var_rotate_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'var_rotate_i32'
@@ -1292,16 +1292,16 @@ define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
;
; SLM-LABEL: 'var_rotate_i32'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; SLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_rotate_i32'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 %c32)
-; GLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_rotate_i32'
@@ -1321,16 +1321,16 @@ define void @var_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512, i16 %c16, <8 x i16> %c128, <16 x i16> %c256, <32 x i16> %c512) {
; SSSE3-LABEL: 'var_rotate_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 176 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 172 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'var_rotate_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_rotate_i16'
@@ -1343,15 +1343,15 @@ define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX2-LABEL: 'var_rotate_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_rotate_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_rotate_i16'
@@ -1364,22 +1364,22 @@ define void @var_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
; AVX512DQ-LABEL: 'var_rotate_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_rotate_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; SLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_rotate_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 %c16)
-; GLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_rotate_i16'
@@ -1406,9 +1406,9 @@ define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
;
; SSE42-LABEL: 'var_rotate_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'var_rotate_i8'
@@ -1421,15 +1421,15 @@ define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX2-LABEL: 'var_rotate_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'var_rotate_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'var_rotate_i8'
@@ -1442,22 +1442,22 @@ define void @var_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %
; AVX512DQ-LABEL: 'var_rotate_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'var_rotate_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; SLM-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; SLM-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; SLM-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; SLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; SLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; SLM-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'var_rotate_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 %c8)
-; GLM-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
-; GLM-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
-; GLM-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
+; GLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %c128)
+; GLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %c256)
+; GLM-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %c512)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'var_rotate_i8'
@@ -1749,7 +1749,7 @@ define void @splatvar_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %u128)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %u256)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'splatvar_rotate_i8'
@@ -1767,7 +1767,7 @@ define void @splatvar_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %u512 = shufflevector <64 x i8> %c512, <64 x i8> undef, <64 x i32> zeroinitializer
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> %u128)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> %u256)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> %u512)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'splatvar_rotate_i8'
@@ -1827,9 +1827,9 @@ define void @constant_rotate_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
;
; AVX2-LABEL: 'constant_rotate_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.fshr.i64(i64 %a64, i64 %a64, i64 7)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> <i64 1, i64 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.fshr.v2i64(<2 x i64> %a128, <2 x i64> %a128, <2 x i64> <i64 1, i64 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.fshr.v4i64(<4 x i64> %a256, <4 x i64> %a256, <4 x i64> <i64 1, i64 7, i64 15, i64 31>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %a512, <8 x i64> %a512, <8 x i64> <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_rotate_i64'
@@ -1885,8 +1885,8 @@ define void @constant_rotate_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
; AVX2-LABEL: 'constant_rotate_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.fshr.i32(i32 %a32, i32 %a32, i32 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I32 = call <4 x i32> @llvm.fshr.v4i32(<4 x i32> %a128, <4 x i32> %a128, <4 x i32> <i32 4, i32 5, i32 6, i32 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.fshr.v8i32(<8 x i32> %a256, <8 x i32> %a256, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %a512, <16 x i32> %a512, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'constant_rotate_i32'
@@ -1934,9 +1934,9 @@ define void @constant_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
;
; SSE42-LABEL: 'constant_rotate_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_rotate_i16'
@@ -1949,15 +1949,15 @@ define void @constant_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX2-LABEL: 'constant_rotate_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_rotate_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_rotate_i16'
@@ -1970,22 +1970,22 @@ define void @constant_rotate_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
; AVX512DQ-LABEL: 'constant_rotate_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_rotate_i16'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_rotate_i16'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.fshr.i16(i16 %a16, i16 %a16, i16 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V8I16 = call <8 x i16> @llvm.fshr.v8i16(<8 x i16> %a128, <8 x i16> %a128, <8 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V16I16 = call <16 x i16> @llvm.fshr.v16i16(<16 x i16> %a256, <16 x i16> %a256, <16 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V32I16 = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %a512, <32 x i16> %a512, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_rotate_i16'
@@ -2012,9 +2012,9 @@ define void @constant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
;
; SSE42-LABEL: 'constant_rotate_i8'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 7)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'constant_rotate_i8'
@@ -2027,15 +2027,15 @@ define void @constant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX2-LABEL: 'constant_rotate_i8'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 7)
; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512F-LABEL: 'constant_rotate_i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 7)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512BW-LABEL: 'constant_rotate_i8'
@@ -2048,22 +2048,22 @@ define void @constant_rotate_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x
; AVX512DQ-LABEL: 'constant_rotate_i8'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 7)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SLM-LABEL: 'constant_rotate_i8'
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 7)
-; SLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; SLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; SLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; GLM-LABEL: 'constant_rotate_i8'
; GLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = call i8 @llvm.fshr.i8(i8 %a8, i8 %a8, i8 7)
-; GLM-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
-; GLM-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16I8 = call <16 x i8> @llvm.fshr.v16i8(<16 x i8> %a128, <16 x i8> %a128, <16 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32I8 = call <32 x i8> @llvm.fshr.v32i8(<32 x i8> %a256, <32 x i8> %a256, <32 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
+; GLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V64I8 = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %a512, <64 x i8> %a512, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>)
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; XOP-LABEL: 'constant_rotate_i8'
diff --git a/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll b/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
index 155d509b37126..815629ed798c7 100644
--- a/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
+++ b/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll
@@ -285,22 +285,22 @@ define void @ctlz(i32 %a, <16 x i32> %va) {
define void @fshl(i32 %a, i32 %b, i32 %c, <16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc) {
; THRU-LABEL: 'fshl'
; THRU-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %s = call i32 @llvm.fshl.i32(i32 %a, i32 %b, i32 %c)
-; THRU-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
+; THRU-NEXT: Cost Model: Found an estimated cost of 140 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
; THRU-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; LATE-LABEL: 'fshl'
; LATE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %s = call i32 @llvm.fshl.i32(i32 %a, i32 %b, i32 %c)
-; LATE-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
+; LATE-NEXT: Cost Model: Found an estimated cost of 142 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
; LATE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SIZE-LABEL: 'fshl'
; SIZE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %s = call i32 @llvm.fshl.i32(i32 %a, i32 %b, i32 %c)
-; SIZE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
+; SIZE-NEXT: Cost Model: Found an estimated cost of 122 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
; SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SIZE_LATE-LABEL: 'fshl'
; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %s = call i32 @llvm.fshl.i32(i32 %a, i32 %b, i32 %c)
-; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
+; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 146 for instruction: %v = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %va, <16 x i32> %vb, <16 x i32> %vc)
; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%s = call i32 @llvm.fshl.i32(i32 %a, i32 %b, i32 %c)
diff --git a/llvm/test/Analysis/CostModel/X86/mul.ll b/llvm/test/Analysis/CostModel/X86/mul.ll
index 4639b5dbde49b..ce9b5e03da30a 100644
--- a/llvm/test/Analysis/CostModel/X86/mul.ll
+++ b/llvm/test/Analysis/CostModel/X86/mul.ll
@@ -90,13 +90,13 @@ define i32 @mul_constpow2() {
;
; AVX2-LABEL: 'mul_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = mul i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2i64 = mul <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4i64 = mul <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i64 = mul <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2i64 = mul <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4i64 = mul <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8i64 = mul <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = mul i32 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4i32 = mul <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8i32 = mul <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16i32 = mul <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8i32 = mul <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i32 = mul <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = mul i16 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = mul <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = mul <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
@@ -119,11 +119,11 @@ define i32 @mul_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = mul i16 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8i16 = mul <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16i16 = mul <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32i16 = mul <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i16 = mul <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I8 = mul i8 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i8 = mul <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32i8 = mul <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V64i8 = mul <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V32i8 = mul <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V64i8 = mul <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'mul_constpow2'
diff --git a/llvm/test/Analysis/CostModel/X86/rem-codesize.ll b/llvm/test/Analysis/CostModel/X86/rem-codesize.ll
index 01b3dc1aade48..696713834647e 100644
--- a/llvm/test/Analysis/CostModel/X86/rem-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/rem-codesize.ll
@@ -272,78 +272,78 @@ define i32 @urem_uniformconst() {
define i32 @srem_constpow2() {
; SSE2-LABEL: 'srem_constpow2'
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 78 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 156 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 106 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 210 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 190 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 378 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 141 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 278 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 552 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SSE42-LABEL: 'srem_constpow2'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 186 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 142 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 282 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 324 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX1-LABEL: 'srem_constpow2'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 93 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 186 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 69 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 154 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 306 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX2-LABEL: 'srem_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 65 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 130 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'srem_constpow2'
@@ -356,13 +356,13 @@ define i32 @srem_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 65 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 65 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 145 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'srem_constpow2'
@@ -379,28 +379,28 @@ define i32 @srem_constpow2() {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 65 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 81 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SLM-LABEL: 'srem_constpow2'
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 186 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 142 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 282 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 324 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = srem i64 undef, 16
diff --git a/llvm/test/Analysis/CostModel/X86/rem-latency.ll b/llvm/test/Analysis/CostModel/X86/rem-latency.ll
index 207ac8616a1c7..196ba38974871 100644
--- a/llvm/test/Analysis/CostModel/X86/rem-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/rem-latency.ll
@@ -272,78 +272,78 @@ define i32 @urem_uniformconst() {
define i32 @srem_constpow2() {
; SSE2-LABEL: 'srem_constpow2'
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 59 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 90 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 178 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 126 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 250 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 200 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 396 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SSE42-LABEL: 'srem_constpow2'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 59 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 49 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 126 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 250 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 85 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 334 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 123 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 242 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 480 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX1-LABEL: 'srem_constpow2'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 204 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 83 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 135 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 268 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX2-LABEL: 'srem_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 74 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 166 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'srem_constpow2'
@@ -356,13 +356,13 @@ define i32 @srem_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 63 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 75 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 110 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'srem_constpow2'
@@ -379,28 +379,28 @@ define i32 @srem_constpow2() {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 65 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 71 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SLM-LABEL: 'srem_constpow2'
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 107 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 132 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 264 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 49 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 126 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 250 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 85 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 334 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 123 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 242 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 480 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = srem i64 undef, 16
diff --git a/llvm/test/Analysis/CostModel/X86/rem-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/rem-sizelatency.ll
index e7768fa517d74..e077602c5cecb 100644
--- a/llvm/test/Analysis/CostModel/X86/rem-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/rem-sizelatency.ll
@@ -272,78 +272,78 @@ define i32 @urem_uniformconst() {
define i32 @srem_constpow2() {
; SSE2-LABEL: 'srem_constpow2'
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 204 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 130 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 258 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 190 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 378 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 284 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 564 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SSE42-LABEL: 'srem_constpow2'
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 59 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 140 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 118 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 234 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 166 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 330 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 200 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 396 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX1-LABEL: 'srem_constpow2'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 35 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 67 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 107 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 214 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 248 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 39 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 69 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 162 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 324 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 221 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 440 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX2-LABEL: 'srem_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 74 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 94 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 83 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 118 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 234 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'srem_constpow2'
@@ -356,13 +356,13 @@ define i32 @srem_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 47 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 83 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 117 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 175 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'srem_constpow2'
@@ -379,28 +379,28 @@ define i32 @srem_constpow2() {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 83 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
; SLM-LABEL: 'srem_constpow2'
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I64 = srem i64 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 29 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 136 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 66 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 130 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 258 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 166 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 330 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 200 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 396 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret i32 undef
;
%I64 = srem i64 undef, 16
diff --git a/llvm/test/Analysis/CostModel/X86/rem.ll b/llvm/test/Analysis/CostModel/X86/rem.ll
index df80e117cf535..91918478389ae 100644
--- a/llvm/test/Analysis/CostModel/X86/rem.ll
+++ b/llvm/test/Analysis/CostModel/X86/rem.ll
@@ -732,17 +732,17 @@ define i32 @srem_constpow2() {
; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 176 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 208 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 69 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 138 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 276 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SSE42-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 126 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 252 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 101 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 204 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 408 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX1-LABEL: 'srem_constpow2'
@@ -766,21 +766,21 @@ define i32 @srem_constpow2() {
;
; AVX2-LABEL: 'srem_constpow2'
; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %I64 = srem i64 undef, 16
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V2i64 = srem <2 x i64> undef, <i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512F-LABEL: 'srem_constpow2'
@@ -794,12 +794,12 @@ define i32 @srem_constpow2() {
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
; AVX512F-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 55 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 113 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 59 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 122 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; AVX512BW-LABEL: 'srem_constpow2'
@@ -827,17 +827,17 @@ define i32 @srem_constpow2() {
; SLM-NEXT: Cost Model: Found an estimated cost of 90 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 180 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 53 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 106 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 212 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 61 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 122 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 244 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; SLM-NEXT: Cost Model: Found an estimated cost of 172 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 70 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 140 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; SLM-NEXT: Cost Model: Found an estimated cost of 280 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; SLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; SLM-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 130 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; SLM-NEXT: Cost Model: Found an estimated cost of 260 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 103 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 208 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; SLM-NEXT: Cost Model: Found an estimated cost of 416 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
; GLM-LABEL: 'srem_constpow2'
@@ -846,17 +846,17 @@ define i32 @srem_constpow2() {
; GLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %V4i64 = srem <4 x i64> undef, <i64 2, i64 4, i64 8, i64 16>
; GLM-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %V8i64 = srem <8 x i64> undef, <i64 2, i64 4, i64 8, i64 16, i64 32, i64 64, i64 128, i64 256>
; GLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I32 = srem i32 undef, 16
-; GLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
-; GLM-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
-; GLM-NEXT: Cost Model: Found an estimated cost of 176 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; GLM-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %V4i32 = srem <4 x i32> undef, <i32 2, i32 4, i32 8, i32 16>
+; GLM-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %V8i32 = srem <8 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
+; GLM-NEXT: Cost Model: Found an estimated cost of 208 for instruction: %V16i32 = srem <16 x i32> undef, <i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256, i32 2, i32 4, i32 8, i32 16, i32 32, i32 64, i32 128, i32 256>
; GLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I16 = srem i16 undef, 16
-; GLM-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; GLM-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
-; GLM-NEXT: Cost Model: Found an estimated cost of 168 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; GLM-NEXT: Cost Model: Found an estimated cost of 69 for instruction: %V8i16 = srem <8 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; GLM-NEXT: Cost Model: Found an estimated cost of 138 for instruction: %V16i16 = srem <16 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
+; GLM-NEXT: Cost Model: Found an estimated cost of 276 for instruction: %V32i16 = srem <32 x i16> undef, <i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256, i16 2, i16 4, i16 8, i16 16, i16 32, i16 64, i16 128, i16 256>
; GLM-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %I8 = srem i8 undef, 16
-; GLM-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; GLM-NEXT: Cost Model: Found an estimated cost of 126 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
-; GLM-NEXT: Cost Model: Found an estimated cost of 252 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; GLM-NEXT: Cost Model: Found an estimated cost of 101 for instruction: %V16i8 = srem <16 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; GLM-NEXT: Cost Model: Found an estimated cost of 204 for instruction: %V32i8 = srem <32 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
+; GLM-NEXT: Cost Model: Found an estimated cost of 408 for instruction: %V64i8 = srem <64 x i8> undef, <i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16, i8 2, i8 4, i8 8, i8 16>
; GLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef
;
%I64 = srem i64 undef, 16
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
index 1fdc3d7c2f041..0743368842c71 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-codesize.ll
@@ -23,108 +23,372 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'var_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <2 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = ashr <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'var_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <4 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <4 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <4 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = ashr <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'var_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <8 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <8 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = ashr <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = ashr <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = ashr <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = ashr <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = ashr <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = ashr <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = ashr <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'var_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = ashr <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'var_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = ashr <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = ashr <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -622,7 +886,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -677,7 +941,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -689,7 +953,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -709,108 +973,372 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = ashr <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
ret <4 x i32> %shift
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %shift
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'constant_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'constant_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 102 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 51 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -1119,13 +1647,9 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
-; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
-;
-; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; XOP-LABEL: 'splatconstant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'splatconstant_shift_v32i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1156,13 +1680,9 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
-; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
-;
-; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; XOP-LABEL: 'splatconstant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'splatconstant_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost-inseltpoison.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost-inseltpoison.ll
index 0836b077b33bf..939586ff80957 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost-inseltpoison.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost-inseltpoison.ll
@@ -32,7 +32,7 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <2 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; XOP-LABEL: 'var_shift_v2i64'
@@ -57,7 +57,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <4 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOP-LABEL: 'var_shift_v4i64'
@@ -82,7 +82,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOP-LABEL: 'var_shift_v8i64'
@@ -98,9 +98,13 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; SSE-LABEL: 'var_shift_v4i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, %b
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v4i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <4 x i32> %a, %b
@@ -123,16 +127,20 @@ define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; SSE-LABEL: 'var_shift_v8i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, %b
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v8i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i32> %a, %b
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v8i32'
@@ -140,7 +148,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v8i32'
@@ -152,16 +160,20 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; SSE-LABEL: 'var_shift_v16i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, %b
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v16i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <16 x i32> %a, %b
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v16i32'
@@ -169,7 +181,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v16i32'
@@ -186,7 +198,7 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <8 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v8i16'
@@ -227,7 +239,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <16 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v16i16'
@@ -235,7 +247,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'var_shift_v16i16'
@@ -243,7 +255,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v16i16'
@@ -251,7 +263,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v16i16'
@@ -268,7 +280,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = ashr <32 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v32i16'
@@ -276,7 +288,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <32 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'var_shift_v32i16'
@@ -284,7 +296,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i16'
@@ -292,7 +304,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i16'
@@ -309,7 +321,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <16 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v16i8'
@@ -350,7 +362,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v32i8'
@@ -358,7 +370,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'var_shift_v32i8'
@@ -366,7 +378,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i8'
@@ -374,7 +386,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i8'
@@ -391,7 +403,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %shift = ashr <64 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v64i8'
@@ -399,7 +411,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <64 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'var_shift_v64i8'
@@ -407,7 +419,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v64i8'
@@ -415,7 +427,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v64i8'
@@ -1021,7 +1033,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> poison, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> poison, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -1033,7 +1045,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> poison, i8 %b, i32 0
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> poison, <64 x i32> zeroinitializer
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
@@ -1062,7 +1074,7 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; XOP-LABEL: 'constant_shift_v2i64'
@@ -1087,7 +1099,7 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOP-LABEL: 'constant_shift_v4i64'
@@ -1112,7 +1124,7 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOP-LABEL: 'constant_shift_v8i64'
@@ -1128,9 +1140,13 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; SSE-LABEL: 'constant_shift_v4i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'constant_shift_v4i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
@@ -1153,25 +1169,25 @@ define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; SSE-LABEL: 'constant_shift_v8i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'constant_shift_v8i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'constant_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1182,25 +1198,25 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; SSE-LABEL: 'constant_shift_v16i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'constant_shift_v16i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'constant_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1216,7 +1232,7 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v8i16'
@@ -1257,7 +1273,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v16i16'
@@ -1265,7 +1281,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v16i16'
@@ -1273,7 +1289,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v16i16'
@@ -1281,7 +1297,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v16i16'
@@ -1298,7 +1314,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v32i16'
@@ -1306,7 +1322,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v32i16'
@@ -1314,7 +1330,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
@@ -1322,7 +1338,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
@@ -1339,7 +1355,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v16i8'
@@ -1380,7 +1396,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v32i8'
@@ -1388,7 +1404,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v32i8'
@@ -1396,7 +1412,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i8'
@@ -1404,7 +1420,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i8'
@@ -1421,7 +1437,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v64i8'
@@ -1429,7 +1445,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v64i8'
@@ -1437,7 +1453,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v64i8'
@@ -1445,7 +1461,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
index 484c5e50d5296..fa057ef67cc58 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-cost.ll
@@ -32,7 +32,7 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <2 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; XOP-LABEL: 'var_shift_v2i64'
@@ -57,7 +57,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <4 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOP-LABEL: 'var_shift_v4i64'
@@ -82,7 +82,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOP-LABEL: 'var_shift_v8i64'
@@ -98,9 +98,13 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; SSE-LABEL: 'var_shift_v4i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, %b
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v4i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <4 x i32> %a, %b
@@ -123,16 +127,20 @@ define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; SSE-LABEL: 'var_shift_v8i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, %b
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v8i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i32> %a, %b
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v8i32'
@@ -140,7 +148,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v8i32'
@@ -152,16 +160,20 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; SSE-LABEL: 'var_shift_v16i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, %b
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v16i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <16 x i32> %a, %b
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v16i32'
@@ -169,7 +181,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v16i32'
@@ -186,7 +198,7 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <8 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v8i16'
@@ -227,7 +239,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <16 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v16i16'
@@ -235,7 +247,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'var_shift_v16i16'
@@ -243,7 +255,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v16i16'
@@ -251,7 +263,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v16i16'
@@ -268,7 +280,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = ashr <32 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v32i16'
@@ -276,7 +288,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <32 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'var_shift_v32i16'
@@ -284,7 +296,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i16'
@@ -292,7 +304,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i16'
@@ -309,7 +321,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <16 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v16i8'
@@ -350,7 +362,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v32i8'
@@ -358,7 +370,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'var_shift_v32i8'
@@ -366,7 +378,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i8'
@@ -374,7 +386,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i8'
@@ -391,7 +403,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %shift = ashr <64 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v64i8'
@@ -399,7 +411,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <64 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'var_shift_v64i8'
@@ -407,7 +419,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v64i8'
@@ -415,7 +427,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v64i8'
@@ -1021,7 +1033,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -1033,7 +1045,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
@@ -1062,7 +1074,7 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; XOP-LABEL: 'constant_shift_v2i64'
@@ -1087,7 +1099,7 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOP-LABEL: 'constant_shift_v4i64'
@@ -1112,7 +1124,7 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOP-LABEL: 'constant_shift_v8i64'
@@ -1128,9 +1140,13 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; SSE-LABEL: 'constant_shift_v4i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'constant_shift_v4i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
@@ -1153,25 +1169,25 @@ define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; SSE-LABEL: 'constant_shift_v8i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'constant_shift_v8i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'constant_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1182,25 +1198,25 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; SSE-LABEL: 'constant_shift_v16i32'
-; SSE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'constant_shift_v16i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'constant_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1216,7 +1232,7 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v8i16'
@@ -1257,7 +1273,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v16i16'
@@ -1265,7 +1281,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v16i16'
@@ -1273,7 +1289,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v16i16'
@@ -1281,7 +1297,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v16i16'
@@ -1298,7 +1314,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v32i16'
@@ -1306,7 +1322,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v32i16'
@@ -1314,7 +1330,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
@@ -1322,7 +1338,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
@@ -1339,7 +1355,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v16i8'
@@ -1380,7 +1396,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v32i8'
@@ -1388,7 +1404,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v32i8'
@@ -1396,7 +1412,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i8'
@@ -1404,7 +1420,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i8'
@@ -1421,7 +1437,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v64i8'
@@ -1429,7 +1445,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v64i8'
@@ -1437,7 +1453,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v64i8'
@@ -1445,7 +1461,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
index ff09f4f26131c..fe9e0edd5a183 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-latency.ll
@@ -1,20 +1,20 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx | FileCheck %s --check-prefixes=CHECK,XOP,XOPAVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=CHECK,XOP,XOPAVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOP,XOPAVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOP,XOPAVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512BW
+;
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1
; Verify the cost of vector logical shift right instructions.
@@ -23,108 +23,380 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'var_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <2 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <2 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = ashr <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'var_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <4 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <4 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <4 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = ashr <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'var_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <8 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = ashr <8 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = ashr <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = ashr <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = ashr <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = ashr <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = ashr <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = ashr <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = ashr <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = ashr <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = ashr <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 90 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = ashr <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -597,7 +869,7 @@ define <16 x i8> @splatvar_shift_v16i8(<16 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v16i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <16 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v16i8'
@@ -652,7 +924,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -707,7 +979,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -719,7 +991,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -739,108 +1011,380 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = ashr <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
ret <4 x i32> %shift
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %shift
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 41 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 82 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 45 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 164 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 90 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 37 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -1165,7 +1709,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v16i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v16i8'
@@ -1202,7 +1746,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1239,7 +1783,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
index b0527a7cf3e54..780ec54ca6785 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-ashr-sizelatency.ll
@@ -23,108 +23,380 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'var_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <2 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <2 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = ashr <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'var_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = ashr <4 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <4 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <4 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = ashr <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'var_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = ashr <8 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <8 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <8 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = ashr <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = ashr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = ashr <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = ashr <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = ashr <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = ashr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = ashr <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = ashr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = ashr <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = ashr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = ashr <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = ashr <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %shift = ashr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <32 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = ashr <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %shift = ashr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 63 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = ashr <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -622,7 +894,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -677,7 +949,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -689,7 +961,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 63 for instruction: %shift = ashr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -709,108 +981,380 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = ashr <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = ashr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = ashr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = ashr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
ret <4 x i32> %shift
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = ashr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = ashr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %shift
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = ashr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = ashr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = ashr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = ashr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = ashr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 216 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 144 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 152 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 63 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = ashr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -1120,7 +1664,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = ashr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1157,7 +1701,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = ashr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
index e0fb19899a7bd..f9094c3d61615 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-codesize.ll
@@ -23,108 +23,380 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'var_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <2 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <2 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = lshr <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'var_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <4 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <4 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = lshr <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'var_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <8 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <8 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = lshr <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = lshr <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = lshr <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = lshr <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = lshr <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = lshr <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = lshr <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'var_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = lshr <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'var_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = lshr <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = lshr <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -616,7 +888,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -665,7 +937,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -677,7 +949,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -697,108 +969,380 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = lshr <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
ret <4 x i32> %shift
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %shift
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 92 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'constant_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'constant_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -1088,7 +1632,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1121,7 +1665,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost-inseltpoison.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost-inseltpoison.ll
index 1875bf7486119..1cb3d919ac9c0 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost-inseltpoison.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost-inseltpoison.ll
@@ -27,21 +27,13 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, %b
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'var_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
-;
-; AVX2-LABEL: 'var_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
-;
-; XOPAVX1-LABEL: 'var_shift_v2i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'var_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOPAVX2-LABEL: 'var_shift_v2i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
@@ -61,7 +53,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOPAVX1-LABEL: 'var_shift_v4i64'
@@ -69,7 +61,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOPAVX2-LABEL: 'var_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v4i64'
@@ -90,7 +82,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOPAVX1-LABEL: 'var_shift_v8i64'
@@ -98,7 +90,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOPAVX2-LABEL: 'var_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v8i64'
@@ -152,7 +144,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v8i32'
@@ -160,7 +152,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v8i32'
@@ -185,7 +177,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v16i32'
@@ -193,7 +185,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v16i32'
@@ -210,7 +202,7 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v8i16'
@@ -251,7 +243,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <16 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v16i16'
@@ -259,7 +251,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'var_shift_v16i16'
@@ -267,7 +259,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v16i16'
@@ -275,7 +267,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v16i16'
@@ -292,7 +284,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = lshr <32 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v32i16'
@@ -300,7 +292,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <32 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'var_shift_v32i16'
@@ -308,7 +300,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i16'
@@ -316,7 +308,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i16'
@@ -333,7 +325,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <16 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v16i8'
@@ -374,7 +366,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = lshr <32 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v32i8'
@@ -382,7 +374,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'var_shift_v32i8'
@@ -390,7 +382,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i8'
@@ -398,7 +390,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i8'
@@ -415,7 +407,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = lshr <64 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v64i8'
@@ -423,7 +415,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <64 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'var_shift_v64i8'
@@ -431,7 +423,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v64i8'
@@ -439,7 +431,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v64i8'
@@ -1015,7 +1007,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> poison, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> poison, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -1027,7 +1019,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> poison, i8 %b, i32 0
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> poison, <64 x i32> zeroinitializer
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
@@ -1051,17 +1043,17 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'constant_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'constant_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX2-LABEL: 'constant_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX1-LABEL: 'constant_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOP-LABEL: 'constant_shift_v2i64'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX2-LABEL: 'constant_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
@@ -1081,16 +1073,12 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v4i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v4i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
@@ -1110,16 +1098,12 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v8i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
@@ -1172,16 +1156,12 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'constant_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1205,16 +1185,12 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'constant_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1230,7 +1206,7 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v8i16'
@@ -1271,7 +1247,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v16i16'
@@ -1279,7 +1255,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v16i16'
@@ -1287,7 +1263,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v16i16'
@@ -1295,7 +1271,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v16i16'
@@ -1312,7 +1288,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v32i16'
@@ -1320,7 +1296,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v32i16'
@@ -1328,7 +1304,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
@@ -1336,7 +1312,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
@@ -1353,7 +1329,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v16i8'
@@ -1394,7 +1370,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v32i8'
@@ -1402,7 +1378,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v32i8'
@@ -1410,7 +1386,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i8'
@@ -1418,7 +1394,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i8'
@@ -1435,7 +1411,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v64i8'
@@ -1443,7 +1419,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v64i8'
@@ -1451,7 +1427,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v64i8'
@@ -1459,7 +1435,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
index 0ad55196db4e9..a246a42f77bb9 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-cost.ll
@@ -27,21 +27,13 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, %b
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'var_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
-;
-; AVX2-LABEL: 'var_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
-;
-; XOPAVX1-LABEL: 'var_shift_v2i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'var_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOPAVX2-LABEL: 'var_shift_v2i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
@@ -61,7 +53,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOPAVX1-LABEL: 'var_shift_v4i64'
@@ -69,7 +61,7 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; XOPAVX2-LABEL: 'var_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v4i64'
@@ -90,7 +82,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOPAVX1-LABEL: 'var_shift_v8i64'
@@ -98,7 +90,7 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; XOPAVX2-LABEL: 'var_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v8i64'
@@ -152,7 +144,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v8i32'
@@ -160,7 +152,7 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v8i32'
@@ -185,7 +177,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX1-LABEL: 'var_shift_v16i32'
@@ -193,7 +185,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; XOPAVX2-LABEL: 'var_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, %b
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v16i32'
@@ -210,7 +202,7 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v8i16'
@@ -251,7 +243,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <16 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v16i16'
@@ -259,7 +251,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'var_shift_v16i16'
@@ -267,7 +259,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v16i16'
@@ -275,7 +267,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v16i16'
@@ -292,7 +284,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = lshr <32 x i16> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'var_shift_v32i16'
@@ -300,7 +292,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <32 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'var_shift_v32i16'
@@ -308,7 +300,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i16'
@@ -316,7 +308,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i16'
@@ -333,7 +325,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <16 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v16i8'
@@ -374,7 +366,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = lshr <32 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v32i8'
@@ -382,7 +374,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'var_shift_v32i8'
@@ -390,7 +382,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i8'
@@ -398,7 +390,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i8'
@@ -415,7 +407,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = lshr <64 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v64i8'
@@ -423,7 +415,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'var_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <64 x i8> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'var_shift_v64i8'
@@ -431,7 +423,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v64i8'
@@ -439,7 +431,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v64i8'
@@ -1015,7 +1007,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -1027,7 +1019,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
@@ -1051,17 +1043,17 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'constant_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'constant_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX2-LABEL: 'constant_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX1-LABEL: 'constant_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOP-LABEL: 'constant_shift_v2i64'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX2-LABEL: 'constant_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
@@ -1081,16 +1073,12 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v4i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v4i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
@@ -1110,16 +1098,12 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v8i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
@@ -1172,16 +1156,12 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'constant_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1205,16 +1185,12 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'constant_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1230,7 +1206,7 @@ define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v8i16'
@@ -1271,7 +1247,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v16i16'
@@ -1279,7 +1255,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v16i16'
@@ -1287,7 +1263,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v16i16'
@@ -1295,7 +1271,7 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v16i16'
@@ -1312,7 +1288,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'constant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v32i16'
@@ -1320,7 +1296,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'constant_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'constant_shift_v32i16'
@@ -1328,7 +1304,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
@@ -1336,7 +1312,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
@@ -1353,7 +1329,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v16i8'
@@ -1394,7 +1370,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v32i8'
@@ -1402,7 +1378,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v32i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v32i8'
@@ -1410,7 +1386,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i8'
@@ -1418,7 +1394,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i8'
@@ -1435,7 +1411,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v64i8'
@@ -1443,7 +1419,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX2-LABEL: 'constant_shift_v64i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; XOP-LABEL: 'constant_shift_v64i8'
@@ -1451,7 +1427,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v64i8'
@@ -1459,7 +1435,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
index 6e73f62d6d1a9..791745306972f 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-latency.ll
@@ -1,20 +1,20 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx | FileCheck %s --check-prefixes=CHECK,XOPAVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=CHECK,XOPAVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOPAVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOPAVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512BW
+;
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=AVX1
; Verify the cost of vector logical shift right instructions.
@@ -23,108 +23,404 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'var_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <2 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <2 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <2 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = lshr <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'var_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <4 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = lshr <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'var_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <8 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = lshr <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = lshr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = lshr <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = lshr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = lshr <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = lshr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = lshr <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i16> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i16> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = lshr <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = lshr <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v32i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v32i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = lshr <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = lshr <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v32i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v32i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'var_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = lshr <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v64i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v64i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = lshr <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -609,7 +905,7 @@ define <16 x i8> @splatvar_shift_v16i8(<16 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v16i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v16i8'
@@ -664,7 +960,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -713,7 +1009,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -725,7 +1021,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -745,108 +1041,404 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'constant_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = lshr <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'constant_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'constant_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v4i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
ret <4 x i32> %shift
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %shift
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v32i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v32i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v32i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v32i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'constant_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v64i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v64i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -857,13 +1449,9 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
;
define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v2i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 7, i64 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v2i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 7, i64 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'splatconstant_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v2i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <2 x i64> %a, <i64 7, i64 7>
@@ -890,13 +1478,9 @@ define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) {
}
define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v4i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v4i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'splatconstant_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v4i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
@@ -923,13 +1507,9 @@ define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) {
}
define <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v8i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v8i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'splatconstant_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v8i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
@@ -956,13 +1536,9 @@ define <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) {
}
define <4 x i32> @splatconstant_shift_v4i32(<4 x i32> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v4i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v4i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE-LABEL: 'splatconstant_shift_v4i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v4i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
@@ -989,13 +1565,9 @@ define <4 x i32> @splatconstant_shift_v4i32(<4 x i32> %a) {
}
define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v8i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v8i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE-LABEL: 'splatconstant_shift_v8i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v8i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
@@ -1022,13 +1594,9 @@ define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) {
}
define <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v16i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v16i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE-LABEL: 'splatconstant_shift_v16i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v16i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
@@ -1055,13 +1623,9 @@ define <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) {
}
define <8 x i16> @splatconstant_shift_v8i16(<8 x i16> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v8i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE-LABEL: 'splatconstant_shift_v8i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v8i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -1088,13 +1652,9 @@ define <8 x i16> @splatconstant_shift_v8i16(<8 x i16> %a) {
}
define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v16i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE-LABEL: 'splatconstant_shift_v16i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v16i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -1125,13 +1685,9 @@ define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) {
}
define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v32i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE-LABEL: 'splatconstant_shift_v32i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v32i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -1162,13 +1718,9 @@ define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) {
}
define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v16i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE-LABEL: 'splatconstant_shift_v16i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v16i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1179,7 +1731,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v16i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v16i8'
@@ -1199,13 +1751,9 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
}
define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v32i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE-LABEL: 'splatconstant_shift_v32i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v32i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1216,7 +1764,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1232,13 +1780,9 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
}
define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v64i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE-LABEL: 'splatconstant_shift_v64i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v64i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1249,7 +1793,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
index 03b66db1229aa..688bdba0d188b 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-lshr-sizelatency.ll
@@ -23,108 +23,384 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'var_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <2 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <2 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = lshr <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'var_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <4 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <4 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = lshr <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'var_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <8 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <8 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = lshr <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = lshr <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = lshr <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = lshr <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = lshr <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = lshr <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = lshr <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = lshr <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = lshr <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = lshr <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = lshr <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = lshr <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = lshr <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'var_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = lshr <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <32 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = lshr <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = lshr <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = lshr <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -616,7 +892,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -665,7 +941,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -677,7 +953,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = lshr <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -697,108 +973,384 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = lshr <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = lshr <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = lshr <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'constant_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = lshr <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
ret <4 x i32> %shift
}
define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = lshr <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <8 x i32> %shift
}
define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
-; CHECK-LABEL: 'constant_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'constant_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'constant_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = lshr <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
ret <16 x i32> %shift
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 31 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = lshr <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 62 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = lshr <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'constant_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 124 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = lshr <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'constant_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = lshr <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 43 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = lshr <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 120 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 86 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = lshr <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -1088,7 +1640,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = lshr <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1125,7 +1677,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = lshr <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
index 8fbff53b215d4..5a30f643bf211 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-codesize.ll
@@ -23,108 +23,380 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'var_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <2 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <2 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = shl <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'var_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <4 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = shl <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'var_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = shl <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = shl <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = shl <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = shl <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = shl <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = shl <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = shl <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'var_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'var_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -622,7 +894,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -671,7 +943,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -683,7 +955,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -703,27 +975,95 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v2i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v2i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = shl <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v4i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v4i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE2-LABEL: 'constant_shift_v8i64'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; SSE42-LABEL: 'constant_shift_v8i64'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
@@ -887,36 +1227,104 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
-; AVX512-LABEL: 'constant_shift_v32i16'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX-LABEL: 'constant_shift_v16i8'
+; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'constant_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -1206,7 +1614,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1239,7 +1647,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-cost-inseltpoison.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-cost-inseltpoison.ll
index d1c2c50da75df..178b3366cad92 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-cost-inseltpoison.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-cost-inseltpoison.ll
@@ -27,17 +27,17 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'var_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, %b
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'var_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX2-LABEL: 'var_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX1-LABEL: 'var_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOP-LABEL: 'var_shift_v2i64'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX2-LABEL: 'var_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
@@ -57,16 +57,12 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v4i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
+; XOP-LABEL: 'var_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v4i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
@@ -86,16 +82,12 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v8i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
+; XOP-LABEL: 'var_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v8i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
@@ -107,11 +99,11 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
; SSE2-LABEL: 'var_shift_v4i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <4 x i32> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; SSE42-LABEL: 'var_shift_v4i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <4 x i32> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v4i32'
@@ -131,7 +123,7 @@ define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; SLM-LABEL: 'var_shift_v4i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <4 x i32> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
%shift = shl <4 x i32> %a, %b
@@ -140,11 +132,11 @@ define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; SSE2-LABEL: 'var_shift_v8i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = shl <8 x i32> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; SSE42-LABEL: 'var_shift_v8i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <8 x i32> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v8i32'
@@ -152,23 +144,19 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'var_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; SLM-LABEL: 'var_shift_v8i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <8 x i32> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
%shift = shl <8 x i32> %a, %b
@@ -177,11 +165,11 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; SSE2-LABEL: 'var_shift_v16i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = shl <16 x i32> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; SSE42-LABEL: 'var_shift_v16i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <16 x i32> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v16i32'
@@ -189,23 +177,19 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'var_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; SLM-LABEL: 'var_shift_v16i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <16 x i32> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
%shift = shl <16 x i32> %a, %b
@@ -214,7 +198,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
; SSE2-LABEL: 'var_shift_v8i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <8 x i16> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v8i16'
@@ -259,7 +243,7 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; SSE2-LABEL: 'var_shift_v16i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <16 x i16> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v16i16'
@@ -271,7 +255,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'var_shift_v16i16'
@@ -279,7 +263,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v16i16'
@@ -287,7 +271,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v16i16'
@@ -304,7 +288,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; SSE2-LABEL: 'var_shift_v32i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = shl <32 x i16> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v32i16'
@@ -316,7 +300,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'var_shift_v32i16'
@@ -324,7 +308,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i16'
@@ -332,7 +316,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i16'
@@ -353,7 +337,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v16i8'
@@ -385,7 +369,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SLM-LABEL: 'var_shift_v16i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, %b
@@ -398,7 +382,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v32i8'
@@ -430,7 +414,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SLM-LABEL: 'var_shift_v32i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, %b
@@ -443,7 +427,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v64i8'
@@ -459,7 +443,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v64i8'
@@ -467,7 +451,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v64i8'
@@ -475,7 +459,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SLM-LABEL: 'var_shift_v64i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, %b
@@ -1113,7 +1097,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> poison, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> poison, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -1125,7 +1109,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> poison, i8 %b, i32 0
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> poison, <64 x i32> zeroinitializer
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
@@ -1155,17 +1139,17 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'constant_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'constant_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX2-LABEL: 'constant_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX1-LABEL: 'constant_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOP-LABEL: 'constant_shift_v2i64'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX2-LABEL: 'constant_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
@@ -1185,16 +1169,12 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v4i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v4i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
@@ -1214,16 +1194,12 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v8i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
@@ -1276,16 +1252,12 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'constant_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1313,16 +1285,12 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'constant_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1440,7 +1408,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
@@ -1448,7 +1416,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
@@ -1469,7 +1437,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v16i8'
@@ -1501,7 +1469,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SLM-LABEL: 'constant_shift_v16i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
@@ -1514,7 +1482,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v32i8'
@@ -1546,7 +1514,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SLM-LABEL: 'constant_shift_v32i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
@@ -1559,7 +1527,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v64i8'
@@ -1575,7 +1543,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v64i8'
@@ -1583,7 +1551,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
@@ -1591,7 +1559,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SLM-LABEL: 'constant_shift_v64i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
@@ -2072,17 +2040,17 @@ define <2 x i64> @test5(<2 x i64> %a) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
-; AVX1-LABEL: 'test5'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
+; AVX-LABEL: 'test5'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
-; AVX2-LABEL: 'test5'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
+; XOPAVX1-LABEL: 'test5'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
-; XOP-LABEL: 'test5'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
+; XOPAVX2-LABEL: 'test5'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
; AVX512-LABEL: 'test5'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
@@ -2171,16 +2139,12 @@ define <8 x i32> @test7(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
;
; AVX2-LABEL: 'test7'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
;
-; XOPAVX1-LABEL: 'test7'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
-;
-; XOPAVX2-LABEL: 'test7'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
+; XOP-LABEL: 'test7'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
;
; AVX512-LABEL: 'test7'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
@@ -2209,16 +2173,12 @@ define <4 x i64> @test8(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
;
; AVX2-LABEL: 'test8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
;
-; XOPAVX1-LABEL: 'test8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
-;
-; XOPAVX2-LABEL: 'test8'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
+; XOP-LABEL: 'test8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
;
; AVX512-LABEL: 'test8'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
@@ -2256,7 +2216,7 @@ define <32 x i16> @test9(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512F-LABEL: 'test9'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512BW-LABEL: 'test9'
@@ -2264,7 +2224,7 @@ define <32 x i16> @test9(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512VL-LABEL: 'test9'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512BWVL-LABEL: 'test9'
@@ -2295,16 +2255,12 @@ define <16 x i32> @test10(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
;
; AVX2-LABEL: 'test10'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
;
-; XOPAVX1-LABEL: 'test10'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
-;
-; XOPAVX2-LABEL: 'test10'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
+; XOP-LABEL: 'test10'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
;
; AVX512-LABEL: 'test10'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
@@ -2333,16 +2289,12 @@ define <8 x i64> @test11(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
;
; AVX2-LABEL: 'test11'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
;
-; XOPAVX1-LABEL: 'test11'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
-;
-; XOPAVX2-LABEL: 'test11'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
+; XOP-LABEL: 'test11'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
;
; AVX512-LABEL: 'test11'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
index f5e246e176743..b8f9e0683fb87 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-cost.ll
@@ -27,17 +27,17 @@ define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'var_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, %b
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'var_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX2-LABEL: 'var_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX1-LABEL: 'var_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOP-LABEL: 'var_shift_v2i64'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX2-LABEL: 'var_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
@@ -57,16 +57,12 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v4i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
+; XOP-LABEL: 'var_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v4i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
@@ -86,16 +82,12 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'var_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v8i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
+; XOP-LABEL: 'var_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'var_shift_v8i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
@@ -107,11 +99,11 @@ define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
; SSE2-LABEL: 'var_shift_v4i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <4 x i32> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; SSE42-LABEL: 'var_shift_v4i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <4 x i32> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v4i32'
@@ -131,7 +123,7 @@ define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
; SLM-LABEL: 'var_shift_v4i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <4 x i32> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift
;
%shift = shl <4 x i32> %a, %b
@@ -140,11 +132,11 @@ define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; SSE2-LABEL: 'var_shift_v8i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = shl <8 x i32> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; SSE42-LABEL: 'var_shift_v8i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <8 x i32> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v8i32'
@@ -152,23 +144,19 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'var_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; SLM-LABEL: 'var_shift_v8i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <8 x i32> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
%shift = shl <8 x i32> %a, %b
@@ -177,11 +165,11 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; SSE2-LABEL: 'var_shift_v16i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 68 for instruction: %shift = shl <16 x i32> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; SSE42-LABEL: 'var_shift_v16i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <16 x i32> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'var_shift_v16i32'
@@ -189,23 +177,19 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'var_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'var_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'var_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, %b
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'var_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'var_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; SLM-LABEL: 'var_shift_v16i32'
-; SLM-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <16 x i32> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
%shift = shl <16 x i32> %a, %b
@@ -214,7 +198,7 @@ define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
; SSE2-LABEL: 'var_shift_v8i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <8 x i16> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v8i16'
@@ -259,7 +243,7 @@ define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; SSE2-LABEL: 'var_shift_v16i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <16 x i16> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v16i16'
@@ -271,7 +255,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v16i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; XOP-LABEL: 'var_shift_v16i16'
@@ -279,7 +263,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v16i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v16i16'
@@ -287,7 +271,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v16i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v16i16'
@@ -304,7 +288,7 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; SSE2-LABEL: 'var_shift_v32i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 100 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = shl <32 x i16> %a, %b
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; SSE42-LABEL: 'var_shift_v32i16'
@@ -316,7 +300,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX2-LABEL: 'var_shift_v32i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i16> %a, %b
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; XOP-LABEL: 'var_shift_v32i16'
@@ -324,7 +308,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'var_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'var_shift_v32i16'
@@ -332,7 +316,7 @@ define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'var_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v32i16'
@@ -353,7 +337,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v16i8'
@@ -385,7 +369,7 @@ define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SLM-LABEL: 'var_shift_v16i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, %b
@@ -398,7 +382,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v32i8'
@@ -430,7 +414,7 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SLM-LABEL: 'var_shift_v32i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, %b
@@ -443,7 +427,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'var_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, %b
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'var_shift_v64i8'
@@ -459,7 +443,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'var_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %b
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'var_shift_v64i8'
@@ -467,7 +451,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'var_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %b
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'var_shift_v64i8'
@@ -475,7 +459,7 @@ define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SLM-LABEL: 'var_shift_v64i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, %b
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, %b
@@ -1113,7 +1097,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -1125,7 +1109,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512VL-LABEL: 'splatvar_shift_v64i8'
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'splatvar_shift_v64i8'
@@ -1155,17 +1139,17 @@ define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX1-LABEL: 'constant_shift_v2i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; AVX-LABEL: 'constant_shift_v2i64'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; AVX2-LABEL: 'constant_shift_v2i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX1-LABEL: 'constant_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
-; XOP-LABEL: 'constant_shift_v2i64'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
+; XOPAVX2-LABEL: 'constant_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v2i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
@@ -1185,16 +1169,12 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v4i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v4i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v4i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
+; XOP-LABEL: 'constant_shift_v4i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v4i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
@@ -1214,16 +1194,12 @@ define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX2-LABEL: 'constant_shift_v8i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i64'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i64'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
+; XOP-LABEL: 'constant_shift_v8i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift
;
; AVX512-LABEL: 'constant_shift_v8i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
@@ -1276,16 +1252,12 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
+; XOP-LABEL: 'constant_shift_v8i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1313,16 +1285,12 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
-; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
-;
-; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
+; XOP-LABEL: 'constant_shift_v16i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -1440,7 +1408,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
@@ -1448,7 +1416,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512VL-LABEL: 'constant_shift_v32i16'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v32i16'
@@ -1469,7 +1437,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v16i8'
@@ -1501,7 +1469,7 @@ define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
; SLM-LABEL: 'constant_shift_v16i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
@@ -1514,7 +1482,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v32i8'
@@ -1546,7 +1514,7 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
; SLM-LABEL: 'constant_shift_v32i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
@@ -1559,7 +1527,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SSE42-LABEL: 'constant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'constant_shift_v64i8'
@@ -1575,7 +1543,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512F-LABEL: 'constant_shift_v64i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'constant_shift_v64i8'
@@ -1583,7 +1551,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512VL-LABEL: 'constant_shift_v64i8'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; AVX512BWVL-LABEL: 'constant_shift_v64i8'
@@ -1591,7 +1559,7 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
; SLM-LABEL: 'constant_shift_v64i8'
-; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
; SLM-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
@@ -2072,17 +2040,17 @@ define <2 x i64> @test5(<2 x i64> %a) {
; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
-; AVX1-LABEL: 'test5'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
+; AVX-LABEL: 'test5'
+; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
+; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
-; AVX2-LABEL: 'test5'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
+; XOPAVX1-LABEL: 'test5'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
-; XOP-LABEL: 'test5'
-; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
-; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
+; XOPAVX2-LABEL: 'test5'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl
;
; AVX512-LABEL: 'test5'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3>
@@ -2171,16 +2139,12 @@ define <8 x i32> @test7(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
;
; AVX2-LABEL: 'test7'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
;
-; XOPAVX1-LABEL: 'test7'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
-;
-; XOPAVX2-LABEL: 'test7'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
+; XOP-LABEL: 'test7'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl
;
; AVX512-LABEL: 'test7'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
@@ -2209,16 +2173,12 @@ define <4 x i64> @test8(<4 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
;
; AVX2-LABEL: 'test8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
;
-; XOPAVX1-LABEL: 'test8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
-;
-; XOPAVX2-LABEL: 'test8'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
+; XOP-LABEL: 'test8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl
;
; AVX512-LABEL: 'test8'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4>
@@ -2256,7 +2216,7 @@ define <32 x i16> @test9(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512F-LABEL: 'test9'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512BW-LABEL: 'test9'
@@ -2264,7 +2224,7 @@ define <32 x i16> @test9(<32 x i16> %a) {
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512VL-LABEL: 'test9'
-; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
+; AVX512VL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11>
; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl
;
; AVX512BWVL-LABEL: 'test9'
@@ -2295,16 +2255,12 @@ define <16 x i32> @test10(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
;
; AVX2-LABEL: 'test10'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
;
-; XOPAVX1-LABEL: 'test10'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
-;
-; XOPAVX2-LABEL: 'test10'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
+; XOP-LABEL: 'test10'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl
;
; AVX512-LABEL: 'test10'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3>
@@ -2333,16 +2289,12 @@ define <8 x i64> @test11(<8 x i64> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
;
; AVX2-LABEL: 'test11'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
;
-; XOPAVX1-LABEL: 'test11'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
-;
-; XOPAVX2-LABEL: 'test11'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
+; XOP-LABEL: 'test11'
+; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
+; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl
;
; AVX512-LABEL: 'test11'
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3>
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
index 185c4161270d0..06b97aeb3c4f9 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-latency.ll
@@ -1,20 +1,20 @@
; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx | FileCheck %s --check-prefixes=CHECK,XOPAVX1
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=CHECK,XOPAVX2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW
-;
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE42
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=CHECK,AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse2 | FileCheck %s --check-prefixes=SSE,SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx | FileCheck %s --check-prefixes=AVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx | FileCheck %s --check-prefixes=XOPAVX1
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=XOPAVX2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512F
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512BW
+;
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=slm | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE42
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mcpu=btver2 | FileCheck %s --check-prefixes=AVX1
; Verify the cost of vector logical shift right instructions.
@@ -23,108 +23,400 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'var_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <2 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <2 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <2 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = shl <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'var_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <4 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <4 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = shl <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'var_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <8 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = shl <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = shl <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = shl <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 76 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
%shift = shl <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i16> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i16> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = shl <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 54 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = shl <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 108 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v32i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v32i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = shl <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v32i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v32i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'var_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v64i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v64i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512-LABEL: 'var_shift_v64i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, %b
ret <64 x i8> %shift
@@ -609,7 +901,7 @@ define <16 x i8> @splatvar_shift_v16i8(<16 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v16i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v16i8'
@@ -658,7 +950,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -707,7 +999,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -719,7 +1011,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -739,27 +1031,87 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'constant_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v2i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v2i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
%shift = shl <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'constant_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
%shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'constant_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
%shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
@@ -779,19 +1131,19 @@ define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v4i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v4i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v4i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v4i32'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
%shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
@@ -812,19 +1164,19 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
%shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
@@ -845,15 +1197,15 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
@@ -865,22 +1217,42 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
}
define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) {
-; CHECK-LABEL: 'constant_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE-LABEL: 'constant_shift_v8i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'constant_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i16'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i16'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
%shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <8 x i16> %shift
}
define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
-; SSE2-LABEL: 'constant_shift_v16i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
-;
-; SSE42-LABEL: 'constant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE-LABEL: 'constant_shift_v16i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v16i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
@@ -891,29 +1263,29 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v16i16'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v16i16'
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
-; AVX512-LABEL: 'constant_shift_v16i16'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; AVX512F-LABEL: 'constant_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
%shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <16 x i16> %shift
}
define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
-; SSE2-LABEL: 'constant_shift_v32i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
-;
-; SSE42-LABEL: 'constant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE-LABEL: 'constant_shift_v32i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'constant_shift_v32i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
@@ -924,7 +1296,7 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v32i16'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v32i16'
@@ -932,11 +1304,11 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; AVX512F-LABEL: 'constant_shift_v32i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; AVX512BW-LABEL: 'constant_shift_v32i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
@@ -944,27 +1316,103 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v16i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v16i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
%shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v32i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v32i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512-LABEL: 'constant_shift_v32i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
%shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 84 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 46 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v64i8'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v64i8'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512-LABEL: 'constant_shift_v64i8'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
%shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
@@ -975,13 +1423,9 @@ define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
;
define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v2i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 7, i64 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v2i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 7, i64 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'splatconstant_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v2i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <2 x i64> %a, <i64 7, i64 7>
@@ -1008,13 +1452,9 @@ define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) {
}
define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v4i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v4i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'splatconstant_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v4i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7>
@@ -1041,13 +1481,9 @@ define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) {
}
define <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v8i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v8i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'splatconstant_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v8i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7>
@@ -1074,13 +1510,9 @@ define <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) {
}
define <4 x i32> @splatconstant_shift_v4i32(<4 x i32> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v4i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v4i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE-LABEL: 'splatconstant_shift_v4i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v4i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5>
@@ -1107,13 +1539,9 @@ define <4 x i32> @splatconstant_shift_v4i32(<4 x i32> %a) {
}
define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v8i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v8i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE-LABEL: 'splatconstant_shift_v8i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v8i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
@@ -1140,13 +1568,9 @@ define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) {
}
define <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v16i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v16i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE-LABEL: 'splatconstant_shift_v16i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v16i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5>
@@ -1173,13 +1597,9 @@ define <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) {
}
define <8 x i16> @splatconstant_shift_v8i16(<8 x i16> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v8i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v8i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE-LABEL: 'splatconstant_shift_v8i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v8i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -1206,13 +1626,9 @@ define <8 x i16> @splatconstant_shift_v8i16(<8 x i16> %a) {
}
define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v16i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v16i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE-LABEL: 'splatconstant_shift_v16i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v16i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -1243,13 +1659,9 @@ define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) {
}
define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v32i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v32i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE-LABEL: 'splatconstant_shift_v32i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v32i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -1280,13 +1692,9 @@ define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) {
}
define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v16i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v16i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE-LABEL: 'splatconstant_shift_v16i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v16i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1297,7 +1705,7 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v16i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v16i8'
@@ -1317,13 +1725,9 @@ define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) {
}
define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v32i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v32i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE-LABEL: 'splatconstant_shift_v32i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v32i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1334,7 +1738,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1350,13 +1754,9 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
}
define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
-; SSE2-LABEL: 'splatconstant_shift_v64i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
-;
-; SSE42-LABEL: 'splatconstant_shift_v64i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE-LABEL: 'splatconstant_shift_v64i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX1-LABEL: 'splatconstant_shift_v64i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
@@ -1367,7 +1767,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
index 9c8ff8dac12c9..abdf29c77b7f0 100644
--- a/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/vshift-shl-sizelatency.ll
@@ -23,109 +23,449 @@
;
define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'var_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <2 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'var_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
+; BTVER2-LABEL: 'var_shift_v2i64'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <2 x i64> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
%shift = shl <2 x i64> %a, %b
ret <2 x i64> %shift
}
define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'var_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <4 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
+; AVX1-LABEL: 'var_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <4 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; BTVER2-LABEL: 'var_shift_v4i64'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <4 x i64> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
%shift = shl <4 x i64> %a, %b
ret <4 x i64> %shift
}
define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) {
-; CHECK-LABEL: 'var_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'var_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <8 x i64> %a, %b
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <8 x i64> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
+; BTVER2-LABEL: 'var_shift_v8i64'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <8 x i64> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
%shift = shl <8 x i64> %a, %b
ret <8 x i64> %shift
}
define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v4i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; SSE2-LABEL: 'var_shift_v4i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v4i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v4i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v4i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
+; XOP-LABEL: 'var_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v4i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; SLM-LABEL: 'var_shift_v4i32'
+; SLM-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+;
+; BTVER2-LABEL: 'var_shift_v4i32'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <4 x i32> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
%shift = shl <4 x i32> %a, %b
ret <4 x i32> %shift
}
define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v8i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+; SSE2-LABEL: 'var_shift_v8i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v8i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <8 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v8i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v8i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
+;
+; SLM-LABEL: 'var_shift_v8i32'
+; SLM-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
+; BTVER2-LABEL: 'var_shift_v8i32'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <8 x i32> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
%shift = shl <8 x i32> %a, %b
ret <8 x i32> %shift
}
define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) {
-; CHECK-LABEL: 'var_shift_v16i32'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+; SSE2-LABEL: 'var_shift_v16i32'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i32'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <16 x i32> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i32'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i32'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX1-LABEL: 'var_shift_v16i32'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <16 x i32> %a, %b
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; XOPAVX2-LABEL: 'var_shift_v16i32'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i32> %a, %b
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; AVX512-LABEL: 'var_shift_v16i32'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
+;
+; SLM-LABEL: 'var_shift_v16i32'
+; SLM-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <16 x i32> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
+; BTVER2-LABEL: 'var_shift_v16i32'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %shift = shl <16 x i32> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
%shift = shl <16 x i32> %a, %b
ret <16 x i32> %shift
}
define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v8i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+; SSE2-LABEL: 'var_shift_v8i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v8i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v8i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v8i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v8i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v8i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
;
+; AVX512BW-LABEL: 'var_shift_v8i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; SLM-LABEL: 'var_shift_v8i16'
+; SLM-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i16> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
+;
+; BTVER2-LABEL: 'var_shift_v8i16'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <8 x i16> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %shift
%shift = shl <8 x i16> %a, %b
ret <8 x i16> %shift
}
define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v16i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+; SSE2-LABEL: 'var_shift_v16i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX1-LABEL: 'var_shift_v16i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v16i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
;
+; AVX512BW-LABEL: 'var_shift_v16i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; SLM-LABEL: 'var_shift_v16i16'
+; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i16> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
+;
+; BTVER2-LABEL: 'var_shift_v16i16'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 25 for instruction: %shift = shl <16 x i16> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %shift
%shift = shl <16 x i16> %a, %b
ret <16 x i16> %shift
}
define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) {
-; CHECK-LABEL: 'var_shift_v32i16'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; SSE2-LABEL: 'var_shift_v32i16'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i16'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i16> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
+; AVX1-LABEL: 'var_shift_v32i16'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i16'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; XOP-LABEL: 'var_shift_v32i16'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <32 x i16> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; SLM-LABEL: 'var_shift_v32i16'
+; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i16> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; BTVER2-LABEL: 'var_shift_v32i16'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 50 for instruction: %shift = shl <32 x i16> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
%shift = shl <32 x i16> %a, %b
ret <32 x i16> %shift
}
define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'var_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
+; AVX1-LABEL: 'var_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <16 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SLM-LABEL: 'var_shift_v16i8'
+; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; BTVER2-LABEL: 'var_shift_v16i8'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <16 x i8> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
%shift = shl <16 x i8> %a, %b
ret <16 x i8> %shift
}
define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'var_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
+; AVX1-LABEL: 'var_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SLM-LABEL: 'var_shift_v32i8'
+; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; BTVER2-LABEL: 'var_shift_v32i8'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <32 x i8> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
%shift = shl <32 x i8> %a, %b
ret <32 x i8> %shift
}
define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) {
-; CHECK-LABEL: 'var_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %b
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'var_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'var_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = shl <64 x i8> %a, %b
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'var_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'var_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'var_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %b
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'var_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'var_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, %b
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SLM-LABEL: 'var_shift_v64i8'
+; SLM-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = shl <64 x i8> %a, %b
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
+; BTVER2-LABEL: 'var_shift_v64i8'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %shift = shl <64 x i8> %a, %b
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
%shift = shl <64 x i8> %a, %b
ret <64 x i8> %shift
}
@@ -682,7 +1022,7 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v32i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v32i8'
@@ -737,7 +1077,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; XOPAVX1-LABEL: 'splatvar_shift_v64i8'
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %splat
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, %splat
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatvar_shift_v64i8'
@@ -749,7 +1089,7 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
; AVX512F-LABEL: 'splatvar_shift_v64i8'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, %splat
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %shift = shl <64 x i8> %a, %splat
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; AVX512BW-LABEL: 'splatvar_shift_v64i8'
@@ -775,28 +1115,93 @@ define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) {
;
define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v2i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+; SSE-LABEL: 'constant_shift_v2i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v2i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v2i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; XOP-LABEL: 'constant_shift_v2i64'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v2i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
;
+; BTVER2-LABEL: 'constant_shift_v2i64'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %shift
%shift = shl <2 x i64> %a, <i64 1, i64 7>
ret <2 x i64> %shift
}
define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v4i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+; SSE-LABEL: 'constant_shift_v4i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v4i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX2-LABEL: 'constant_shift_v4i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v4i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
;
+; XOPAVX2-LABEL: 'constant_shift_v4i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v4i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
+;
+; BTVER2-LABEL: 'constant_shift_v4i64'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %shift
%shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31>
ret <4 x i64> %shift
}
define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) {
-; CHECK-LABEL: 'constant_shift_v8i64'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+; SSE-LABEL: 'constant_shift_v8i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX1-LABEL: 'constant_shift_v8i64'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
;
+; AVX2-LABEL: 'constant_shift_v8i64'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX1-LABEL: 'constant_shift_v8i64'
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; XOPAVX2-LABEL: 'constant_shift_v8i64'
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; AVX512-LABEL: 'constant_shift_v8i64'
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
+;
+; BTVER2-LABEL: 'constant_shift_v8i64'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %shift
%shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31>
ret <8 x i64> %shift
}
@@ -810,24 +1215,16 @@ define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) {
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
-; AVX1-LABEL: 'constant_shift_v4i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
-;
-; AVX2-LABEL: 'constant_shift_v4i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
-;
-; XOPAVX1-LABEL: 'constant_shift_v4i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; AVX-LABEL: 'constant_shift_v4i32'
+; AVX-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
-; XOPAVX2-LABEL: 'constant_shift_v4i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
+; XOP-LABEL: 'constant_shift_v4i32'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v4i32'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %shift
;
; SLM-LABEL: 'constant_shift_v4i32'
@@ -852,19 +1249,19 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v8i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v8i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v8i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v8i32'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %shift
;
; SLM-LABEL: 'constant_shift_v8i32'
@@ -889,15 +1286,15 @@ define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) {
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; AVX2-LABEL: 'constant_shift_v16i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; XOPAVX1-LABEL: 'constant_shift_v16i32'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; XOPAVX2-LABEL: 'constant_shift_v16i32'
-; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
+; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %shift
;
; AVX512-LABEL: 'constant_shift_v16i32'
@@ -971,37 +1368,134 @@ define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) {
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
-; AVX512-LABEL: 'constant_shift_v32i16'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
-; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+; AVX512F-LABEL: 'constant_shift_v32i16'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i16'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %shift
;
%shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7>
ret <32 x i16> %shift
}
define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v16i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v16i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v16i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v16i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v16i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v16i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v16i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v16i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
+;
+; SLM-LABEL: 'constant_shift_v16i8'
+; SLM-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
;
+; BTVER2-LABEL: 'constant_shift_v16i8'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %shift
%shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <16 x i8> %shift
}
define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v32i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v32i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v32i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v32i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v32i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v32i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v32i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v32i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
+;
+; SLM-LABEL: 'constant_shift_v32i8'
+; SLM-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
+; BTVER2-LABEL: 'constant_shift_v32i8'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
%shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <32 x i8> %shift
}
define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) {
-; CHECK-LABEL: 'constant_shift_v64i8'
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
-; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+; SSE2-LABEL: 'constant_shift_v64i8'
+; SSE2-NEXT: Cost Model: Found an estimated cost of 112 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SSE42-LABEL: 'constant_shift_v64i8'
+; SSE42-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX1-LABEL: 'constant_shift_v64i8'
+; AVX1-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX2-LABEL: 'constant_shift_v64i8'
+; AVX2-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; XOP-LABEL: 'constant_shift_v64i8'
+; XOP-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512F-LABEL: 'constant_shift_v64i8'
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; AVX512BW-LABEL: 'constant_shift_v64i8'
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
+;
+; SLM-LABEL: 'constant_shift_v64i8'
+; SLM-NEXT: Cost Model: Found an estimated cost of 88 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; SLM-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
+; BTVER2-LABEL: 'constant_shift_v64i8'
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 80 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
+; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
%shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0>
ret <64 x i8> %shift
}
@@ -1258,7 +1752,7 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v32i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v32i8'
@@ -1291,7 +1785,7 @@ define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) {
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX1-LABEL: 'splatconstant_shift_v64i8'
-; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
+; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3>
; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %shift
;
; XOPAVX2-LABEL: 'splatconstant_shift_v64i8'
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/alternate-int-inseltpoison.ll b/llvm/test/Transforms/SLPVectorizer/X86/alternate-int-inseltpoison.ll
index 99b3173c92851..e07764e623595 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/alternate-int-inseltpoison.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/alternate-int-inseltpoison.ll
@@ -178,10 +178,12 @@ define <8 x i32> @ashr_shl_v8i32_const(<8 x i32> %a) {
; SSE-NEXT: ret <8 x i32> [[R71]]
;
; SLM-LABEL: @ashr_shl_v8i32_const(
-; SLM-NEXT: [[TMP1:%.*]] = ashr <8 x i32> [[A:%.*]], <i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
-; SLM-NEXT: [[TMP2:%.*]] = shl <8 x i32> [[A]], <i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
-; SLM-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> [[TMP2]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>
-; SLM-NEXT: ret <8 x i32> [[TMP3]]
+; SLM-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; SLM-NEXT: [[TMP2:%.*]] = ashr <4 x i32> [[TMP1]], <i32 2, i32 2, i32 2, i32 2>
+; SLM-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; SLM-NEXT: [[TMP4:%.*]] = shl <4 x i32> [[TMP3]], <i32 3, i32 3, i32 3, i32 3>
+; SLM-NEXT: [[R71:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP4]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; SLM-NEXT: ret <8 x i32> [[R71]]
;
; AVX1-LABEL: @ashr_shl_v8i32_const(
; AVX1-NEXT: [[TMP1:%.*]] = ashr <8 x i32> [[A:%.*]], <i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
diff --git a/llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll b/llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll
index a89bb203ff63d..64e98ddde1f1a 100644
--- a/llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll
+++ b/llvm/test/Transforms/SLPVectorizer/X86/alternate-int.ll
@@ -178,10 +178,12 @@ define <8 x i32> @ashr_shl_v8i32_const(<8 x i32> %a) {
; SSE-NEXT: ret <8 x i32> [[R71]]
;
; SLM-LABEL: @ashr_shl_v8i32_const(
-; SLM-NEXT: [[TMP1:%.*]] = ashr <8 x i32> [[A:%.*]], <i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
-; SLM-NEXT: [[TMP2:%.*]] = shl <8 x i32> [[A]], <i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
-; SLM-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> [[TMP2]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>
-; SLM-NEXT: ret <8 x i32> [[TMP3]]
+; SLM-NEXT: [[TMP1:%.*]] = shufflevector <8 x i32> [[A:%.*]], <8 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; SLM-NEXT: [[TMP2:%.*]] = ashr <4 x i32> [[TMP1]], <i32 2, i32 2, i32 2, i32 2>
+; SLM-NEXT: [[TMP3:%.*]] = shufflevector <8 x i32> [[A]], <8 x i32> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; SLM-NEXT: [[TMP4:%.*]] = shl <4 x i32> [[TMP3]], <i32 3, i32 3, i32 3, i32 3>
+; SLM-NEXT: [[R71:%.*]] = shufflevector <4 x i32> [[TMP2]], <4 x i32> [[TMP4]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; SLM-NEXT: ret <8 x i32> [[R71]]
;
; AVX1-LABEL: @ashr_shl_v8i32_const(
; AVX1-NEXT: [[TMP1:%.*]] = ashr <8 x i32> [[A:%.*]], <i32 2, i32 2, i32 2, i32 2, i32 3, i32 3, i32 3, i32 3>
More information about the llvm-commits
mailing list