[llvm] 20ad05f - [CostModel][X86] Add CostKinds handling for abs ops
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Mon Sep 12 08:34:56 PDT 2022
Author: Simon Pilgrim
Date: 2022-09-12T16:34:37+01:00
New Revision: 20ad05f9b462e23502681f4857de0e850f76fb2c
URL: https://github.com/llvm/llvm-project/commit/20ad05f9b462e23502681f4857de0e850f76fb2c
DIFF: https://github.com/llvm/llvm-project/commit/20ad05f9b462e23502681f4857de0e850f76fb2c.diff
LOG: [CostModel][X86] Add CostKinds handling for abs ops
This was achieved with an updated version of the 'cost-tables vs llvm-mca' script D103695
Added:
Modified:
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
llvm/test/Analysis/CostModel/X86/abs-codesize.ll
llvm/test/Analysis/CostModel/X86/abs-latency.ll
llvm/test/Analysis/CostModel/X86/abs-sizelatency.ll
llvm/test/Analysis/CostModel/X86/abs.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 373be67d5b4e..36b75011832c 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -3153,8 +3153,8 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::CTLZ, MVT::v16i8, { 4 } },
};
static const CostKindTblEntry AVX512BWCostTbl[] = {
- { ISD::ABS, MVT::v32i16, { 1 } },
- { ISD::ABS, MVT::v64i8, { 1 } },
+ { ISD::ABS, MVT::v32i16, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v64i8, { 1, 1, 1, 1 } },
{ ISD::BITREVERSE, MVT::v8i64, { 3 } },
{ ISD::BITREVERSE, MVT::v16i32, { 3 } },
{ ISD::BITREVERSE, MVT::v32i16, { 3 } },
@@ -3212,12 +3212,15 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::USUBSAT, MVT::v64i8, { 1 } },
};
static const CostKindTblEntry AVX512CostTbl[] = {
- { ISD::ABS, MVT::v8i64, { 1 } },
- { ISD::ABS, MVT::v16i32, { 1 } },
- { ISD::ABS, MVT::v32i16, { 2 } },
- { ISD::ABS, MVT::v64i8, { 2 } },
- { ISD::ABS, MVT::v4i64, { 1 } },
- { ISD::ABS, MVT::v2i64, { 1 } },
+ { ISD::ABS, MVT::v8i64, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v4i64, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v2i64, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v16i32, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v8i32, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v32i16, { 2, 7, 4, 4 } },
+ { ISD::ABS, MVT::v16i16, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v64i8, { 2, 7, 4, 4 } },
+ { ISD::ABS, MVT::v32i8, { 1, 1, 1, 1 } },
{ ISD::BITREVERSE, MVT::v8i64, { 36 } },
{ ISD::BITREVERSE, MVT::v16i32, { 24 } },
{ ISD::BITREVERSE, MVT::v32i16, { 10 } },
@@ -3338,10 +3341,14 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::ROTR, MVT::v16i8, { 2 } }
};
static const CostKindTblEntry AVX2CostTbl[] = {
- { ISD::ABS, MVT::v4i64, { 2 } }, // VBLENDVPD(X,VPSUBQ(0,X),X)
- { ISD::ABS, MVT::v8i32, { 1 } },
- { ISD::ABS, MVT::v16i16, { 1 } },
- { ISD::ABS, MVT::v32i8, { 1 } },
+ { ISD::ABS, MVT::v2i64, { 2, 4, 3, 5 } }, // VBLENDVPD(X,VPSUBQ(0,X),X)
+ { ISD::ABS, MVT::v4i64, { 2, 4, 3, 5 } }, // VBLENDVPD(X,VPSUBQ(0,X),X)
+ { ISD::ABS, MVT::v4i32, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v8i32, { 1, 1, 1, 2 } },
+ { ISD::ABS, MVT::v8i16, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v16i16, { 1, 1, 1, 2 } },
+ { ISD::ABS, MVT::v16i8, { 1, 1, 1, 1 } },
+ { ISD::ABS, MVT::v32i8, { 1, 1, 1, 2 } },
{ ISD::BITREVERSE, MVT::v2i64, { 3 } },
{ ISD::BITREVERSE, MVT::v4i64, { 3 } },
{ ISD::BITREVERSE, MVT::v4i32, { 3 } },
@@ -3409,10 +3416,10 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::FSQRT, MVT::v4f64, { 28, 35, 1, 3 } }, // vsqrtpd
};
static const CostKindTblEntry AVX1CostTbl[] = {
- { ISD::ABS, MVT::v4i64, { 5 } }, // VBLENDVPD(X,VPSUBQ(0,X),X)
- { ISD::ABS, MVT::v8i32, { 3 } },
- { ISD::ABS, MVT::v16i16, { 3 } },
- { ISD::ABS, MVT::v32i8, { 3 } },
+ { ISD::ABS, MVT::v4i64, { 6, 8, 6, 12 } }, // VBLENDVPD(X,VPSUBQ(0,X),X)
+ { ISD::ABS, MVT::v8i32, { 3, 6, 4, 5 } },
+ { ISD::ABS, MVT::v16i16, { 3, 6, 4, 5 } },
+ { ISD::ABS, MVT::v32i8, { 3, 6, 4, 5 } },
{ ISD::BITREVERSE, MVT::v4i64, { 12 } }, // 2 x 128-bit Op + extract/insert
{ ISD::BITREVERSE, MVT::v8i32, { 12 } }, // 2 x 128-bit Op + extract/insert
{ ISD::BITREVERSE, MVT::v16i16, { 12 } }, // 2 x 128-bit Op + extract/insert
@@ -3490,7 +3497,7 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::FSQRT, MVT::v4f32, { 18, 18, 1, 1 } }, // Nehalem from http://www.agner.org/
};
static const CostKindTblEntry SSE41CostTbl[] = {
- { ISD::ABS, MVT::v2i64, { 2 } }, // BLENDVPD(X,PSUBQ(0,X),X)
+ { ISD::ABS, MVT::v2i64, { 3, 4, 3, 5 } }, // BLENDVPD(X,PSUBQ(0,X),X)
{ ISD::SMAX, MVT::v4i32, { 1 } },
{ ISD::SMAX, MVT::v16i8, { 1 } },
{ ISD::SMIN, MVT::v4i32, { 1 } },
@@ -3501,9 +3508,9 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::UMIN, MVT::v8i16, { 1 } },
};
static const CostKindTblEntry SSSE3CostTbl[] = {
- { ISD::ABS, MVT::v4i32, { 1 } },
- { ISD::ABS, MVT::v8i16, { 1 } },
- { ISD::ABS, MVT::v16i8, { 1 } },
+ { ISD::ABS, MVT::v4i32, { 1, 2, 1, 1 } },
+ { ISD::ABS, MVT::v8i16, { 1, 2, 1, 1 } },
+ { ISD::ABS, MVT::v16i8, { 1, 2, 1, 1 } },
{ ISD::BITREVERSE, MVT::v2i64, { 5 } },
{ ISD::BITREVERSE, MVT::v4i32, { 5 } },
{ ISD::BITREVERSE, MVT::v8i16, { 5 } },
@@ -3525,10 +3532,10 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::CTTZ, MVT::v16i8, { 9 } }
};
static const CostKindTblEntry SSE2CostTbl[] = {
- { ISD::ABS, MVT::v2i64, { 4 } },
- { ISD::ABS, MVT::v4i32, { 3 } },
- { ISD::ABS, MVT::v8i16, { 2 } },
- { ISD::ABS, MVT::v16i8, { 2 } },
+ { ISD::ABS, MVT::v2i64, { 3, 6, 5, 5 } },
+ { ISD::ABS, MVT::v4i32, { 1, 4, 4, 4 } },
+ { ISD::ABS, MVT::v8i16, { 1, 2, 3, 3 } },
+ { ISD::ABS, MVT::v16i8, { 1, 2, 3, 3 } },
{ ISD::BITREVERSE, MVT::v2i64, { 29 } },
{ ISD::BITREVERSE, MVT::v4i32, { 27 } },
{ ISD::BITREVERSE, MVT::v8i16, { 27 } },
@@ -3598,7 +3605,7 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::CTPOP, MVT::i8, { 1, 1, 2, 2 } }, // popcnt(zext())
};
static const CostKindTblEntry X64CostTbl[] = { // 64-bit targets
- { ISD::ABS, MVT::i64, { 2 } }, // SUB+CMOV
+ { ISD::ABS, MVT::i64, { 1, 2, 3, 4 } }, // SUB+CMOV
{ ISD::BITREVERSE, MVT::i64, { 14 } },
{ ISD::BSWAP, MVT::i64, { 1 } },
{ ISD::CTLZ, MVT::i64, { 4 } }, // BSR+XOR or BSR+XOR+CMOV
@@ -3612,9 +3619,9 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::UMULO, MVT::i64, { 2 } }, // mulq + seto
};
static const CostKindTblEntry X86CostTbl[] = { // 32 or 64-bit targets
- { ISD::ABS, MVT::i32, { 2 } }, // SUB+XOR+SRA or SUB+CMOV
- { ISD::ABS, MVT::i16, { 2 } }, // SUB+XOR+SRA or SUB+CMOV
- { ISD::ABS, MVT::i8, { 2 } }, // SUB+XOR+SRA
+ { ISD::ABS, MVT::i32, { 1, 2, 3, 4 } }, // SUB+XOR+SRA or SUB+CMOV
+ { ISD::ABS, MVT::i16, { 2, 2, 3, 4 } }, // SUB+XOR+SRA or SUB+CMOV
+ { ISD::ABS, MVT::i8, { 2, 4, 4, 4 } }, // SUB+XOR+SRA
{ ISD::BITREVERSE, MVT::i32, { 14 } },
{ ISD::BITREVERSE, MVT::i16, { 14 } },
{ ISD::BITREVERSE, MVT::i8, { 11 } },
diff --git a/llvm/test/Analysis/CostModel/X86/abs-codesize.ll b/llvm/test/Analysis/CostModel/X86/abs-codesize.ll
index e33b81703f26..fd944df26dcf 100644
--- a/llvm/test/Analysis/CostModel/X86/abs-codesize.ll
+++ b/llvm/test/Analysis/CostModel/X86/abs-codesize.ll
@@ -16,43 +16,43 @@ define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64>
; SSE2-LABEL: 'cost_abs_i64'
; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i64'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i64'
; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I64 = call i64 @llvm.abs.i64(i64 %a64, i1 0)
@@ -65,44 +65,44 @@ define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64>
define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_abs_i32'
; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i32'
; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I32 = call i32 @llvm.abs.i32(i32 %a32, i1 0)
@@ -115,58 +115,58 @@ define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32>
define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
; SSE2-LABEL: 'cost_abs_i16'
; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i16'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i16'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i16'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i16'
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I16 = call i16 @llvm.abs.i16(i16 %a16, i1 0)
@@ -178,59 +178,59 @@ define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16
define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
; SSE2-LABEL: 'cost_abs_i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i8'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i8'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i8'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I8 = call i8 @llvm.abs.i8(i8 %a8, i1 0)
@@ -248,43 +248,43 @@ define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
; SSE2-LABEL: 'cost_abs_i64_poison'
; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i64_poison'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i64_poison'
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64_poison'
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64_poison'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i64_poison'
; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I64 = call i64 @llvm.abs.i64(i64 %a64, i1 -1)
@@ -297,44 +297,44 @@ define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_abs_i32_poison'
; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i32_poison'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i32_poison'
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32_poison'
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32_poison'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i32_poison'
; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I32 = call i32 @llvm.abs.i32(i32 %a32, i1 -1)
@@ -347,58 +347,58 @@ define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
; SSE2-LABEL: 'cost_abs_i16_poison'
; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i16_poison'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i16_poison'
; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16_poison'
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i16_poison'
; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i16_poison'
; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i16_poison'
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i16_poison'
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I16 = call i16 @llvm.abs.i16(i16 %a16, i1 -1)
@@ -410,59 +410,59 @@ define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
define void @cost_abs_i8_poison(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
; SSE2-LABEL: 'cost_abs_i8_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i8_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i8_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i8_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i8_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i8_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i8_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I8 = call i8 @llvm.abs.i8(i8 %a8, i1 -1)
diff --git a/llvm/test/Analysis/CostModel/X86/abs-latency.ll b/llvm/test/Analysis/CostModel/X86/abs-latency.ll
index 297ec40dbc29..a33153f26f21 100644
--- a/llvm/test/Analysis/CostModel/X86/abs-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/abs-latency.ll
@@ -1,7 +1,7 @@
; 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=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=latency -mattr=+ssse3 | FileCheck %s -check-prefixes=SSSE3
-; 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=+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=+ssse3 | FileCheck %s -check-prefixes=SSE,SSSE3
+; 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=+avx512f | FileCheck %s -check-prefixes=AVX512,AVX512F
@@ -14,45 +14,45 @@
define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
; SSE2-LABEL: 'cost_abs_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i64'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I64 = call i64 @llvm.abs.i64(i64 %a64, i1 0)
@@ -64,45 +64,45 @@ define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64>
define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_abs_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i32'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I32 = call i32 @llvm.abs.i32(i32 %a32, i1 0)
@@ -113,60 +113,46 @@ define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32>
}
define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_abs_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i16'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I16 = call i16 @llvm.abs.i16(i16 %a16, i1 0)
@@ -177,60 +163,46 @@ define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16
}
define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
-; SSE2-LABEL: 'cost_abs_i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i8'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i8'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i8'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I8 = call i8 @llvm.abs.i8(i8 %a8, i1 0)
@@ -246,45 +218,45 @@ define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a5
define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
; SSE2-LABEL: 'cost_abs_i64_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i64_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i64_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i64_poison'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I64 = call i64 @llvm.abs.i64(i64 %a64, i1 -1)
@@ -296,45 +268,45 @@ define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_abs_i32_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i32_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i32_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i32_poison'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I32 = call i32 @llvm.abs.i32(i32 %a32, i1 -1)
@@ -345,60 +317,46 @@ define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
}
define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_abs_i16_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i16_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i16_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i16_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i16_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i16_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i16_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i16_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I16 = call i16 @llvm.abs.i16(i16 %a16, i1 -1)
@@ -409,60 +367,46 @@ define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
}
define void @cost_abs_i8_poison(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
-; SSE2-LABEL: 'cost_abs_i8_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i8_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i8_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i8_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i8_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i8_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i8_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i8_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I8 = call i8 @llvm.abs.i8(i8 %a8, i1 -1)
diff --git a/llvm/test/Analysis/CostModel/X86/abs-sizelatency.ll b/llvm/test/Analysis/CostModel/X86/abs-sizelatency.ll
index ae2c303b8635..4498d499745a 100644
--- a/llvm/test/Analysis/CostModel/X86/abs-sizelatency.ll
+++ b/llvm/test/Analysis/CostModel/X86/abs-sizelatency.ll
@@ -1,7 +1,7 @@
; 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=size-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=size-latency -mattr=+ssse3 | FileCheck %s -check-prefixes=SSSE3
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-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=size-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=size-latency -mattr=+ssse3 | FileCheck %s -check-prefixes=SSE,SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=size-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=size-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=size-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=size-latency -mattr=+avx512f | FileCheck %s -check-prefixes=AVX512F
@@ -13,60 +13,46 @@
;
define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_abs_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i64'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i64'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i64'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I64 = call i64 @llvm.abs.i64(i64 %a64, i1 0)
@@ -78,59 +64,59 @@ define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64>
define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_abs_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i32'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i32'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i32'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I32 = call i32 @llvm.abs.i32(i32 %a32, i1 0)
@@ -142,59 +128,59 @@ define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32>
define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
; SSE2-LABEL: 'cost_abs_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i16'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i16'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i16'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i16'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I16 = call i16 @llvm.abs.i16(i16 %a16, i1 0)
@@ -206,59 +192,59 @@ define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16
define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
; SSE2-LABEL: 'cost_abs_i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i8'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i8'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i8'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i8'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i8'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I8 = call i8 @llvm.abs.i8(i8 %a8, i1 0)
@@ -273,60 +259,46 @@ define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a5
;
define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_abs_i64_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i64_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i64_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i64_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i64_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i64_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i64_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I64 = call i64 @llvm.abs.i64(i64 %a64, i1 -1)
@@ -338,59 +310,59 @@ define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
; SSE2-LABEL: 'cost_abs_i32_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i32_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i32_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i32_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i32_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i32_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I32 = call i32 @llvm.abs.i32(i32 %a32, i1 -1)
@@ -402,59 +374,59 @@ define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
; SSE2-LABEL: 'cost_abs_i16_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i16_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i16_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i16_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i16_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i16_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i16_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I16 = call i16 @llvm.abs.i16(i16 %a16, i1 -1)
@@ -466,59 +438,59 @@ define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
define void @cost_abs_i8_poison(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
; SSE2-LABEL: 'cost_abs_i8_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSSE3-LABEL: 'cost_abs_i8_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_abs_i8_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 19 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i8_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512F-LABEL: 'cost_abs_i8_poison'
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512DQ-LABEL: 'cost_abs_i8_poison'
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512DQ-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX512BW-LABEL: 'cost_abs_i8_poison'
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%I8 = call i8 @llvm.abs.i8(i8 %a8, i1 -1)
diff --git a/llvm/test/Analysis/CostModel/X86/abs.ll b/llvm/test/Analysis/CostModel/X86/abs.ll
index 43cfdfde42d2..7815fdfd2bdd 100644
--- a/llvm/test/Analysis/CostModel/X86/abs.ll
+++ b/llvm/test/Analysis/CostModel/X86/abs.ll
@@ -1,7 +1,7 @@
; 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 -mattr=+sse2 | FileCheck %s -check-prefixes=SSE2
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s -check-prefixes=SSSE3
-; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -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 -mattr=+sse2 | FileCheck %s -check-prefixes=SSE
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s -check-prefixes=SSE
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s -check-prefixes=SSE
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx | FileCheck %s -check-prefixes=AVX1
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx2 | FileCheck %s -check-prefixes=AVX2
; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+avx512f | FileCheck %s -check-prefixes=AVX512,AVX512F
@@ -13,43 +13,29 @@
;
define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_abs_i64'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i64'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i64'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i64'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i64'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 false)
@@ -63,43 +49,29 @@ define void @cost_abs_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64>
}
define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
-; SSE2-LABEL: 'cost_abs_i32'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i32'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i32'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i32'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i32'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 false)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 false)
@@ -113,26 +85,12 @@ define void @cost_abs_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32>
}
define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_abs_i16'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i16'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i16'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i16'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 false)
@@ -177,26 +135,12 @@ define void @cost_abs_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16
}
define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
-; SSE2-LABEL: 'cost_abs_i8'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i8'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i8'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i8'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 false)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 false)
@@ -245,43 +189,29 @@ define void @cost_abs_i8(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a5
;
define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64> %a512) {
-; SSE2-LABEL: 'cost_abs_i64_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i64_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i64_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i64_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i64_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i64_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i64_poison'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.abs.i64(i64 %a64, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I64 = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %a128, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I64 = call <4 x i64> @llvm.abs.v4i64(<4 x i64> %a256, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I64 = call <8 x i64> @llvm.abs.v8i64(<8 x i64> %a512, i1 true)
@@ -295,43 +225,29 @@ define void @cost_abs_i64_poison(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8
}
define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i32> %a512) {
-; SSE2-LABEL: 'cost_abs_i32_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i32_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i32_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i32_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i32_poison'
-; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX2-LABEL: 'cost_abs_i32_poison'
-; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX512-LABEL: 'cost_abs_i32_poison'
-; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
+; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.abs.i32(i32 %a32, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2I32 = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %a128, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4I32 = call <8 x i32> @llvm.abs.v8i32(<8 x i32> %a256, i1 true)
; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I32 = call <16 x i32> @llvm.abs.v16i32(<16 x i32> %a512, i1 true)
@@ -345,26 +261,12 @@ define void @cost_abs_i32_poison(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16
}
define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i16> %a512) {
-; SSE2-LABEL: 'cost_abs_i16_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i16_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i16_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i16_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8I16 = call <8 x i16> @llvm.abs.v8i16(<8 x i16> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I16 = call <16 x i16> @llvm.abs.v16i16(<16 x i16> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I16 = call <32 x i16> @llvm.abs.v32i16(<32 x i16> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i16_poison'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.abs.i16(i16 %a16, i1 true)
@@ -409,26 +311,12 @@ define void @cost_abs_i16_poison(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <3
}
define void @cost_abs_i8_poison(i8 %a8, <16 x i8> %a128, <32 x i8> %a256, <64 x i8> %a512) {
-; SSE2-LABEL: 'cost_abs_i8_poison'
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
-; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSSE3-LABEL: 'cost_abs_i8_poison'
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
-;
-; SSE42-LABEL: 'cost_abs_i8_poison'
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
+; SSE-LABEL: 'cost_abs_i8_poison'
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16I8 = call <16 x i8> @llvm.abs.v16i8(<16 x i8> %a128, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V32I8 = call <32 x i8> @llvm.abs.v32i8(<32 x i8> %a256, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V64I8 = call <64 x i8> @llvm.abs.v64i8(<64 x i8> %a512, i1 true)
+; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_abs_i8_poison'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I8 = call i8 @llvm.abs.i8(i8 %a8, i1 true)
More information about the llvm-commits
mailing list