[llvm] 5842dfe - [CostModel][X86] Update SSSE3/AVX1 BSWAP costs
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Tue Jan 2 04:07:17 PST 2024
Author: Simon Pilgrim
Date: 2024-01-02T12:06:48Z
New Revision: 5842dfe34d89d9cc664d14f511ecb415d6609237
URL: https://github.com/llvm/llvm-project/commit/5842dfe34d89d9cc664d14f511ecb415d6609237
DIFF: https://github.com/llvm/llvm-project/commit/5842dfe34d89d9cc664d14f511ecb415d6609237.diff
LOG: [CostModel][X86] Update SSSE3/AVX1 BSWAP costs
Drop atom/slm costs from the default bswap costs, and update the avx1 latency costs based off latest codegen.
Based off analysis report from https://github.com/RKSimon/llvm-scripts/check_cost_tables.py
Fixes #62659
Added:
Modified:
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
llvm/test/Analysis/CostModel/X86/bswap-latency.ll
llvm/test/Analysis/CostModel/X86/bswap.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index e09dc7ff02a070..49631f38017a14 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -3733,10 +3733,10 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::BITREVERSE, MVT::v8i16, { 8, 13, 10, 16 } },
{ ISD::BITREVERSE, MVT::v32i8, { 13, 15, 17, 26 } }, // 2 x 128-bit Op + extract/insert
{ ISD::BITREVERSE, MVT::v16i8, { 7, 7, 9, 13 } },
- { ISD::BSWAP, MVT::v4i64, { 5, 7, 5, 10 } },
- { ISD::BSWAP, MVT::v2i64, { 2, 3, 1, 3 } },
- { ISD::BSWAP, MVT::v8i32, { 5, 7, 5, 10 } },
- { ISD::BSWAP, MVT::v4i32, { 2, 3, 1, 3 } },
+ { ISD::BSWAP, MVT::v4i64, { 5, 6, 5, 10 } },
+ { ISD::BSWAP, MVT::v2i64, { 2, 2, 1, 3 } },
+ { ISD::BSWAP, MVT::v8i32, { 5, 6, 5, 10 } },
+ { ISD::BSWAP, MVT::v4i32, { 2, 2, 1, 3 } },
{ ISD::BSWAP, MVT::v16i16, { 5, 6, 5, 10 } },
{ ISD::BSWAP, MVT::v8i16, { 2, 2, 1, 3 } },
{ ISD::CTLZ, MVT::v4i64, { 29, 33, 49, 58 } }, // 2 x 128-bit Op + extract/insert
@@ -3813,6 +3813,9 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::FSQRT, MVT::v2f64, { 67, 71, 1, 5 } }, // sqrtpd
};
static const CostKindTblEntry SLMCostTbl[] = {
+ { ISD::BSWAP, MVT::v2i64, { 5, 5, 1, 5 } },
+ { ISD::BSWAP, MVT::v4i32, { 5, 5, 1, 5 } },
+ { ISD::BSWAP, MVT::v8i16, { 5, 5, 1, 5 } },
{ ISD::FSQRT, MVT::f32, { 20, 20, 1, 1 } }, // sqrtss
{ ISD::FSQRT, MVT::v4f32, { 40, 41, 1, 5 } }, // sqrtps
{ ISD::FSQRT, MVT::f64, { 35, 35, 1, 1 } }, // sqrtsd
@@ -3851,9 +3854,9 @@ X86TTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
{ ISD::BITREVERSE, MVT::v4i32, { 16, 20, 11, 21 } },
{ ISD::BITREVERSE, MVT::v8i16, { 16, 20, 11, 21 } },
{ ISD::BITREVERSE, MVT::v16i8, { 11, 12, 10, 16 } },
- { ISD::BSWAP, MVT::v2i64, { 5, 5, 1, 5 } },
- { ISD::BSWAP, MVT::v4i32, { 5, 5, 1, 5 } },
- { ISD::BSWAP, MVT::v8i16, { 5, 5, 1, 5 } },
+ { ISD::BSWAP, MVT::v2i64, { 2, 3, 1, 5 } },
+ { ISD::BSWAP, MVT::v4i32, { 2, 3, 1, 5 } },
+ { ISD::BSWAP, MVT::v8i16, { 2, 3, 1, 5 } },
{ ISD::CTLZ, MVT::v2i64, { 18, 28, 28, 35 } },
{ ISD::CTLZ, MVT::v4i32, { 15, 20, 22, 28 } },
{ ISD::CTLZ, MVT::v8i16, { 13, 17, 16, 22 } },
diff --git a/llvm/test/Analysis/CostModel/X86/bswap-latency.ll b/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
index 2d4e6604425816..592541f4816f25 100644
--- a/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
+++ b/llvm/test/Analysis/CostModel/X86/bswap-latency.ll
@@ -24,23 +24,23 @@ define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
;
; SSSE3-LABEL: 'cost_bswap_i64'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_bswap_i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_bswap_i64'
; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_bswap_i64'
@@ -95,23 +95,23 @@ define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
;
; SSSE3-LABEL: 'cost_bswap_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_bswap_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_bswap_i32'
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX2-LABEL: 'cost_bswap_i32'
@@ -166,16 +166,16 @@ define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
;
; SSSE3-LABEL: 'cost_bswap_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; SSE42-LABEL: 'cost_bswap_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
; AVX1-LABEL: 'cost_bswap_i16'
diff --git a/llvm/test/Analysis/CostModel/X86/bswap.ll b/llvm/test/Analysis/CostModel/X86/bswap.ll
index 1c1497965e839e..d1c7daa04cfa60 100644
--- a/llvm/test/Analysis/CostModel/X86/bswap.ll
+++ b/llvm/test/Analysis/CostModel/X86/bswap.ll
@@ -24,16 +24,16 @@ define void @cost_bswap_i64(i64 %a64, <2 x i64> %a128, <4 x i64> %a256, <8 x i64
;
; SSSE3-LABEL: 'cost_bswap_i64'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'cost_bswap_i64'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I64 = call i64 @llvm.bswap.i64(i64 %a64)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I64 = call <2 x i64> @llvm.bswap.v2i64(<2 x i64> %a128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I64 = call <4 x i64> @llvm.bswap.v4i64(<4 x i64> %a256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I64 = call <8 x i64> @llvm.bswap.v8i64(<8 x i64> %a512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_bswap_i64'
@@ -95,16 +95,16 @@ define void @cost_bswap_i32(i32 %a32, <4 x i32> %a128, <8 x i32> %a256, <16 x i3
;
; SSSE3-LABEL: 'cost_bswap_i32'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'cost_bswap_i32'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I32 = call i32 @llvm.bswap.i32(i32 %a32)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2I32 = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %a128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4I32 = call <8 x i32> @llvm.bswap.v8i32(<8 x i32> %a256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8I32 = call <16 x i32> @llvm.bswap.v16i32(<16 x i32> %a512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_bswap_i32'
@@ -166,16 +166,16 @@ define void @cost_bswap_i16(i16 %a16, <8 x i16> %a128, <16 x i16> %a256, <32 x i
;
; SSSE3-LABEL: 'cost_bswap_i16'
; SSSE3-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSSE3-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSSE3-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
; SSSE3-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; SSE42-LABEL: 'cost_bswap_i16'
; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %I16 = call i16 @llvm.bswap.i16(i16 %a16)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
-; SSE42-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8I16 = call <8 x i16> @llvm.bswap.v8i16(<8 x i16> %a128)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16I16 = call <16 x i16> @llvm.bswap.v16i16(<16 x i16> %a256)
+; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V32I16 = call <32 x i16> @llvm.bswap.v32i16(<32 x i16> %a512)
; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void
;
; AVX1-LABEL: 'cost_bswap_i16'
More information about the llvm-commits
mailing list