From: Simon Pilgrim Date: Thu, 13 Apr 2023 10:54:44 +0000 (+0100) Subject: [TTI][X86] getMinMaxCost - use existing float minnum/maxnum intrinsic cost values... X-Git-Tag: upstream/17.0.6~11773 X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=2bfd7a07b39c756fd857651fe827a46b7bd9af14;p=platform%2Fupstream%2Fllvm.git [TTI][X86] getMinMaxCost - use existing float minnum/maxnum intrinsic cost values instead of maintaining a duplicate cost table Without fastmath (nnan) flags, minnum/maxnum must perform isnan handling as well as fmin/fmax - meaning the costs are notably higher, this is correctly handled in getIntrinsicInstrCost but was missing from the getMinMaxCost cost tables (which assumed fastmath). Followup to 63c3895327839ba5b57f5b99ec9e888abf976ac6 which handled the integer cases --- diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp index 1db80a4..fcc6427 100644 --- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp @@ -5193,62 +5193,17 @@ X86TTIImpl::getArithmeticReductionCost(unsigned Opcode, VectorType *ValTy, InstructionCost X86TTIImpl::getMinMaxCost(Type *Ty, Type *CondTy, TTI::TargetCostKind CostKind, bool IsUnsigned, FastMathFlags FMF) { + Intrinsic::ID Id; if (Ty->isIntOrIntVectorTy()) { - Intrinsic::ID Id = IsUnsigned ? Intrinsic::umin : Intrinsic::smin; - IntrinsicCostAttributes ICA(Id, Ty, {Ty, Ty}, FMF); - return getIntrinsicInstrCost(ICA, CostKind); + Id = IsUnsigned ? Intrinsic::umin : Intrinsic::smin; + } else { + assert(Ty->isFPOrFPVectorTy() && + "Expected float point or integer vector type."); + Id = Intrinsic::minnum; } - // TODO: Use getIntrinsicInstrCost once ISD::FMINNUM costs are improved. - assert(Ty->isFPOrFPVectorTy() && - "Expected float point or integer vector type."); - std::pair LT = getTypeLegalizationCost(Ty); - MVT MTy = LT.second; - int ISD = ISD::FMINNUM; - - static const CostTblEntry SSE1CostTbl[] = { - {ISD::FMINNUM, MVT::v4f32, 1}, - }; - - static const CostTblEntry SSE2CostTbl[] = { - {ISD::FMINNUM, MVT::v2f64, 1}, - }; - - static const CostTblEntry AVX1CostTbl[] = { - {ISD::FMINNUM, MVT::v8f32, 1}, - {ISD::FMINNUM, MVT::v4f64, 1}, - }; - - static const CostTblEntry AVX512CostTbl[] = { - {ISD::FMINNUM, MVT::v16f32, 1}, - {ISD::FMINNUM, MVT::v8f64, 1}, - }; - - // If we have a native MIN/MAX instruction for this type, use it. - if (ST->hasAVX512()) - if (const auto *Entry = CostTableLookup(AVX512CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; - - if (ST->hasAVX()) - if (const auto *Entry = CostTableLookup(AVX1CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; - - if (ST->hasSSE2()) - if (const auto *Entry = CostTableLookup(SSE2CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; - - if (ST->hasSSE1()) - if (const auto *Entry = CostTableLookup(SSE1CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; - - // Otherwise fall back to cmp+select. - unsigned CmpOpcode = Instruction::FCmp; - InstructionCost Result = - getCmpSelInstrCost(CmpOpcode, Ty, CondTy, CmpInst::BAD_ICMP_PREDICATE, - CostKind) + - getCmpSelInstrCost(Instruction::Select, Ty, CondTy, - CmpInst::BAD_ICMP_PREDICATE, CostKind); - return Result; + IntrinsicCostAttributes ICA(Id, Ty, {Ty, Ty}, FMF); + return getIntrinsicInstrCost(ICA, CostKind); } InstructionCost diff --git a/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll b/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll index 5cc2861..93429ee 100644 --- a/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll +++ b/llvm/test/Analysis/CostModel/X86/intrinsic-cost-kinds.ll @@ -352,19 +352,19 @@ define void @maskedscatter(<16 x float> %va, <16 x ptr> %vb, <16 x i1> %vc) { define void @reduce_fmax(<16 x float> %va) { ; THRU-LABEL: 'reduce_fmax' -; THRU-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) +; THRU-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) ; THRU-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret void ; ; LATE-LABEL: 'reduce_fmax' -; LATE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) +; LATE-NEXT: Cost Model: Found an estimated cost of 247 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) ; LATE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void ; ; SIZE-LABEL: 'reduce_fmax' -; SIZE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) +; SIZE-NEXT: Cost Model: Found an estimated cost of 67 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) ; SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void ; ; SIZE_LATE-LABEL: 'reduce_fmax' -; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) +; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 247 for instruction: %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) ; SIZE_LATE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void ; %v = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> %va) diff --git a/llvm/test/Analysis/CostModel/X86/reduce-fmax.ll b/llvm/test/Analysis/CostModel/X86/reduce-fmax.ll index cdeb915..15495be6 100644 --- a/llvm/test/Analysis/CostModel/X86/reduce-fmax.ll +++ b/llvm/test/Analysis/CostModel/X86/reduce-fmax.ll @@ -3,8 +3,8 @@ ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE -; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX -; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX +; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1 +; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2 ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512 ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512 ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512 @@ -12,26 +12,34 @@ define i32 @reduce_f64(i32 %arg) { ; SSE-LABEL: 'reduce_f64' ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmax.v1f64(<1 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX-LABEL: 'reduce_f64' -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmax.v1f64(<1 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX1-LABEL: 'reduce_f64' +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmax.v1f64(<1 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'reduce_f64' +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmax.v1f64(<1 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'reduce_f64' ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmax.v1f64(<1 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2 = call double @llvm.vector.reduce.fmax.v2f64(<2 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4 = call double @llvm.vector.reduce.fmax.v4f64(<4 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8 = call double @llvm.vector.reduce.fmax.v8f64(<8 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16 = call double @llvm.vector.reduce.fmax.v16f64(<16 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %V1 = call double @llvm.vector.reduce.fmax.v1f64(<1 x double> undef) @@ -45,29 +53,38 @@ define i32 @reduce_f64(i32 %arg) { define i32 @reduce_f32(i32 %arg) { ; SSE-LABEL: 'reduce_f32' ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmax.v1f32(<1 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX-LABEL: 'reduce_f32' -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmax.v1f32(<1 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX1-LABEL: 'reduce_f32' +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmax.v1f32(<1 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'reduce_f32' +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmax.v1f32(<1 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'reduce_f32' ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmax.v1f32(<1 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2 = call float @llvm.vector.reduce.fmax.v2f32(<2 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4 = call float @llvm.vector.reduce.fmax.v4f32(<4 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8 = call float @llvm.vector.reduce.fmax.v8f32(<8 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16 = call float @llvm.vector.reduce.fmax.v16f32(<16 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32 = call float @llvm.vector.reduce.fmax.v32f32(<32 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %V1 = call float @llvm.vector.reduce.fmax.v1f32(<1 x float> undef) diff --git a/llvm/test/Analysis/CostModel/X86/reduce-fmin.ll b/llvm/test/Analysis/CostModel/X86/reduce-fmin.ll index 91b27a2..eb41dbc 100644 --- a/llvm/test/Analysis/CostModel/X86/reduce-fmin.ll +++ b/llvm/test/Analysis/CostModel/X86/reduce-fmin.ll @@ -3,8 +3,8 @@ ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+ssse3 | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse4.1 | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE -; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX -; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX +; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1 +; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2 ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512 ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512 ; RUN: opt < %s -passes="print" -mtriple=x86_64-apple-darwin 2>&1 -disable-output -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=AVX512 @@ -12,26 +12,34 @@ define i32 @reduce_f64(i32 %arg) { ; SSE-LABEL: 'reduce_f64' ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmin.v1f64(<1 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 33 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX-LABEL: 'reduce_f64' -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmin.v1f64(<1 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX1-LABEL: 'reduce_f64' +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmin.v1f64(<1 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 23 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'reduce_f64' +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmin.v1f64(<1 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'reduce_f64' ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call double @llvm.vector.reduce.fmin.v1f64(<1 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2 = call double @llvm.vector.reduce.fmin.v2f64(<2 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4 = call double @llvm.vector.reduce.fmin.v4f64(<4 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8 = call double @llvm.vector.reduce.fmin.v8f64(<8 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V16 = call double @llvm.vector.reduce.fmin.v16f64(<16 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %V1 = call double @llvm.vector.reduce.fmin.v1f64(<1 x double> undef) @@ -45,29 +53,38 @@ define i32 @reduce_f64(i32 %arg) { define i32 @reduce_f32(i32 %arg) { ; SSE-LABEL: 'reduce_f32' ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmin.v1f32(<1 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 38 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX-LABEL: 'reduce_f32' -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmin.v1f32(<1 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) -; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX1-LABEL: 'reduce_f32' +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmin.v1f32(<1 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 27 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) +; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; +; AVX2-LABEL: 'reduce_f32' +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmin.v1f32(<1 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) +; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'reduce_f32' ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: %V1 = call float @llvm.vector.reduce.fmin.v1f32(<1 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2 = call float @llvm.vector.reduce.fmin.v2f32(<2 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V4 = call float @llvm.vector.reduce.fmin.v4f32(<4 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %V8 = call float @llvm.vector.reduce.fmin.v8f32(<8 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16 = call float @llvm.vector.reduce.fmin.v16f32(<16 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %V32 = call float @llvm.vector.reduce.fmin.v32f32(<32 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %V1 = call float @llvm.vector.reduce.fmin.v1f32(<1 x float> undef)