diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp --- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp +++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp @@ -2802,93 +2802,105 @@ return LT.first * Cost; } + auto adjustTableCost = [](const CostTblEntry &Entry, int LegalizationCost, + FastMathFlags FMF) { + // If there are no NANs to deal with, then these are reduced to a + // single MIN** or MAX** instruction instead of the MIN/CMP/SELECT that we + // assume is used in the non-fast case. + if (Entry.ISD == ISD::FMAXNUM || Entry.ISD == ISD::FMINNUM) { + if (FMF.noNaNs()) + return LegalizationCost * 1; + } + return LegalizationCost * (int)Entry.Cost; + }; + if (ST->useGLMDivSqrtCosts()) if (const auto *Entry = CostTableLookup(GLMCostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->isSLM()) if (const auto *Entry = CostTableLookup(SLMCostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasCDI()) if (const auto *Entry = CostTableLookup(AVX512CDCostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasBWI()) if (const auto *Entry = CostTableLookup(AVX512BWCostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasAVX512()) if (const auto *Entry = CostTableLookup(AVX512CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasXOP()) if (const auto *Entry = CostTableLookup(XOPCostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasAVX2()) if (const auto *Entry = CostTableLookup(AVX2CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasAVX()) if (const auto *Entry = CostTableLookup(AVX1CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasSSE42()) if (const auto *Entry = CostTableLookup(SSE42CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasSSE41()) if (const auto *Entry = CostTableLookup(SSE41CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasSSSE3()) if (const auto *Entry = CostTableLookup(SSSE3CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasSSE2()) if (const auto *Entry = CostTableLookup(SSE2CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasSSE1()) if (const auto *Entry = CostTableLookup(SSE1CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (ST->hasBMI()) { if (ST->is64Bit()) if (const auto *Entry = CostTableLookup(BMI64CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (const auto *Entry = CostTableLookup(BMI32CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); } if (ST->hasLZCNT()) { if (ST->is64Bit()) if (const auto *Entry = CostTableLookup(LZCNT64CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (const auto *Entry = CostTableLookup(LZCNT32CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); } if (ST->hasPOPCNT()) { if (ST->is64Bit()) if (const auto *Entry = CostTableLookup(POPCNT64CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (const auto *Entry = CostTableLookup(POPCNT32CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); } // TODO - add BMI (TZCNT) scalar handling if (ST->is64Bit()) if (const auto *Entry = CostTableLookup(X64CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); if (const auto *Entry = CostTableLookup(X86CostTbl, ISD, MTy)) - return LT.first * Entry->Cost; + return adjustTableCost(*Entry, LT.first, ICA.getFlags()); } return BaseT::getIntrinsicInstrCost(ICA, CostKind); diff --git a/llvm/test/Analysis/CostModel/X86/fmaxnum.ll b/llvm/test/Analysis/CostModel/X86/fmaxnum.ll --- a/llvm/test/Analysis/CostModel/X86/fmaxnum.ll +++ b/llvm/test/Analysis/CostModel/X86/fmaxnum.ll @@ -1,14 +1,14 @@ ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE -; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX1 -; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2 +; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1 +; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512 ; ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=slm | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=goldmont | FileCheck %s --check-prefixes=SSE -; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=btver2 | FileCheck %s --check-prefixes=AVX1 +; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1 define i32 @f32(i32 %arg) { ; SSE-LABEL: 'f32' @@ -94,35 +94,27 @@ define i32 @f32_nnan(i32 %arg) { ; SSE-LABEL: 'f32_nnan' -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX1-LABEL: 'f32_nnan' -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef -; -; AVX2-LABEL: 'f32_nnan' -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX-LABEL: 'f32_nnan' +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'f32_nnan' -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call nnan <2 x float> @llvm.maxnum.v2f32(<2 x float> undef, <2 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> undef, <4 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call nnan <8 x float> @llvm.maxnum.v8f32(<8 x float> undef, <8 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call nnan <16 x float> @llvm.maxnum.v16f32(<16 x float> undef, <16 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %F32 = call nnan float @llvm.maxnum.f32(float undef, float undef) @@ -135,35 +127,27 @@ define i32 @f64_nnan(i32 %arg) { ; SSE-LABEL: 'f64_nnan' -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX1-LABEL: 'f64_nnan' -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef -; -; AVX2-LABEL: 'f64_nnan' -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX-LABEL: 'f64_nnan' +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'f64_nnan' -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> undef, <2 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4f64 = call nnan <4 x double> @llvm.maxnum.v4f64(<4 x double> undef, <4 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8f64 = call nnan <8 x double> @llvm.maxnum.v8f64(<8 x double> undef, <8 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16f64 = call nnan <16 x double> @llvm.maxnum.v16f64(<16 x double> undef, <16 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %f64 = call nnan double @llvm.maxnum.f64(double undef, double undef) diff --git a/llvm/test/Analysis/CostModel/X86/fminnum.ll b/llvm/test/Analysis/CostModel/X86/fminnum.ll --- a/llvm/test/Analysis/CostModel/X86/fminnum.ll +++ b/llvm/test/Analysis/CostModel/X86/fminnum.ll @@ -1,14 +1,14 @@ ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+sse2 | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+sse4.2 | FileCheck %s --check-prefixes=SSE -; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX1 -; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX2 +; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVX1 +; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVX2 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512 ; ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=slm | FileCheck %s --check-prefixes=SSE ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=goldmont | FileCheck %s --check-prefixes=SSE -; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=btver2 | FileCheck %s --check-prefixes=AVX1 +; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-- -mcpu=btver2 | FileCheck %s --check-prefixes=AVX,AVX1 define i32 @f32(i32 %arg) { ; SSE-LABEL: 'f32' @@ -94,35 +94,27 @@ define i32 @f32_nnan(i32 %arg) { ; SSE-LABEL: 'f32_nnan' -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX1-LABEL: 'f32_nnan' -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef -; -; AVX2-LABEL: 'f32_nnan' -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX-LABEL: 'f32_nnan' +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'f32_nnan' -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2F32 = call nnan <2 x float> @llvm.minnum.v2f32(<2 x float> undef, <2 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4F32 = call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> undef, <4 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8F32 = call nnan <8 x float> @llvm.minnum.v8f32(<8 x float> undef, <8 x float> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V16F32 = call nnan <16 x float> @llvm.minnum.v16f32(<16 x float> undef, <16 x float> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %F32 = call nnan float @llvm.minnum.f32(float undef, float undef) @@ -135,35 +127,27 @@ define i32 @f64_nnan(i32 %arg) { ; SSE-LABEL: 'f64_nnan' -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) -; SSE-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) +; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; -; AVX1-LABEL: 'f64_nnan' -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) -; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef -; -; AVX2-LABEL: 'f64_nnan' -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) -; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef +; AVX-LABEL: 'f64_nnan' +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) +; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; ; AVX512-LABEL: 'f64_nnan' -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) -; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call nnan double @llvm.minnum.f64(double undef, double undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V2f64 = call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> undef, <2 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V4f64 = call nnan <4 x double> @llvm.minnum.v4f64(<4 x double> undef, <4 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %V8f64 = call nnan <8 x double> @llvm.minnum.v8f64(<8 x double> undef, <8 x double> undef) +; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %V16f64 = call nnan <16 x double> @llvm.minnum.v16f64(<16 x double> undef, <16 x double> undef) ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 undef ; %f64 = call nnan double @llvm.minnum.f64(double undef, double undef)