aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Generate FMA instructions with AVX2 (when possible).

Browse Source 
Issue #320.
This commit is contained in:
Matt Pharr
2012-08-03 10:43:41 -07:00
parent e6aec96e05
commit 19d8f2e258
3 changed files with 22 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
ispc.cpp
View File

@@ -366,7 +366,11 @@ Target::GetTarget(const char *arch, const char *cpu, const char *isa,
t->isa = Target::AVX2;
t->nativeVectorWidth = 8;
t->vectorWidth = 8;
t->attributes = "+avx2,+popcnt,+cmov,+f16c,+rdrand";
t->attributes = "+avx2,+popcnt,+cmov,+f16c,+rdrand"
#ifndef LLVM_3_1
",+fma"
#endif // !LLVM_3_1
;
t->maskingIsFree = false;
t->maskBitCount = 32;
t->hasHalf = true;
@@ -380,7 +384,11 @@ Target::GetTarget(const char *arch, const char *cpu, const char *isa,
t->isa = Target::AVX2;
t->nativeVectorWidth = 16;
t->vectorWidth = 16;
t->attributes = "+avx2,+popcnt,+cmov,+f16c,+rdrand";
t->attributes = "+avx2,+popcnt,+cmov,+f16c,+rdrand"
#ifndef LLVM_3_1
",+fma"
#endif // !LLVM_3_1
;
t->maskingIsFree = false;
t->maskBitCount = 32;
t->hasHalf = true;
@@ -477,6 +485,10 @@ Target::GetTargetMachine() const {
#else
std::string featuresString = attributes;
llvm::TargetOptions options;
#if !defined(LLVM_3_1)
if (g->opt.disableFMA == false)
options.AllowFPOpFusion = llvm::FPOpFusion::Fast;
#endif // !LLVM_3_1
llvm::TargetMachine *targetMachine =
target->createTargetMachine(triple, cpu, featuresString, options,
relocModel);
@@ -627,6 +639,7 @@ Opt::Opt() {
force32BitAddressing = true;
unrollLoops = true;
disableAsserts = false;
disableFMA = false;
disableMaskAllOnOptimizations = false;
disableHandlePseudoMemoryOps = false;
disableBlendedMaskedStores = false;