This is an archive of the discontinued LLVM Phabricator instance.

Differential D50834

AMDGPU: Add feature for fast f32 denormals
ClosedPublic

Authored by arsenm on Aug 16 2018, 1:35 AM.

Details

Reviewers
rampitec
kzhuravl
b-sumner

Diff Detail

Revision Contents

PathSize
lib/
Target/
AMDGPU/
10 lines
1 line
1 line

Diff 160975

lib/Target/AMDGPU/AMDGPU.td

Show All 15 Lines
//===------------------------------------------------------------===// //===------------------------------------------------------------===//
def FeatureFastFMAF32 : SubtargetFeature<"fast-fmaf", def FeatureFastFMAF32 : SubtargetFeature<"fast-fmaf",
"FastFMAF32", "FastFMAF32",
"true", "true",
"Assuming f32 fma is at least as fast as mul + add" "Assuming f32 fma is at least as fast as mul + add"
>; >;
def FeatureFastDenormalF32 : SubtargetFeature<"fast-denormal-f32",
"FastDenormalF32",
"true",
"Enabling denormals does not cause f32 instructions to run at f64 rates"
>;
def FeatureMIMG_R128 : SubtargetFeature<"mimg-r128", def FeatureMIMG_R128 : SubtargetFeature<"mimg-r128",
"MIMG_R128", "MIMG_R128",
"true", "true",
"Support 128-bit texture resources" "Support 128-bit texture resources"
>; >;
def HalfRate64Ops : SubtargetFeature<"half-rate-64-ops", def HalfRate64Ops : SubtargetFeature<"half-rate-64-ops",
"HalfRate64Ops", "HalfRate64Ops",
▲ Show 20 Lines • Show All 394 Lines • ▼ Show 20 Lines
def FeatureVolcanicIslands : GCNSubtargetFeatureGeneration<"VOLCANIC_ISLANDS", def FeatureVolcanicIslands : GCNSubtargetFeatureGeneration<"VOLCANIC_ISLANDS",
[FeatureFP64, FeatureLocalMemorySize65536, FeatureMIMG_R128, [FeatureFP64, FeatureLocalMemorySize65536, FeatureMIMG_R128,
FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN, FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN,
FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts, FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts,
FeatureSMemRealTime, FeatureVGPRIndexMode, FeatureMovrel, FeatureSMemRealTime, FeatureVGPRIndexMode, FeatureMovrel,
FeatureScalarStores, FeatureInv2PiInlineImm, FeatureScalarStores, FeatureInv2PiInlineImm,
FeatureSDWA, FeatureSDWAOutModsVOPC, FeatureSDWAMac, FeatureDPP, FeatureSDWA, FeatureSDWAOutModsVOPC, FeatureSDWAMac, FeatureDPP,
FeatureIntClamp FeatureIntClamp, FeatureFastDenormalF32
] ]
>; >;
def FeatureGFX9 : GCNSubtargetFeatureGeneration<"GFX9", def FeatureGFX9 : GCNSubtargetFeatureGeneration<"GFX9",
[FeatureFP64, FeatureLocalMemorySize65536, [FeatureFP64, FeatureLocalMemorySize65536,
FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN, FeatureWavefrontSize64, FeatureFlatAddressSpace, FeatureGCN,
FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts, FeatureGCN3Encoding, FeatureCIInsts, FeatureVIInsts, Feature16BitInsts,
FeatureSMemRealTime, FeatureScalarStores, FeatureInv2PiInlineImm, FeatureSMemRealTime, FeatureScalarStores, FeatureInv2PiInlineImm,
FeatureApertureRegs, FeatureGFX9Insts, FeatureVOP3P, FeatureVGPRIndexMode, FeatureApertureRegs, FeatureGFX9Insts, FeatureVOP3P, FeatureVGPRIndexMode,
FeatureFastFMAF32, FeatureDPP, FeatureIntClamp, FeatureFastFMAF32, FeatureDPP, FeatureIntClamp,
FeatureSDWA, FeatureSDWAOmod, FeatureSDWAScalar, FeatureSDWASdst, FeatureSDWA, FeatureSDWAOmod, FeatureSDWAScalar, FeatureSDWASdst,
FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts, FeatureFlatInstOffsets, FeatureFlatGlobalInsts, FeatureFlatScratchInsts,
FeatureAddNoCarryInsts, FeatureScalarAtomics FeatureAddNoCarryInsts, FeatureScalarAtomics, FeatureFastDenormalF32
] ]
>; >;
class SubtargetFeatureISAVersion <int Major, int Minor, int Stepping, class SubtargetFeatureISAVersion <int Major, int Minor, int Stepping,
list<SubtargetFeature> Implies> list<SubtargetFeature> Implies>
: SubtargetFeature < : SubtargetFeature <
"isaver"#Major#"."#Minor#"."#Stepping, "isaver"#Major#"."#Minor#"."#Stepping,
"IsaVersion", "IsaVersion",
▲ Show 20 Lines • Show All 288 Lines • Show Last 20 Lines

lib/Target/AMDGPU/AMDGPUSubtarget.h

Show First 20 Lines • Show All 304 Lines • ▼ Show 20 Linesprotected:
Triple TargetTriple; Triple TargetTriple;
unsigned Gen; unsigned Gen;
unsigned IsaVersion; unsigned IsaVersion;
int LDSBankCount; int LDSBankCount;
unsigned MaxPrivateElementSize; unsigned MaxPrivateElementSize;
// Possibly statically set by tablegen, but may want to be overridden. // Possibly statically set by tablegen, but may want to be overridden.
bool FastFMAF32; bool FastFMAF32;
bool FastDenormalF32;
bool HalfRate64Ops; bool HalfRate64Ops;
// Dynamially set bits that enable features. // Dynamially set bits that enable features.
bool FP64FP16Denormals; bool FP64FP16Denormals;
bool DX10Clamp; bool DX10Clamp;
bool FlatForGlobal; bool FlatForGlobal;
bool AutoWaitcntBeforeBarrier; bool AutoWaitcntBeforeBarrier;
bool CodeObjectV3; bool CodeObjectV3;
▲ Show 20 Lines • Show All 751 Lines • Show Last 20 Lines

lib/Target/AMDGPU/AMDGPUSubtarget.cpp

Show First 20 Lines • Show All 148 Lines • ▼ Show 20 LinesGCNSubtarget::GCNSubtarget(const Triple &TT, StringRef GPU, StringRef FS,
AMDGPUSubtarget(TT, getFeatureBits()), AMDGPUSubtarget(TT, getFeatureBits()),
TargetTriple(TT), TargetTriple(TT),
Gen(SOUTHERN_ISLANDS), Gen(SOUTHERN_ISLANDS),
IsaVersion(ISAVersion0_0_0), IsaVersion(ISAVersion0_0_0),
LDSBankCount(0), LDSBankCount(0),
MaxPrivateElementSize(0), MaxPrivateElementSize(0),
FastFMAF32(false), FastFMAF32(false),
FastDenormalF32(false),
HalfRate64Ops(false), HalfRate64Ops(false),
FP64FP16Denormals(false), FP64FP16Denormals(false),
DX10Clamp(false), DX10Clamp(false),
FlatForGlobal(false), FlatForGlobal(false),
AutoWaitcntBeforeBarrier(false), AutoWaitcntBeforeBarrier(false),
CodeObjectV3(false), CodeObjectV3(false),
UnalignedScratchAccess(false), UnalignedScratchAccess(false),
▲ Show 20 Lines • Show All 531 Lines • Show Last 20 Lines