6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-12-12 13:30:51 +00:00
Code Issues Projects Releases Wiki Activity

R600/SI: Add subtarget feature for if f32 fma is fast

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@227483 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Matt Arsenault 2015-01-29 19:34:25 +00:00
parent 30a5d21457
commit c416d94735
5 changed files with 23 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
lib/Target/R600/AMDGPU.td
View File

@ -48,6 +48,12 @@ def FeatureFP64Denormals : SubtargetFeature<"fp64-denormals",
"Enable double precision denormal handling",
[FeatureFP64]>;
def FeatureFastFMAF32 : SubtargetFeature<"fast-fmaf",
"FastFMAF32",
"true",
"Assuming f32 fma is at least as fast as mul + add",
[]>;
// Some instructions do not support denormals despite this flag. Using
// fp32 denormals also causes instructions to run at the double
// precision rate for the device.

3
lib/Target/R600/AMDGPUSubtarget.cpp
View File

@ -65,7 +65,8 @@ AMDGPUSubtarget::AMDGPUSubtarget(StringRef TT, StringRef GPU, StringRef FS,
: AMDGPUGenSubtargetInfo(TT, GPU, FS), DevName(GPU), Is64bit(false),
DumpCode(false), R600ALUInst(false), HasVertexCache(false),
TexVTXClauseSize(0), Gen(AMDGPUSubtarget::R600), FP64(false),
FP64Denormals(false), FP32Denormals(false), CaymanISA(false),
FP64Denormals(false), FP32Denormals(false),
FastFMAF32(false), CaymanISA(false),
FlatAddressSpace(false), EnableIRStructurizer(true),
EnablePromoteAlloca(false), EnableIfCvt(true),
EnableLoadStoreOpt(false), WavefrontSize(0), CFALUBug(false), LocalMemorySize(0),

5
lib/Target/R600/AMDGPUSubtarget.h
View File

@ -55,6 +55,7 @@ private:
bool FP64;
bool FP64Denormals;
bool FP32Denormals;
bool FastFMAF32;
bool CaymanISA;
bool FlatAddressSpace;
bool EnableIRStructurizer;
@ -127,6 +128,10 @@ public:
return FP64Denormals;
}
bool hasFastFMAF32() const {
return FastFMAF32;
}
bool hasFlatAddressSpace() const {
return FlatAddressSpace;
}

12
lib/Target/R600/Processors.td
View File

@ -83,9 +83,13 @@ def : Proc<"cayman", R600_VLIW4_Itin,
// Southern Islands
//===----------------------------------------------------------------------===//
def : ProcessorModel<"SI", SIFullSpeedModel, [FeatureSouthernIslands]>;
def : ProcessorModel<"SI", SIFullSpeedModel,
[FeatureSouthernIslands, FeatureFastFMAF32]
>;
def : ProcessorModel<"tahiti", SIFullSpeedModel, [FeatureSouthernIslands]>;
def : ProcessorModel<"tahiti", SIFullSpeedModel,
[FeatureSouthernIslands, FeatureFastFMAF32]
>;
def : ProcessorModel<"pitcairn", SIQuarterSpeedModel, [FeatureSouthernIslands]>;
@ -105,7 +109,9 @@ def : ProcessorModel<"kabini", SIQuarterSpeedModel, [FeatureSeaIslands]>;
def : ProcessorModel<"kaveri", SIQuarterSpeedModel, [FeatureSeaIslands]>;
def : ProcessorModel<"hawaii", SIFullSpeedModel, [FeatureSeaIslands]>;
def : ProcessorModel<"hawaii", SIFullSpeedModel,
[FeatureSeaIslands, FeatureFastFMAF32]
>;
def : ProcessorModel<"mullins", SIQuarterSpeedModel, [FeatureSeaIslands]>;

2
lib/Target/R600/SIISelLowering.cpp
View File

@ -655,7 +655,7 @@ bool SITargetLowering::isFMAFasterThanFMulAndFAdd(EVT VT) const {
switch (VT.getSimpleVT().SimpleTy) {
case MVT::f32:
return false; /* There is V_MAD_F32 for f32 */
return Subtarget->hasFastFMAF32();
case MVT::f64:
return true;
default: