derek/gem5
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

arch-vega: Fix V_FMAC_F32 data type

Browse Source 
The datatype is U32 but should be F32. This is causing an implicit cast
leading to incorrect results. This fixes nn.Dropout in PyTorch.

Change-Id: I546aa917fde1fd6bc832d9d0fa9ffe66505e87dd
This commit is contained in:
Matthew Poremba
2024-03-18 19:02:10 -05:00
parent ba2f5615ba
commit 3f8d0e1ef8
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/arch/amdgpu/vega/insts/vop2.cc
View File

@@ -2167,9 +2167,9 @@ namespace VegaISA
Inst_VOP2__V_FMAC_F32::execute(GPUDynInstPtr gpuDynInst)
{
Wavefront *wf = gpuDynInst->wavefront();
ConstVecOperandU32 src0(gpuDynInst, instData.SRC0);
ConstVecOperandU32 src1(gpuDynInst, instData.VSRC1);
VecOperandU32 vdst(gpuDynInst, instData.VDST);
ConstVecOperandF32 src0(gpuDynInst, instData.SRC0);
ConstVecOperandF32 src1(gpuDynInst, instData.VSRC1);
VecOperandF32 vdst(gpuDynInst, instData.VDST);
src0.readSrc();
src1.read();