[vdi] Fix calculation of MaxWaves
- Consider the case where `usedVGPRs` is zero.
- This fixes [SWDEV-228537](http://ontrack-internal.amd.com/browse/SWDEV-228537)
Change-Id: I8675311f5fe24fb59c5d45bada122afefb55b128
[ROCm/clr commit: 55d869df99]
Этот коммит содержится в:
@@ -700,17 +700,16 @@ hipError_t ihipOccupancyMaxActiveBlocksPerMultiprocessor(int* numBlocks, int* nu
|
||||
}
|
||||
// Find threads accupancy per CU => simd_per_cu * GPR usage
|
||||
constexpr size_t MaxWavesPerSimd = 8; // Limited by SPI 32 per CU, hence 8 per SIMD
|
||||
size_t VgprWaves = wrkGrpInfo->availableVGPRs_ / amd::alignUp(wrkGrpInfo->usedVGPRs_, 4);
|
||||
|
||||
size_t GprWaves;
|
||||
size_t VgprWaves = MaxWavesPerSimd;
|
||||
if (wrkGrpInfo->usedVGPRs_ > 0) {
|
||||
VgprWaves = wrkGrpInfo->availableVGPRs_ / amd::alignUp(wrkGrpInfo->usedVGPRs_, 4);
|
||||
}
|
||||
size_t GprWaves = VgprWaves;
|
||||
if (wrkGrpInfo->usedSGPRs_ > 0) {
|
||||
const size_t maxSGPRs = (device->info().gfxipVersion_ < 800) ? 512 : 800;
|
||||
size_t SgprWaves = maxSGPRs / amd::alignUp(wrkGrpInfo->usedSGPRs_, 16);
|
||||
GprWaves = std::min(VgprWaves, SgprWaves);
|
||||
}
|
||||
else {
|
||||
GprWaves = VgprWaves;
|
||||
}
|
||||
|
||||
size_t alu_accupancy = device->info().simdPerCU_ * std::min(MaxWavesPerSimd, GprWaves);
|
||||
alu_accupancy *= wrkGrpInfo->wavefrontSize_;
|
||||
|
||||
Ссылка в новой задаче
Block a user