P4 to Git Change 1397617 by gandryey@gera-w8 on 2017/04/12 16:11:38
SWDEV-118846 - [OCL][Vega] clGetKernelInfoAMD is returning zero for CL_KERNELINFO_AVAILABLE_LDS_SIZE
- Pass LDS size from PAL device properties.
Affected files ...
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palkernel.cpp#33 edit
[ROCm/clr commit: 4f113f6a2a]
Этот коммит содержится в:
@@ -429,12 +429,14 @@ HSAILKernel::aqlCreateHWInfo(amd::hsa::loader::Symbol *sym)
|
||||
workGroupInfo_.usedVGPRs_ = akc->workitem_vgpr_count;
|
||||
|
||||
if (!prog().isNull()) {
|
||||
workGroupInfo_.availableLDSSize_ = dev().properties().gfxipProperties.shaderCore.ldsSizePerCu;
|
||||
workGroupInfo_.availableSGPRs_ = dev().properties().gfxipProperties.shaderCore.numAvailableSgprs;
|
||||
workGroupInfo_.availableVGPRs_ = dev().properties().gfxipProperties.shaderCore.numAvailableVgprs;
|
||||
workGroupInfo_.preferredSizeMultiple_ =
|
||||
workGroupInfo_.wavefrontPerSIMD_ = dev().properties().gfxipProperties.shaderCore.wavefrontSize;
|
||||
}
|
||||
else {
|
||||
workGroupInfo_.availableLDSSize_ = 64 * Ki;
|
||||
workGroupInfo_.availableSGPRs_ = 104;
|
||||
workGroupInfo_.availableVGPRs_ = 256;
|
||||
workGroupInfo_.preferredSizeMultiple_ =
|
||||
|
||||
Ссылка в новой задаче
Block a user