intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-26 06:15:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use Kernels maxKernelWorkGroupSize to compute WG size

Browse Source 
Change-Id: I947b5612f36cd2437e1ead98712c914ffe4b5b15
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2019-09-16 20:07:04 +02:00
committed by sys_ocldev
parent 6d21c2a422
commit 9d6dcd6fff
6 changed files with 15 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
runtime/kernel/kernel.cpp
View File

@@ -562,7 +562,7 @@ cl_int Kernel::getWorkGroupInfo(cl_device_id device, cl_kernel_work_group_info p
switch (paramName) {
case CL_KERNEL_WORK_GROUP_SIZE:
maxWorkgroupSize = this->device.getDeviceInfo().maxWorkGroupSize;
maxWorkgroupSize = this->maxKernelWorkGroupSize;
if (DebugManager.flags.UseMaxSimdSizeToDeduceMaxWorkgroupSize.get()) {
auto divisionSize = 32 / patchInfo.executionEnvironment->LargestCompiledSIMDSize;
maxWorkgroupSize /= divisionSize;
@@ -614,7 +614,7 @@ cl_int Kernel::getSubGroupInfo(cl_kernel_sub_group_info paramName,
size_t numDimensions = 0;
size_t WGS = 1;
auto maxSimdSize = static_cast<size_t>(getKernelInfo().getMaxSimdSize());
auto maxRequiredWorkGroupSize = static_cast<size_t>(getKernelInfo().getMaxRequiredWorkGroupSize(device.getDeviceInfo().maxWorkGroupSize));
auto maxRequiredWorkGroupSize = static_cast<size_t>(getKernelInfo().getMaxRequiredWorkGroupSize(maxKernelWorkGroupSize));
auto largestCompiledSIMDSize = static_cast<size_t>(getKernelInfo().patchInfo.executionEnvironment->LargestCompiledSIMDSize);
GetInfoHelper info(paramValue, paramValueSize, paramValueSizeRet);