intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: improve reporting max cooperative group count

Browse Source 
Related-To: NEO-8210

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2024-01-22 11:57:53 +00:00
committed by Compute-Runtime-Automation
parent 8d56f8fb6b
commit b77e1a6a71
7 changed files with 97 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
level_zero/core/source/kernel/kernel_imp.cpp
View File

@@ -8,6 +8,7 @@
#include "level_zero/core/source/kernel/kernel_imp.h"
#include "shared/source/assert_handler/assert_handler.h"
#include "shared/source/command_container/implicit_scaling.h"
#include "shared/source/debugger/debugger_l0.h"
#include "shared/source/execution_environment/root_device_environment.h"
#include "shared/source/gmm_helper/gmm_helper.h"
@@ -482,8 +483,18 @@ ze_result_t KernelImp::suggestMaxCooperativeGroupCount(uint32_t *totalGroupCount
const uint32_t workDim = 3;
const size_t localWorkSize[] = {groupSize[0], groupSize[1], groupSize[2]};
uint32_t numSubDevicesForExecution = 1;
bool platformImplicitScaling = helper.platformSupportsImplicitScaling(rootDeviceEnvironment);
auto deviceBitfield = module->getDevice()->getNEODevice()->getDeviceBitfield();
if (NEO::ImplicitScalingHelper::isImplicitScalingEnabled(deviceBitfield, platformImplicitScaling)) {
numSubDevicesForExecution = static_cast<uint32_t>(deviceBitfield.count());
}
*totalGroupCount = NEO::KernelHelper::getMaxWorkGroupCount(rootDeviceEnvironment,
descriptor,
numSubDevicesForExecution,
usedSlmSize,
workDim,
localWorkSize,