intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-07 12:42:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Adjust maxWorkGroupSize value"

Browse Source 
This reverts commit f7685a93e4.

Signed-off-by: Compute-Runtime-Validation <compute-runtime-validation@intel.com>
This commit is contained in:
Compute-Runtime-Validation
2023-02-20 11:46:37 +01:00
committed by Compute-Runtime-Automation
parent 83eb52591d
commit 678e47de2d
14 changed files with 21 additions and 83 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
level_zero/core/source/kernel/kernel_imp.cpp
View File

@@ -737,9 +737,7 @@ ze_result_t KernelImp::getProperties(ze_kernel_properties_t *pKernelProperties)
memset(pKernelProperties->uuid.kid, 0, ZE_MAX_KERNEL_UUID_SIZE);
memset(pKernelProperties->uuid.mid, 0, ZE_MAX_MODULE_UUID_SIZE);
const auto &gfxCoreHelper = this->module->getDevice()->getGfxCoreHelper();
uint32_t maxKernelWorkGroupSize = static_cast<uint32_t>(this->module->getDevice()->getNEODevice()->getDeviceInfo().maxWorkGroupSize);
maxKernelWorkGroupSize = gfxCoreHelper.adjustMaxWorkGroupSize(kernelDescriptor.kernelAttributes.numGrfRequired, kernelDescriptor.kernelAttributes.simdSize, maxKernelWorkGroupSize);
pKernelProperties->maxNumSubgroups = maxKernelWorkGroupSize / kernelDescriptor.kernelAttributes.simdSize;
void *pNext = pKernelProperties->pNext;
@@ -750,6 +748,7 @@ ze_result_t KernelImp::getProperties(ze_kernel_properties_t *pKernelProperties)
reinterpret_cast<ze_kernel_preferred_group_size_properties_t *>(extendedProperties);
preferredGroupSizeProperties->preferredMultiple = this->kernelImmData->getKernelInfo()->getMaxSimdSize();
auto &gfxCoreHelper = this->module->getDevice()->getGfxCoreHelper();
if (gfxCoreHelper.isFusedEuDispatchEnabled(this->module->getDevice()->getHwInfo(), kernelDescriptor.kernelAttributes.flags.requiresDisabledEUFusion)) {
preferredGroupSizeProperties->preferredMultiple *= 2;
}

16
level_zero/core/test/unit_tests/sources/cmdlist/test_cmdlist_7.cpp
View File

@@ -319,7 +319,15 @@ HWTEST2_F(CommandListAppendLaunchKernel, givenSignalEventWhenAppendLaunchIndirec
context->freeMem(alloc);
}
HWTEST2_F(CommandListAppendLaunchKernel, GivenComputeModePropertiesWhenUpdateStreamPropertiesIsCalledTwiceThenChangedFieldsAreDirty, IsAtLeastGen12lp) {
struct ProgramChangedFieldsInComputeMode {
template <PRODUCT_FAMILY productFamily>
static constexpr bool isMatched() {
if (productFamily == IGFX_BROADWELL)
return false;
return TestTraits<NEO::ToGfxCoreFamily<productFamily>::get()>::programOnlyChangedFieldsInComputeStateMode;
}
};
HWTEST2_F(CommandListAppendLaunchKernel, GivenComputeModePropertiesWhenUpdateStreamPropertiesIsCalledTwiceThenChangedFieldsAreDirty, ProgramChangedFieldsInComputeMode) {
DebugManagerStateRestore restorer;
auto &productHelper = device->getProductHelper();
@@ -352,11 +360,7 @@ HWTEST2_F(CommandListAppendLaunchKernel, GivenComputeModePropertiesWhenUpdateStr
const_cast<NEO::KernelDescriptor *>(&kernel.getKernelDescriptor())->kernelAttributes.numGrfRequired = 0x80;
commandList->updateStreamProperties(kernel, false, &launchKernelArgs, false);
if constexpr (TestTraits<gfxCoreFamily>::largeGrfModeInStateComputeModeSupported) {
EXPECT_EQ(productHelper.isGrfNumReportedWithScm(), commandList->finalStreamState.stateComputeMode.largeGrfMode.isDirty);
} else {
EXPECT_EQ(0, commandList->finalStreamState.stateComputeMode.largeGrfMode.isDirty);
}
EXPECT_EQ(productHelper.isGrfNumReportedWithScm(), commandList->finalStreamState.stateComputeMode.largeGrfMode.isDirty);
if (productHelper.getScmPropertyCoherencyRequiredSupport()) {
EXPECT_EQ(0, commandList->finalStreamState.stateComputeMode.isCoherencyRequired.value);
} else {