intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-20 08:53:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Disable adjusting tg dispatch size on pvc

Browse Source 
Disable only on asymmetric SKUs

Disabling because of issues with specaccel

Related-To: NEO-6989

Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This commit is contained in:
Dominik Dabek
2022-10-26 08:05:59 +00:00
committed by Compute-Runtime-Automation
parent 26144d38a6
commit 040dd8302c
2 changed files with 28 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp
View File

@@ -36,7 +36,7 @@ void EncodeDispatchKernel<Family>::adjustInterfaceDescriptorData(INTERFACE_DESCR
if (hwInfoConfig.isDisableOverdispatchAvailable(hwInfo)) { if (hwInfoConfig.isDisableOverdispatchAvailable(hwInfo)) {
interfaceDescriptor.setThreadGroupDispatchSize(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1); interfaceDescriptor.setThreadGroupDispatchSize(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1);
bool adjustTGDispatchSize = true; bool adjustTGDispatchSize = hwInfo.gtSystemInfo.MaxDualSubSlicesSupported == hwInfo.gtSystemInfo.DualSubSliceCount;
if (DebugManager.flags.AdjustThreadGroupDispatchSize.get() != -1) { if (DebugManager.flags.AdjustThreadGroupDispatchSize.get() != -1) {
adjustTGDispatchSize = !!DebugManager.flags.AdjustThreadGroupDispatchSize.get(); adjustTGDispatchSize = !!DebugManager.flags.AdjustThreadGroupDispatchSize.get();
} }
@@ -49,7 +49,6 @@ void EncodeDispatchKernel<Family>::adjustInterfaceDescriptorData(INTERFACE_DESCR
uint32_t availableThreadCount = hwHelper.calculateAvailableThreadCount(hwInfo, numGrf); uint32_t availableThreadCount = hwHelper.calculateAvailableThreadCount(hwInfo, numGrf);
uint32_t numberOfThreadsInThreadGroup = interfaceDescriptor.getNumberOfThreadsInGpgpuThreadGroup(); uint32_t numberOfThreadsInThreadGroup = interfaceDescriptor.getNumberOfThreadsInGpgpuThreadGroup();
uint32_t dispatchedTotalThreadCount = numberOfThreadsInThreadGroup * threadGroupCount; uint32_t dispatchedTotalThreadCount = numberOfThreadsInThreadGroup * threadGroupCount;
UNRECOVERABLE_IF(numberOfThreadsInThreadGroup == 0u); UNRECOVERABLE_IF(numberOfThreadsInThreadGroup == 0u);
if (dispatchedTotalThreadCount <= availableThreadCount) { if (dispatchedTotalThreadCount <= availableThreadCount) {

26
shared/test/unit_test/xe_hpc_core/test_encode_xe_hpc_core.cpp
View File

@@ -510,6 +510,28 @@ XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenCleanHeapsAndSlmNotChangedAndU
(gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER_CACHELINE_MISALIGNED))); (gmmHelper->getMOCS(GMM_RESOURCE_USAGE_OCL_BUFFER_CACHELINE_MISALIGNED)));
} }
XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenInterfaceDescriptorDataAndNonSymmetricalSkuWhenAdjustInterfaceDescriptorDataIsCalledThenThreadGroupDispatchSizeIsCorrectlySet) {
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
INTERFACE_DESCRIPTOR_DATA iddArg = FamilyType::cmdInitInterfaceDescriptorData;
const auto &hwInfoConfig = *HwInfoConfig::get(productFamily);
auto hwInfo = pDevice->getHardwareInfo();
hwInfo.gtSystemInfo.DualSubSliceCount = 32u;
ASSERT_NE(hwInfo.gtSystemInfo.DualSubSliceCount, hwInfo.gtSystemInfo.MaxDualSubSlicesSupported);
for (const auto &revision : {REVISION_A0, REVISION_B}) {
hwInfo.platform.usRevId = hwInfoConfig.getHwRevIdFromStepping(revision, hwInfo);
EncodeDispatchKernel<FamilyType>::adjustInterfaceDescriptorData(iddArg, hwInfo, 0, 0);
if (hwInfoConfig.isDisableOverdispatchAvailable(hwInfo)) {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1, iddArg.getThreadGroupDispatchSize());
} else {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_8, iddArg.getThreadGroupDispatchSize());
}
}
}
XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenDispatchSizeSmallerOrEqualToAvailableThreadCountWhenAdjustInterfaceDescriptorDataIsCalledThenThreadGroupDispatchSizeIsCorrectlySet) { XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenDispatchSizeSmallerOrEqualToAvailableThreadCountWhenAdjustInterfaceDescriptorDataIsCalledThenThreadGroupDispatchSizeIsCorrectlySet) {
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA; using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
@@ -548,7 +570,11 @@ XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenNumberOfThreadsInThreadGroupWh
EncodeDispatchKernel<FamilyType>::adjustInterfaceDescriptorData(iddArg, hwInfo, threadGroupCount, numGrf); EncodeDispatchKernel<FamilyType>::adjustInterfaceDescriptorData(iddArg, hwInfo, threadGroupCount, numGrf);
if (hwInfo.gtSystemInfo.MaxDualSubSlicesSupported == hwInfo.gtSystemInfo.DualSubSliceCount) {
EXPECT_EQ(expectedThreadGroupDispatchSize, iddArg.getThreadGroupDispatchSize()); EXPECT_EQ(expectedThreadGroupDispatchSize, iddArg.getThreadGroupDispatchSize());
} else {
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1, iddArg.getThreadGroupDispatchSize());
}
} }
} }