intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-24 21:18:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

[perf] Remove divisible condition from heuristics.

Browse Source 
Related-To: NEO-6989
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This commit is contained in:
Michal Mrozek
2023-03-01 19:34:40 +00:00
committed by Compute-Runtime-Automation
parent 535debff2d
commit 7028d3c0d3
2 changed files with 0 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
shared/source/xe_hpc_core/command_encoder_xe_hpc_core.cpp
View File

@@ -64,13 +64,6 @@ void EncodeDispatchKernel<Family>::adjustInterfaceDescriptorData(INTERFACE_DESCR
} else {
interfaceDescriptor.setThreadGroupDispatchSize(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_2);
}
uint32_t exponent = INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1 - interfaceDescriptor.getThreadGroupDispatchSize();
uint32_t threadGroupDispatchSize = 1u << exponent;
if ((dispatchedTotalThreadCount % (numberOfThreadsInThreadGroup * threadGroupDispatchSize)) != 0) {
interfaceDescriptor.setThreadGroupDispatchSize(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1);
}
}
}

17
shared/test/unit_test/xe_hpc_core/test_encode_xe_hpc_core.cpp
View File

@@ -642,23 +642,6 @@ XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenNumberOfThreadsInThreadGroupAn
}
}
XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenIndivisibleDispatchSizeWhenCallingAdjustInterfaceDescriptorDataThenThreadGroupDispatchSizeIsCorrectlySet) {
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;
const auto &productHelper = pDevice->getProductHelper();
auto hwInfo = pDevice->getHardwareInfo();
hwInfo.platform.usRevId = productHelper.getHwRevIdFromStepping(REVISION_B, hwInfo);
const uint32_t threadGroupCount = 343u;
const uint32_t numGrf = GrfConfig::DefaultGrfNumber;
INTERFACE_DESCRIPTOR_DATA iddArg = FamilyType::cmdInitInterfaceDescriptorData;
iddArg.setNumberOfThreadsInGpgpuThreadGroup(18u);
EncodeDispatchKernel<FamilyType>::adjustInterfaceDescriptorData(iddArg, *pDevice, hwInfo, threadGroupCount, numGrf);
EXPECT_EQ(INTERFACE_DESCRIPTOR_DATA::THREAD_GROUP_DISPATCH_SIZE_TG_SIZE_1, iddArg.getThreadGroupDispatchSize());
}
XE_HPC_CORETEST_F(EncodeKernelXeHpcCoreTest, givenThreadGroupCountZeroWhenCallingAdjustInterfaceDescriptorDataThenThreadGroupDispatchSizeIsSetToDefault) {
using INTERFACE_DESCRIPTOR_DATA = typename FamilyType::INTERFACE_DESCRIPTOR_DATA;