Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use DSS count to callculate max work group size

Browse Source 
This reverts commit 89b7a4c9cc0a69cfd2eb5a0bd138356f34c52f0d.

Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka
2021-07-02 17:37:22 +00:00
committed by Compute-Runtime-Automation
parent 233aab88e7
commit bc121c09fa
8 changed files with 68 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
opencl/test/unit_test/device/device_caps_tests.cpp
View File

@ -1685,3 +1685,50 @@ HWTEST_F(QueueFamilyNameTest, givenBcsWhenGettingQueueFamilyNameThenReturnProper
HWTEST_F(QueueFamilyNameTest, givenInvalidEngineGroupWhenGettingQueueFamilyNameThenReturnEmptyName) {
verify(EngineGroupType::MaxEngineGroups, "");
}
HWCMDTEST_F(IGFX_GEN8_CORE, DeviceGetCapsTest, givenSysInfoWhenDeviceCreatedThenMaxWorkGroupCalculatedCorrectly) {
HardwareInfo myHwInfo = *defaultHwInfo;
GT_SYSTEM_INFO &mySysInfo = myHwInfo.gtSystemInfo;
PLATFORM &myPlatform = myHwInfo.platform;
mySysInfo.EUCount = 16;
mySysInfo.SubSliceCount = 4;
mySysInfo.DualSubSliceCount = 2;
mySysInfo.ThreadCount = 16 * 8;
myPlatform.usRevId = 0x4;
auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(&myHwInfo));
auto minSimd = 8;
auto expectedWG = (mySysInfo.ThreadCount / mySysInfo.EUCount) * (mySysInfo.EUCount / mySysInfo.SubSliceCount) * minSimd;
EXPECT_EQ(expectedWG, device->sharedDeviceInfo.maxWorkGroupSize);
}
HWTEST_F(DeviceGetCapsTest, givenDSSDifferentThanZeroWhenDeviceCreatedThenDualSubSliceCountIsDifferentThanSubSliceCount) {
HardwareInfo myHwInfo = *defaultHwInfo;
GT_SYSTEM_INFO &mySysInfo = myHwInfo.gtSystemInfo;
PLATFORM &myPlatform = myHwInfo.platform;
mySysInfo.EUCount = 16;
mySysInfo.SubSliceCount = 4;
mySysInfo.DualSubSliceCount = 2;
mySysInfo.ThreadCount = 16 * 8;
myPlatform.usRevId = 0x4;
auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(&myHwInfo));
EXPECT_NE(device->sharedDeviceInfo.maxNumEUsPerSubSlice, device->sharedDeviceInfo.maxNumEUsPerDualSubSlice);
}
HWTEST_F(DeviceGetCapsTest, givenDSSCountEqualZeroWhenDeviceCreatedThenMaxEuPerDSSEqualMaxEuPerSS) {
HardwareInfo myHwInfo = *defaultHwInfo;
GT_SYSTEM_INFO &mySysInfo = myHwInfo.gtSystemInfo;
PLATFORM &myPlatform = myHwInfo.platform;
mySysInfo.EUCount = 16;
mySysInfo.SubSliceCount = 4;
mySysInfo.DualSubSliceCount = 0;
mySysInfo.ThreadCount = 16 * 8;
myPlatform.usRevId = 0x4;
auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(&myHwInfo));
EXPECT_EQ(device->sharedDeviceInfo.maxNumEUsPerSubSlice, device->sharedDeviceInfo.maxNumEUsPerDualSubSlice);
}

4
opencl/test/unit_test/kernel/kernel_tests.cpp
View File

@ -2515,6 +2515,7 @@ HWTEST_F(KernelTest, givenKernelWhenDebugFlagToUseMaxSimdForCalculationsIsUsedTh
mySysInfo.EUCount = 24;
mySysInfo.SubSliceCount = 3;
mySysInfo.DualSubSliceCount = 3;
mySysInfo.ThreadCount = 24 * 7;
auto device = std::make_unique<MockClDevice>(MockDevice::createWithNewExecutionEnvironment<MockDevice>(&myHwInfo));
@ -3166,7 +3167,8 @@ TEST_F(KernelTests, givenKernelWithSimdEqual1WhenKernelCreatedThenMaxWorgGroupSi
std::unique_ptr<MockKernel> pKernel(new MockKernel(pProgram, *pKernelInfo, *pClDevice));
auto deviceMaxWorkGroupSize = pDevice->getDeviceInfo().maxWorkGroupSize;
auto maxThreadsPerWG = HwHelper::get(pKernel->getHardwareInfo().platform.eRenderCoreFamily).getMaxThreadsForWorkgroup(pKernel->getHardwareInfo(), static_cast<uint32_t>(pClDevice->getDevice().getDeviceInfo().maxNumEUsPerSubSlice));
auto deviceInfo = pClDevice->getDevice().getDeviceInfo();
auto maxThreadsPerWG = HwHelper::get(pKernel->getHardwareInfo().platform.eRenderCoreFamily).getMaxThreadsForWorkgroupInDSSOrSS(pKernel->getHardwareInfo(), static_cast<uint32_t>(deviceInfo.maxNumEUsPerSubSlice), static_cast<uint32_t>(deviceInfo.maxNumEUsPerDualSubSlice));
EXPECT_LT(pKernel->getMaxKernelWorkGroupSize(), deviceMaxWorkGroupSize);
EXPECT_EQ(pKernel->getMaxKernelWorkGroupSize(), maxThreadsPerWG);