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

Correct dimension order in local ids generated for implicit args

Browse Source 
when local ids are generated by HW, use same dim order for runtime generation
move common logic to separated file

Related-To: NEO-5081
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This commit is contained in:
Mateusz Jablonski
2022-02-03 17:00:25 +00:00
committed by Compute-Runtime-Automation
parent ca5a8162eb
commit b697d75695
15 changed files with 282 additions and 35 deletions
Download Patch File Download Diff File Expand all files Collapse all files

72
opencl/test/unit_test/helpers/hardware_commands_helper_tests.cpp
View File

@ -1261,7 +1261,9 @@ HWTEST_F(HardwareCommandsImplicitArgsTests, givenKernelWithImplicitArgsWhenSendi
pDevice->getMemoryManager()->freeGraphicsMemory(indirectHeapAllocation);
}
HWTEST_F(HardwareCommandsImplicitArgsTests, givenKernelWithImplicitArgsWhenSendingIndirectStateThenLocalIdsAreGeneratedAndCorrectlyProgrammedInCrossThreadData) {
HWTEST_F(HardwareCommandsImplicitArgsTests, givenKernelWithImplicitArgsAndRuntimeLocalIdsGenerationWhenSendingIndirectStateThenLocalIdsAreGeneratedAndCorrectlyProgrammedInCrossThreadData) {
DebugManagerStateRestore restorer;
DebugManager.flags.EnableHwGenerationLocalIds.set(0);
auto pKernelInfo = std::make_unique<MockKernelInfo>();
uint32_t simd = 32;
pKernelInfo->kernelDescriptor.kernelAttributes.simdSize = simd;
@ -1330,6 +1332,74 @@ HWTEST_F(HardwareCommandsImplicitArgsTests, givenKernelWithImplicitArgsWhenSendi
EXPECT_EQ(ioh.getGraphicsAllocation()->getGpuAddress(), pImplicitArgs->localIdTablePtr);
}
HWCMDTEST_F(IGFX_XE_HP_CORE, HardwareCommandsImplicitArgsTests, givenKernelWithImplicitArgsAndHwLocalIdsGenerationWhenSendingIndirectStateThenLocalIdsAreGeneratedAndCorrectlyProgrammedInCrossThreadData) {
auto pKernelInfo = std::make_unique<MockKernelInfo>();
uint32_t simd = 32;
pKernelInfo->kernelDescriptor.kernelAttributes.simdSize = simd;
pKernelInfo->kernelDescriptor.kernelAttributes.flags.requiresImplicitArgs = true;
pKernelInfo->kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[0] = 2;
pKernelInfo->kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[1] = 1;
pKernelInfo->kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[2] = 0;
MockContext context(pClDevice);
CommandQueueHw<FamilyType> cmdQ(&context, pClDevice, 0, false);
MockProgram program(&context, false, toClDeviceVector(*pClDevice));
MockKernel kernel(&program, *pKernelInfo, *pClDevice);
ASSERT_EQ(CL_SUCCESS, kernel.initialize());
const size_t localWorkSizeX = 2;
const size_t localWorkSizeY = 3;
const size_t localWorkSizeZ = 4;
const size_t localWorkSizes[3]{localWorkSizeX, localWorkSizeY, localWorkSizeZ};
std::array<uint8_t, 3> expectedDimOrder = {0, 2, 1};
auto &commandStream = cmdQ.getCS(1024);
auto pWalkerCmd = reinterpret_cast<typename FamilyType::WALKER_TYPE *>(commandStream.getSpace(0));
auto &dsh = cmdQ.getIndirectHeap(IndirectHeap::Type::DYNAMIC_STATE, 8192);
auto &ioh = cmdQ.getIndirectHeap(IndirectHeap::Type::INDIRECT_OBJECT, 8192);
auto &ssh = cmdQ.getIndirectHeap(IndirectHeap::Type::SURFACE_STATE, 8192);
dsh.align(EncodeStates<FamilyType>::alignInterfaceDescriptorData);
auto interfaceDescriptor = reinterpret_cast<typename FamilyType::INTERFACE_DESCRIPTOR_DATA *>(dsh.getSpace(0));
uint32_t interfaceDescriptorIndex = 0u;
HardwareCommandsHelper<FamilyType>::sendIndirectState(
commandStream,
dsh,
ioh,
ssh,
kernel,
0u,
simd,
localWorkSizes,
0u,
interfaceDescriptorIndex,
pDevice->getPreemptionMode(),
pWalkerCmd,
interfaceDescriptor,
false,
*pDevice);
constexpr uint32_t grfSize = sizeof(typename FamilyType::GRF);
size_t localWorkSize = localWorkSizeX * localWorkSizeY * localWorkSizeZ;
size_t expectedIohSize = PerThreadDataHelper::getPerThreadDataSizeTotal(simd, grfSize, 3u, localWorkSize);
ASSERT_LE(expectedIohSize, ioh.getUsed());
auto expectedLocalIds = alignedMalloc(expectedIohSize, 64);
generateLocalIDs(expectedLocalIds, simd,
std::array<uint16_t, 3>{{localWorkSizeX, localWorkSizeY, localWorkSizeZ}},
expectedDimOrder,
false, grfSize);
EXPECT_EQ(0, memcmp(expectedLocalIds, ioh.getCpuBase(), expectedIohSize));
alignedFree(expectedLocalIds);
auto pImplicitArgs = reinterpret_cast<ImplicitArgs *>(ptrOffset(ioh.getCpuBase(), alignUp(expectedIohSize, MemoryConstants::cacheLineSize)));
EXPECT_EQ(ioh.getGraphicsAllocation()->getGpuAddress(), pImplicitArgs->localIdTablePtr);
}
using HardwareCommandsTestXeHpAndLater = HardwareCommandsTest;
HWCMDTEST_F(IGFX_XE_HP_CORE, HardwareCommandsTestXeHpAndLater, givenIndirectHeapNotAllocatedFromInternalPoolWhenSendCrossThreadDataIsCalledThenOffsetZeroIsReturned) {