intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-06 10:26:29 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refactor KernelInfo

Browse Source 
Using kernelDescriptor's workgroupWalkOrder, workgroupDimensionsOrder
and requiresWorkgroupWalkOrder flag.

Related-To: NEO-3739
This commit is contained in:
Krystian Chmielewski
2020-11-04 10:34:42 +01:00
committed by Compute-Runtime-Automation
parent bfbc0e07ba
commit 2ef52ff39c
8 changed files with 45 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
opencl/test/unit_test/helpers/hardware_commands_helper_tests.cpp
View File

@@ -546,13 +546,13 @@ HWCMDTEST_F(IGFX_GEN8_CORE, HardwareCommandsTest, whenSendingIndirectStateThenKe
KernelInfo modifiedKernelInfo = {};
modifiedKernelInfo.patchInfo = kernel->getKernelInfo().patchInfo;
modifiedKernelInfo.workgroupWalkOrder[0] = 2;
modifiedKernelInfo.workgroupWalkOrder[1] = 1;
modifiedKernelInfo.workgroupWalkOrder[2] = 0;
modifiedKernelInfo.workgroupDimensionsOrder[0] = 2;
modifiedKernelInfo.workgroupDimensionsOrder[1] = 1;
modifiedKernelInfo.workgroupDimensionsOrder[2] = 0;
MockKernel mockKernel{kernel->getProgram(), modifiedKernelInfo, false};
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupWalkOrder[0] = 2;
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupWalkOrder[1] = 1;
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupWalkOrder[2] = 0;
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[0] = 2;
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[1] = 1;
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[2] = 0;
MockKernel mockKernel(kernel->getProgram(), modifiedKernelInfo, false);
uint32_t interfaceDescriptorIndex = 0;
auto isCcsUsed = EngineHelpers::isCcs(cmdQ.getGpgpuEngine().osContext->getEngineType());
auto kernelUsesLocalIds = HardwareCommandsHelper<FamilyType>::kernelUsesLocalIds(mockKernel);
@@ -582,7 +582,10 @@ HWCMDTEST_F(IGFX_GEN8_CORE, HardwareCommandsTest, whenSendingIndirectStateThenKe
uint32_t grfSize = sizeof(typename FamilyType::GRF);
generateLocalIDs(expectedLocalIds, modifiedKernelInfo.getMaxSimdSize(),
std::array<uint16_t, 3>{{localWorkSizeX, localWorkSizeY, localWorkSizeZ}},
std::array<uint8_t, 3>{{modifiedKernelInfo.workgroupDimensionsOrder[0], modifiedKernelInfo.workgroupDimensionsOrder[1], modifiedKernelInfo.workgroupDimensionsOrder[2]}}, false, grfSize);
std::array<uint8_t, 3>{{modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[0],
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[1],
modifiedKernelInfo.kernelDescriptor.kernelAttributes.workgroupDimensionsOrder[2]}},
false, grfSize);
EXPECT_EQ(0, memcmp(expectedLocalIds, ioh.getCpuBase(), expectedIohSize));
alignedFree(expectedLocalIds);
}

18
opencl/test/unit_test/helpers/per_thread_data_tests.cpp
View File

@@ -88,8 +88,8 @@ HWTEST_F(PerThreadDataXYZTests, Given256x1x1WhenSendingPerThreadDataThenCorrectA
MockGraphicsAllocation gfxAllocation(indirectHeapMemory, indirectHeapMemorySize);
LinearStream indirectHeap(&gfxAllocation);
size_t localWorkSizes[3] = {256, 1, 1};
auto localWorkSize = localWorkSizes[0] * localWorkSizes[1] * localWorkSizes[2];
const std::array<uint16_t, 3> localWorkSizes = {{256, 1, 1}};
size_t localWorkSize = localWorkSizes[0] * localWorkSizes[1] * localWorkSizes[2];
auto offsetPerThreadData = PerThreadDataHelper::sendPerThreadData(
indirectHeap,
@@ -109,7 +109,7 @@ HWTEST_F(PerThreadDataXYZTests, Given2x4x8WhenSendingPerThreadDataThenCorrectAmo
MockGraphicsAllocation gfxAllocation(indirectHeapMemory, indirectHeapMemorySize);
LinearStream indirectHeap(&gfxAllocation);
const size_t localWorkSizes[3]{2, 4, 8};
const std::array<uint16_t, 3> localWorkSizes = {{2, 4, 8}};
auto offsetPerThreadData = PerThreadDataHelper::sendPerThreadData(
indirectHeap,
simd,
@@ -163,7 +163,7 @@ HWTEST_F(PerThreadDataNoIdsTests, GivenThreadPaylodDataWithoutLocalIdsWhenSendin
MockGraphicsAllocation gfxAllocation(indirectHeapMemory, indirectHeapMemorySize);
LinearStream indirectHeap(&gfxAllocation);
size_t localWorkSizes[3] = {256, 1, 1};
const std::array<uint16_t, 3> localWorkSizes = {{256, 1, 1}};
auto offsetPerThreadData = PerThreadDataHelper::sendPerThreadData(
indirectHeap,
simd,
@@ -226,7 +226,8 @@ TEST(PerThreadDataTest, WhenSettingLocalIdsInPerThreadDataThenIdsAreSetInCorrect
uint32_t numChannels = 3;
uint32_t localWorkSize = 24;
size_t localWorkSizes[3] = {24, 1, 1};
const std::array<uint16_t, 3> localWorkSizes = {{24, 1, 1}};
const std::array<uint8_t, 3> workgroupWalkOrder = {{0, 1, 2}};
auto sizePerThreadDataTotal = PerThreadDataHelper::getPerThreadDataSizeTotal(simd, numChannels, localWorkSize, grfSize);
@@ -245,7 +246,7 @@ TEST(PerThreadDataTest, WhenSettingLocalIdsInPerThreadDataThenIdsAreSetInCorrect
grfSize,
numChannels,
localWorkSizes,
{{0, 1, 2}},
workgroupWalkOrder,
false);
// Check if buffer overrun happend, only first sizePerThreadDataTotal bytes can be overwriten, following should be same as reference.
@@ -264,7 +265,8 @@ TEST(PerThreadDataTest, givenSimdEqualOneWhenSettingLocalIdsInPerThreadDataThenI
uint32_t numChannels = 3;
uint32_t localWorkSize = 24;
size_t localWorkSizes[3] = {3, 4, 2};
const std::array<uint16_t, 3> localWorkSizes = {{3, 4, 2}};
const std::array<uint8_t, 3> workgroupWalkOrder = {{0, 1, 2}};
auto sizePerThreadDataTotal = PerThreadDataHelper::getPerThreadDataSizeTotal(simd, grfSize, numChannels, localWorkSize);
@@ -283,7 +285,7 @@ TEST(PerThreadDataTest, givenSimdEqualOneWhenSettingLocalIdsInPerThreadDataThenI
grfSize,
numChannels,
localWorkSizes,
{{0, 1, 2}},
workgroupWalkOrder,
false);
auto bufferPtr = buffer;