intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 14:55:24 +08:00
Code Issues Packages Projects Releases Wiki Activity

add support for packed simd1 dispatch

Browse Source 
Change-Id: I3f2bf8e62e0a38d358fb87f02c88c387c874f6b3
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This commit is contained in:
Maciej Plewka
2019-10-23 09:36:37 +02:00
committed by sys_ocldev
parent 2eafa99342
commit 51dcf2b6d2
19 changed files with 233 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
unit_tests/command_queue/dispatch_walker_tests.cpp
View File

@@ -124,6 +124,27 @@ HWTEST_F(DispatchWalkerTest, WhenGettingComputeDimensionsThenCorrectNumberOfDime
EXPECT_EQ(3u, computeDimensions(workItems3D));
}
HWTEST_F(DispatchWalkerTest, givenSimd1WhenSetGpgpuWalkerThreadDataThenSimdInWalkerIsSetTo32Value) {
uint32_t pCmdBuffer[1024];
MockGraphicsAllocation gfxAllocation((void *)pCmdBuffer, sizeof(pCmdBuffer));
LinearStream linearStream(&gfxAllocation);
using WALKER_TYPE = typename FamilyType::WALKER_TYPE;
WALKER_TYPE *computeWalker = static_cast<WALKER_TYPE *>(linearStream.getSpace(sizeof(WALKER_TYPE)));
*computeWalker = FamilyType::cmdInitGpgpuWalker;
size_t globalOffsets[] = {0, 0, 0};
size_t startWorkGroups[] = {0, 0, 0};
size_t numWorkGroups[] = {1, 1, 1};
size_t localWorkSizesIn[] = {32, 1, 1};
uint32_t simd = 1;
iOpenCL::SPatchThreadPayload threadPayload;
GpgpuWalkerHelper<FamilyType>::setGpgpuWalkerThreadData(
computeWalker, globalOffsets, startWorkGroups, numWorkGroups, localWorkSizesIn, simd, 3, true, false, threadPayload, 5u);
EXPECT_EQ(computeWalker->getSimdSize(), 32 >> 4);
}
HWTEST_F(DispatchWalkerTest, WhenDispatchingWalkerThenCommandStreamMemoryIsntChanged) {
MockKernel kernel(program.get(), kernelInfo, *pDevice);
ASSERT_EQ(CL_SUCCESS, kernel.initialize());

6
unit_tests/command_queue/local_id_tests.cpp
View File

@@ -59,6 +59,12 @@ TEST(LocalID, PerThreadSizeLocalIDs_SIMD32) {
EXPECT_EQ(6 * sizeof(GRF), getPerThreadSizeLocalIDs(simd));
}
TEST(LocalID, PerThreadSizeLocalIDs_SIMD1) {
uint32_t simd = 1;
EXPECT_EQ(sizeof(GRF), getPerThreadSizeLocalIDs(simd));
}
struct LocalIDFixture : public ::testing::TestWithParam<std::tuple<int, int, int, int>> {
void SetUp() override {
simd = std::get<0>(GetParam());

47
unit_tests/helpers/per_thread_data_tests.cpp
View File

@@ -252,3 +252,50 @@ TEST(PerThreadDataTest, generateLocalIDs) {
alignedFree(buffer);
alignedFree(reference);
}
TEST(PerThreadDataTest, givenSimdEqualOneWhenSetingLocalIdsInPerThreadDataThenIdsAreSetInCorrectOrder) {
uint32_t simd = 1;
uint32_t numChannels = 3;
uint32_t localWorkSize = 24;
size_t localWorkSizes[3] = {3, 4, 2};
auto sizePerThreadDataTotal = PerThreadDataHelper::getPerThreadDataSizeTotal(simd, numChannels, localWorkSize);
auto sizeOverSizedBuffer = sizePerThreadDataTotal * 4;
auto buffer = static_cast<char *>(alignedMalloc(sizeOverSizedBuffer, 16));
memset(buffer, 0, sizeOverSizedBuffer);
// Setup reference filled with zeros
auto reference = static_cast<char *>(alignedMalloc(sizePerThreadDataTotal, 16));
memset(reference, 0, sizePerThreadDataTotal);
LinearStream stream(buffer, sizeOverSizedBuffer / 2);
PerThreadDataHelper::sendPerThreadData(
stream,
simd,
numChannels,
localWorkSizes,
{{0, 1, 2}},
false);
auto bufferPtr = buffer;
for (uint16_t i = 0; i < localWorkSizes[2]; i++) {
for (uint16_t j = 0; j < localWorkSizes[1]; j++) {
for (uint16_t k = 0; k < localWorkSizes[0]; k++) {
uint16_t ids[] = {k, j, i};
int result = memcmp(bufferPtr, ids, sizeof(uint16_t) * 3);
EXPECT_EQ(0, result);
bufferPtr += sizeof(GRF);
}
}
}
// Check if buffer overrun happend, only first sizePerThreadDataTotal bytes can be overwriten, following should be same as reference.
for (auto i = sizePerThreadDataTotal; i < sizeOverSizedBuffer; i += sizePerThreadDataTotal) {
int result = memcmp(buffer + i, reference, sizePerThreadDataTotal);
EXPECT_EQ(0, result);
}
alignedFree(buffer);
alignedFree(reference);
}

10
unit_tests/kernel/kernel_tests.cpp
View File

@@ -2163,6 +2163,16 @@ TEST_F(KernelExecutionEnvironmentTest, getMaxSimdReturns1WhenExecutionEnvironmen
this->pKernelInfo->patchInfo.executionEnvironment = oldExcEnv;
}
TEST_F(KernelExecutionEnvironmentTest, getMaxSimdReturns1WhenLargestCompilledSimdSizeEqualOne) {
executionEnvironment.LargestCompiledSIMDSize = 1;
auto oldExcEnv = this->pKernelInfo->patchInfo.executionEnvironment;
EXPECT_EQ(1U, this->pKernelInfo->getMaxSimdSize());
this->pKernelInfo->patchInfo.executionEnvironment = oldExcEnv;
}
TEST_F(KernelExecutionEnvironmentTest, getMaxRequiredWorkGroupSizeWhenCompiledWorkGroupSizeIsZero) {
auto maxWorkGroupSize = pDevice->getDeviceInfo().maxWorkGroupSize;
auto oldRequiredWorkGroupSizeX = this->pKernelInfo->patchInfo.executionEnvironment->RequiredWorkGroupSizeX;