Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-10 12:53:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

Print warning when kernel uses too much SLM

Browse Source 
Instead of just returning proper error code in case of exceeding
available Shared Local Memory size we also want to print error message
to make debugging easier.

Related-To: NEO-7280
Signed-off-by: Fabian Zwolinski <fabian.zwolinski@intel.com>
This commit is contained in:
Fabian Zwolinski
2022-10-07 10:44:43 +00:00
committed by Compute-Runtime-Automation
parent d4cddc7ecd
commit 7953d15826
15 changed files with 101 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
opencl/test/unit_test/api/cl_enqueue_nd_range_kernel_tests.inl
View File

@ -39,7 +39,10 @@ TEST_F(clEnqueueNDRangeKernelTests, GivenValidParametersWhenExecutingKernelThenS
EXPECT_EQ(CL_SUCCESS, retVal);
}
TEST_F(clEnqueueNDRangeKernelTests, GivenKernelWithSlmSizeExceedingLocalMemorySizeWhenExecutingKernelThenOutOfResourcesIsReturned) {
TEST_F(clEnqueueNDRangeKernelTests, GivenKernelWithSlmSizeExceedingLocalMemorySizeWhenExecutingKernelThenDebugMsgErrIsPrintedAndOutOfResourcesIsReturned) {
DebugManagerStateRestore dbgRestorer;
DebugManager.flags.PrintDebugMessages.set(true);
cl_uint workDim = 1;
size_t globalWorkOffset[3] = {0, 0, 0};
size_t globalWorkSize[3] = {1, 1, 1};
@ -48,6 +51,8 @@ TEST_F(clEnqueueNDRangeKernelTests, GivenKernelWithSlmSizeExceedingLocalMemorySi
cl_event *eventWaitList = nullptr;
cl_event *event = nullptr;
::testing::internal::CaptureStderr();
auto localMemSize = static_cast<uint32_t>(pDevice->getDevice().getDeviceInfo().localMemSize);
pProgram->mockKernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize = localMemSize - 10u;
@ -64,6 +69,11 @@ TEST_F(clEnqueueNDRangeKernelTests, GivenKernelWithSlmSizeExceedingLocalMemorySi
EXPECT_EQ(CL_SUCCESS, retVal);
std::string output = testing::internal::GetCapturedStderr();
EXPECT_EQ(std::string(""), output);
::testing::internal::CaptureStderr();
pProgram->mockKernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize = localMemSize + 10u;
retVal = clEnqueueNDRangeKernel(
pCommandQueue,
@ -77,6 +87,11 @@ TEST_F(clEnqueueNDRangeKernelTests, GivenKernelWithSlmSizeExceedingLocalMemorySi
event);
EXPECT_EQ(CL_OUT_OF_RESOURCES, retVal);
output = testing::internal::GetCapturedStderr();
const auto &slmInlineSize = pProgram->mockKernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize;
std::string expectedOutput = "Size of SLM (" + std::to_string(slmInlineSize) + ") larger than available (" + std::to_string(localMemSize) + ")\n";
EXPECT_EQ(expectedOutput, output);
}
TEST_F(clEnqueueNDRangeKernelTests, GivenQueueIncapableWhenExecutingKernelThenInvalidOperationIsReturned) {

17
opencl/test/unit_test/kernel/kernel_tests.cpp
View File

@ -3005,10 +3005,15 @@ TEST_F(KernelCreateTest, whenInitFailedThenReturnNull) {
EXPECT_EQ(nullptr, ret);
}
TEST_F(KernelCreateTest, whenSlmSizeExceedsLocalMemorySizeThenReturnOutOfResources) {
TEST_F(KernelCreateTest, whenSlmSizeExceedsLocalMemorySizeThenDebugMsgErrIsPrintedAndOutOfResourcesIsReturned) {
DebugManagerStateRestore dbgRestorer;
DebugManager.flags.PrintDebugMessages.set(true);
KernelInfo info{};
cl_int retVal{};
::testing::internal::CaptureStderr();
auto localMemSize = static_cast<uint32_t>(mockProgram.mDevice.getDevice().getDeviceInfo().localMemSize);
info.kernelDescriptor.kernelAttributes.slmInlineSize = localMemSize - 10u;
@ -3016,6 +3021,11 @@ TEST_F(KernelCreateTest, whenSlmSizeExceedsLocalMemorySizeThenReturnOutOfResourc
EXPECT_EQ(nullptr, ret);
EXPECT_NE(CL_OUT_OF_RESOURCES, retVal);
std::string output = testing::internal::GetCapturedStderr();
EXPECT_EQ(std::string(""), output);
::testing::internal::CaptureStderr();
retVal = 0;
info.kernelDescriptor.kernelAttributes.slmInlineSize = localMemSize + 10u;
@ -3023,6 +3033,11 @@ TEST_F(KernelCreateTest, whenSlmSizeExceedsLocalMemorySizeThenReturnOutOfResourc
EXPECT_EQ(nullptr, ret);
EXPECT_EQ(CL_OUT_OF_RESOURCES, retVal);
output = testing::internal::GetCapturedStderr();
const auto &slmInlineSize = info.kernelDescriptor.kernelAttributes.slmInlineSize;
std::string expectedOutput = "Size of SLM (" + std::to_string(slmInlineSize) + ") larger than available (" + std::to_string(localMemSize) + ")\n";
EXPECT_EQ(expectedOutput, output);
ret = Kernel::create<MockKernel>(&mockProgram, info, mockProgram.mDevice, nullptr);
EXPECT_EQ(nullptr, ret);
}

14
opencl/test/unit_test/program/program_tests.cpp
View File

@ -2063,16 +2063,28 @@ TEST_F(ProgramTests, whenCreatingFromZebinThenAppendAllowZebinFlagToBuildOptions
EXPECT_STREQ(expectedOptions.c_str(), program->options.c_str());
}
TEST_F(ProgramTests, givenProgramFromGenBinaryWhenSLMSizeIsBiggerThenDeviceLimitThenReturnError) {
TEST_F(ProgramTests, givenProgramFromGenBinaryWhenSLMSizeIsBiggerThenDeviceLimitThenPrintDebugMsgAndReturnError) {
DebugManagerStateRestore dbgRestorer;
DebugManager.flags.PrintDebugMessages.set(true);
PatchTokensTestData::ValidProgramWithKernelUsingSlm patchtokensProgram;
patchtokensProgram.slmMutable->TotalInlineLocalMemorySize = static_cast<uint32_t>(pDevice->getDeviceInfo().localMemSize * 2);
patchtokensProgram.recalcTokPtr();
auto program = std::make_unique<MockProgram>(nullptr, false, toClDeviceVector(*pClDevice));
program->buildInfos[rootDeviceIndex].unpackedDeviceBinary = makeCopy(patchtokensProgram.storage.data(), patchtokensProgram.storage.size());
program->buildInfos[rootDeviceIndex].unpackedDeviceBinarySize = patchtokensProgram.storage.size();
::testing::internal::CaptureStderr();
auto retVal = program->processGenBinary(*pClDevice);
EXPECT_EQ(CL_OUT_OF_RESOURCES, retVal);
std::string output = testing::internal::GetCapturedStderr();
const auto &slmInlineSize = patchtokensProgram.slmMutable->TotalInlineLocalMemorySize;
const auto &localMemSize = pDevice->getDeviceInfo().localMemSize;
std::string expectedOutput = "Size of SLM (" + std::to_string(slmInlineSize) + ") larger than available (" + std::to_string(localMemSize) + ")\n";
EXPECT_EQ(expectedOutput, output);
}
TEST_F(ProgramTests, givenExistingConstantSurfacesWhenProcessGenBinaryThenCleanupTheSurfaceOnlyForSpecificDevice) {