intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-04 23:56:39 +08:00
Code Issues Packages Projects Releases Wiki Activity

Include dynamic SLM in clGetKernelWorkGroupInfo

Browse Source 
Current implementation only takes static slmInlineSize into account.
With this change we also include dynamic SLM passed as a kernel arguments.

Related-To: NEO-5761
Signed-off-by: Fabian Zwolinski <fabian.zwolinski@intel.com>
This commit is contained in:
Fabian Zwolinski
2022-09-26 19:01:33 +00:00
committed by Compute-Runtime-Automation
parent 67b670c5b9
commit ccdb5aaa2a
3 changed files with 53 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
opencl/source/kernel/kernel.cpp
View File

@@ -75,7 +75,8 @@ Kernel::Kernel(Program *programArg, const KernelInfo &kernelInfoArg, ClDevice &c
} else { } else {
maxKernelWorkGroupSize = static_cast<uint32_t>(deviceInfo.maxWorkGroupSize); maxKernelWorkGroupSize = static_cast<uint32_t>(deviceInfo.maxWorkGroupSize);
} }
slmTotalSize = kernelInfoArg.kernelDescriptor.kernelAttributes.slmInlineSize;
slmTotalSize = slmTotalSum = kernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize;
} }
Kernel::~Kernel() { Kernel::~Kernel() {
@@ -527,7 +528,7 @@ cl_int Kernel::getWorkGroupInfo(cl_kernel_work_group_info paramName,
struct size_t3 { struct size_t3 {
size_t val[3]; size_t val[3];
} requiredWorkGroupSize; } requiredWorkGroupSize;
cl_ulong localMemorySize; size_t totalLocalMemorySize = static_cast<size_t>(slmTotalSum);
const auto &kernelDescriptor = kernelInfo.kernelDescriptor; const auto &kernelDescriptor = kernelInfo.kernelDescriptor;
size_t preferredWorkGroupSizeMultiple = 0; size_t preferredWorkGroupSizeMultiple = 0;
cl_ulong scratchSize; cl_ulong scratchSize;
@@ -558,9 +559,8 @@ cl_int Kernel::getWorkGroupInfo(cl_kernel_work_group_info paramName,
break; break;
case CL_KERNEL_LOCAL_MEM_SIZE: case CL_KERNEL_LOCAL_MEM_SIZE:
localMemorySize = kernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize; srcSize = sizeof(totalLocalMemorySize);
srcSize = sizeof(localMemorySize); pSrc = &totalLocalMemorySize;
pSrc = &localMemorySize;
break; break;
case CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE: case CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE:
@@ -1376,6 +1376,14 @@ cl_int Kernel::setArgLocal(uint32_t argIndexIn,
slmTotalSize = kernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize + alignUp(slmOffset, KB); slmTotalSize = kernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize + alignUp(slmOffset, KB);
uint32_t slmSum = 0;
for (const auto &kernelArg : kernelArguments) {
if (kernelArg.type == SLM_OBJ) {
slmSum += static_cast<uint32_t>(kernelArg.size);
}
}
slmTotalSum = kernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize + alignUp(slmSum, KB);
return CL_SUCCESS; return CL_SUCCESS;
} }

1
opencl/source/kernel/kernel.h
View File

@@ -519,6 +519,7 @@ class Kernel : public ReferenceTrackedObject<Kernel> {
uint32_t additionalKernelExecInfo = AdditionalKernelExecInfo::DisableOverdispatch; uint32_t additionalKernelExecInfo = AdditionalKernelExecInfo::DisableOverdispatch;
uint32_t maxKernelWorkGroupSize = 0; uint32_t maxKernelWorkGroupSize = 0;
uint32_t slmTotalSize = 0u; uint32_t slmTotalSize = 0u;
uint32_t slmTotalSum = 0u;
uint32_t sshLocalSize = 0u; uint32_t sshLocalSize = 0u;
uint32_t crossThreadDataSize = 0u; uint32_t crossThreadDataSize = 0u;

41
opencl/test/unit_test/kernel/kernel_tests.cpp
View File

@@ -312,7 +312,7 @@ TEST_F(KernelTests, GivenKernelCompileWorkGroupSizeWhenGettingWorkGroupInfoThenC
EXPECT_EQ(paramValueSize, paramValueSizeRet); EXPECT_EQ(paramValueSize, paramValueSizeRet);
} }
TEST_F(KernelTests, GivenRequiredDisabledEUFusionFlagWhenGettingPrefferedWorkGroupSizeMultipleThenCorectValueIsReturned) { TEST_F(KernelTests, GivenRequiredDisabledEUFusionFlagWhenGettingPrefferedWorkGroupSizeMultipleThenCorrectValueIsReturned) {
KernelInfo kernelInfo = {}; KernelInfo kernelInfo = {};
kernelInfo.kernelDescriptor.kernelAttributes.flags.requiresDisabledEUFusion = true; kernelInfo.kernelDescriptor.kernelAttributes.flags.requiresDisabledEUFusion = true;
MockKernel kernel(pProgram, kernelInfo, *pClDevice); MockKernel kernel(pProgram, kernelInfo, *pClDevice);
@@ -337,7 +337,44 @@ TEST_F(KernelTests, GivenRequiredDisabledEUFusionFlagWhenGettingPrefferedWorkGro
EXPECT_EQ(expectedValue, paramValue); EXPECT_EQ(expectedValue, paramValue);
} }
TEST_F(KernelTests, GivenCFEFusedEUDispatchEnabledAndRequiredDisabledUEFusionWhenGettingPrefferedWorkGroupSizeMultipleThenCorectValueIsReturned) { TEST_F(KernelTests, GivenSlmInlineSizeAndSlmOffsetWhenGettingWorkGroupInfoThenCorrectValueIsReturned) {
MockKernelInfo kernelInfo = {};
kernelInfo.kernelDescriptor.kernelAttributes.slmInlineSize = 100u;
kernelInfo.addArgLocal(0, 0x10, 0x1);
kernelInfo.addArgBuffer(1, 0x20, sizeof(void *));
kernelInfo.addArgBuffer(2, 0x20, sizeof(void *));
kernelInfo.addArgLocal(3, 0x30, 0x10);
MockKernel kernel(pProgram, kernelInfo, *pClDevice);
kernel.kernelArguments.resize(4);
kernel.slmSizes.resize(4);
uint32_t crossThreadData[0x40]{};
crossThreadData[0x20 / sizeof(uint32_t)] = 0x12344321;
kernel.setCrossThreadData(crossThreadData, sizeof(crossThreadData));
kernel.setArgLocal(0, 4096, nullptr);
kernel.setArgLocal(3, 0, nullptr);
cl_kernel_info paramName = CL_KERNEL_LOCAL_MEM_SIZE;
size_t paramValue;
size_t paramValueSize = sizeof(paramValue);
size_t paramValueSizeRet = 0;
size_t expectedValue = 4096 + 0 + 100;
retVal = kernel.getWorkGroupInfo(
paramName,
paramValueSize,
&paramValue,
&paramValueSizeRet);
EXPECT_EQ(CL_SUCCESS, retVal);
EXPECT_EQ(paramValueSize, paramValueSizeRet);
EXPECT_EQ(expectedValue, paramValue);
}
TEST_F(KernelTests, GivenCFEFusedEUDispatchEnabledAndRequiredDisabledUEFusionWhenGettingPrefferedWorkGroupSizeMultipleThenCorrectValueIsReturned) {
DebugManagerStateRestore dbgRestorer; DebugManagerStateRestore dbgRestorer;
DebugManager.flags.CFEFusedEUDispatch.set(0); DebugManager.flags.CFEFusedEUDispatch.set(0);