intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-25 13:33:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add adjustNumberOfThreadsInThreadGroup method to EncodeDispatchKernel

Browse Source 
Related-To: NEO-7357

Signed-off-by: Rafal Maziejuk <rafal.maziejuk@intel.com>
This commit is contained in:
Rafal Maziejuk
2023-01-24 15:58:16 +00:00
committed by Compute-Runtime-Automation
parent 73bd55ad40
commit b7380237c2
8 changed files with 30 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
opencl/source/helpers/hardware_commands_helper.h
View File

@@ -48,6 +48,7 @@ struct HardwareCommandsHelper : public PerThreadDataHelper {
uint32_t numSamplers,
const uint32_t threadGroupCount,
uint32_t numThreadsPerThreadGroup,
uint32_t simd,
const Kernel &kernel,
uint32_t bindingTablePrefetchSize,
PreemptionMode preemptionMode,

3
opencl/source/helpers/hardware_commands_helper_base.inl
View File

@@ -118,6 +118,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(
uint32_t numSamplers,
const uint32_t threadGroupCount,
uint32_t threadsPerThreadGroup,
uint32_t simd,
const Kernel &kernel,
uint32_t bindingTablePrefetchSize,
PreemptionMode preemptionMode,
@@ -149,6 +150,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(
productHelper.updateIddCommand(&interfaceDescriptor, kernelDescriptor.kernelAttributes.numGrfRequired,
kernelDescriptor.kernelAttributes.threadArbitrationPolicy);
EncodeDispatchKernel<GfxFamily>::adjustNumberOfThreadsInThreadGroup(interfaceDescriptor, simd, threadsPerThreadGroup);
EncodeDispatchKernel<GfxFamily>::appendAdditionalIDDFields(&interfaceDescriptor, device.getRootDeviceEnvironment(), threadsPerThreadGroup,
slmTotalSize, SlmPolicy::SlmPolicyNone);
@@ -285,6 +287,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendIndirectState(
samplerCount,
threadGroupCount,
threadsPerThreadGroup,
simd,
kernel,
bindingTablePrefetchSize,
preemptionMode,

2
opencl/test/unit_test/helpers/hardware_commands_helper_tests.cpp
View File

@@ -94,7 +94,7 @@ HWCMDTEST_F(IGFX_GEN8_CORE, HardwareCommandsTest, WhenProgramInterfaceDescriptor
const uint32_t threadGroupCount = 1u;
size_t crossThreadDataSize = kernel->getCrossThreadDataSize();
HardwareCommandsHelper<FamilyType>::sendInterfaceDescriptorData(
indirectHeap, 0, 0, crossThreadDataSize, 64, 0, 0, 0, threadGroupCount, 1, *kernel, 0, pDevice->getPreemptionMode(), nullptr, *pDevice);
indirectHeap, 0, 0, crossThreadDataSize, 64, 0, 0, 0, threadGroupCount, 1, 1, *kernel, 0, pDevice->getPreemptionMode(), nullptr, *pDevice);
auto usedIndirectHeapAfter = indirectHeap.getUsed();
EXPECT_EQ(sizeof(INTERFACE_DESCRIPTOR_DATA), usedIndirectHeapAfter - usedIndirectHeapBefore);

2
opencl/test/unit_test/kernel/kernel_slm_tests.cpp
View File

@@ -83,6 +83,7 @@ HWCMDTEST_P(IGFX_GEN8_CORE, KernelSLMAndBarrierTest, GivenStaticSlmSizeWhenProgr
0,
threadGroupCount,
1,
1,
kernel,
4u,
pDevice->getPreemptionMode(),
@@ -173,6 +174,7 @@ HWTEST_F(KernelSLMAndBarrierTest, GivenInterfaceDescriptorProgrammedWhenOverride
0,
threadGroupCount,
1,
1,
kernel,
4u,
pDevice->getPreemptionMode(),