intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-12 17:33:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

Apply heuristics when setting TG dispatch size on XE_HPC_CORE

Browse Source 
The default TG dispatch size can be changed
to a better value based on number of threads in TG or
currently available amount of threads on GPU.
Decision on what TG dispatch size should be are based on
implemented heuristics.

Signed-off-by: Rafal Maziejuk <rafal.maziejuk@intel.com>
Related-To: NEO-6989
This commit is contained in:
Rafal Maziejuk
2022-08-03 12:22:30 +00:00
committed by Compute-Runtime-Automation
parent 52133e61ce
commit ed0c36117e
21 changed files with 188 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
opencl/source/helpers/hardware_commands_helper.h
View File

@@ -47,6 +47,7 @@ struct HardwareCommandsHelper : public PerThreadDataHelper {
size_t bindingTablePointer,
[[maybe_unused]] size_t offsetSamplerState,
uint32_t numSamplers,
const uint32_t threadGroupCount,
uint32_t numThreadsPerThreadGroup,
const Kernel &kernel,
uint32_t bindingTablePrefetchSize,
@@ -79,6 +80,7 @@ struct HardwareCommandsHelper : public PerThreadDataHelper {
uint64_t kernelStartOffset,
uint32_t simd,
const size_t localWorkSize[3],
const uint32_t threadGroupCount,
const uint64_t offsetInterfaceDescriptorTable,
uint32_t &interfaceDescriptorIndex,
PreemptionMode preemptionMode,

6
opencl/source/helpers/hardware_commands_helper_base.inl
View File

@@ -113,6 +113,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(
size_t bindingTablePointer,
[[maybe_unused]] size_t offsetSamplerState,
uint32_t numSamplers,
const uint32_t threadGroupCount,
uint32_t threadsPerThreadGroup,
const Kernel &kernel,
uint32_t bindingTablePrefetchSize,
@@ -169,7 +170,8 @@ size_t HardwareCommandsHelper<GfxFamily>::sendInterfaceDescriptorData(
hardwareInfo);
PreemptionHelper::programInterfaceDescriptorDataPreemption<GfxFamily>(&interfaceDescriptor, preemptionMode);
EncodeDispatchKernel<GfxFamily>::adjustInterfaceDescriptorData(interfaceDescriptor, hardwareInfo);
EncodeDispatchKernel<GfxFamily>::adjustInterfaceDescriptorData(interfaceDescriptor, hardwareInfo, threadGroupCount, kernelDescriptor.kernelAttributes.numGrfRequired);
*pInterfaceDescriptor = interfaceDescriptor;
return (size_t)offsetInterfaceDescriptor;
@@ -185,6 +187,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendIndirectState(
uint64_t kernelStartOffset,
uint32_t simd,
const size_t localWorkSize[3],
const uint32_t threadGroupCount,
const uint64_t offsetInterfaceDescriptorTable,
uint32_t &interfaceDescriptorIndex,
PreemptionMode preemptionMode,
@@ -263,6 +266,7 @@ size_t HardwareCommandsHelper<GfxFamily>::sendIndirectState(
dstBindingTablePointer,
samplerStateOffset,
samplerCount,
threadGroupCount,
threadsPerThreadGroup,
kernel,
bindingTablePrefetchSize,