intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: avoid reading from gfx memory when modifying walker command

Browse Source 
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2024-09-11 15:25:33 +00:00
committed by Compute-Runtime-Automation
parent 487b02a2ac
commit 8bb92ff445
8 changed files with 68 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
opencl/source/command_queue/hardware_interface_xehp_and_later.inl
View File

@@ -162,25 +162,25 @@ inline void HardwareInterface<GfxFamily>::programWalker(
uint32_t partitionCount = 0u;
RequiredPartitionDim requiredPartitionDim = kernel.usesImages() ? RequiredPartitionDim::x : RequiredPartitionDim::none;
void *outWalker = nullptr;
ImplicitScalingDispatchCommandArgs implicitScalingArgs{
workPartitionAllocationGpuVa, // workPartitionAllocationGpuVa
&hwInfo, // hwInfo
&outWalker, // outWalkerPtr
nullptr, // outWalkerPtr
requiredPartitionDim, // requiredPartitionDim
partitionCount, // partitionCount
workgroupSize, // workgroupSize
maxWgCountPerTile, // maxWgCountPerTile
false, // useSecondaryBatchBuffer
false, // apiSelfCleanup
queueCsr.getDcFlushSupport(), // dcFlush
kernel.isSingleSubdevicePreferred(), // forceExecutionOnSingleTile
false}; // blockDispatchToCommandBuffer
false, // blockDispatchToCommandBuffer
requiredWalkOrder != 0}; // isRequiredWorkGroupOrder
ImplicitScalingDispatch<GfxFamily>::template dispatchCommands<WalkerType>(commandStream,
walkerCmd,
devices,
implicitScalingArgs);
EncodeDispatchKernel<GfxFamily>::setWalkerRegionSettings(*static_cast<WalkerType *>(outWalker), hwInfo, implicitScalingArgs.partitionCount, workgroupSize, maxWgCountPerTile, requiredWalkOrder != 0);
if (queueCsr.isStaticWorkPartitioningEnabled()) {
queueCsr.setActivePartitions(std::max(queueCsr.getActivePartitions(), implicitScalingArgs.partitionCount));