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

refactor: pass additional enqueue params

Browse Source 
Related-To: NEO-12591

Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
This commit is contained in:
Bartosz Dunajski
2024-10-21 15:11:18 +00:00
committed by Compute-Runtime-Automation
parent 18d828421d
commit 4d6265a46c
1 changed files with 14 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
opencl/source/command_queue/hardware_interface_xehp_and_later.inl
View File

@@ -60,7 +60,9 @@ inline void HardwareInterface<GfxFamily>::programWalker(
uint32_t dim = dispatchInfo.getDim(); uint32_t dim = dispatchInfo.getDim();
uint32_t simd = kernelInfo.getMaxSimdSize(); uint32_t simd = kernelInfo.getMaxSimdSize();
auto numChannels = kernelInfo.kernelDescriptor.kernelAttributes.numLocalIdChannels; const auto &kernelAttributes = kernelInfo.kernelDescriptor.kernelAttributes;
auto numChannels = kernelAttributes.numLocalIdChannels;
size_t startWorkGroups[3] = {walkerArgs.startOfWorkgroups->x, walkerArgs.startOfWorkgroups->y, walkerArgs.startOfWorkgroups->z}; size_t startWorkGroups[3] = {walkerArgs.startOfWorkgroups->x, walkerArgs.startOfWorkgroups->y, walkerArgs.startOfWorkgroups->z};
size_t numWorkGroups[3] = {walkerArgs.numberOfWorkgroups->x, walkerArgs.numberOfWorkgroups->y, walkerArgs.numberOfWorkgroups->z}; size_t numWorkGroups[3] = {walkerArgs.numberOfWorkgroups->x, walkerArgs.numberOfWorkgroups->y, walkerArgs.numberOfWorkgroups->z};
@@ -71,10 +73,10 @@ inline void HardwareInterface<GfxFamily>::programWalker(
bool localIdsGenerationByRuntime = kernelUsesLocalIds && EncodeDispatchKernel<GfxFamily>::isRuntimeLocalIdsGenerationRequired( bool localIdsGenerationByRuntime = kernelUsesLocalIds && EncodeDispatchKernel<GfxFamily>::isRuntimeLocalIdsGenerationRequired(
numChannels, numChannels,
walkerArgs.localWorkSizes, walkerArgs.localWorkSizes,
std::array<uint8_t, 3>{{kernelInfo.kernelDescriptor.kernelAttributes.workgroupWalkOrder[0], std::array<uint8_t, 3>{{kernelAttributes.workgroupWalkOrder[0],
kernelInfo.kernelDescriptor.kernelAttributes.workgroupWalkOrder[1], kernelAttributes.workgroupWalkOrder[1],
kernelInfo.kernelDescriptor.kernelAttributes.workgroupWalkOrder[2]}}, kernelAttributes.workgroupWalkOrder[2]}},
kernelInfo.kernelDescriptor.kernelAttributes.flags.requiresWorkgroupWalkOrder, kernelAttributes.flags.requiresWorkgroupWalkOrder,
requiredWalkOrder, requiredWalkOrder,
simd); simd);
@@ -142,7 +144,9 @@ inline void HardwareInterface<GfxFamily>::programWalker(
} }
bool requiredSystemFence = kernelSystemAllocation && walkerArgs.event != nullptr; bool requiredSystemFence = kernelSystemAllocation && walkerArgs.event != nullptr;
auto maxFrontEndThreads = device.getDeviceInfo().maxFrontEndThreads; auto maxFrontEndThreads = device.getDeviceInfo().maxFrontEndThreads;
EncodeWalkerArgs encodeWalkerArgs{kernel.getExecutionType(), requiredSystemFence, kernelInfo.kernelDescriptor, NEO::RequiredDispatchWalkOrder::none, 0, maxFrontEndThreads};
EncodeWalkerArgs encodeWalkerArgs{kernel.getExecutionType(), requiredSystemFence, kernelInfo.kernelDescriptor, kernelAttributes.walkOrder, kernelAttributes.additionalSize, maxFrontEndThreads};
EncodeDispatchKernel<GfxFamily>::template encodeAdditionalWalkerFields<WalkerType>(rootDeviceEnvironment, walkerCmd, encodeWalkerArgs); EncodeDispatchKernel<GfxFamily>::template encodeAdditionalWalkerFields<WalkerType>(rootDeviceEnvironment, walkerCmd, encodeWalkerArgs);
auto devices = queueCsr.getOsContext().getDeviceBitfield(); auto devices = queueCsr.getOsContext().getDeviceBitfield();
@@ -162,6 +166,10 @@ inline void HardwareInterface<GfxFamily>::programWalker(
uint32_t partitionCount = 0u; uint32_t partitionCount = 0u;
RequiredPartitionDim requiredPartitionDim = kernel.usesImages() ? RequiredPartitionDim::x : RequiredPartitionDim::none; RequiredPartitionDim requiredPartitionDim = kernel.usesImages() ? RequiredPartitionDim::x : RequiredPartitionDim::none;
if (kernelAttributes.partitionDim != NEO::RequiredPartitionDim::none) {
requiredPartitionDim = kernelAttributes.partitionDim;
}
ImplicitScalingDispatchCommandArgs implicitScalingArgs{ ImplicitScalingDispatchCommandArgs implicitScalingArgs{
workPartitionAllocationGpuVa, // workPartitionAllocationGpuVa workPartitionAllocationGpuVa, // workPartitionAllocationGpuVa
&hwInfo, // hwInfo &hwInfo, // hwInfo