intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-30 01:35:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: set walker params prior to adjusting interface descriptor

Browse Source 
Signed-off-by: Mrozek, Michal <michal.mrozek@intel.com>
This commit is contained in:
Mrozek, Michal
2023-10-04 15:55:40 +00:00
committed by Compute-Runtime-Automation
parent ad2701ad26
commit 6f58ec9d2c
3 changed files with 36 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -77,6 +77,11 @@ inline void HardwareInterface<GfxFamily>::programWalker(
auto isCcsUsed = EngineHelpers::isCcs(commandQueue.getGpgpuEngine().osContext->getEngineType());
auto kernelUsesLocalIds = HardwareCommandsHelper<GfxFamily>::kernelUsesLocalIds(kernel);
GpgpuWalkerHelper<GfxFamily>::setGpgpuWalkerThreadData(&walkerCmd, kernel.getKernelInfo().kernelDescriptor,
globalOffsets, startWorkGroups,
numWorkGroups, walkerArgs.localWorkSizes, simd, dim,
false, false, 0u);
HardwareCommandsHelper<GfxFamily>::sendIndirectState(
commandStream,
dsh,
@@ -95,11 +100,6 @@ inline void HardwareInterface<GfxFamily>::programWalker(
kernelUsesLocalIds,
commandQueue.getDevice());
GpgpuWalkerHelper<GfxFamily>::setGpgpuWalkerThreadData(&walkerCmd, kernel.getKernelInfo().kernelDescriptor,
globalOffsets, startWorkGroups,
numWorkGroups, walkerArgs.localWorkSizes, simd, dim,
false, false, 0u);
EncodeWalkerArgs encodeWalkerArgs{kernel.getExecutionType(), false, kernel.getKernelInfo().kernelDescriptor};
EncodeDispatchKernel<GfxFamily>::encodeAdditionalWalkerFields(rootDeviceEnvironment, walkerCmd, encodeWalkerArgs);
*walkerCmdBuf = walkerCmd;

7
opencl/source/command_queue/hardware_interface_xehp_and_later.inl
View File

@@ -96,6 +96,10 @@ inline void HardwareInterface<GfxFamily>::programWalker(
EncodeMemoryPrefetch<GfxFamily>::programMemoryPrefetch(commandStream, *kernelAllocation, kernelInfo.heapInfo.kernelHeapSize, 0, rootDeviceEnvironment);
}
GpgpuWalkerHelper<GfxFamily>::setGpgpuWalkerThreadData(&walkerCmd, kernelInfo.kernelDescriptor, globalOffsets, startWorkGroups,
numWorkGroups, walkerArgs.localWorkSizes, simd, dim,
localIdsGenerationByRuntime, inlineDataProgrammingRequired, requiredWalkOrder);
HardwareCommandsHelper<GfxFamily>::sendIndirectState(
commandStream,
dsh,
@@ -114,9 +118,6 @@ inline void HardwareInterface<GfxFamily>::programWalker(
localIdsGenerationByRuntime,
commandQueue.getDevice());
GpgpuWalkerHelper<GfxFamily>::setGpgpuWalkerThreadData(&walkerCmd, kernelInfo.kernelDescriptor, globalOffsets, startWorkGroups,
numWorkGroups, walkerArgs.localWorkSizes, simd, dim,
localIdsGenerationByRuntime, inlineDataProgrammingRequired, requiredWalkOrder);
bool kernelSystemAllocation = false;
if (kernel.isBuiltIn) {
kernelSystemAllocation = kernel.getDestinationAllocationInSystemMemory();