Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-10 12:53:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

RelaxedOrdering: Improve dependencies tracking

Browse Source 
Avoid not needed scheduler programming
Related-To: NEO-7458

Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2022-11-26 20:10:32 +00:00
committed by Compute-Runtime-Automation
parent ad6237478f
commit 3f962bf3e8
23 changed files with 734 additions and 161 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
opencl/source/command_queue/enqueue_common.h
View File

@ -804,7 +804,8 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueNonBlocked(
kernel->areMultipleSubDevicesInContext(), // areMultipleSubDevicesInContext
kernel->requiresMemoryMigration(), // memoryMigrationRequired
isTextureCacheFlushNeeded(commandType), // textureCacheFlush
false); // hasStallingCmds
false, // hasStallingCmds
false); // hasRelaxedOrderingDependencies
dispatchFlags.pipelineSelectArgs.mediaSamplerRequired = mediaSamplerRequired;
dispatchFlags.pipelineSelectArgs.systolicPipelineSelectMode = systolicPipelineSelectMode;
@ -1050,7 +1051,8 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueCommandWithoutKernel(
context->containsMultipleSubDevices(rootDeviceIndex), // areMultipleSubDevicesInContext
false, // memoryMigrationRequired
false, // textureCacheFlush
false); // hasStallingCmds
false, // hasStallingCmds
false); // hasRelaxedOrderingDependencies
const bool isHandlingBarrier = getGpgpuCommandStreamReceiver().isStallingCommandsOnNextFlushRequired();

9
opencl/source/helpers/task_information.cpp
View File

@ -80,7 +80,8 @@ CompletionStamp &CommandMapUnmap::submit(uint32_t taskLevel, bool terminated) {
false, // areMultipleSubDevicesInContext
false, // memoryMigrationRequired
false, // textureCacheFlush
false); // hasStallingCmds
false, // hasStallingCmds
false); // hasRelaxedOrderingDependencies
DEBUG_BREAK_IF(taskLevel >= CompletionStamp::notReady);
@ -210,7 +211,8 @@ CompletionStamp &CommandComputeKernel::submit(uint32_t taskLevel, bool terminate
kernel->areMultipleSubDevicesInContext(), // areMultipleSubDevicesInContext
kernel->requiresMemoryMigration(), // memoryMigrationRequired
commandQueue.isTextureCacheFlushNeeded(this->commandType), // textureCacheFlush
false); // hasStallingCmds
false, // hasStallingCmds
false); // hasRelaxedOrderingDependencies
if (commandQueue.getContext().getRootDeviceIndices().size() > 1) {
eventsRequest.fillCsrDependenciesForTaskCountContainer(dispatchFlags.csrDependencies, commandStreamReceiver);
@ -385,7 +387,8 @@ CompletionStamp &CommandWithoutKernel::submit(uint32_t taskLevel, bool terminate
commandQueue.getContext().containsMultipleSubDevices(rootDeviceIndex), // areMultipleSubDevicesInContext
false, // memoryMigrationRequired
false, // textureCacheFlush
false); // hasStallingCmds
false, // hasStallingCmds
false); // hasRelaxedOrderingDependencies
if (commandQueue.getContext().getRootDeviceIndices().size() > 1) {
eventsRequest.fillCsrDependenciesForTaskCountContainer(dispatchFlags.csrDependencies, commandStreamReceiver);