intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-25 13:33:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

performance: Skip fence after PC when no post sync write

Browse Source 
Related-To: NEO-14642

Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2025-05-22 15:59:42 +00:00
committed by Compute-Runtime-Automation
parent 8e5b29f55e
commit 86e7d5b276
32 changed files with 58 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
shared/source/command_container/command_encoder_xehp_and_later.inl
View File

@@ -338,7 +338,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
if (!args.makeCommandView) {
if (NEO::PauseOnGpuProperties::pauseModeAllowed(NEO::debugManager.flags.PauseOnEnqueue.get(), args.device->debugExecutionCounter.load(), NEO::PauseOnGpuProperties::PauseMode::BeforeWorkload)) {
void *commandBuffer = listCmdBufferStream->getSpace(MemorySynchronizationCommands<Family>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment));
void *commandBuffer = listCmdBufferStream->getSpace(MemorySynchronizationCommands<Family>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment, false));
args.additionalCommands->push_back(commandBuffer);
EncodeSemaphore<Family>::applyMiSemaphoreWaitCommand(*listCmdBufferStream, *args.additionalCommands);
@@ -475,7 +475,7 @@ void EncodeDispatchKernel<Family>::encode(CommandContainer &container, EncodeDis
if (!args.makeCommandView) {
if (NEO::PauseOnGpuProperties::pauseModeAllowed(NEO::debugManager.flags.PauseOnEnqueue.get(), args.device->debugExecutionCounter.load(), NEO::PauseOnGpuProperties::PauseMode::AfterWorkload)) {
void *commandBuffer = listCmdBufferStream->getSpace(MemorySynchronizationCommands<Family>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment));
void *commandBuffer = listCmdBufferStream->getSpace(MemorySynchronizationCommands<Family>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment, false));
args.additionalCommands->push_back(commandBuffer);
EncodeSemaphore<Family>::applyMiSemaphoreWaitCommand(*listCmdBufferStream, *args.additionalCommands);

4
shared/source/command_container/walker_partition_xehp_and_later.h
View File

@@ -345,7 +345,7 @@ void programPostSyncPipeControlCommand(void *&inputAddress,
rootDeviceEnvironment,
flushArgs);
totalBytesProgrammed += static_cast<uint32_t>(NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment));
totalBytesProgrammed += static_cast<uint32_t>(NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment, true));
}
template <typename GfxFamily>
@@ -828,7 +828,7 @@ uint64_t computeBarrierControlSectionOffset(WalkerPartitionArgs &args,
}
if (args.usePostSync) {
offset += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment);
offset += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForBarrierWithPostSyncOperation(rootDeviceEnvironment, true);
} else {
offset += NEO::MemorySynchronizationCommands<GfxFamily>::getSizeForSingleBarrier();
}