intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-09 22:43:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

refactor: Improving information transfer about the copy engine

Browse Source 
Related-To: NEO-11934

Signed-off-by: Andrzej Koska <andrzej.koska@intel.com>
This commit is contained in:
Andrzej Koska
2024-09-04 21:24:17 +00:00
committed by Compute-Runtime-Automation
parent bf5953c091
commit b0e7a11e9a
25 changed files with 183 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
opencl/source/command_queue/enqueue_common.h
View File

@@ -316,7 +316,7 @@ cl_int CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
}
if (relaxedOrderingEnabled) {
RelaxedOrderingHelper::encodeRegistersBeforeDependencyCheckers<GfxFamily>(commandStream);
RelaxedOrderingHelper::encodeRegistersBeforeDependencyCheckers<GfxFamily>(commandStream, isCopyOnly);
}
TimestampPacketHelper::programCsrDependenciesForTimestampPacketContainer<GfxFamily>(commandStream, csrDeps, relaxedOrderingEnabled, isCopyOnly);
@@ -717,17 +717,17 @@ void CommandQueueHw<GfxFamily>::processDispatchForMarkerWithTimestampPacket(Comm
auto timestampContextStartGpuAddress = TimestampPacketHelper::getContextStartGpuAddress(*currentTimestampPacketNode);
auto timestampGlobalStartAddress = TimestampPacketHelper::getGlobalStartGpuAddress(*currentTimestampPacketNode);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::gpThreadTimeRegAddressOffsetLow, timestampContextStartGpuAddress, false, nullptr);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::globalTimestampLdw, timestampGlobalStartAddress, false, nullptr);
MemorySynchronizationCommands<GfxFamily>::encodeAdditionalTimestampOffsets(*commandStream, timestampContextStartGpuAddress, timestampGlobalStartAddress);
bool isBcs = NEO::EngineHelpers::isBcs(getGpgpuCommandStreamReceiver().getOsContext().getEngineType());
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::gpThreadTimeRegAddressOffsetLow, timestampContextStartGpuAddress, false, nullptr, isBcs);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::globalTimestampLdw, timestampGlobalStartAddress, false, nullptr, isBcs);
MemorySynchronizationCommands<GfxFamily>::encodeAdditionalTimestampOffsets(*commandStream, timestampContextStartGpuAddress, timestampGlobalStartAddress, isBcs);
auto timestampContextEndGpuAddress = TimestampPacketHelper::getContextEndGpuAddress(*currentTimestampPacketNode);
auto timestampGlobalEndAddress = TimestampPacketHelper::getGlobalEndGpuAddress(*currentTimestampPacketNode);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::gpThreadTimeRegAddressOffsetLow, timestampContextEndGpuAddress, false, nullptr);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::globalTimestampLdw, timestampGlobalEndAddress, false, nullptr);
MemorySynchronizationCommands<GfxFamily>::encodeAdditionalTimestampOffsets(*commandStream, timestampContextEndGpuAddress, timestampGlobalEndAddress);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::gpThreadTimeRegAddressOffsetLow, timestampContextEndGpuAddress, false, nullptr, isBcs);
EncodeStoreMMIO<GfxFamily>::encode(*commandStream, RegisterOffsets::globalTimestampLdw, timestampGlobalEndAddress, false, nullptr, isBcs);
MemorySynchronizationCommands<GfxFamily>::encodeAdditionalTimestampOffsets(*commandStream, timestampContextEndGpuAddress, timestampGlobalEndAddress, isBcs);
}
template <typename GfxFamily>

2
opencl/source/command_queue/hardware_interface_base.inl
View File

@@ -114,7 +114,7 @@ void HardwareInterface<GfxFamily>::dispatchWalker(
}
if (walkerArgs.relaxedOrderingEnabled) {
RelaxedOrderingHelper::encodeRegistersBeforeDependencyCheckers<GfxFamily>(*commandStream);
RelaxedOrderingHelper::encodeRegistersBeforeDependencyCheckers<GfxFamily>(*commandStream, false);
}
TimestampPacketHelper::programCsrDependenciesForTimestampPacketContainer<GfxFamily>(*commandStream, csrDependencies, walkerArgs.relaxedOrderingEnabled, commandQueue.isBcs());