intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-28 16:48:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

Avoid ReadModifyWrite problem on devices with coherent L3.

Browse Source 
Setting one of bitfields requires read from local memory which is very slow.
This is not needed for devices that have coherent L3.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This commit is contained in:
Michal Mrozek
2022-03-04 14:07:37 +00:00
committed by Compute-Runtime-Automation
parent c5ff07237b
commit f293c9ab25
1 changed files with 8 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@@ -759,10 +759,14 @@ inline bool CommandStreamReceiverHw<GfxFamily>::flushBatchedSubmissions() {
}
//make sure we flush DC if needed
if (epiloguePipeControlLocation) {
bool flushDcInEpilogue = MemorySynchronizationCommands<GfxFamily>::getDcFlushEnable(
!DebugManager.flags.DisableDcFlushInEpilogue.get(), hwInfo);
((PIPE_CONTROL *)epiloguePipeControlLocation)->setDcFlushEnable(flushDcInEpilogue);
if (epiloguePipeControlLocation && MemorySynchronizationCommands<GfxFamily>::getDcFlushEnable(true, hwInfo)) {
auto emitDcFlush = true;
if (DebugManager.flags.DisableDcFlushInEpilogue.get()) {
emitDcFlush = false;
}
((PIPE_CONTROL *)epiloguePipeControlLocation)->setDcFlushEnable(emitDcFlush);
}
primaryCmdBuffer->batchBuffer.endCmdPtr = currentBBendLocation;