Media/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-09-15 13:01:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

optimize batching

Browse Source 
- removes redundant MI_BATCH_BUFFER_START when target is within same command
buffer
- this way we do not have jump in command buffer

Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This commit is contained in:
Michal Mrozek
2021-07-13 11:56:41 +00:00
committed by Compute-Runtime-Automation
parent c389db6f1c
commit 65396fd2e5
1 changed files with 13 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@ -702,7 +702,9 @@ inline bool CommandStreamReceiverHw<GfxFamily>::flushBatchedSubmissions() {
if (DebugManager.flags.FlattenBatchBufferForAUBDump.get()) { if (DebugManager.flags.FlattenBatchBufferForAUBDump.get()) {
flatBatchBufferHelper->registerCommandChunk(primaryCmdBuffer.get()->batchBuffer, sizeof(MI_BATCH_BUFFER_START)); flatBatchBufferHelper->registerCommandChunk(primaryCmdBuffer.get()->batchBuffer, sizeof(MI_BATCH_BUFFER_START));
} }
while (nextCommandBuffer && nextCommandBuffer->inspectionId == primaryCmdBuffer->inspectionId) { while (nextCommandBuffer && nextCommandBuffer->inspectionId == primaryCmdBuffer->inspectionId) {
//noop pipe control //noop pipe control
if (currentPipeControlForNooping) { if (currentPipeControlForNooping) {
if (DebugManager.flags.AddPatchInfoCommentsForAUBDump.get()) { if (DebugManager.flags.AddPatchInfoCommentsForAUBDump.get()) {
@ -718,7 +720,16 @@ inline bool CommandStreamReceiverHw<GfxFamily>::flushBatchedSubmissions() {
flushStampUpdateHelper.insert(nextCommandBuffer->flushStamp->getStampReference()); flushStampUpdateHelper.insert(nextCommandBuffer->flushStamp->getStampReference());
auto nextCommandBufferAddress = nextCommandBuffer->batchBuffer.commandBufferAllocation->getGpuAddress(); auto nextCommandBufferAddress = nextCommandBuffer->batchBuffer.commandBufferAllocation->getGpuAddress();
auto offsetedCommandBuffer = (uint64_t)ptrOffset(nextCommandBufferAddress, nextCommandBuffer->batchBuffer.startOffset); auto offsetedCommandBuffer = (uint64_t)ptrOffset(nextCommandBufferAddress, nextCommandBuffer->batchBuffer.startOffset);
addBatchBufferStart((MI_BATCH_BUFFER_START *)currentBBendLocation, offsetedCommandBuffer, false); auto cpuAddressForCommandBufferDestination = ptrOffset(nextCommandBuffer->batchBuffer.commandBufferAllocation->getUnderlyingBuffer(), nextCommandBuffer->batchBuffer.startOffset);
auto cpuAddressForCurrentCommandBufferEndingSection = alignUp(ptrOffset(currentBBendLocation, sizeof(MI_BATCH_BUFFER_START)), MemoryConstants::cacheLineSize);
//if we point to exact same command buffer, then batch buffer start is not needed at all
if (cpuAddressForCurrentCommandBufferEndingSection == cpuAddressForCommandBufferDestination) {
memset(currentBBendLocation, 0u, ptrDiff(cpuAddressForCurrentCommandBufferEndingSection, currentBBendLocation));
} else {
addBatchBufferStart((MI_BATCH_BUFFER_START *)currentBBendLocation, offsetedCommandBuffer, false);
}
if (DebugManager.flags.FlattenBatchBufferForAUBDump.get()) { if (DebugManager.flags.FlattenBatchBufferForAUBDump.get()) {
flatBatchBufferHelper->registerCommandChunk(nextCommandBuffer->batchBuffer, sizeof(MI_BATCH_BUFFER_START)); flatBatchBufferHelper->registerCommandChunk(nextCommandBuffer->batchBuffer, sizeof(MI_BATCH_BUFFER_START));
} }
@ -726,6 +737,7 @@ inline bool CommandStreamReceiverHw<GfxFamily>::flushBatchedSubmissions() {
currentBBendLocation = nextCommandBuffer->batchBufferEndLocation; currentBBendLocation = nextCommandBuffer->batchBufferEndLocation;
lastTaskCount = nextCommandBuffer->taskCount; lastTaskCount = nextCommandBuffer->taskCount;
nextCommandBuffer = nextCommandBuffer->next; nextCommandBuffer = nextCommandBuffer->next;
commandBufferList.removeFrontOne(); commandBufferList.removeFrontOne();
} }
surfacesForSubmit.reserve(resourcePackage.size() + 1); surfacesForSubmit.reserve(resourcePackage.size() + 1);