intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-05 17:20:26 +08:00
Code Issues Packages Projects Releases Wiki Activity

feature: add support to dispatch epilogue commands into dedicated stream

Browse Source 
Related-To: NEO-10356

Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
This commit is contained in:
Zbigniew Zdanowicz
2025-02-28 18:22:33 +00:00
committed by Compute-Runtime-Automation
parent c1a986c71c
commit db99c25c79
2 changed files with 86 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
shared/source/command_stream/command_stream_receiver_hw_base.inl
View File

@@ -214,6 +214,8 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushBcsTask(LinearStream &c
NEO::EncodeDummyBlitWaArgs waArgs{false, const_cast<RootDeviceEnvironment *>(&(this->peekRootDeviceEnvironment()))};
LinearStream &epilogueCommandStream = dispatchBcsFlags.optionalEpilogueCmdStream != nullptr ? *dispatchBcsFlags.optionalEpilogueCmdStream : commandStreamTask;
if (dispatchBcsFlags.flushTaskCount) {
uint64_t postSyncAddress = getTagAllocation()->getGpuAddress();
TaskCountType postSyncData = peekTaskCount() + 1;
@@ -222,7 +224,7 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushBcsTask(LinearStream &c
args.notifyEnable = isUsedNotifyEnableForPostSync();
args.tlbFlush |= (debugManager.flags.ForceTlbFlushWithTaskCountAfterCopy.get() == 1);
NEO::EncodeMiFlushDW<GfxFamily>::programWithWa(commandStreamTask, postSyncAddress, postSyncData, args);
NEO::EncodeMiFlushDW<GfxFamily>::programWithWa(epilogueCommandStream, postSyncAddress, postSyncData, args);
}
auto &commandStreamCSR = getCS(getRequiredCmdStreamSizeAligned(dispatchBcsFlags));
@@ -245,12 +247,15 @@ CompletionStamp CommandStreamReceiverHw<GfxFamily>::flushBcsTask(LinearStream &c
makeResident(*getTagAllocation());
makeResident(*commandStreamTask.getGraphicsAllocation());
if (dispatchBcsFlags.optionalEpilogueCmdStream != nullptr) {
makeResident(*dispatchBcsFlags.optionalEpilogueCmdStream->getGraphicsAllocation());
}
bool submitCSR = (commandStreamStartCSR != commandStreamCSR.getUsed());
void *bbEndLocation = nullptr;
programEndingCmd(commandStreamTask, &bbEndLocation, isBlitterDirectSubmissionEnabled(), dispatchBcsFlags.hasRelaxedOrderingDependencies, true);
EncodeNoop<GfxFamily>::alignToCacheLine(commandStreamTask);
programEndingCmd(epilogueCommandStream, &bbEndLocation, isBlitterDirectSubmissionEnabled(), dispatchBcsFlags.hasRelaxedOrderingDependencies, true);
EncodeNoop<GfxFamily>::alignToCacheLine(epilogueCommandStream);
if (submitCSR) {
auto bbStart = reinterpret_cast<MI_BATCH_BUFFER_START *>(commandStreamCSR.getSpace(sizeof(MI_BATCH_BUFFER_START)));
@@ -2178,6 +2183,8 @@ template <typename GfxFamily>
void CommandStreamReceiverHw<GfxFamily>::dispatchImmediateFlushClientBufferCommands(ImmediateDispatchFlags &dispatchFlags,
LinearStream &immediateCommandStream,
ImmediateFlushData &flushData) {
LinearStream &epilogueCommandStream = dispatchFlags.optionalEpilogueCmdStream != nullptr ? *dispatchFlags.optionalEpilogueCmdStream : immediateCommandStream;
if (dispatchFlags.blockingAppend || dispatchFlags.requireTaskCountUpdate) {
auto address = getTagAllocation()->getGpuAddress();
@@ -2187,7 +2194,7 @@ void CommandStreamReceiverHw<GfxFamily>::dispatchImmediateFlushClientBufferComma
args.notifyEnable = isUsedNotifyEnableForPostSync();
args.workloadPartitionOffset = isMultiTileOperationEnabled();
MemorySynchronizationCommands<GfxFamily>::addBarrierWithPostSyncOperation(
immediateCommandStream,
epilogueCommandStream,
PostSyncMode::immediateData,
address,
this->taskCount + 1,
@@ -2196,9 +2203,12 @@ void CommandStreamReceiverHw<GfxFamily>::dispatchImmediateFlushClientBufferComma
}
makeResident(*immediateCommandStream.getGraphicsAllocation());
if (dispatchFlags.optionalEpilogueCmdStream != nullptr) {
makeResident(*dispatchFlags.optionalEpilogueCmdStream->getGraphicsAllocation());
}
programEndingCmd(immediateCommandStream, &flushData.endPtr, isDirectSubmissionEnabled(), dispatchFlags.hasRelaxedOrderingDependencies, EngineHelpers::isBcs(this->osContext->getEngineType()));
EncodeNoop<GfxFamily>::alignToCacheLine(immediateCommandStream);
programEndingCmd(epilogueCommandStream, &flushData.endPtr, isDirectSubmissionEnabled(), dispatchFlags.hasRelaxedOrderingDependencies, EngineHelpers::isBcs(this->osContext->getEngineType()));
EncodeNoop<GfxFamily>::alignToCacheLine(epilogueCommandStream);
}
template <typename GfxFamily>

70
shared/test/unit_test/command_stream/command_stream_receiver_tests.cpp
View File

@@ -6078,3 +6078,73 @@ HWTEST_F(CommandStreamReceiverHwTest, givenRequiredFlushTaskCountWhenFlushBcsTas
EXPECT_TRUE(commandStreamReceiver.latestFlushedBatchBuffer.hasStallingCmds);
}
HWTEST_F(CommandStreamReceiverHwTest, givenEpilogueStreamAvailableWhenFlushBcsTaskCalledThenDispachEpilogueCommandsIntoEpilogueStream) {
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
DispatchBcsFlags dispatchBcsFlags(false, false, false);
// first flush can carry preamble, no interest in flags here
commandStreamReceiver.flushBcsTask(commandStream,
commandStream.getUsed(),
dispatchBcsFlags,
pDevice->getHardwareInfo());
// regular dispatch here
GraphicsAllocation *commandBuffer = commandStreamReceiver.getMemoryManager()->allocateGraphicsMemoryWithProperties(MockAllocationProperties{commandStreamReceiver.getRootDeviceIndex(), MemoryConstants::pageSize});
ASSERT_NE(nullptr, commandBuffer);
LinearStream epilogueStream(commandBuffer);
commandStreamReceiver.storeMakeResidentAllocations = true;
dispatchBcsFlags.flushTaskCount = true;
dispatchBcsFlags.optionalEpilogueCmdStream = &epilogueStream;
commandStreamReceiver.flushBcsTask(commandStream,
commandStream.getUsed(),
dispatchBcsFlags,
pDevice->getHardwareInfo());
EXPECT_TRUE(commandStreamReceiver.isMadeResident(commandBuffer));
HardwareParse hwParser;
hwParser.parseCommands<FamilyType>(epilogueStream, 0);
auto cmdIterator = find<typename FamilyType::MI_FLUSH_DW *>(hwParser.cmdList.begin(), hwParser.cmdList.end());
EXPECT_NE(hwParser.cmdList.end(), cmdIterator);
commandStreamReceiver.getMemoryManager()->freeGraphicsMemoryImpl(commandBuffer);
}
HWTEST_F(CommandStreamReceiverHwTest, givenEpilogueStreamAvailableWhenFlushImmediateTaskCalledThenDispachEpilogueCommandsIntoEpilogueStream) {
auto &commandStreamReceiver = pDevice->getUltCommandStreamReceiver<FamilyType>();
// first flush can carry preamble, no interest in flags here
commandStreamReceiver.flushImmediateTask(commandStream,
commandStream.getUsed(),
immediateFlushTaskFlags,
*pDevice);
// regular dispatch here
GraphicsAllocation *commandBuffer = commandStreamReceiver.getMemoryManager()->allocateGraphicsMemoryWithProperties(MockAllocationProperties{commandStreamReceiver.getRootDeviceIndex(), MemoryConstants::pageSize});
ASSERT_NE(nullptr, commandBuffer);
LinearStream epilogueStream(commandBuffer);
commandStreamReceiver.storeMakeResidentAllocations = true;
immediateFlushTaskFlags.requireTaskCountUpdate = true;
immediateFlushTaskFlags.optionalEpilogueCmdStream = &epilogueStream;
commandStreamReceiver.flushImmediateTask(commandStream,
commandStream.getUsed(),
immediateFlushTaskFlags,
*pDevice);
EXPECT_TRUE(commandStreamReceiver.isMadeResident(commandBuffer));
HardwareParse hwParser;
hwParser.parseCommands<FamilyType>(epilogueStream, 0);
auto cmdIterator = find<typename FamilyType::PIPE_CONTROL *>(hwParser.cmdList.begin(), hwParser.cmdList.end());
EXPECT_NE(hwParser.cmdList.end(), cmdIterator);
commandStreamReceiver.getMemoryManager()->freeGraphicsMemoryImpl(commandBuffer);
}