intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 23:03:02 +08:00
Code Issues Packages Projects Releases Wiki Activity

Setup enqueue dependencies in blocked path

Browse Source 
Change-Id: Ia6d549924e3e557bae74dbdac691987450cbd3b7
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
This commit is contained in:
Dunajski, Bartosz
2019-11-18 13:35:44 +01:00
committed by sys_ocldev
parent b31e25dbb6
commit 2d5169cf65
6 changed files with 280 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
runtime/helpers/blit_commands_helper.cpp
View File

@@ -109,4 +109,27 @@ BlitterConstants::BlitDirection BlitProperties::obtainBlitDirection(uint32_t com
: BlitterConstants::BlitDirection::BufferToHostPtr;
}
void BlitProperties::setupDependenciesForAuxTranslation(BlitPropertiesContainer &blitPropertiesContainer, TimestampPacketDependencies &timestampPacketDependencies,
TimestampPacketContainer &kernelTimestamps, const EventsRequest &eventsRequest,
CommandStreamReceiver &gpguCsr, CommandStreamReceiver &bcsCsr) {
auto numObjects = blitPropertiesContainer.size() / 2;
for (size_t i = 0; i < numObjects; i++) {
blitPropertiesContainer[i].outputTimestampPacket = timestampPacketDependencies.auxToNonAuxNodes.peekNodes()[i];
blitPropertiesContainer[i + numObjects].outputTimestampPacket = timestampPacketDependencies.nonAuxToAuxNodes.peekNodes()[i];
}
gpguCsr.requestStallingPipeControlOnNextFlush();
auto nodesAllocator = gpguCsr.getTimestampPacketAllocator();
timestampPacketDependencies.barrierNodes.add(nodesAllocator->getTag());
// wait for barrier and events before AuxToNonAux
blitPropertiesContainer[0].csrDependencies.push_back(&timestampPacketDependencies.barrierNodes);
blitPropertiesContainer[0].csrDependencies.fillFromEventsRequest(eventsRequest, bcsCsr,
CsrDependencies::DependenciesType::All);
// wait for NDR before NonAuxToAux
blitPropertiesContainer[numObjects].csrDependencies.push_back(&kernelTimestamps);
}
} // namespace NEO

10
runtime/helpers/blit_commands_helper.h
View File

@@ -24,6 +24,10 @@ struct TimestampPacketStorage;
template <typename TagType>
struct TagNode;
struct BlitProperties;
struct TimestampPacketDependencies;
using BlitPropertiesContainer = StackVec<BlitProperties, 32>;
struct BlitProperties {
static BlitProperties constructPropertiesForReadWriteBuffer(BlitterConstants::BlitDirection blitDirection,
CommandStreamReceiver &commandStreamReceiver,
@@ -42,6 +46,10 @@ struct BlitProperties {
static BlitProperties constructPropertiesForAuxTranslation(AuxTranslationDirection auxTranslationDirection,
GraphicsAllocation *allocation);
static void setupDependenciesForAuxTranslation(BlitPropertiesContainer &blitPropertiesContainer, TimestampPacketDependencies &timestampPacketDependencies,
TimestampPacketContainer &kernelTimestamps, const EventsRequest &eventsRequest,
CommandStreamReceiver &gpguCsr, CommandStreamReceiver &bcsCsr);
static BlitterConstants::BlitDirection obtainBlitDirection(uint32_t commandType);
TagNode<TimestampPacketStorage> *outputTimestampPacket = nullptr;
@@ -56,8 +64,6 @@ struct BlitProperties {
uint64_t copySize = 0;
};
using BlitPropertiesContainer = StackVec<BlitProperties, 32>;
template <typename GfxFamily>
struct BlitCommandsHelper {
static size_t estimateBlitCommandsSize(uint64_t copySize, const CsrDependencies &csrDependencies, bool updateTimestampPacket);

13
runtime/helpers/task_information.cpp
View File

@@ -188,6 +188,16 @@ CompletionStamp &CommandComputeKernel::submit(uint32_t taskLevel, bool terminate
this->kernel->getProgram()->getBlockKernelManager()->makeInternalAllocationsResident(commandStreamReceiver);
}
if (kernelOperation->blitPropertiesContainer.size() > 0) {
auto &bcsCsr = *commandQueue.getBcsCommandStreamReceiver();
BlitProperties::setupDependenciesForAuxTranslation(kernelOperation->blitPropertiesContainer, *timestampPacketDependencies,
*currentTimestampPacketNodes, eventsRequest,
commandQueue.getGpgpuCommandStreamReceiver(), bcsCsr);
auto bcsTaskCount = bcsCsr.blitBuffer(kernelOperation->blitPropertiesContainer, false);
commandQueue.updateBcsTaskCount(bcsTaskCount);
}
DispatchFlags dispatchFlags(
{}, //csrDependencies
nullptr, //barrierTimestampPacketNodes
@@ -211,8 +221,9 @@ CompletionStamp &CommandComputeKernel::submit(uint32_t taskLevel, bool terminate
false //epilogueRequired
);
if (commandStreamReceiver.peekTimestampPacketWriteEnabled()) {
if (timestampPacketDependencies) {
dispatchFlags.csrDependencies.fillFromEventsRequest(eventsRequest, commandStreamReceiver, CsrDependencies::DependenciesType::OutOfCsr);
dispatchFlags.barrierTimestampPacketNodes = &timestampPacketDependencies->barrierNodes;
}
dispatchFlags.pipelineSelectArgs.specialPipelineSelectMode = kernel->requiresSpecialPipelineSelectMode();
if (anyUncacheableArgs) {