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

Refactor dispatching blit enqueue

Browse Source 
Change-Id: Ibe499e4815a16d5884510c6804221d2b74dbffd4
Signed-off-by: Dunajski, Bartosz <bartosz.dunajski@intel.com>
Related-To: NEO-3020
This commit is contained in:
Dunajski, Bartosz
2019-08-30 09:37:44 +02:00
committed by sys_ocldev
parent 094068807e
commit 77e22bd81b
6 changed files with 130 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
runtime/command_queue/command_queue_hw.h
View File

@ -347,7 +347,7 @@ class CommandQueueHw : public CommandQueue {
LinearStream &commandStream,
size_t commandStreamStart,
bool &blocking,
bool blitEnqueue,
const EnqueueProperties &enqueueProperties,
TimestampPacketContainer *previousTimestampPacketNodes,
EventsRequest &eventsRequest,
EventBuilder &eventBuilder,
@ -356,11 +356,11 @@ class CommandQueueHw : public CommandQueue {
size_t numSurfaces,
LinearStream *commandStream,
CsrDependencies &csrDeps);
void processDispatchForBlitEnqueue(const MultiDispatchInfo &multiDispatchInfo,
TimestampPacketContainer &previousTimestampPacketNodes,
const EventsRequest &eventsRequest,
LinearStream &commandStream,
uint32_t commandType);
BlitProperties processDispatchForBlitEnqueue(const MultiDispatchInfo &multiDispatchInfo,
TimestampPacketContainer &previousTimestampPacketNodes,
const EventsRequest &eventsRequest,
LinearStream &commandStream,
uint32_t commandType);
void submitCacheFlush(Surface **surfaces,
size_t numSurfaces,
LinearStream *commandStream,

44
runtime/command_queue/enqueue_common.h
View File

@ -190,6 +190,7 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
TimestampPacketContainer previousTimestampPacketNodes;
EventsRequest eventsRequest(numEventsInWaitList, eventWaitList, event);
CsrDependencies csrDeps;
BlitProperties blitProperties;
if (getGpgpuCommandStreamReceiver().peekTimestampPacketWriteEnabled()) {
csrDeps.fillFromEventsRequestAndMakeResident(eventsRequest, getGpgpuCommandStreamReceiver(), CsrDependencies::DependenciesType::OnCsr);
@ -218,7 +219,7 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
bool flushDependenciesForNonKernelCommand = false;
if (blitEnqueue) {
processDispatchForBlitEnqueue(multiDispatchInfo, previousTimestampPacketNodes, eventsRequest, commandStream, commandType);
blitProperties = processDispatchForBlitEnqueue(multiDispatchInfo, previousTimestampPacketNodes, eventsRequest, commandStream, commandType);
} else if (multiDispatchInfo.empty() == false) {
processDispatchForKernels<commandType>(multiDispatchInfo, printfHandler, eventBuilder.getEvent(),
hwTimeStamps, blockQueue, devQueueHw, csrDeps, blockedCommandsData.get(),
@ -245,14 +246,15 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
}
CompletionStamp completionStamp = {Event::eventNotReady, taskLevel, 0};
EnqueueProperties enqueueProperties(blitEnqueue, !multiDispatchInfo.empty(), isCacheFlushCommand(commandType),
flushDependenciesForNonKernelCommand, &blitProperties);
if (!blockQueue) {
if (parentKernel) {
processDeviceEnqueue(devQueueHw, multiDispatchInfo, hwTimeStamps, blocking);
}
auto kernelSubmissionRequired = !isCommandWithoutKernel(commandType) && !blitEnqueue;
if (kernelSubmissionRequired) {
if (enqueueProperties.operation == EnqueueProperties::Operation::GpuKernel) {
completionStamp = enqueueNonBlocked<commandType>(
surfacesForResidency,
numSurfaceForResidency,
@ -284,19 +286,20 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
devQueueHw->getDebugQueue());
}
}
} else if (isCacheFlushCommand(commandType) || blitEnqueue || flushDependenciesForNonKernelCommand) {
} else if (enqueueProperties.isFlushWithoutKernelRequired()) {
completionStamp = enqueueCommandWithoutKernel(
surfacesForResidency,
numSurfaceForResidency,
commandStream,
commandStreamStart,
blocking,
blitEnqueue,
enqueueProperties,
&previousTimestampPacketNodes,
eventsRequest,
eventBuilder,
taskLevel);
} else {
UNRECOVERABLE_IF(enqueueProperties.operation != EnqueueProperties::Operation::EnqueueWithoutSubmission);
auto maxTaskCount = this->taskCount;
for (auto eventId = 0u; eventId < numEventsInWaitList; eventId++) {
auto event = castToObject<Event>(eventWaitList[eventId]);
@ -432,11 +435,11 @@ void CommandQueueHw<GfxFamily>::processDispatchForKernels(const MultiDispatchInf
}
template <typename GfxFamily>
void CommandQueueHw<GfxFamily>::processDispatchForBlitEnqueue(const MultiDispatchInfo &multiDispatchInfo,
TimestampPacketContainer &previousTimestampPacketNodes,
const EventsRequest &eventsRequest,
LinearStream &commandStream,
uint32_t commandType) {
BlitProperties CommandQueueHw<GfxFamily>::processDispatchForBlitEnqueue(const MultiDispatchInfo &multiDispatchInfo,
TimestampPacketContainer &previousTimestampPacketNodes,
const EventsRequest &eventsRequest,
LinearStream &commandStream,
uint32_t commandType) {
auto blitDirection = BlitProperties::obtainBlitDirection(commandType);
auto blitCommandStreamReceiver = getBcsCommandStreamReceiver();
@ -450,12 +453,10 @@ void CommandQueueHw<GfxFamily>::processDispatchForBlitEnqueue(const MultiDispatc
blitProperties.csrDependencies.push_back(&previousTimestampPacketNodes);
blitProperties.outputTimestampPacket = timestampPacketContainer.get();
previousTimestampPacketNodes.makeResident(*blitCommandStreamReceiver);
timestampPacketContainer->makeResident(*blitCommandStreamReceiver);
blitCommandStreamReceiver->blitBuffer(blitProperties);
auto currentTimestampPacketNode = timestampPacketContainer->peekNodes().at(0);
TimestampPacketHelper::programSemaphoreWithImplicitDependency<GfxFamily>(commandStream, *currentTimestampPacketNode);
return blitProperties;
}
template <typename GfxFamily>
@ -809,7 +810,7 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueCommandWithoutKernel(
LinearStream &commandStream,
size_t commandStreamStart,
bool &blocking,
bool blitEnqueue,
const EnqueueProperties &enqueueProperties,
TimestampPacketContainer *previousTimestampPacketNodes,
EventsRequest &eventsRequest,
EventBuilder &eventBuilder,
@ -826,11 +827,20 @@ CompletionStamp CommandQueueHw<GfxFamily>::enqueueCommandWithoutKernel(
requiresCoherency |= surface->IsCoherent;
}
if (enqueueProperties.operation == EnqueueProperties::Operation::Blit) {
UNRECOVERABLE_IF(!enqueueProperties.blitProperties);
auto bcsCsr = getBcsCommandStreamReceiver();
previousTimestampPacketNodes->makeResident(*bcsCsr);
timestampPacketContainer->makeResident(*bcsCsr);
bcsCsr->blitBuffer(*enqueueProperties.blitProperties);
}
DispatchFlags dispatchFlags = {};
dispatchFlags.blocking = blocking;
dispatchFlags.multiEngineQueue = multiEngineQueue;
dispatchFlags.preemptionMode = device->getPreemptionMode();
dispatchFlags.implicitFlush = blitEnqueue;
dispatchFlags.implicitFlush = (enqueueProperties.operation == EnqueueProperties::Operation::Blit);
dispatchFlags.guardCommandBufferWithPipeControl = true;
dispatchFlags.outOfOrderExecutionAllowed = getGpgpuCommandStreamReceiver().isNTo1SubmissionModelEnabled();