intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2025-12-29 09:03:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add timestamp wait support for BCS

Browse Source 
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
This commit is contained in:
Lukasz Jobczyk
2022-04-15 12:41:12 +00:00
committed by Compute-Runtime-Automation
parent b15c8e971c
commit 7e4262839a
2 changed files with 30 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
opencl/source/command_queue/command_queue.cpp
View File

@@ -122,8 +122,8 @@ CommandQueue::~CommandQueue() {
}
timestampPacketContainer.reset();
//for normal queue, decrement ref count on context
//special queue is owned by context so ref count doesn't have to be decremented
// for normal queue, decrement ref count on context
// special queue is owned by context so ref count doesn't have to be decremented
if (context && !isSpecialCommandQueue) {
context->decRefInternal();
}
@@ -269,16 +269,20 @@ WaitStatus CommandQueue::waitUntilComplete(uint32_t gpgpuTaskCountToWait, Range<
if (gtpinIsGTPinInitialized()) {
gtpinNotifyTaskCompletion(gpgpuTaskCountToWait);
}
for (const CopyEngineState &copyEngine : copyEnginesToWait) {
auto bcsCsr = getBcsCommandStreamReceiver(copyEngine.engineType);
waitStatus = bcsCsr->waitForTaskCountWithKmdNotifyFallback(copyEngine.taskCount, 0, false, this->getThrottle());
if (waitStatus == WaitStatus::GpuHang) {
return WaitStatus::GpuHang;
}
}
}
for (const CopyEngineState &copyEngine : copyEnginesToWait) {
auto bcsCsr = getBcsCommandStreamReceiver(copyEngine.engineType);
waitStatus = bcsCsr->waitForTaskCountWithKmdNotifyFallback(copyEngine.taskCount, 0, false, this->getThrottle());
if (waitStatus == WaitStatus::GpuHang) {
return WaitStatus::GpuHang;
}
waitStatus = bcsCsr->waitForTaskCountAndCleanTemporaryAllocationList(copyEngine.taskCount);
if (waitStatus == WaitStatus::GpuHang) {
return WaitStatus::GpuHang;
@@ -296,7 +300,7 @@ WaitStatus CommandQueue::waitUntilComplete(uint32_t gpgpuTaskCountToWait, Range<
bool CommandQueue::isQueueBlocked() {
TakeOwnershipWrapper<CommandQueue> takeOwnershipWrapper(*this);
//check if we have user event and if so, if it is in blocked state.
// check if we have user event and if so, if it is in blocked state.
if (this->virtualEvent) {
auto executionStatus = this->virtualEvent->peekExecutionStatus();
if (executionStatus <= CL_SUBMITTED) {
@@ -311,7 +315,7 @@ bool CommandQueue::isQueueBlocked() {
taskLevel++;
}
} else {
//at this point we may reset queue TaskCount, since all command previous to this were aborted
// at this point we may reset queue TaskCount, since all command previous to this were aborted
taskCount = 0;
flushStamp->setStamp(0);
taskLevel = getGpgpuCommandStreamReceiver().peekTaskLevel();
@@ -319,7 +323,7 @@ bool CommandQueue::isQueueBlocked() {
FileLoggerInstance().log(DebugManager.flags.EventsDebugEnable.get(), "isQueueBlocked taskLevel change from", taskLevel, "to new from virtualEvent", this->virtualEvent, "new tasklevel", this->virtualEvent->taskLevel.load());
//close the access to virtual event, driver added only 1 ref count.
// close the access to virtual event, driver added only 1 ref count.
this->virtualEvent->decRefInternal();
this->virtualEvent = nullptr;
return false;
@@ -625,11 +629,11 @@ void CommandQueue::enqueueBlockedMapUnmapOperation(const cl_event *eventWaitList
eventBuilder = &internalEventBuilder;
}
//store task data in event
// store task data in event
auto cmd = std::unique_ptr<Command>(new CommandMapUnmap(opType, *memObj, copySize, copyOffset, readOnly, *this));
eventBuilder->getEvent()->setCommand(std::move(cmd));
//bind output event with input events
// bind output event with input events
eventBuilder->addParentEvents(ArrayRef<const cl_event>(eventWaitList, numEventsInWaitlist));
eventBuilder->addParentEvent(this->virtualEvent);
eventBuilder->finalize();
@@ -772,12 +776,12 @@ bool CommandQueue::bufferCpuCopyAllowed(Buffer *buffer, cl_command_type commandT
debugVariableSet = true;
}
//if we are blocked by user events, we can't service the call on CPU
// if we are blocked by user events, we can't service the call on CPU
if (Event::checkUserEventDependencies(numEventsInWaitList, eventWaitList)) {
return false;
}
//check if buffer is compatible
// check if buffer is compatible
if (!buffer->isReadWriteOnCpuAllowed(device->getDevice())) {
return false;
}
@@ -790,18 +794,18 @@ bool CommandQueue::bufferCpuCopyAllowed(Buffer *buffer, cl_command_type commandT
return true;
}
//non blocking transfers are not expected to be serviced by CPU
//we do not want to artifically stall the pipeline to allow CPU access
// non blocking transfers are not expected to be serviced by CPU
// we do not want to artifically stall the pipeline to allow CPU access
if (blocking == CL_FALSE) {
return false;
}
//check if it is beneficial to do transfer on CPU
// check if it is beneficial to do transfer on CPU
if (!buffer->isReadWriteOnCpuPreferred(ptr, size, getDevice())) {
return false;
}
//make sure that event wait list is empty
// make sure that event wait list is empty
if (numEventsInWaitList == 0) {
return true;
}
@@ -988,7 +992,7 @@ void CommandQueue::aubCaptureHook(bool &blocking, bool &clearAllDependencies, co
void CommandQueue::assignDataToOverwrittenBcsNode(TagNodeBase *node) {
std::array<uint32_t, 8u> timestampData;
timestampData.fill(std::numeric_limits<uint32_t>::max());
if (node->refCountFetchSub(0) <= 2) { //One ref from deferred container and one from bcs barrier container it is going to be released from
if (node->refCountFetchSub(0) <= 2) { // One ref from deferred container and one from bcs barrier container it is going to be released from
for (uint32_t i = 0; i < node->getPacketsUsed(); i++) {
node->assignDataToAllTimestamps(i, timestampData.data());
}