intel/compute-runtime
1
0
Fork 0
mirror of https://github.com/intel/compute-runtime.git synced 2026-01-03 06:49:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

Refactor deducing blocked state and task level.

Browse Source 
- Do this with one helper function.

Change-Id: I81dd3107a98db7e45a691ba6d5e708d98eabe3d2
This commit is contained in:
Mrozek, Michal
2017-12-22 16:05:10 +01:00
parent ec59a900e1
commit a8b91c8c99
6 changed files with 27 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
runtime/command_queue/command_queue.cpp
View File

@@ -352,6 +352,9 @@ cl_int CommandQueue::enqueueReleaseSharedObjects(cl_uint numObjects, const cl_me
}
void CommandQueue::updateFromCompletionStamp(const CompletionStamp &completionStamp) {
DEBUG_BREAK_IF(this->taskLevel > completionStamp.taskLevel);
DEBUG_BREAK_IF(this->taskCount > completionStamp.taskCount);
taskCount = completionStamp.taskCount;
flushStamp->setStamp(completionStamp.flushStamp);
this->taskLevel = completionStamp.taskLevel;

2
runtime/command_queue/command_queue_hw.h
View File

@@ -385,7 +385,7 @@ class CommandQueueHw : public CommandQueue {
private:
bool isTaskLevelUpdateRequired(const uint32_t &taskLevel, const cl_event *eventWaitList, const cl_uint &numEventsInWaitList, unsigned int commandType);
void obtainTaskLevelAndBlockedStatus(unsigned int &taskLevel, cl_uint &numEventsInWaitList, const cl_event *&eventWaitList, bool &blockQueue, unsigned int commandType);
void forceDispatchScheduler(OCLRT::MultiDispatchInfo &multiDispatchInfo);
};
} // namespace OCLRT

13
runtime/command_queue/cpu_data_transfer_handler.h
View File

@@ -38,7 +38,6 @@ void *CommandQueueHw<GfxFamily>::cpuDataTransferHandler(MemObj *memObj,
cl_event *event,
cl_int &retVal) {
EventBuilder eventBuilder;
bool blockQueue = false;
bool eventCompleted = false;
ErrorCodeHelper err(&retVal, CL_SUCCESS);
@@ -52,22 +51,16 @@ void *CommandQueueHw<GfxFamily>::cpuDataTransferHandler(MemObj *memObj,
TakeOwnershipWrapper<Device> deviceOwnership(*device);
TakeOwnershipWrapper<CommandQueueHw<GfxFamily>> queueOwnership(*this);
auto taskLevel = getTaskLevelFromWaitList(this->taskLevel, numEventsInWaitList, eventWaitList);
auto updateTaskLevel = isTaskLevelUpdateRequired(taskLevel, eventWaitList, numEventsInWaitList, cmdType);
auto blockQueue = false;
auto taskLevel = 0u;
obtainTaskLevelAndBlockedStatus(taskLevel, numEventsInWaitList, eventWaitList, blockQueue, cmdType);
DBG_LOG(LogTaskCounts, __FUNCTION__, "taskLevel", taskLevel);
if (updateTaskLevel) {
taskLevel++;
this->taskLevel = taskLevel;
}
if (event) {
eventBuilder.getEvent()->taskLevel = taskLevel;
}
blockQueue = ((taskLevel == Event::eventNotReady) || isQueueBlocked());
if (blockQueue &&
(cmdType == CL_COMMAND_MAP_BUFFER || cmdType == CL_COMMAND_UNMAP_MEM_OBJECT)) {

28
runtime/command_queue/enqueue_common.h
View File

@@ -182,17 +182,14 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
bool slmUsed = false;
TakeOwnershipWrapper<CommandQueueHw<GfxFamily>> queueOwnership(*this);
auto taskLevel = getTaskLevelFromWaitList(this->taskLevel, numEventsInWaitList, eventWaitList);
auto blockQueue = (taskLevel == Event::eventNotReady) || isQueueBlocked();
auto blockQueue = false;
auto taskLevel = 0u;
obtainTaskLevelAndBlockedStatus(taskLevel, numEventsInWaitList, eventWaitList, blockQueue, commandType);
// isQueueBlocked() may use commandStream resolving events tree, get start offset after the call
auto &commandStream = getCommandStream<GfxFamily, commandType>(*this, profilingRequired, perfCountersRequired, multiDispatchInfo);
auto commandStreamStart = commandStream.getUsed();
auto &commandStreamReceiver = device->getCommandStreamReceiver();
// isQueueBlocked may unblock queue, get new taskLevel
taskLevel = getTaskLevelFromWaitList(this->taskLevel, numEventsInWaitList, eventWaitList);
DBG_LOG(EventsDebugEnable, "blockQueue", blockQueue, "virtualEvent", virtualEvent, "taskLevel", taskLevel);
if (DebugManager.flags.MakeEachEnqueueBlocking.get()) {
@@ -204,12 +201,6 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
;
}
auto updateTaskLevel = isTaskLevelUpdateRequired(taskLevel, eventWaitList, numEventsInWaitList, commandType);
if (updateTaskLevel) {
taskLevel++;
}
enqueueHandlerHook(commandType, multiDispatchInfo);
if (multiDispatchInfo.empty() == false) {
@@ -423,6 +414,19 @@ void CommandQueueHw<GfxFamily>::enqueueHandler(Surface **surfacesForResidency,
}
}
template <typename GfxFamily>
void CommandQueueHw<GfxFamily>::obtainTaskLevelAndBlockedStatus(unsigned int &taskLevel, cl_uint &numEventsInWaitList, const cl_event *&eventWaitList, bool &blockQueue, unsigned int commandType) {
auto isQueueBlockedStatus = isQueueBlocked();
taskLevel = getTaskLevelFromWaitList(this->taskLevel, numEventsInWaitList, eventWaitList);
blockQueue = (taskLevel == Event::eventNotReady) || isQueueBlockedStatus;
auto updateTaskLevel = isTaskLevelUpdateRequired(taskLevel, eventWaitList, numEventsInWaitList, commandType);
if (updateTaskLevel) {
taskLevel++;
this->taskLevel = taskLevel;
}
}
template <typename GfxFamily>
bool CommandQueueHw<GfxFamily>::isTaskLevelUpdateRequired(const uint32_t &taskLevel, const cl_event *eventWaitList, const cl_uint &numEventsInWaitList, unsigned int commandType) {
bool updateTaskLevel = true;

7
runtime/command_queue/enqueue_map_image.h
View File

@@ -39,7 +39,6 @@ void *CommandQueueHw<GfxFamily>::enqueueMapImage(cl_mem image, cl_bool blockingM
size_t *imageSlicePitch, cl_uint numEventsInWaitList,
const cl_event *eventWaitList, cl_event *event,
cl_int &errcodeRet) {
bool blockQueue = false;
auto pImage = castToObject<Image>(image);
void *ptrToReturn = nullptr;
if (context->isProvidingPerformanceHints()) {
@@ -93,16 +92,16 @@ void *CommandQueueHw<GfxFamily>::enqueueMapImage(cl_mem image, cl_bool blockingM
EventBuilder eventBuilder;
TakeOwnershipWrapper<Device> deviceOwnership(*device);
TakeOwnershipWrapper<CommandQueueHw<GfxFamily>> queueOwnership(*this);
auto blockQueue = false;
auto taskLevel = 0u;
obtainTaskLevelAndBlockedStatus(taskLevel, numEventsInWaitList, eventWaitList, blockQueue, CL_COMMAND_MAP_IMAGE);
auto taskLevel = getTaskLevelFromWaitList(this->taskLevel, numEventsInWaitList, eventWaitList);
if (event) {
eventBuilder.create<Event>(this, CL_COMMAND_MAP_IMAGE, taskLevel, Event::eventNotReady);
*event = eventBuilder.getEvent();
eventBuilder.getEvent()->setQueueTimeStamp();
}
blockQueue = ((taskLevel == Event::eventNotReady) || isQueueBlocked());
if (blockQueue) {
addMapUnmapToWaitlistEventsDependencies(eventWaitList,
static_cast<size_t>(numEventsInWaitList),

3
runtime/event/event.h
View File

@@ -149,11 +149,10 @@ class Event : public BaseObject<_cl_event>, public IDNode<Event> {
static cl_int waitForEvents(cl_uint numEvents,
const cl_event *eventList);
std::unique_ptr<Command> setCommand(std::unique_ptr<Command> newCmd) {
void setCommand(std::unique_ptr<Command> newCmd) {
UNRECOVERABLE_IF(cmdToSubmit.load());
cmdToSubmit.exchange(newCmd.release());
eventWithoutCommand = false;
return nullptr;
}
Command *peekCommand() {
return cmdToSubmit;