Skip to content

[NFC][SYCL] Pass queue_impl by raw ptr/ref in event_impl.hpp #18983

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Jun 16, 2025
Merged

[NFC][SYCL] Pass queue_impl by raw ptr/ref in event_impl.hpp #18983

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
76dd39e
[NFC][SYCL] Pass `queue_impl` by raw ptr/ref in `event_impl.hpp`
aelovikov-intel Jun 13, 2025
118a184
`auto` -> `std::shared_ptr<queue_impl>`
aelovikov-intel Jun 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
22 changes: 11 additions & 11 deletions sycl/source/detail/event_impl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -215,7 +215,7 @@ void event_impl::setQueue(queue_impl &Queue) {
void event_impl::setSubmittedQueue(std::weak_ptr<queue_impl> SubmittedQueue) {
MSubmittedQueue = std::move(SubmittedQueue);
if (MHostProfilingInfo) {
if (auto QueuePtr = MSubmittedQueue.lock()) {
if (std::shared_ptr<queue_impl> QueuePtr = MSubmittedQueue.lock()) {
device_impl &Device = QueuePtr->getDeviceImpl();
MHostProfilingInfo->setDevice(&Device);
}
Expand Down Expand Up @@ -248,7 +248,7 @@ void *event_impl::instrumentationProlog(std::string &Name, int32_t StreamID,
// queue is available with the wait events. We check to see if the
// TraceEvent is available in the Queue object.
void *TraceEvent = nullptr;
if (QueueImplPtr Queue = MQueue.lock()) {
if (std::shared_ptr<queue_impl> Queue = MQueue.lock()) {
TraceEvent = Queue->getTraceEvent();
WaitEvent =
(TraceEvent ? static_cast<xpti_td *>(TraceEvent) : GSYCLGraphEvent);
Expand Down Expand Up @@ -317,7 +317,7 @@ void event_impl::wait_and_throw(
std::shared_ptr<sycl::detail::event_impl> Self) {
wait(Self);

if (QueueImplPtr SubmittedQueue = MSubmittedQueue.lock())
if (std::shared_ptr<queue_impl> SubmittedQueue = MSubmittedQueue.lock())
SubmittedQueue->throw_asynchronous();
}

Expand Down Expand Up @@ -462,7 +462,7 @@ event_impl::get_backend_info<info::platform::version>() const {
"the info::platform::version info descriptor can "
"only be queried with an OpenCL backend");
}
if (QueueImplPtr Queue = MQueue.lock()) {
if (std::shared_ptr<queue_impl> Queue = MQueue.lock()) {
return Queue->getDeviceImpl()
.get_platform()
.get_info<info::platform::version>();
Expand All @@ -485,7 +485,7 @@ event_impl::get_backend_info<info::device::version>() const {
"the info::device::version info descriptor can only "
"be queried with an OpenCL backend");
}
if (QueueImplPtr Queue = MQueue.lock()) {
if (std::shared_ptr<queue_impl> Queue = MQueue.lock()) {
return Queue->getDeviceImpl().get_info<info::device::version>();
}
return ""; // If the queue has been released, no device will be associated so
Expand Down Expand Up @@ -552,21 +552,21 @@ std::vector<EventImplPtr> event_impl::getWaitList() {
return Result;
}

void event_impl::flushIfNeeded(const QueueImplPtr &UserQueue) {
void event_impl::flushIfNeeded(queue_impl *UserQueue) {
// Some events might not have a native handle underneath even at this point,
// e.g. those produced by memset with 0 size (no UR call is made).
auto Handle = this->getHandle();
if (MIsFlushed || !Handle)
return;

QueueImplPtr Queue = MQueue.lock();
std::shared_ptr<queue_impl> Queue = MQueue.lock();
// If the queue has been released, all of the commands have already been
// implicitly flushed by urQueueRelease.
if (!Queue) {
MIsFlushed = true;
return;
}
if (Queue == UserQueue)
if (Queue.get() == UserQueue)
return;

// Check if the task for this event has already been submitted.
Expand Down Expand Up @@ -604,9 +604,9 @@ void event_impl::setSubmissionTime() {
if (!MIsProfilingEnabled && !MProfilingTagEvent)
return;

std::weak_ptr<queue_impl> Queue = isHost() ? MSubmittedQueue : MQueue;
if (QueueImplPtr QueuePtr = Queue.lock()) {
device_impl &Device = QueuePtr->getDeviceImpl();
if (std::shared_ptr<queue_impl> Queue =
isHost() ? MSubmittedQueue.lock() : MQueue.lock()) {
device_impl &Device = Queue->getDeviceImpl();
MSubmitTime = getTimestamp(&Device);
}
}
Expand Down
11 changes: 7 additions & 4 deletions sycl/source/detail/event_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,7 +31,6 @@ class Adapter;
class context_impl;
using ContextImplPtr = std::shared_ptr<sycl::detail::context_impl>;
class queue_impl;
using QueueImplPtr = std::shared_ptr<sycl::detail::queue_impl>;
class event_impl;
using EventImplPtr = std::shared_ptr<sycl::detail::event_impl>;

Expand Down Expand Up @@ -242,7 +241,7 @@ class event_impl : public std::enable_shared_from_this<event_impl> {
/// Performs a flush on the queue associated with this event if the user queue
/// is different and the task associated with this event hasn't been submitted
/// to the device yet.
void flushIfNeeded(const QueueImplPtr &UserQueue);
void flushIfNeeded(queue_impl *UserQueue);

/// Cleans dependencies of this event_impl.
void cleanupDependencyEvents();
Expand All @@ -262,7 +261,9 @@ class event_impl : public std::enable_shared_from_this<event_impl> {
///
/// @return shared_ptr to MWorkerQueue, please be aware it can be empty
/// pointer
QueueImplPtr getWorkerQueue() { return MWorkerQueue.lock(); };
std::shared_ptr<sycl::detail::queue_impl> getWorkerQueue() {
return MWorkerQueue.lock();
};

/// Sets worker queue for command.
///
Expand All @@ -289,7 +290,9 @@ class event_impl : public std::enable_shared_from_this<event_impl> {
/// @return Submission time for command associated with this event
uint64_t getSubmissionTime();

QueueImplPtr getSubmittedQueue() const { return MSubmittedQueue.lock(); };
std::shared_ptr<sycl::detail::queue_impl> getSubmittedQueue() const {
return MSubmittedQueue.lock();
};

/// Checks if this event is complete.
///
Expand Down
21 changes: 7 additions & 14 deletions sycl/source/detail/scheduler/commands.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -323,13 +323,6 @@ bool Command::isFusable() const {
}
#endif // __INTEL_PREVIEW_BREAKING_CHANGES

static void flushCrossQueueDeps(const std::vector<EventImplPtr> &EventImpls,
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Moved to be a member function of Command

const QueueImplPtr &Queue) {
for (auto &EventImpl : EventImpls) {
EventImpl->flushIfNeeded(Queue);
}
}

namespace {

struct EnqueueNativeCommandData {
Expand Down Expand Up @@ -553,7 +546,7 @@ void Command::waitForEvents(QueueImplPtr Queue,
}
} else {
std::vector<ur_event_handle_t> RawEvents = getUrEvents(EventImpls);
flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);
const AdapterPtr &Adapter = Queue->getAdapter();

Adapter->call<UrApiKind::urEnqueueEventsWait>(
Expand Down Expand Up @@ -1397,7 +1390,7 @@ ur_result_t MapMemObject::enqueueImp() {
waitForPreparedHostEvents();
std::vector<EventImplPtr> EventImpls = MPreparedDepsEvents;
std::vector<ur_event_handle_t> RawEvents = getUrEvents(EventImpls);
flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);

ur_event_handle_t UREvent = nullptr;
if (auto Result = callMemOpHelperRet(
Expand Down Expand Up @@ -1480,7 +1473,7 @@ ur_result_t UnMapMemObject::enqueueImp() {
waitForPreparedHostEvents();
std::vector<EventImplPtr> EventImpls = MPreparedDepsEvents;
std::vector<ur_event_handle_t> RawEvents = getUrEvents(EventImpls);
flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);

ur_event_handle_t UREvent = nullptr;
if (auto Result =
Expand Down Expand Up @@ -1590,7 +1583,7 @@ ur_result_t MemCpyCommand::enqueueImp() {
ur_event_handle_t UREvent = nullptr;

auto RawEvents = getUrEvents(EventImpls);
flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);

if (auto Result = callMemOpHelper(
MemoryManager::copy, MSrcAllocaCmd->getSYCLMemObj(),
Expand Down Expand Up @@ -1751,7 +1744,7 @@ ur_result_t MemCpyCommandHost::enqueueImp() {
return UR_RESULT_SUCCESS;
}

flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);

if (auto Result = callMemOpHelper(
MemoryManager::copy, MSrcAllocaCmd->getSYCLMemObj(),
Expand Down Expand Up @@ -2850,7 +2843,7 @@ ur_result_t ExecCGCommand::enqueueImpCommandBuffer() {
// submissions of the command buffer itself will not receive dependencies on
// them, e.g. initial copies from host to device
std::vector<EventImplPtr> EventImpls = MPreparedDepsEvents;
flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);
std::vector<ur_event_handle_t> RawEvents = getUrEvents(EventImpls);
if (!RawEvents.empty()) {
MQueue->getAdapter()->call<UrApiKind::urEventWait>(RawEvents.size(),
Expand Down Expand Up @@ -3130,7 +3123,7 @@ ur_result_t ExecCGCommand::enqueueImpQueue() {
waitForPreparedHostEvents();
std::vector<EventImplPtr> EventImpls = MPreparedDepsEvents;
auto RawEvents = getUrEvents(EventImpls);
flushCrossQueueDeps(EventImpls, MWorkerQueue);
flushCrossQueueDeps(EventImpls);

// We can omit creating a UR event and create a "discarded" event if the
// command has been explicitly marked as not needing an event, e.g. if the
Expand Down
6 changes: 6 additions & 0 deletions sycl/source/detail/scheduler/commands.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -272,6 +272,12 @@ class Command {

void waitForPreparedHostEvents() const;

void flushCrossQueueDeps(const std::vector<EventImplPtr> &EventImpls) {
for (auto &EventImpl : EventImpls) {
EventImpl->flushIfNeeded(MWorkerQueue.get());
}
}

/// Perform glueing of events from different contexts
/// \param DepEvent event this commands should depend on
/// \param Dep optional DepDesc to perform connection of events properly
Expand Down