Skip to content

Handler-less kernel submit API #19294

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

Draft
wants to merge 19 commits into
base: sycl
Choose a base branch
from
Draft

Handler-less kernel submit API #19294

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
3223842
[SYCL] Handler-less kernel submit API
slawekptak Jul 3, 2025
fde19ca
Fix formatting
slawekptak Jul 3, 2025
13424de
Fix formatting
slawekptak Jul 4, 2025
fbc789d
Change the ExtendedSubmissionInfo to KernelRuntimeInfo,
slawekptak Jul 7, 2025
591b3ec
Added copy/move constructor and assignment operator
slawekptak Jul 8, 2025
d235b7c
Merge branch 'sycl' into no_handler_lib_entry
slawekptak Jul 8, 2025
6641601
Add a no event submit and no handler compile flag
slawekptak Jul 11, 2025
0f41d5a
Merge branch 'sycl' into no_handler_lib_entry
slawekptak Jul 14, 2025
a6e711e
Added a new configure option to build no handler submit path, changed
slawekptak Jul 14, 2025
9c8040e
Host task dependency test
slawekptak Jul 17, 2025
31cbdb9
Add a check for special captures
slawekptak Jul 18, 2025
c5cd091
Merge branch 'sycl' into no_handler_lib_entry
slawekptak Aug 11, 2025
998d592
Switch to the common kernel wrappers, fix the KRInfo function call
slawekptak Aug 11, 2025
4000c07
Enable no handler in the preview lib build, add no handler unit
slawekptak Aug 12, 2025
f8e9cd6
Merge branch 'sycl' into no_handler_lib_entry
slawekptak Aug 12, 2025
01af8bb
Unused argument fix and IsTopCodeLoc assignment
slawekptak Aug 12, 2025
4469e59
Implemented the barrier and un-enqueued commands synchronization
slawekptak Aug 13, 2025
ac1a5cf
Fix formatting
slawekptak Aug 13, 2025
5865f3a
Fixed #ifdef, added comment to a new function.
slawekptak Aug 13, 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
95 changes: 95 additions & 0 deletions sycl/include/sycl/queue.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,6 +149,37 @@ class __SYCL_EXPORT SubmissionInfo {
ext::oneapi::experimental::event_mode_enum::none;
};

using KernelParamDescGetterFuncPtr = detail::kernel_param_desc_t (*)(int);

class __SYCL_EXPORT ExtendedSubmissionInfo : public SubmissionInfo {
public:
ExtendedSubmissionInfo() {}

std::string_view &KernelName() { return MKernelName; }
std::unique_ptr<detail::HostKernelBase> &HostKernel() { return MHostKernel; }
const std::unique_ptr<detail::HostKernelBase> &HostKernel() const {
return MHostKernel;
}
int &KernelNumArgs() { return MKernelNumArgs; }
KernelParamDescGetterFuncPtr &KernelParamDescGetter() {
return MKernelParamDescGetter;
}
bool &KernelIsESIMD() { return MKernelIsESIMD; }
bool &KernelHasSpecialCaptures() { return MKernelHasSpecialCaptures; }
detail::KernelNameBasedCacheT *&KernelNameBasedCachePtr() {
return MKernelNameBasedCachePtr;
}

private:
std::string_view MKernelName;
std::unique_ptr<detail::HostKernelBase> MHostKernel;
int MKernelNumArgs = 0;
KernelParamDescGetterFuncPtr MKernelParamDescGetter = nullptr;
bool MKernelIsESIMD = false;
bool MKernelHasSpecialCaptures = true;
detail::KernelNameBasedCacheT *MKernelNameBasedCachePtr = nullptr;
};

} // namespace v1
} // namespace detail

Expand Down Expand Up @@ -3609,6 +3640,38 @@ class __SYCL_EXPORT queue : public detail::OwnerLessBase<queue> {
}
}

template <int Dims, typename LambdaArgType> struct TransformUserItemType {
using type = std::conditional_t<
std::is_convertible_v<nd_item<Dims>, LambdaArgType>, nd_item<Dims>,
std::conditional_t<std::is_convertible_v<item<Dims>, LambdaArgType>,
item<Dims>, LambdaArgType>>;
};

template <typename PropertiesT, typename KernelName, typename KernelType,
int Dims>
void ProcessExtendedSubmitProperties(
PropertiesT Props, const KernelType &KernelFunc,
detail::v1::ExtendedSubmissionInfo &SI) const {
ProcessSubmitProperties(Props, SI);

using NameT =
typename detail::get_kernel_name_t<KernelName, KernelType>::name;
using LambdaArgType = sycl::detail::lambda_arg_type<KernelType, item<Dims>>;
using TransformedArgType = std::conditional_t<
std::is_integral<LambdaArgType>::value && Dims == 1, item<Dims>,
typename TransformUserItemType<Dims, LambdaArgType>::type>;

SI.HostKernel().reset(
new detail::HostKernel<KernelType, TransformedArgType, Dims>(
std::forward<KernelType>(KernelFunc)));
SI.KernelName() = detail::getKernelName<NameT>();
SI.KernelNumArgs() = detail::getKernelNumParams<NameT>();
SI.KernelParamDescGetter() = &(detail::getKernelParamDesc<NameT>);
SI.KernelIsESIMD() = detail::isKernelESIMD<NameT>();
SI.KernelHasSpecialCaptures() = detail::hasSpecialCaptures<NameT>();
SI.KernelNameBasedCachePtr() = detail::getKernelNameBasedCache<NameT>();
}

#ifndef __INTEL_PREVIEW_BREAKING_CHANGES
/// TODO: Unused. Remove these when ABI-break window is open.
/// Not using `type_erased_cgfo_ty` on purpose.
Expand Down Expand Up @@ -3680,6 +3743,21 @@ class __SYCL_EXPORT queue : public detail::OwnerLessBase<queue> {
const detail::code_location &CodeLoc,
bool IsTopCodeLoc) const;

event submit_with_event_impl(
Copy link
Contributor

Choose a reason for hiding this comment

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

What about eventless? It is not done yet, right?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Yes, I think it would be similar, so I've skipped it for now.

Copy link
Contributor

Choose a reason for hiding this comment

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

What is the disadvantage of returning optional<event> and having somewhere (probably, in SubmissionInfo, as this is mode of submission) a flag, pointing out is it event or eventless mode? I think about bunch of functions that pass arguments by chain and about duplicating them (for event and for eventless) and this is not looks good. What do you think?

Copy link
Contributor

Choose a reason for hiding this comment

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

I am not sure if returning std::optional<event> is a good idea because of ABI concerns. It might not have a stable ABI across compiler versions or even different standard libraries (libstdc++ vs libc++).

Copy link
Contributor

Choose a reason for hiding this comment

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

I am not sure if returning std::optional<event> is a good idea because of ABI concerns. It might not have a stable ABI across compiler versions or even different standard libraries (libstdc++ vs libc++).

Yes, good point.

Copy link
Contributor

Choose a reason for hiding this comment

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

Probably, sycl::detail::optional might be considered.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

sycl::detail::optional might work, good idea

Copy link
Contributor

Choose a reason for hiding this comment

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

But still we need to care about the stable layout of the sycl::detail::optional. I am not sure that we are doing it today.

I think having two versions (that return sycl::event and return void) might be a good alternative.

Copy link
Contributor

Choose a reason for hiding this comment

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

We don't generally do that and we rely on backward compatibility guarantees of the C++ library we use (GNU libstdc++ on Linux/MSVC on Windows). The only exception is pre-C++11 ABI of GNU libstdc++ that pyTorch used to use (see https://gcc.gnu.org/onlinedocs/libstdc++/manual/using_dual_abi.html). I don't see std::optional listed on that page, so we should be safe to use it.

Another possible caveat is if some STL's implementation of it isn't is_sycl_device_copyable. I think that might have been a reason why we added sycl::detail::optional (or maybe it was simply added when we used C++14, `std::optional' is C++17 and above).

Anyway, unless you have a known case when it doesn't work, the current approach in the rest of the project is to use std::optional, AFAIK.

const nd_range<1> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) const;

event submit_with_event_impl(
const nd_range<2> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) const;

event submit_with_event_impl(
const nd_range<3> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) const;

/// A template-free version of submit_without_event as const member function.
void submit_without_event_impl(const detail::type_erased_cgfo_ty &CGH,
const detail::v1::SubmissionInfo &SubmitInfo,
Expand Down Expand Up @@ -3763,6 +3841,23 @@ class __SYCL_EXPORT queue : public detail::OwnerLessBase<queue> {
TlsCodeLocCapture.isToplevel());
}

template <bool UseFallbackAssert, typename PropertiesT, typename KernelName,
typename KernelType, int Dims>
event submit_with_event(PropertiesT Props, const nd_range<Dims> Range,
const KernelType &KernelFunc,
const detail::code_location &CodeLoc =
detail::code_location::current()) const {
detail::tls_code_loc_t TlsCodeLocCapture(CodeLoc);
detail::v1::ExtendedSubmissionInfo SI{};
ProcessExtendedSubmitProperties<KernelName, KernelType>(Props, KernelFunc,
SI);

// TODO UseFallbackAssert

return submit_with_event_impl(Range, SI, TlsCodeLocCapture.query(),
TlsCodeLocCapture.isToplevel());
}

/// Submits a command group function object to the queue, in order to be
/// scheduled for execution on the device.
///
Expand Down
33 changes: 33 additions & 0 deletions sycl/source/detail/queue_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,6 +370,39 @@ class queue_impl : public std::enable_shared_from_this<queue_impl> {
return createSyclObjFromImpl<event>(ResEvent);
}

event
submit_with_event(const nd_range<1> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) {
(void)Range;
(void)ExtSubmitInfo;
(void)CodeLoc;
(void)IsTopCodeLoc;
return event();
}

event
submit_with_event(const nd_range<2> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) {
(void)Range;
(void)ExtSubmitInfo;
(void)CodeLoc;
(void)IsTopCodeLoc;
return event();
}

event
submit_with_event(const nd_range<3> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) {
(void)Range;
(void)ExtSubmitInfo;
(void)CodeLoc;
(void)IsTopCodeLoc;
return event();
}

void submit_without_event(const detail::type_erased_cgfo_ty &CGF,
const v1::SubmissionInfo &SubmitInfo,
const detail::code_location &Loc,
Expand Down
21 changes: 21 additions & 0 deletions sycl/source/queue.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -312,6 +312,27 @@ event queue::submit_with_event_impl(
return impl->submit_with_event(CGH, SubmitInfo, CodeLoc, IsTopCodeLoc);
}

event queue::submit_with_event_impl(
const nd_range<1> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) const {
return impl->submit_with_event(Range, ExtSubmitInfo, CodeLoc, IsTopCodeLoc);
}

event queue::submit_with_event_impl(
const nd_range<2> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) const {
return impl->submit_with_event(Range, ExtSubmitInfo, CodeLoc, IsTopCodeLoc);
}

event queue::submit_with_event_impl(
const nd_range<3> Range,
const detail::v1::ExtendedSubmissionInfo &ExtSubmitInfo,
const detail::code_location &CodeLoc, bool IsTopCodeLoc) const {
return impl->submit_with_event(Range, ExtSubmitInfo, CodeLoc, IsTopCodeLoc);
}

void queue::submit_without_event_impl(
const detail::type_erased_cgfo_ty &CGH,
const detail::v1::SubmissionInfo &SubmitInfo,
Expand Down
Loading