Skip to content

[SYCL][NFC] More sub_group_mask.hpp cleanup #18975

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 3 commits into from
Sep 12, 2025
+14 −16
Merged

[SYCL][NFC] More sub_group_mask.hpp cleanup #18975

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
eb3962a
[SYCL][NFC] Optimize dependencies of atomic_ref
AlexeySachkov Jun 13, 2025
3b2ec59
Clang-format and missing forward declarations
AlexeySachkov Jun 16, 2025
c229c73
Merge remote-tracking branch 'origin/sycl' into private/asachkov/simp…
AlexeySachkov Sep 11, 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
10 changes: 8 additions & 2 deletions sycl/include/sycl/detail/spirv.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,7 +9,6 @@
#pragma once

#ifdef __SYCL_DEVICE_ONLY__

// Some __spirv_* inrinsics are automatically forward-declared by the compiler,
// but not all of them. For example:
// __spirv_AtomicStore(unsigned long long*, ...)
Expand All @@ -18,7 +17,10 @@
#include <sycl/__spirv/spirv_ops.hpp>
#include <sycl/__spirv/spirv_types.hpp>

#include <sycl/ext/oneapi/experimental/non_uniform_groups.hpp> // for IdToMaskPosition
#include <sycl/access/access.hpp>
#include <sycl/detail/generic_type_traits.hpp>
#include <sycl/id.hpp>
#include <sycl/multi_ptr.hpp>

#if defined(__NVPTX__)
#include <sycl/ext/oneapi/experimental/cuda/masked_shuffles.hpp>
Expand All @@ -33,6 +35,7 @@ struct sub_group;
namespace ext {
namespace oneapi {
struct sub_group;
struct sub_group_mask;
namespace experimental {
template <typename ParentGroup> class fragment;

Expand Down Expand Up @@ -61,6 +64,9 @@ GetMultiPtrDecoratedAs(multi_ptr<FromT, Space, IsDecorated> MPtr) {

template <typename NonUniformGroup>
inline uint32_t IdToMaskPosition(NonUniformGroup Group, uint32_t Id);
template <typename NonUniformGroup>
inline ext::oneapi::sub_group_mask GetMask(NonUniformGroup Group);
inline sycl::vec<unsigned, 4> ExtractMask(ext::oneapi::sub_group_mask Mask);

namespace spirv {

Expand Down
20 changes: 6 additions & 14 deletions sycl/include/sycl/ext/oneapi/sub_group_mask.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,10 +9,10 @@

#include <sycl/detail/helpers.hpp> // for Builder
#include <sycl/detail/memcpy.hpp> // detail::memcpy
#include <sycl/exception.hpp> // for errc, exception
#include <sycl/feature_test.hpp> // for SYCL_EXT_ONEAPI_SUB_GROUP_MASK
#include <sycl/id.hpp> // for id
#include <sycl/marray.hpp> // for marray
#include <sycl/detail/spirv.hpp>
#include <sycl/feature_test.hpp> // for SYCL_EXT_ONEAPI_SUB_GROUP_MASK
#include <sycl/id.hpp> // for id
#include <sycl/marray.hpp> // for marray
#include <sycl/sub_group.hpp>
#include <sycl/vector.hpp> // for vec

Expand Down Expand Up @@ -378,19 +378,11 @@ group_ballot([[maybe_unused]] Group g, [[maybe_unused]] bool predicate) {
#ifdef __SYCL_DEVICE_ONLY__
return sycl::detail::commonGroupBallotImpl(g, predicate);
#else
throw exception{errc::feature_not_supported,
"Sub-group mask is not supported on host device"};
// Groups are not user-constructible, this call should not be reachable from
Copy link
Contributor

Choose a reason for hiding this comment

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

Would it be worth adding llvm_unreachable here?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Two concerns here:

  1. llvm_unreachable comes from LLVM headers that we don't ship (and probably don't want to)
  2. Any "early exits" in form of exception/unreachable may affect host compilation - I'm afraid of unintended side effects like we saw in [SYCL] Fix SYCL_EXTERNAL device code when linking with a static lib #14256

// host and therefore we do nothing here.
#endif
}

} // namespace ext::oneapi
} // namespace _V1
} // namespace sycl

// We have a cyclic dependency with
// sub_group_mask.hpp
// detail/spirv.hpp
// non_uniform_groups.hpp
// "Break" it by including this at the end (instead of beginning). Ideally, we
// should refactor this somehow...
#include <sycl/detail/spirv.hpp>