Skip to content

[SYCL] Add forward declarations for Subgroup Shuffle functions #8450

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
Mar 1, 2023
Merged

[SYCL] Add forward declarations for Subgroup Shuffle functions #8450

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f396050
[SYCL] Add forward declarations for Subgroup Shuffle functions
maksimsab Feb 24, 2023
f568ca2
[SYCL] use std::enable_if_t instead of detail::enable_if_t
maksimsab Feb 27, 2023
b9bde57
[SYCL] use detail::is_arithmetic::value instead of detail::is_arithme…
maksimsab Feb 27, 2023
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
110 changes: 68 additions & 42 deletions sycl/include/sycl/detail/spirv.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -537,6 +537,49 @@ using EnableIfVectorShuffle =
std::enable_if_t<detail::is_vector_arithmetic<T>::value, T>;
#endif // ifndef __NVPTX__

// Bitcast shuffles can be implemented using a single SubgroupShuffle
// intrinsic, but require type-punning via an appropriate integer type
#ifndef __NVPTX__
template <typename T>
using EnableIfBitcastShuffle =
std::enable_if_t<!detail::is_arithmetic<T>::value &&
(std::is_trivially_copyable_v<T> &&
(sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4 ||
sizeof(T) == 8)),
T>;
#else
template <typename T>
using EnableIfBitcastShuffle =
std::enable_if_t<!(std::is_integral_v<T> &&
(sizeof(T) <= sizeof(int32_t))) &&
!detail::is_vector_arithmetic<T>::value &&
(std::is_trivially_copyable_v<T> &&
(sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4)),
T>;
#endif // ifndef __NVPTX__

// Generic shuffles may require multiple calls to SubgroupShuffle
// intrinsics, and should use the fewest shuffles possible:
// - Loop over 64-bit chunks until remaining bytes < 64-bit
// - At most one 32-bit, 16-bit and 8-bit chunk left over
#ifndef __NVPTX__
template <typename T>
using EnableIfGenericShuffle =
std::enable_if_t<!detail::is_arithmetic<T>::value &&
!(std::is_trivially_copyable_v<T> &&
(sizeof(T) == 1 || sizeof(T) == 2 ||
sizeof(T) == 4 || sizeof(T) == 8)),
T>;
#else
template <typename T>
using EnableIfGenericShuffle = std::enable_if_t<
!(std::is_integral<T>::value && (sizeof(T) <= sizeof(int32_t))) &&
!detail::is_vector_arithmetic<T>::value &&
!(std::is_trivially_copyable_v<T> &&
(sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4)),
T>;
#endif

#ifdef __NVPTX__
inline uint32_t membermask() {
// use a full mask as sync operations are required to be convergent and exited
Expand All @@ -545,6 +588,31 @@ inline uint32_t membermask() {
}
#endif

// Forward declarations for template overloadings
template <typename T>
EnableIfBitcastShuffle<T> SubgroupShuffle(T x, id<1> local_id);

template <typename T>
EnableIfBitcastShuffle<T> SubgroupShuffleXor(T x, id<1> local_id);

template <typename T>
EnableIfBitcastShuffle<T> SubgroupShuffleDown(T x, id<1> local_id);

template <typename T>
EnableIfBitcastShuffle<T> SubgroupShuffleUp(T x, id<1> local_id);

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffle(T x, id<1> local_id);

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffleXor(T x, id<1> local_id);

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffleDown(T x, id<1> local_id);

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffleUp(T x, id<1> local_id);

template <typename T>
EnableIfNativeShuffle<T> SubgroupShuffle(T x, id<1> local_id) {
#ifndef __NVPTX__
Expand Down Expand Up @@ -623,26 +691,6 @@ EnableIfVectorShuffle<T> SubgroupShuffleUp(T x, uint32_t delta) {
return result;
}

// Bitcast shuffles can be implemented using a single SubgroupShuffle
// intrinsic, but require type-punning via an appropriate integer type
#ifndef __NVPTX__
template <typename T>
using EnableIfBitcastShuffle =
detail::enable_if_t<!detail::is_arithmetic<T>::value &&
(std::is_trivially_copyable<T>::value &&
(sizeof(T) == 1 || sizeof(T) == 2 ||
sizeof(T) == 4 || sizeof(T) == 8)),
T>;
#else
template <typename T>
using EnableIfBitcastShuffle = detail::enable_if_t<
!(std::is_integral<T>::value && (sizeof(T) <= sizeof(int32_t))) &&
!detail::is_vector_arithmetic<T>::value &&
(std::is_trivially_copyable<T>::value &&
(sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4)),
T>;
#endif

template <typename T>
using ConvertToNativeShuffleType_t = select_cl_scalar_integral_unsigned_t<T>;

Expand Down Expand Up @@ -699,28 +747,6 @@ EnableIfBitcastShuffle<T> SubgroupShuffleUp(T x, uint32_t delta) {
return bit_cast<T>(Result);
}

// Generic shuffles may require multiple calls to SubgroupShuffle
// intrinsics, and should use the fewest shuffles possible:
// - Loop over 64-bit chunks until remaining bytes < 64-bit
// - At most one 32-bit, 16-bit and 8-bit chunk left over
#ifndef __NVPTX__
template <typename T>
using EnableIfGenericShuffle =
detail::enable_if_t<!detail::is_arithmetic<T>::value &&
!(std::is_trivially_copyable<T>::value &&
(sizeof(T) == 1 || sizeof(T) == 2 ||
sizeof(T) == 4 || sizeof(T) == 8)),
T>;
#else
template <typename T>
using EnableIfGenericShuffle = detail::enable_if_t<
!(std::is_integral<T>::value && (sizeof(T) <= sizeof(int32_t))) &&
!detail::is_vector_arithmetic<T>::value &&
!(std::is_trivially_copyable<T>::value &&
(sizeof(T) == 1 || sizeof(T) == 2 || sizeof(T) == 4)),
T>;
#endif

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffle(T x, id<1> local_id) {
T Result;
Expand Down