Skip to content

[SYCL] Remove two-input sub-group shuffles #2614

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
Oct 9, 2020
Merged

[SYCL] Remove two-input sub-group shuffles #2614

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
61f6d17
[SYCL] Remove two-input sub-group shuffles
Pennycook Oct 8, 2020
8a99e4c
[SYCL] Remove unused variable to suppress warning
Pennycook Oct 8, 2020
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
50 changes: 2 additions & 48 deletions sycl/include/CL/sycl/ONEAPI/sub_group.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,7 +191,7 @@ struct sub_group {

template <typename T> T shuffle_down(T x, uint32_t delta) const {
#ifdef __SYCL_DEVICE_ONLY__
return sycl::detail::spirv::SubgroupShuffleDown(x, x, delta);
return sycl::detail::spirv::SubgroupShuffleDown(x, delta);
#else
(void)x;
(void)delta;
Expand All @@ -202,7 +202,7 @@ struct sub_group {

template <typename T> T shuffle_up(T x, uint32_t delta) const {
#ifdef __SYCL_DEVICE_ONLY__
return sycl::detail::spirv::SubgroupShuffleUp(x, x, delta);
return sycl::detail::spirv::SubgroupShuffleUp(x, delta);
#else
(void)x;
(void)delta;
Expand All @@ -222,52 +222,6 @@ struct sub_group {
#endif
}

/* --- two-input shuffles --- */
/* indices in [0 , 2 * sub_group size) */

template <typename T>
__SYCL_DEPRECATED("Two-input sub-group shuffles are deprecated.")
T shuffle(T x, T y, id_type local_id) const {
#ifdef __SYCL_DEVICE_ONLY__
return sycl::detail::spirv::SubgroupShuffleDown(
x, y, (local_id - get_local_id()).get(0));
#else
(void)x;
(void)y;
(void)local_id;
throw runtime_error("Sub-groups are not supported on host device.",
PI_INVALID_DEVICE);
#endif
}

template <typename T>
__SYCL_DEPRECATED("Two-input sub-group shuffles are deprecated.")
T shuffle_down(T current, T next, uint32_t delta) const {
#ifdef __SYCL_DEVICE_ONLY__
return sycl::detail::spirv::SubgroupShuffleDown(current, next, delta);
#else
(void)current;
(void)next;
(void)delta;
throw runtime_error("Sub-groups are not supported on host device.",
PI_INVALID_DEVICE);
#endif
}

template <typename T>
__SYCL_DEPRECATED("Two-input sub-group shuffles are deprecated.")
T shuffle_up(T previous, T current, uint32_t delta) const {
#ifdef __SYCL_DEVICE_ONLY__
return sycl::detail::spirv::SubgroupShuffleUp(previous, current, delta);
#else
(void)previous;
(void)current;
(void)delta;
throw runtime_error("Sub-groups are not supported on host device.",
PI_INVALID_DEVICE);
#endif
}

/* --- sub_group load/stores --- */
/* these can map to SIMD or block read/write hardware where available */

Expand Down
34 changes: 14 additions & 20 deletions sycl/include/CL/sycl/detail/spirv.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -443,16 +443,16 @@ EnableIfNativeShuffle<T> SubgroupShuffleXor(T x, id<1> local_id) {
}

template <typename T>
EnableIfNativeShuffle<T> SubgroupShuffleDown(T x, T y, id<1> local_id) {
EnableIfNativeShuffle<T> SubgroupShuffleDown(T x, id<1> local_id) {
using OCLT = detail::ConvertToOpenCLType_t<T>;
return __spirv_SubgroupShuffleDownINTEL(
OCLT(x), OCLT(y), static_cast<uint32_t>(local_id.get(0)));
OCLT(x), OCLT(x), static_cast<uint32_t>(local_id.get(0)));
}

template <typename T>
EnableIfNativeShuffle<T> SubgroupShuffleUp(T x, T y, id<1> local_id) {
EnableIfNativeShuffle<T> SubgroupShuffleUp(T x, id<1> local_id) {
using OCLT = detail::ConvertToOpenCLType_t<T>;
return __spirv_SubgroupShuffleUpINTEL(OCLT(x), OCLT(y),
return __spirv_SubgroupShuffleUpINTEL(OCLT(x), OCLT(x),
static_cast<uint32_t>(local_id.get(0)));
}

Expand Down Expand Up @@ -488,22 +488,20 @@ EnableIfBitcastShuffle<T> SubgroupShuffleXor(T x, id<1> local_id) {
}

template <typename T>
EnableIfBitcastShuffle<T> SubgroupShuffleDown(T x, T y, id<1> local_id) {
EnableIfBitcastShuffle<T> SubgroupShuffleDown(T x, id<1> local_id) {
using ShuffleT = ConvertToNativeShuffleType_t<T>;
auto ShuffleX = detail::bit_cast<ShuffleT>(x);
auto ShuffleY = detail::bit_cast<ShuffleT>(y);
ShuffleT Result = __spirv_SubgroupShuffleDownINTEL(
ShuffleX, ShuffleY, static_cast<uint32_t>(local_id.get(0)));
ShuffleX, ShuffleX, static_cast<uint32_t>(local_id.get(0)));
return detail::bit_cast<T>(Result);
}

template <typename T>
EnableIfBitcastShuffle<T> SubgroupShuffleUp(T x, T y, id<1> local_id) {
EnableIfBitcastShuffle<T> SubgroupShuffleUp(T x, id<1> local_id) {
using ShuffleT = ConvertToNativeShuffleType_t<T>;
auto ShuffleX = detail::bit_cast<ShuffleT>(x);
auto ShuffleY = detail::bit_cast<ShuffleT>(y);
ShuffleT Result = __spirv_SubgroupShuffleUpINTEL(
ShuffleX, ShuffleY, static_cast<uint32_t>(local_id.get(0)));
ShuffleX, ShuffleX, static_cast<uint32_t>(local_id.get(0)));
return detail::bit_cast<T>(Result);
}

Expand Down Expand Up @@ -550,33 +548,29 @@ EnableIfGenericShuffle<T> SubgroupShuffleXor(T x, id<1> local_id) {
}

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffleDown(T x, T y, id<1> local_id) {
EnableIfGenericShuffle<T> SubgroupShuffleDown(T x, id<1> local_id) {
T Result;
char *XBytes = reinterpret_cast<char *>(&x);
char *YBytes = reinterpret_cast<char *>(&y);
char *ResultBytes = reinterpret_cast<char *>(&Result);
auto ShuffleBytes = [=](size_t Offset, size_t Size) {
uint64_t ShuffleX, ShuffleY, ShuffleResult;
uint64_t ShuffleX, ShuffleResult;
detail::memcpy(&ShuffleX, XBytes + Offset, Size);
detail::memcpy(&ShuffleY, YBytes + Offset, Size);
ShuffleResult = SubgroupShuffleDown(ShuffleX, ShuffleY, local_id);
ShuffleResult = SubgroupShuffleDown(ShuffleX, local_id);
detail::memcpy(ResultBytes + Offset, &ShuffleResult, Size);
};
GenericCall<T>(ShuffleBytes);
return Result;
}

template <typename T>
EnableIfGenericShuffle<T> SubgroupShuffleUp(T x, T y, id<1> local_id) {
EnableIfGenericShuffle<T> SubgroupShuffleUp(T x, id<1> local_id) {
T Result;
char *XBytes = reinterpret_cast<char *>(&x);
char *YBytes = reinterpret_cast<char *>(&y);
char *ResultBytes = reinterpret_cast<char *>(&Result);
auto ShuffleBytes = [=](size_t Offset, size_t Size) {
uint64_t ShuffleX, ShuffleY, ShuffleResult;
uint64_t ShuffleX, ShuffleResult;
detail::memcpy(&ShuffleX, XBytes + Offset, Size);
detail::memcpy(&ShuffleY, YBytes + Offset, Size);
ShuffleResult = SubgroupShuffleUp(ShuffleX, ShuffleY, local_id);
ShuffleResult = SubgroupShuffleUp(ShuffleX, local_id);
detail::memcpy(ResultBytes + Offset, &ShuffleResult, Size);
};
GenericCall<T>(ShuffleBytes);
Expand Down