Skip to content

[ESIMD] Fix an issue with copy_to that incorrectly copies char buffers in some circumstances #6298

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 13 commits into from
Jun 30, 2022
Merged

[ESIMD] Fix an issue with copy_to that incorrectly copies char buffers in some circumstances #6298

Changes from all commits
Commits
Show all changes
13 commits
Select commit Hold shift + click to select a range
0f53d6d
Fix an issue with copy_to that incorrectly copies char buffers in som…
fineg74 Jun 11, 2022
f65fca7
Fix a build break
fineg74 Jun 13, 2022
516cae1
Fix a test break
fineg74 Jun 13, 2022
73f1a92
Address PR comments
fineg74 Jun 14, 2022
bc09856
Fix clang-format issues
fineg74 Jun 14, 2022
1d8ebea
Address PR comments
fineg74 Jun 21, 2022
748a63a
Merge remote-tracking branch 'origin/sycl' into CopyToFix
fineg74 Jun 22, 2022
68318bf
Address an issue of copying of extra data
fineg74 Jun 22, 2022
c2b6ba1
Fix accessor version of copy_to
fineg74 Jun 22, 2022
5a89a27
Add comments to remove the fix when the issue is fixed on GPU side
fineg74 Jun 24, 2022
6c72d4e
Add handling of unaligned buffer.
fineg74 Jun 28, 2022
0692eeb
Remove missed experimentation changes
fineg74 Jun 28, 2022
56e3512
Address PR comments
fineg74 Jun 29, 2022
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
31 changes: 28 additions & 3 deletions sycl/include/sycl/ext/intel/esimd/memory.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1353,9 +1353,34 @@ void simd_obj_impl<T, N, T1, SFINAE>::copy_to(
if constexpr (RemN == 1) {
Addr[NumChunks * ChunkSize] = Tmp[NumChunks * ChunkSize];
} else if constexpr (RemN == 8 || RemN == 16) {
simd<uint32_t, RemN> Offsets(0u, sizeof(T));
scatter<UT, RemN>(Addr + (NumChunks * ChunkSize), Offsets,
Tmp.template select<RemN, 1>(NumChunks * ChunkSize));
// TODO: GPU runtime may handle scatter of 16 byte elements incorrectly.
// The code below is a workaround which must be deleted once GPU runtime
// is fixed.
if constexpr (sizeof(T) == 1 && RemN == 16) {
if constexpr (Align % OperandSize::DWORD > 0) {
ForHelper<RemN>::unroll([Addr, &Tmp](unsigned Index) {
Addr[Index + NumChunks * ChunkSize] =
Tmp[Index + NumChunks * ChunkSize];
});
} else {
simd_mask_type<8> Pred(0);
simd<int32_t, 8> Vals;
Pred.template select<4, 1>() = 1;
Vals.template select<4, 1>() =
Tmp.template bit_cast_view<int32_t>().template select<4, 1>(
NumChunks * ChunkSize);

simd<uint32_t, 8> Offsets(0u, sizeof(int32_t));
scatter<int32_t, 8>(
reinterpret_cast<int32_t *>(Addr + (NumChunks * ChunkSize)),
Offsets, Vals, Pred);
}
} else {
simd<uint32_t, RemN> Offsets(0u, sizeof(T));
scatter<UT, RemN>(
Addr + (NumChunks * ChunkSize), Offsets,
Tmp.template select<RemN, 1>(NumChunks * ChunkSize));
}
} else {
constexpr int N1 = RemN < 8 ? 8 : RemN < 16 ? 16 : 32;
simd_mask_type<N1> Pred(0);
Expand Down