Skip to content
This repository has been archived by the owner on Mar 28, 2023. It is now read-only.

[SYCL] Fix CUDA tests using bfloat16 #1421

Merged
merged 2 commits into from
Nov 30, 2022
Merged

[SYCL] Fix CUDA tests using bfloat16 #1421

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
59377fa
[SYCL] Fix CUDA tests using bfloat16
steffenlarsen Nov 29, 2022
92ff283
Add missing using in element_wise_wi_marray_legacy
steffenlarsen Nov 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
1 change: 1 addition & 0 deletions SYCL/BFloat16/bfloat16_builtins.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@

using namespace sycl;
using namespace sycl::ext::oneapi;
using namespace sycl::ext::oneapi::experimental;

constexpr int N = 60; // divisible by all tested array sizes
constexpr float bf16_eps = 0.00390625;
Expand Down
1 change: 1 addition & 0 deletions SYCL/Matrix/element_wise_all_ops_bf16.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,6 +17,7 @@

using namespace sycl;
using namespace sycl::ext::intel;
using namespace sycl::ext::oneapi;
using namespace sycl::ext::oneapi::experimental::matrix;

#define SG_SZ 16
Expand Down
5 changes: 0 additions & 5 deletions SYCL/Matrix/joint_matrix_tensorcores_legacy.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -66,13 +66,8 @@ T2 matrix_ref_mn(const int &m, const int &n, T1 *A, T1 *B, T2 *C) {
if constexpr (std::is_same<T1, uint16_t>::value) {
for (int k = 0; k < Big_K; k++)
res += make_fp32(A[m * Big_K + k]) * make_fp32(B[k * Big_N + n]);
} else if constexpr (std::is_same<T1, bfloat16>::value) {
for (int k = 0; k < Big_K; k++)
res +=
make_fp32(A[m * Big_K + k].raw()) * make_fp32(B[k * Big_N + n].raw());
} else {
for (int k = 0; k < Big_K; k++)

res +=
static_cast<T2>(A[m * Big_K + k]) * static_cast<T2>(B[k * Big_N + n]);
}
Expand Down