diff options
| author | Renato Arantes <renato.arantes@arm.com> | 2024-01-26 17:31:18 +0000 |
|---|---|---|
| committer | Renato Barros Arantes <renato.arantes@arm.com> | 2024-03-21 11:15:30 +0000 |
| commit | 36a75dafdbe6d6a3a6f50bd075fe01f5b7dace38 (patch) | |
| tree | 0701d615ef30444b9d0789db691b59b81fd9e86e /support | |
| parent | d2191150736dde66d79eb97e0c8ee506eef3c8fc (diff) | |
| download | ComputeLibrary-36a75dafdbe6d6a3a6f50bd075fe01f5b7dace38.tar.gz | |
[ONCPUML-1451] Add matmul kernel to enable bf16 to bf16 operations via PyTorch® autocast() function
The full range of tests must be added with [MLINFSW-482] epic due to the lack of reordering kernels implemented in Acl.
Co-Authored-By: David Mansell <David.Mansell@arm.com>
Change-Id: I820d316295a1ec94fdc89c37e4144a268f914c36
Signed-off-by: Renato Arantes <renato.arantes@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11169
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'support')
| -rw-r--r-- | support/Bfloat16.h | 18 |
1 files changed, 13 insertions, 5 deletions
diff --git a/support/Bfloat16.h b/support/Bfloat16.h index 17013294e2..02772898a8 100644 --- a/support/Bfloat16.h +++ b/support/Bfloat16.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2020-2022 Arm Limited. + * Copyright (c) 2020-2022,2024 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -21,12 +21,12 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ -#ifndef ARM_COMPUTE_BFLOAT16_H -#define ARM_COMPUTE_BFLOAT16_H +#ifndef ACL_SUPPORT_BFLOAT16_H +#define ACL_SUPPORT_BFLOAT16_H #include <cstdint> #include <cstring> - +#include <ostream> namespace arm_compute { namespace @@ -131,8 +131,16 @@ public: return val; } + bfloat16 &operator+=(float v) + { + value = float_to_bf16(bf16_to_float(value) + v); + return *this; + } + + friend std::ostream &operator<<(std::ostream &os, const bfloat16 &arg); + private: uint16_t value; }; } // namespace arm_compute -#endif /* ARM_COMPUTE_BFLOAT16_H */ +#endif // ACL_SUPPORT_BFLOAT16_H |