diff options
| author | Ramy Elgammal <ramy.elgammal@arm.com> | 2023-03-09 21:15:37 +0000 |
|---|---|---|
| committer | Gunes Bayir <gunes.bayir@arm.com> | 2023-03-17 11:00:57 +0000 |
| commit | 2b6ebfe4270b06b09e45f306e8384950aeca7e4e (patch) | |
| tree | 44a65551aa32352a57090bf550b96cbf5b54222f /src/core/CL/cl_kernels/tile_helpers.h | |
| parent | 40aab11ff88ed432d5028478e531f8d0fc404d4c (diff) | |
| download | ComputeLibrary-2b6ebfe4270b06b09e45f306e8384950aeca7e4e.tar.gz | |
Implement OpenCL MatMul for Lhs NT Rhs T/NT FP32/16
- Implement ClNativeMatMulKernel class
- Implement opencl kernel for LHS non-transposed and RHS non-transposed
- Implement opencl kernel for LHS non-transposed and RHS transposed
- Add test fixture and dataset for matmul
- Implement transpose_tensor() for reference implementation to transpose high dimensional tensors
Resolves: COMPMID-5944, COMPMID-5951
Co-authored-by: Gunes Bayir <gunes.bayir@arm.com>
Co-authored-by: Ramy Elgammal <ramy.elgammal@arm.com>
Change-Id: I1d5b8978f41be27baddb3153ade880472141573f
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9333
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/cl_kernels/tile_helpers.h')
| -rw-r--r-- | src/core/CL/cl_kernels/tile_helpers.h | 21 |
1 files changed, 18 insertions, 3 deletions
diff --git a/src/core/CL/cl_kernels/tile_helpers.h b/src/core/CL/cl_kernels/tile_helpers.h index 1e4dddd2db..5d397ad333 100644 --- a/src/core/CL/cl_kernels/tile_helpers.h +++ b/src/core/CL/cl_kernels/tile_helpers.h @@ -21,8 +21,8 @@ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. */ -#ifndef SRC_CORE_CL_CL_KERNELS_TILE_HELPERS -#define SRC_CORE_CL_CL_KERNELS_TILE_HELPERS +#ifndef ACL_SRC_CORE_CL_CL_KERNELS_TILE_HELPERS +#define ACL_SRC_CORE_CL_CL_KERNELS_TILE_HELPERS // *INDENT-OFF* // clang-format off @@ -1282,6 +1282,21 @@ }) \ } +#define T_MMUL_NT_NT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_NT_##LHS_DATA_TYPE##_##RHS_DATA_TYPE##_##DST_DATA_TYPE(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) +#define T_MMUL_NT_NT_float_float_float(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_NT_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) +#define T_MMUL_NT_NT_half_half_float(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_NT_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) +#define T_MMUL_NT_NT_half_half_half(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_NT_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) +#define T_MMUL_NT_NT_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) \ + { \ + LOOP_UNROLLING(int, _m, 0, 1, M0, \ + { \ + LOOP_UNROLLING(int, _k, 0, 1, K0, \ + { \ + dst[_m].v = fma((DST_DATA_TYPE)(lhs[_m].s[_k]), (rhs[_k].v), dst[_m].v); \ + }) \ + }) \ + } + #define T_MMUL_NT_T_INTEGER8(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) \ ({ \ LOOP_UNROLLING(int, _m, 0, 1, M0, \ @@ -1293,4 +1308,4 @@ }) \ }) -#endif /* SRC_CORE_CL_CL_KERNELS_TILE_HELPERS */ +#endif /* ACL_SRC_CORE_CL_CL_KERNELS_TILE_HELPERS */ |