diff options
| author | ramelg01 <ramy.elgammal@arm.com> | 2021-10-29 10:52:53 +0100 |
|---|---|---|
| committer | ramy.elgammal <ramy.elgammal@arm.com> | 2021-11-04 11:10:56 +0000 |
| commit | 6049edadf0c89a026b3fcd1927ee7531d3c40278 (patch) | |
| tree | c12fcea637e41cdb9e1f72dc734e4a87d0b31981 /src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h | |
| parent | 71cbd28b7cf5115b0451d43e5c84cce4ae4d8ec7 (diff) | |
| download | ComputeLibrary-6049edadf0c89a026b3fcd1927ee7531d3c40278.tar.gz | |
Add PRelu to supported PostOps in:
- ClGemmMatrixMultiplyReshapedKernel
- ClGemmMatrixMultiplyNativeKernel
- ClGemmMatrixMultiplyReshapedOnlyRhsKernel
Resolves: COMPMID-4713
Change-Id: I3adcb1b3d4af37ebcbc3bee19cc1845885d08600
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6553
Reviewed-by: SiCong Li <sicong.li@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h')
| -rw-r--r-- | src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h | 12 |
1 files changed, 12 insertions, 0 deletions
diff --git a/src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h b/src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h index 9ddf51a13c..b584251c2a 100644 --- a/src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h +++ b/src/core/CL/cl_kernels/common/experimental/gemm_fused_post_ops/fp_elementwise_op_helpers.h @@ -45,7 +45,13 @@ #if VEC_SIZE == 1 #define PRELU_X_POS_0(x, y) (x > 0 ? x : x * y) #else // VEC_SIZE == 1 + +#if defined(MIXED_PRECISION) +#define PRELU_X_POS_0(x, y) (select(y * x, x, CONVERT((x > (DATA_TYPE_ACCUMULATOR)0), SELECT_VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, VEC_SIZE)))) +#else // MIXED_PRECISION #define PRELU_X_POS_0(x, y) (select(y * x, x, CONVERT((x > (DATA_TYPE)0), SELECT_VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)))) +#endif // MIXED_PRECISION + #endif // VEC_SIZE == 1 #define DIV_X_POS_0(x, y) (x / y) #define AND_X_POS_0(x, y) (CONVERT((x && y), VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)) & ((VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE))1)) @@ -60,7 +66,13 @@ #if VEC_SIZE == 1 #define PRELU_X_POS_1(x, y) (y > 0 ? y : y * x) #else // VEC_SIZE == 1 + +#if defined(MIXED_PRECISION) +#define PRELU_X_POS_1(x, y) (select(x * y, y, CONVERT((y > (DATA_TYPE_ACCUMULATOR)0), SELECT_VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, VEC_SIZE)))) +#else // MIXED_PRECISION #define PRELU_X_POS_1(x, y) (select(x * y, y, CONVERT((y > (DATA_TYPE)0), SELECT_VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)))) +#endif // MIXED_PRECISION + #endif // VEC_SIZE == 1 #define DIV_X_POS_1(x, y) (y / x) #define AND_X_POS_1(x, y) AND_X_POS_0(x, y) |