diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-10-14 19:03:09 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-10-23 12:08:12 +0000 |
commit | 48b3ef89de5f21a0169d8416e3d54081f82c7bf8 (patch) | |
tree | f857d733ccf446c704823dc7ac796a96eb55095e /src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp | |
parent | 1dce3101ef8d77c8cf0af7dfd4af6595a0136b91 (diff) | |
download | ComputeLibrary-48b3ef89de5f21a0169d8416e3d54081f82c7bf8.tar.gz |
COMPMID-2577: Fuse bias addition and activation in gemm assembly kernels
Change-Id: I7f52112d2d05b1ea3d3f3d4b19b8eafab05d6c44
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2141
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez <pablo.tello@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp | 27 |
1 files changed, 21 insertions, 6 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp index 560593958a..1ce934d413 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_fp32_mla_16x4.hpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2019 Arm Limited. + * Copyright (c) 2018-2019 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -32,8 +32,8 @@ namespace arm_gemm { // Actual kernel implementations -void a64_hybrid_fp32_mla_16x4(const float *, int, const float *, float *, int, float, int, int, int); -void a64_hybrid_fp32_mla_16x4_a55(const float *, int, const float *, float *, int, float, int, int, int); +void a64_hybrid_fp32_mla_16x4(const float *, int, const float *, float *, int, int, int, int, const float *, Activation, bool); +void a64_hybrid_fp32_mla_16x4_a55(const float *, int, const float *, float *, int, int, int, int, const float *, Activation, bool); class hybrid_fp32_mla_16x4 { @@ -41,10 +41,10 @@ public: typedef float operand_type; typedef float result_type; - typedef void (*kern_type)(const float *, int, const float *, float *, int, float, int, int, int); + typedef void (*kern_type)(const float *, int, const float *, float *, int, int, int, int, const float *, Activation, bool); /* Kernel blocking parameters */ - static unsigned int out_height() + static constexpr unsigned int out_height() { return 4; } @@ -54,11 +54,26 @@ public: return 16; } - static unsigned int k_unroll() + static constexpr unsigned int k_unroll() { return 1; } + static constexpr bool supports_append() + { + return true; + } + + static constexpr bool supports_bias() + { + return true; + } + + static constexpr bool supports_activation() + { + return true; + } + StdTransformsFixed<operand_type, result_type, 4, 16, 1> transforms = {}; // Default to the generic kernel |