diff options
author | David Mansell <David.Mansell@arm.com> | 2018-07-06 14:52:52 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-11-02 16:54:10 +0000 |
commit | d93991e290618a685b67506c78090350e6aee43f (patch) | |
tree | 1d5c3b3017cfccd3f0ec3f24e8e11334cf977ce3 /src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp | |
parent | dec32a9edd4b3c6dc55c60d7436e79af6be58c3d (diff) | |
download | ComputeLibrary-d93991e290618a685b67506c78090350e6aee43f.tar.gz |
COMPMID-1380: Pre-work for SVE support.
This patch makes the needed infrastructure changes to allow SVE
kernels to be added later on.
Change-Id: Ide5bccac2f47278e93fff3d648231aee2d5f8c2e
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/139070
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp | 27 |
1 files changed, 15 insertions, 12 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp index b5b07b2c56..be7ead9f48 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp @@ -25,6 +25,8 @@ #ifdef __aarch64__ +#include "../std_transforms_fixed.hpp" + namespace arm_gemm { // Load the actual kernel @@ -39,20 +41,21 @@ public: typedef void (*kern_type)(const int8_t *, const int8_t *, int32_t *, int, int, int); - /* Describes the data layout for A input */ - static const int A_interleave = 4; - static const int A_block = 16; - static const bool A_transpose = false; + /* Kernel blocking parameters */ + static int out_width() { + return 4; + } + + static int out_height() { + return 4; + } - /* Same for B input */ - static const int B_interleave = 4; - static const int B_block = 16; - static const bool B_transpose = true; + static int k_unroll() { + return 16; + } - /* Kernel blocking parameters */ - static const int out_width = 4; - static const int out_height = 4; - static const int k_unroll = 16; + // Use the standard fixed size transforms. + StdTransformsFixed<operand_type, result_type, 4, 4, 16> transforms = {}; kern_type kernel=a64_gemm_s8_4x4; |