diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_12x8.hpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_12x8.hpp | 27 |
1 files changed, 15 insertions, 12 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_12x8.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_12x8.hpp index cb97270c24..fdc0200435 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_12x8.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_12x8.hpp @@ -27,6 +27,8 @@ #include "arm_gemm.hpp" +#include "../std_transforms_fixed.hpp" + namespace arm_gemm { // Load the actual kernel @@ -40,20 +42,21 @@ 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 = 8; - static const int A_block = 4; - static const bool A_transpose = false; + /* Kernel blocking parameters */ + static int out_width() { + return 12; + } - /* Same for B input */ - static const int B_interleave = 12; - static const int B_block = 4; - static const bool B_transpose = true; + static int out_height() { + return 8; + } - /* Kernel blocking parameters */ - static const int out_width = 12; - static const int out_height = 8; - static const int k_unroll = 4; + static int k_unroll() { + return 4; + } + + // Use the standard fixed size transforms. + StdTransformsFixed<operand_type, result_type, 8, 12, 4> transforms = {}; kern_type kernel = a64_gemm_s8_12x8; |