diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-07-02 20:02:20 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-07-06 16:51:32 +0000 |
commit | 5aa1a0b7ca5eed010e4b297a95b1c4851f741328 (patch) | |
tree | ba882de9e86589dfdd33937d538a89bbdf01c40e /src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp | |
parent | 42550c039105597ff6acd4e5efc0ee3c7c20b08e (diff) | |
download | ComputeLibrary-5aa1a0b7ca5eed010e4b297a95b1c4851f741328.tar.gz |
COMPID-3324: Clean GEMM kernels
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I170de1671e061a78740caee31fb4a1b8642c1369
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3505
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp | 21 |
1 files changed, 15 insertions, 6 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp index 9f06a48ff5..797ab74498 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_hybrid_s8s32_dot_16x4/generic.cpp @@ -1,5 +1,5 @@ /* - * Copyright (c) 2018-2019 Arm Limited. + * Copyright (c) 2018-2020 Arm Limited. * * SPDX-License-Identifier: MIT * @@ -32,9 +32,7 @@ namespace arm_gemm { -void a64_hybrid_s8s32_dot_16x4(const int8_t *A, int lda, const int8_t *B, int32_t *C, int ldc, int M, int N, int K, const int32_t *bias, Activation act, bool append) { - UNUSED(bias); - UNUSED(act); +void a64_hybrid_s8s32_dot_16x4(const int8_t *A, int lda, const int8_t *B, int32_t *C, int ldc, int M, int N, int K, const int32_t *, Activation , bool append) { const int K_stride = ((K + 3) / 4) * 4; const long loops_count = ((K + 16) / 32) - 1; K -= loops_count * 32; @@ -43,12 +41,23 @@ void a64_hybrid_s8s32_dot_16x4(const int8_t *A, int lda, const int8_t *B, int32_ const long blocks_count = K / 4; const long odds_count = K - (blocks_count * 4); - for (int y=0; y<M; y+=4) { + int rows_to_compute; + + for (int y=0; y<M; y+=rows_to_compute) { const int8_t * const a_ptr0_base = A + (y * lda); const unsigned long ldab = lda * sizeof(int8_t); int32_t *c_ptr0 = C + (y * ldc); + rows_to_compute = M-y; + if (rows_to_compute > 4) { + if (rows_to_compute % 4) { + rows_to_compute = 4 - 1; + } else { + rows_to_compute = 4; + } + } + for (int x0=0; x0<N; x0+=16ul) { const long width = std::min((unsigned long)N-x0, 16ul); long loops = loops_count; @@ -72,7 +81,7 @@ void a64_hybrid_s8s32_dot_16x4(const int8_t *A, int lda, const int8_t *B, int32_ c_ptr0 = result_buffer; } - switch(M-y) { + switch(rows_to_compute) { case 1: __asm __volatile ( "cbnz %[append], 1f\n" |