diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4/generic.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4/generic.cpp index 2b846c7f10..89a16f75a0 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4/generic.cpp @@ -56,8 +56,8 @@ void a64_sgemm_native_16x4(const float *A, int lda, const float *B, int ldb, flo float *c_ptr0 = C + (y * ldc); float *c_ptr1 = (activerows > 1) ? c_ptr0 + ldc : dummy_buffer; - float *c_ptr2 = (activerows > 1) ? c_ptr1 + ldc : dummy_buffer; - float *c_ptr3 = (activerows > 1) ? c_ptr2 + ldc : dummy_buffer; + float *c_ptr2 = (activerows > 2) ? c_ptr1 + ldc : dummy_buffer; + float *c_ptr3 = (activerows > 3) ? c_ptr2 + ldc : dummy_buffer; for (int x0=0; x0<N; x0+=16) { const float *a_ptr0 = a_ptr0_base; |