diff options
| -rw-r--r-- | src/core/NEON/kernels/NEDepthwiseConvolutionLayerNativeKernel.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/src/core/NEON/kernels/NEDepthwiseConvolutionLayerNativeKernel.cpp b/src/core/NEON/kernels/NEDepthwiseConvolutionLayerNativeKernel.cpp index 90a81b30c9..87315909d8 100644 --- a/src/core/NEON/kernels/NEDepthwiseConvolutionLayerNativeKernel.cpp +++ b/src/core/NEON/kernels/NEDepthwiseConvolutionLayerNativeKernel.cpp @@ -372,7 +372,7 @@ void depthwise_loop_multiplier1_quantized(const ITensor *input, const ITensor *w out_of_bound_vector; const auto weights_vals = wrapper::vload(reinterpret_cast<TW *>(weights_ptr + w * run_info.weights_stride_y) + x); - for(size_t i = 0; i < run_info.x_step; ++i) + for(size_t i = 0; i < element_per_vector; ++i) { acc.at(i) += input_vals[i] * weights_vals[i]; in_sum.at(i) += input_vals[i]; @@ -387,7 +387,7 @@ void depthwise_loop_multiplier1_quantized(const ITensor *input, const ITensor *w } VectorType out_vals = wrapper::vdup_n(static_cast<T>(0), TagType{}); - for(size_t i = 0; i < run_info.x_step; ++i) + for(size_t i = 0; i < element_per_vector; ++i) { acc.at(i) -= in_sum.at(i) * weights_qoffset; acc.at(i) -= we_sum.at(i) * input_qoffset; |