ashokei commented on a change in pull request #10317: [MXNET-264] Improve performance of MKLDNN in small batch sizes. URL: https://github.com/apache/incubator-mxnet/pull/10317#discussion_r179306524
########## File path: src/operator/nn/mkldnn/mkldnn_batch_norm-inl.h ########## @@ -234,20 +234,15 @@ void MKLDNNBatchNormForward(const OpContext &ctx, const BatchNormParam ¶m, DType* weight_ptr = gamma.data().dptr<DType>(); DType* bias_ptr = beta.data().dptr<DType>(); if (!param.fix_gamma) { -#pragma omp parallel for - for (int i = 0; i < channels_; i++) { - weight_buf[i] = weight_ptr[i]; - weight_buf[channels_ + i] = bias_ptr[i]; // bias - } + memcpy(weight_buf, weight_ptr, sizeof(weight_buf[0]) * channels_); + memcpy(&weight_buf[channels_], bias_ptr, sizeof(weight_buf[0]) * channels_); Review comment: got it, thanks! we noticed the same performance issue for smaller networks too (eg: mnist) . Lower OMP_NUM_THREADS (eg: 4 -vs- 36) was giving better performance. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services