Merge pull request BVLC#2016 from shelhamer/share-col-buffer

Share convolution buffers to reduce memory usage
DR08 · Jun 27, 2017 · d76465e · d76465e
2 parents 4efdf7e + 839b050
commit d76465e
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 1 deletion.
diff --git a/include/caffe/layers/base_conv_layer.hpp b/include/caffe/layers/base_conv_layer.hpp
@@ -165,7 +165,9 @@ class BaseConvolutionLayer : public Layer<Dtype> {
   int col_offset_;
   int output_offset_;
 
-  Blob<Dtype> col_buffer_;
+  // N.B. sharing the col buffer reduces memory but interferes with
+  // ND conv and parallelism
+  static Blob<Dtype> col_buffer_;
   Blob<Dtype> bias_multiplier_;
 };
 

diff --git a/src/caffe/layers/base_conv_layer.cpp b/src/caffe/layers/base_conv_layer.cpp
@@ -8,6 +8,9 @@
 
 namespace caffe {
 
+template <typename Dtype>
+Blob<Dtype> BaseConvolutionLayer<Dtype>::col_buffer_;
+
 template <typename Dtype>
 void BaseConvolutionLayer<Dtype>::LayerSetUp(const vector<Blob<Dtype>*>& bottom,
       const vector<Blob<Dtype>*>& top) {