add Adopt{Data,Diff} methods to blobs to enable "virtual copying"

diff --git a/include/caffe/blob.hpp b/include/caffe/blob.hpp
index 3185263..7b8989b 100644 (file)
--- a/include/caffe/blob.hpp
+++ b/include/caffe/blob.hpp
@@ -17,7 +17,6 @@ class Blob {
        diff_() {}
   explicit Blob(const int num, const int channels, const int height,
     const int width);
-  virtual ~Blob() {}
   void Reshape(const int num, const int channels, const int height,
     const int width);
   inline int num() const { return num_; }
@@ -52,6 +51,16 @@ class Blob {
     return *(cpu_diff() + offset(n, c, h, w));
   }
 
+  inline const shared_ptr<SyncedMemory>& data() const {
+    CHECK(data_);
+    return data_;
+  }
+
+  inline const shared_ptr<SyncedMemory>& diff() const {
+    CHECK(diff_);
+    return diff_;
+  }
+
   const Dtype* cpu_data() const;
   const Dtype* gpu_data() const;
   const Dtype* cpu_diff() const;
@@ -64,6 +73,9 @@ class Blob {
   void FromProto(const BlobProto& proto);
   void ToProto(BlobProto* proto, bool write_diff = false) const;
 
+  void AdoptData(const Blob& other);
+  void AdoptDiff(const Blob& other);
+
  protected:
   shared_ptr<SyncedMemory> data_;
   shared_ptr<SyncedMemory> diff_;

diff --git a/src/caffe/blob.cpp b/src/caffe/blob.cpp
index 5356d05..e17b8bf 100644 (file)
--- a/src/caffe/blob.cpp
+++ b/src/caffe/blob.cpp
@@ -86,6 +86,18 @@ Dtype* Blob<Dtype>::mutable_gpu_diff() {
 }
 
 template <typename Dtype>
+void Blob<Dtype>::AdoptData(const Blob& other) {
+  CHECK_EQ(count_, other.count());
+  data_ = other.data();
+}
+
+template <typename Dtype>
+void Blob<Dtype>::AdoptDiff(const Blob& other) {
+  CHECK_EQ(count_, other.count());
+  diff_ = other.diff();
+}
+
+template <typename Dtype>
 void Blob<Dtype>::Update() {
   // We will perform update based on where the data is located.
   switch (data_->head()) {

diff --git a/src/caffe/layers/flatten_layer.cpp b/src/caffe/layers/flatten_layer.cpp
index 78e4d0c..83d0502 100644 (file)
--- a/src/caffe/layers/flatten_layer.cpp
+++ b/src/caffe/layers/flatten_layer.cpp
@@ -24,18 +24,14 @@ void FlattenLayer<Dtype>::SetUp(const vector<Blob<Dtype>*>& bottom,
 template <typename Dtype>
 Dtype FlattenLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
       vector<Blob<Dtype>*>* top) {
-  const Dtype* bottom_data = bottom[0]->cpu_data();
-  Dtype* top_data = (*top)[0]->mutable_cpu_data();
-  caffe_copy(count_, bottom_data, top_data);
+  (*top)[0]->AdoptData(*bottom[0]);
   return Dtype(0.);
 }
 
 template <typename Dtype>
 void FlattenLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
       const bool propagate_down, vector<Blob<Dtype>*>* bottom) {
-  const Dtype* top_diff = top[0]->cpu_diff();
-  Dtype* bottom_diff = (*bottom)[0]->mutable_cpu_diff();
-  caffe_copy(count_, top_diff, bottom_diff);
+  (*bottom)[0]->AdoptDiff(*top[0]);
 }
 
 INSTANTIATE_CLASS(FlattenLayer);

diff --git a/src/caffe/layers/flatten_layer.cu b/src/caffe/layers/flatten_layer.cu
index c1f5e48..a28018d 100644 (file)
--- a/src/caffe/layers/flatten_layer.cu
+++ b/src/caffe/layers/flatten_layer.cu
@@ -11,18 +11,14 @@ namespace caffe {
 template <typename Dtype>
 Dtype FlattenLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
       vector<Blob<Dtype>*>* top) {
-  const Dtype* bottom_data = bottom[0]->gpu_data();
-  Dtype* top_data = (*top)[0]->mutable_gpu_data();
-  caffe_gpu_copy(count_, bottom_data, top_data);
+  (*top)[0]->AdoptData(*bottom[0]);
   return Dtype(0.);
 }
 
 template <typename Dtype>
 void FlattenLayer<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top,
       const bool propagate_down, vector<Blob<Dtype>*>* bottom) {
-  const Dtype* top_diff = top[0]->gpu_diff();
-  Dtype* bottom_diff = (*bottom)[0]->mutable_gpu_diff();
-  caffe_gpu_copy(count_, top_diff, bottom_diff);
+  (*bottom)[0]->AdoptDiff(*top[0]);
 }
 
 INSTANTIATE_CLASS(FlattenLayer);

diff --git a/src/caffe/layers/split_layer.cpp b/src/caffe/layers/split_layer.cpp
index 5ca95f3..59a6604 100644 (file)
--- a/src/caffe/layers/split_layer.cpp
+++ b/src/caffe/layers/split_layer.cpp
@@ -30,13 +30,8 @@ void SplitLayer<Dtype>::SetUp(const vector<Blob<Dtype>*>& bottom,
 template <typename Dtype>
 Dtype SplitLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom,
       vector<Blob<Dtype>*>* top) {
-  const Dtype* bottom_data = bottom[0]->cpu_data();
   for (int i = 0; i < top->size(); ++i) {
-    if (i == 0 && (*top)[i] == bottom[0]) {
-      continue;
-    }
-    Dtype* top_data = (*top)[i]->mutable_cpu_data();
-    caffe_copy(count_, bottom_data, top_data);
+    (*top)[i]->AdoptData(*bottom[0]);
   }
   return Dtype(0.);
 }
@@ -45,17 +40,11 @@ template <typename Dtype>
 void SplitLayer<Dtype>::Backward_cpu(const vector<Blob<Dtype>*>& top,
       const bool propagate_down, vector<Blob<Dtype>*>* bottom) {
   if (propagate_down) {
-    const Dtype* top_diff = top[0]->cpu_diff();
-    Dtype* bottom_diff = (*bottom)[0]->mutable_cpu_diff();
-    // Initialize by copying first top blob diff to our diff, unless we're
-    // doing in-place computation for the first blob, in which case the diff is
-    // already initialized.
-    if (top[0] != (*bottom)[0]) {
-      caffe_copy(count_, top_diff, bottom_diff);
-    }
+    (*bottom)[0]->AdoptDiff(*top[0]);
     // Add remaining top blob diffs.
+    Dtype* bottom_diff = (*bottom)[0]->mutable_cpu_diff();
     for (int i = 1; i < top.size(); ++i) {
-      top_diff = top[i]->cpu_diff();
+      const Dtype* top_diff = top[i]->cpu_diff();
       caffe_axpy(count_, Dtype(1.), top_diff, bottom_diff);
     }
   }

diff --git a/src/caffe/layers/split_layer.cu b/src/caffe/layers/split_layer.cu
index 5b2814b..a50e3e8 100644 (file)
--- a/src/caffe/layers/split_layer.cu
+++ b/src/caffe/layers/split_layer.cu
@@ -11,13 +11,8 @@ namespace caffe {
 template <typename Dtype>
 Dtype SplitLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
       vector<Blob<Dtype>*>* top) {
-  const Dtype* bottom_data = bottom[0]->gpu_data();
   for (int i = 0; i < top->size(); ++i) {
-    if (i == 0 && (*top)[i] == bottom[0]) {
-      continue;
-    }
-    Dtype* top_data = (*top)[i]->mutable_gpu_data();
-    caffe_gpu_copy(count_, bottom_data, top_data);
+    (*top)[i]->AdoptData(*bottom[0]);
   }
   return Dtype(0.);
 }
@@ -26,17 +21,11 @@ template <typename Dtype>
 void SplitLayer<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top,
       const bool propagate_down, vector<Blob<Dtype>*>* bottom) {
   if (propagate_down) {
-    const Dtype* top_diff = top[0]->gpu_diff();
-    Dtype* bottom_diff = (*bottom)[0]->mutable_gpu_diff();
-    // Initialize by copying first top blob diff to our diff, unless we're
-    // doing in-place computation for the first blob, in which case the diff is
-    // already initialized.
-    if (top[0] != (*bottom)[0]) {
-      caffe_gpu_copy(count_, top_diff, bottom_diff);
-    }
+    (*bottom)[0]->AdoptDiff(*top[0]);
     // Add remaining top blob diffs.
+    Dtype* bottom_diff = (*bottom)[0]->mutable_gpu_diff();
     for (int i = 1; i < top.size(); ++i) {
-      top_diff = top[i]->gpu_diff();
+      const Dtype* top_diff = top[i]->gpu_diff();
       caffe_gpu_axpy(count_, Dtype(1.), top_diff, bottom_diff);
     }
   }