caffe源码分析--Blob类代码研究
作者:linger
转自须注明转自：http://blog.csdn.net/lingerlanlan/article/details/24379689
数据成员
shared_ptr<SyncedMemory>data_;//data数据，指向SyncedMemory的智能指针
shared_ptr<SyncedMemory>diff_;//表示“差”，用于更新data_
intnum_;
intchannels_;
intheight_;
intwidth_;
intcount_;
构造函数
Blob():num_(0),channels_(0),height_(0),width_(0),count_(0),data_(),diff_(){}
功能：简单的初始化
explicitBlob(constintnum,constintchannels,constintheight,constintwidth);
功能：调用Reshape函数，初始化数据成员
template<typenameDtype>
Blob<Dtype>::Blob(constintnum,constintchannels,constintheight,
constintwidth) {
Reshape(num,channels, height, width);
}
析构函数
virtual~Blob(){}
功能：啥都没做？
voidReshape(constintnum,constintheight,
constintwidth,constintchannels);
功能：初始化数据成员，智能指针指向SyncedMemory对象。此时SyncedMemory对象其实并没有为自己的“数据”申请内存，只是自己“数据”的大小（size）。
template<typenameDtype>
voidBlob<Dtype>::Reshape(constintnum,constintchannels,constintheight,
constintwidth) {
CHECK_GE(num,0);
CHECK_GE(channels,0);
CHECK_GE(height,0);
CHECK_GE(width,0);
num_= num;
channels_= channels;
height_= height;
width_= width;
count_=num_*channels_*height_*width_;
if(count_){
data_.reset(newSyncedMemory(count_*sizeof(Dtype)));
diff_.reset(newSyncedMemory(count_*sizeof(Dtype)));
}else{
data_.reset(reinterpret_cast<SyncedMemory*>(NULL));
diff_.reset(reinterpret_cast<SyncedMemory*>(NULL));
}
}
成员访问函数
功能：就是返回一些成员变量
inlineintnum()const{returnnum_;}
inlineintchannels()const{returnchannels_;}
inlineintheight()const{returnheight_;}
inlineintwidth()const{returnwidth_;}
inlineintcount()const{returncount_;}
inlineintoffset(constintn,constintc = 0, constinth = 0,constintw = 0) const{
return((n * channels_+ c) *height_+ h) *width_+ w;
//计算偏移量，因为数据在内存是一维数组形式的，所以需要计算偏移量来访问
}
“数据”指针返回函数
功能：其实这些函数就是调用SyncedMemory的函数，来返回数据的指针
constDtype*cpu_data()const;
constDtype*gpu_data()const;
constDtype*cpu_diff()const;
constDtype*gpu_diff()const;
Dtype*mutable_cpu_data();
Dtype*mutable_gpu_data();
Dtype*mutable_cpu_diff();
Dtype*mutable_gpu_diff();
inlineDtypedata_at(constintn,constintc,constinth,
constintw)const{
//从cpu访问数据data
return*(cpu_data()+ offset(n, c, h, w));
}
inlineDtypediff_at(constintn,constintc,constinth,
constintw)const{
//从cpu访问数据diff
return*(cpu_diff() + offset(n, c, h, w));
}
函数voidUpdate()
功能：更新data_的数据，就是减去diff_的数据。
template<typenameDtype>
voidBlob<Dtype>::Update(){
//We will perform update based on where the data is located.
switch(data_->head()){
caseSyncedMemory::HEAD_AT_CPU:
//perform computation on CPU
caffe_axpy<Dtype>(count_,Dtype(-1),
reinterpret_cast<constDtype*>(diff_->cpu_data()),
reinterpret_cast<Dtype*>(data_->mutable_cpu_data()));
//在math_functions.cpp可以找到该函数的实现，其实这函数也是封装了mkl的函数。这里调用是为了实现了两个向量的减法。
break;
caseSyncedMemory::HEAD_AT_GPU:
caseSyncedMemory::SYNCED:
//perform computation on GPU
caffe_gpu_axpy<Dtype>(count_,Dtype(-1),
reinterpret_cast<constDtype*>(diff_->gpu_data()),
reinterpret_cast<Dtype*>(data_->mutable_gpu_data()));
//在math_functions.cpp可以找到该函数的实现，其实这函数也是封装了cublas的函数。这里调用是为了实现了两个向量的减法。
break;
default:
LOG(FATAL)<<"Syncedmemnot initialized.";
}
}
函数voidCopyFrom(constBlob<Dtype>&source,boolcopy_diff = false,boolreshape = false);
功能：从source拷贝数据。copy_diff作为标志来区分是拷贝data还是拷贝diff。
template<typenameDtype>
voidBlob<Dtype>::CopyFrom(constBlob&source,boolcopy_diff,boolreshape) {
if(num_!= source.num() || channels_!= source.channels() ||
height_!= source.height() || width_!= source.width()) {
if(reshape) {
Reshape(source.num(),source.channels(), source.height(), source.width());
}else{
LOG(FATAL)<<"Tryingto copy blobs of different sizes.";
}
}
switch(Caffe::mode()){
caseCaffe::GPU:
if(copy_diff){
CUDA_CHECK(cudaMemcpy(diff_->mutable_gpu_data(),source.gpu_diff(),
sizeof(Dtype)*count_,cudaMemcpyDeviceToDevice));
}else{
CUDA_CHECK(cudaMemcpy(data_->mutable_gpu_data(),source.gpu_data(),
sizeof(Dtype)*count_,cudaMemcpyDeviceToDevice));
}
break;
caseCaffe::CPU:
if(copy_diff){
memcpy(diff_->mutable_cpu_data(),source.cpu_diff(),
sizeof(Dtype)*count_);
}else{
memcpy(data_->mutable_cpu_data(),source.cpu_data(),
sizeof(Dtype)*count_);
}
break;
default:
LOG(FATAL)<<"Unknowncaffemode.";
}
}
函数voidFromProto(constBlobProto&proto);
功能：从proto读数据进来，其实就是反序列化
template<typenameDtype>
voidBlob<Dtype>::FromProto(constBlobProto&proto){
Reshape(proto.num(),proto.channels(),proto.height(),proto.width());
//copy data
Dtype*data_vec = mutable_cpu_data();
for(inti = 0; i < count_;++i) {
data_vec[i]=proto.data(i);
}
if(proto.diff_size()> 0) {
Dtype*diff_vec = mutable_cpu_diff();
for(inti = 0; i < count_;++i) {
diff_vec[i]=proto.diff(i);
}
}
}
函数voidToProto(BlobProto*proto,boolwrite_diff = false)const;
功能：序列化到proto保存
template<typenameDtype>
voidBlob<Dtype>::ToProto(BlobProto*proto,boolwrite_diff)const{
proto->set_num(num_);
proto->set_channels(channels_);
proto->set_height(height_);
proto->set_width(width_);
proto->clear_data();
proto->clear_diff();
constDtype*data_vec = cpu_data();
for(inti = 0; i < count_;++i) {
proto->add_data(data_vec[i]);
}
if(write_diff) {
constDtype*diff_vec = cpu_diff();
for(inti = 0; i < count_;++i) {
proto->add_diff(diff_vec[i]);
}
}
}