void BaseConvolutionLayer<Dtype>::weight_gpu_gemm(const Dtype* input,
const Dtype* output, Dtype* weights) {
const Dtype* col_buff = input;
if (!is_1x1_) {
conv_im2col_gpu(input, col_buffer_.mutable_gpu_data());
col_buff = col_buffer_.gpu_data();
}
for (int g = 0; g < group_; ++g) {
caffe_gpu_gemm<Dtype>(CblasNoTrans, CblasTrans, conv_out_channels_ / group_,
kernel_dim_, conv_out_spatial_dim_,
(Dtype)1., output + output_offset_ * g, col_buff + col_offset_ * g,
(Dtype)1., weights + weight_offset_ * g);
}
}
void BaseConvolutionLayer<Dtype>::forward_gpu_gemm(const Dtype* input,
const int_tp input_off,
const Dtype* weights,
Dtype* output,
const int_tp output_off,
bool skip_im2col) {
const Dtype* col_buff = input;
if (this->device_->backend() == BACKEND_CUDA) {
#ifdef USE_CUDA
if (!is_1x1_) {
if (!skip_im2col) {
conv_im2col_gpu(input + input_off, col_buffer()->mutable_gpu_data());
}
col_buff = col_buffer()->gpu_data();
}
for (int_tp g = 0; g < group_; ++g) {
caffe_gpu_gemm<Dtype>(
CblasNoTrans, CblasNoTrans, conv_out_channels_ / group_,
conv_out_spatial_dim_, kernel_dim_, (Dtype) 1.,
weights + weight_offset_ * g,
col_buff + (is_1x1_ ? input_off : 0) + col_offset_ * g, (Dtype) 0.,
output + output_off + output_offset_ * g);
}
#endif // USE_CUDA
} else {
#ifdef USE_GREENTEA
if (!is_1x1_) {
if (!skip_im2col) {
greentea_conv_im2col_gpu(input, input_off,
col_buffer()->mutable_gpu_data(), 0);
}
col_buff = col_buffer()->gpu_data();
}
for (int_tp g = 0; g < group_; ++g) {
greentea_gpu_gemm<Dtype>(this->device_->id(), CblasNoTrans,
CblasNoTrans, conv_out_channels_ / group_,
conv_out_spatial_dim_, kernel_dim_,
(Dtype) 1., (cl_mem) weights, weight_offset_ * g,
(cl_mem) col_buff,
(is_1x1_ ? input_off : 0) + col_offset_ * g,
(Dtype) 0., (cl_mem) output,
output_off + output_offset_ * g);
}
#endif // USE_GREENTEA
}
}
void BaseConvolutionNDLayer<Dtype>::weight_gpu_gemm(const Dtype* input,
const Dtype* output, Dtype* weights) {
const Dtype* col_buff = input;
if (this->device_context_->backend() == BACKEND_CUDA) {
#ifdef USE_CUDA
if (!is_1x1_) {
conv_im2col_gpu(input, col_buffer_.mutable_gpu_data());
col_buff = col_buffer_.gpu_data();
}
for (int g = 0; g < group_; ++g) {
caffe_gpu_gemm<Dtype>(CblasNoTrans, CblasTrans,
conv_out_channels_ / group_,
kernel_dim_ / group_, conv_out_spatial_dim_,
(Dtype)1., output + output_offset_ * g, col_buff + col_offset_ * g,
(Dtype)1., weights + weight_offset_ * g);
}
#endif // USE_CUDA
}
}