diff options
| -rw-r--r-- | src/caffe/acl_layer.cpp | 2 | ||||
| -rw-r--r-- | src/caffe/layers/acl_batch_norm_layer.cpp | 6 | ||||
| -rw-r--r-- | src/caffe/layers/acl_concat_layer.cpp | 1 |
3 files changed, 5 insertions, 4 deletions
diff --git a/src/caffe/acl_layer.cpp b/src/caffe/acl_layer.cpp index 84f201d3..879b6701 100644 --- a/src/caffe/acl_layer.cpp +++ b/src/caffe/acl_layer.cpp @@ -116,7 +116,7 @@ int BaseTensor<ACLTensor>::tensor_copy(void * mem,bool toTensor) #endif //USE_PROFILING arm_compute::Window window; ACLTensor* tensor=this; - window.use_tensor_dimensions(tensor->info(), /* first_dimension =*/Window::DimY); // Iterate through the rows (not each element) + window.use_tensor_dimensions(tensor->info()->tensor_shape(), /* first_dimension =*/Window::DimY); // Iterate through the rows (not each element) int width = tensor->info()->tensor_shape()[0]; //->dimension(0); //window.x().end() - window.x().start(); // + 1; int height = tensor->info()->tensor_shape()[1]; //->dimension(1); //window.y().end() - window.y().start(); // + 1; int deepth = tensor->info()->tensor_shape()[2]; diff --git a/src/caffe/layers/acl_batch_norm_layer.cpp b/src/caffe/layers/acl_batch_norm_layer.cpp index b117343c..a6bc16d1 100644 --- a/src/caffe/layers/acl_batch_norm_layer.cpp +++ b/src/caffe/layers/acl_batch_norm_layer.cpp @@ -115,13 +115,13 @@ void ACLBatchNormLayer<Dtype>::Forward_cpu(const vector<Blob<Dtype>*>& bottom, Forward_gpu(bottom, top); return; } +#ifdef USE_PROFILING + logtime_util log_time(ACL_BN_INFO); +#endif //USE_PROFILING if (this->force_bypass_acl_path_||!this->use_global_stats_) { BatchNormLayer<Dtype>::Forward_cpu(bottom,top); return; } -#ifdef USE_PROFILING - logtime_util log_time(ACL_BN_INFO); -#endif //USE_PROFILING const Dtype* bottom_data = bottom[0]->cpu_data(); Dtype* top_data = top[0]->mutable_cpu_data(); SetupACLLayer(bottom,top); diff --git a/src/caffe/layers/acl_concat_layer.cpp b/src/caffe/layers/acl_concat_layer.cpp index 4750cb4b..57a14126 100644 --- a/src/caffe/layers/acl_concat_layer.cpp +++ b/src/caffe/layers/acl_concat_layer.cpp @@ -113,6 +113,7 @@ void ACLConcatLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom, template <typename Dtype> ACLConcatLayer<Dtype>::~ACLConcatLayer() { + if(this->force_bypass_acl_path_)return; for (int i =0; i < cpu_vectors.size(); i ++) { delete cpu_vectors[i]; } |