fix format. test=develop

revert-16190-refine_parallel_executor
dengkaipeng 6 years ago
parent b33e6bf5ef
commit 0ff9a403d0

@ -84,15 +84,15 @@ class YoloBoxOpCUDAKernel : public framework::OpKernel<T> {
int input_size = downsample_ratio * h; int input_size = downsample_ratio * h;
auto& dev_ctx = ctx.cuda_device_context(); auto& dev_ctx = ctx.cuda_device_context();
auto& allocator = auto& allocator =
platform::DeviceTemporaryAllocator::Instance().Get(dev_ctx); platform::DeviceTemporaryAllocator::Instance().Get(dev_ctx);
int bytes = sizeof(int) * anchors.size(); int bytes = sizeof(int) * anchors.size();
auto anchors_ptr = allocator.Allocate(sizeof(int) * anchors.size()); auto anchors_ptr = allocator.Allocate(sizeof(int) * anchors.size());
int* anchors_data = reinterpret_cast<int*>(anchors_ptr->ptr()); int* anchors_data = reinterpret_cast<int*>(anchors_ptr->ptr());
const auto gplace = boost::get<platform::CUDAPlace>(ctx.GetPlace()); const auto gplace = boost::get<platform::CUDAPlace>(ctx.GetPlace());
const auto cplace = platform::CPUPlace(); const auto cplace = platform::CPUPlace();
memory::Copy(gplace, anchors_data, cplace, anchors.data(), bytes, memory::Copy(gplace, anchors_data, cplace, anchors.data(), bytes,
dev_ctx.stream()); dev_ctx.stream());
const T* input_data = input->data<T>(); const T* input_data = input->data<T>();
const int* imgsize_data = img_size->data<int>(); const int* imgsize_data = img_size->data<int>();

Loading…
Cancel
Save