728x90
반응형

 

Cudnn Error in nvinfer1::rt::CommonContext::configure: 7 (CUDNN_STATUS_MAPPING_ERROR)

Cudnn Error in nvinfer1::rt::CommonContext::configure: 7 (CUDNN_STATUS_MAPPING_ERROR)

Cudnn Error in nvinfer1::rt::CommonContext::configure: 7 (CUDNN_STATUS_MAPPING_ERROR)

Cudnn Error in nvinfer1::rt::CommonContext::configure: 7 (CUDNN_STATUS_MAPPING_ERROR)

Cudnn Error in nvinfer1::rt::CommonContext::configure: 7 (CUDNN_STATUS_MAPPING_ERROR)

 

cuDNN status Error in: file:

C:/Workspace/src/convolutional_kernels.cu : forward_convolutional_layer_gpu() : line: 541 : build time: Nov  7 2019 - 09:52:32 cuDNN Error: CUDNN_STATUS_EXECUTION_FAILED

 

 

TensorRT + TensorFlow 를 같이 사용할 때 발생할 수 있는 CUDA 및 cuDNN 이슈로서, 

데이터의 복사가 일어나기 전 (TensorRT)를 사용하기 전에

반드시 코드상에서 모델을 로드하기 전 단계에서 cudaSetDevice 설정을 해주어야한다. 

 

cudaSetDevice(0)

 

 

 

 

 

 

또한, cudaSetDevice 설정으로 에러가 해결되지 않을 경우, Darknet 을 빌드해서 사용할 때

darknet/src/yolo_console_dll.cpp 에서 340번째 줄의 주석을 풀고 빌드해야한다.

//init_params.sdk_cuda_ctx = (CUcontext)detector.get_cuda_context();

 

//init_params.sdk_cuda_ctx = (CUcontext)detector.get_cuda_context();

 

 

 

 

 

참고자료 1 :

https://github.com/NVIDIA/TensorRT/issues/219

 

Unable to run separate instances in multiple threads · Issue #219 · NVIDIA/TensorRT

I've built a library that loads an engine, and this works well. Now I am testing on two cards. In the library, when initialized, I call setCudaDevice(which GPU) to assign to net to the correct ...

github.com

 

참고자료 2 : 

https://github.com/AlexeyAB/darknet/issues/3577

 

CUDA error : ZED cam application dependency on libdarknet.so · Issue #3577 · AlexeyAB/darknet

I built libdarknet.so by enabling cuda, cudnn, opencv, libso and zed camera flags and built an independent ZED SVO reader C++ CMake application by linking it with libdarknet library. When I launch ...

github.com

 

참고자료 3 :

https://github.com/facebookresearch/flashlight/issues/73

 

CUDNN_STATUS_MAPPING_ERROR when using fl::Module from a different thread · Issue #73 · facebookresearch/flashlight

I'm using wav2letter that based on Flashlight. I'm trying to embed wav2letter decoder (fl::Sequential) in gRPC service and use it simultaneously in concurrent requests. All works fine when ...

github.com

 

728x90
반응형