Nvidia’s cuDNN installation doc recommends testing successful
install by compiling the mnistCUDNN project from cudnn_samples_v7.
This generates a chain of errors, documented here:
Suggested fix:
run sudo emacs /usr/local/cuda/include/cuda_runtime_api.h
Before this line:
typedef void (CUDART_CB *cudaStreamCallback_t)(cudaStream_t stream, cudaError_t status, void *userData);
add this code
#ifdef _WIN32
#define CUDART_CB __stdcall
#else
#define CUDART_CB
#endif
and here:
and here:
Suggested fix:
Open the file:
/usr/include/cudnn.h
And change the line:
#include “driver_types.h”
to:
#include <driver_types.h>
Judging from comments above, this seems to have persisted over a year or so.
Is there a reason why the errors are still present in the sample code?