Add cuMemsetD8Async

This function is needed by the vf_thumbnail_cuda filter in ffmpeg.

Author: philipl

include/ffnvcodec/dynlink_cuda.h

@@ -323,6 +323,7 @@ typedef CUresult CUDAAPI tcuCtxPopCurrent_v2(CUcontext *pctx);
 typedef CUresult CUDAAPI tcuCtxDestroy_v2(CUcontext ctx);
 typedef CUresult CUDAAPI tcuMemAlloc_v2(CUdeviceptr *dptr, size_t bytesize);
 typedef CUresult CUDAAPI tcuMemAllocPitch_v2(CUdeviceptr *dptr, size_t *pPitch, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes);
+typedef CUresult CUDAAPI tcuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream);
 typedef CUresult CUDAAPI tcuMemFree_v2(CUdeviceptr dptr);
 typedef CUresult CUDAAPI tcuMemcpy2D_v2(const CUDA_MEMCPY2D *pcopy);
 typedef CUresult CUDAAPI tcuMemcpy2DAsync_v2(const CUDA_MEMCPY2D *pcopy, CUstream hStream);

include/ffnvcodec/dynlink_loader.h

@@ -150,6 +150,7 @@ typedef struct CudaFunctions {
     tcuCtxDestroy_v2 *cuCtxDestroy;
     tcuMemAlloc_v2 *cuMemAlloc;
     tcuMemAllocPitch_v2 *cuMemAllocPitch;
+    tcuMemsetD8Async *cuMemsetD8Async;
     tcuMemFree_v2 *cuMemFree;
     tcuMemcpy2D_v2 *cuMemcpy2D;
     tcuMemcpy2DAsync_v2 *cuMemcpy2DAsync;
@@ -273,6 +274,7 @@ static inline int cuda_load_functions(CudaFunctions **functions, void *logctx)
     LOAD_SYMBOL(cuCtxDestroy, tcuCtxDestroy_v2, "cuCtxDestroy_v2");
     LOAD_SYMBOL(cuMemAlloc, tcuMemAlloc_v2, "cuMemAlloc_v2");
     LOAD_SYMBOL(cuMemAllocPitch, tcuMemAllocPitch_v2, "cuMemAllocPitch_v2");
+    LOAD_SYMBOL(cuMemsetD8Async, tcuMemsetD8Async, "cuMemsetD8Async");
     LOAD_SYMBOL(cuMemFree, tcuMemFree_v2, "cuMemFree_v2");
     LOAD_SYMBOL(cuMemcpy2D, tcuMemcpy2D_v2, "cuMemcpy2D_v2");
     LOAD_SYMBOL(cuMemcpy2DAsync, tcuMemcpy2DAsync_v2, "cuMemcpy2DAsync_v2");