55 lines · c
1/* Minimal declarations for CUDA support. Testing purposes only. */2 3#include <stddef.h>4 5// Make this file work with nvcc, for testing compatibility.6 7#ifndef __NVCC__8#define __constant__ __attribute__((constant))9#define __device__ __attribute__((device))10#define __global__ __attribute__((global))11#define __host__ __attribute__((host))12#define __shared__ __attribute__((shared))13#define __managed__ __attribute__((managed))14#define __launch_bounds__(...) __attribute__((launch_bounds(__VA_ARGS__)))15 16struct dim3 {17 unsigned x, y, z;18 __host__ __device__ dim3(unsigned x, unsigned y = 1, unsigned z = 1) : x(x), y(y), z(z) {}19};20 21#ifdef __HIP__22typedef struct hipStream *hipStream_t;23typedef enum hipError {} hipError_t;24int hipConfigureCall(dim3 gridSize, dim3 blockSize, size_t sharedSize = 0,25 hipStream_t stream = 0);26extern "C" hipError_t __hipPushCallConfiguration(dim3 gridSize, dim3 blockSize,27 size_t sharedSize = 0,28 hipStream_t stream = 0);29extern "C" hipError_t hipLaunchKernel(const void *func, dim3 gridDim,30 dim3 blockDim, void **args,31 size_t sharedMem,32 hipStream_t stream);33#else34typedef struct cudaStream *cudaStream_t;35typedef enum cudaError {} cudaError_t;36 37extern "C" int cudaConfigureCall(dim3 gridSize, dim3 blockSize,38 size_t sharedSize = 0,39 cudaStream_t stream = 0);40extern "C" int __cudaPushCallConfiguration(dim3 gridSize, dim3 blockSize,41 size_t sharedSize = 0,42 cudaStream_t stream = 0);43extern "C" cudaError_t cudaLaunchKernel(const void *func, dim3 gridDim,44 dim3 blockDim, void **args,45 size_t sharedMem, cudaStream_t stream);46#endif47 48// Host- and device-side placement new overloads.49void *operator new(__SIZE_TYPE__, void *p) { return p; }50void *operator new[](__SIZE_TYPE__, void *p) { return p; }51__device__ void *operator new(__SIZE_TYPE__, void *p) { return p; }52__device__ void *operator new[](__SIZE_TYPE__, void *p) { return p; }53 54#endif // !__NVCC__55