#pragma once

#ifdef __HIP_PLATFORM_AMD__
  #include <hip/hip_runtime.h>
  #include <hip/hip_runtime_api.h>
  #define cudaError_t            hipError_t
  #define cudaSuccess            hipSuccess
  #define cudaMalloc             hipMalloc
  #define cudaFree               hipFree
  #define cudaMemcpy             hipMemcpy
  #define cudaMemcpyHostToDevice hipMemcpyHostToDevice
  #define cudaMemcpyDeviceToHost hipMemcpyDeviceToHost
  #define cudaDeviceSynchronize  hipDeviceSynchronize
  #define cudaGetErrorString     hipGetErrorString
  #define cudaGetLastError       hipGetLastError
  #define cudaMemGetInfo         hipMemGetInfo
  #define cudaDeviceReset        hipDeviceReset
  #define __global__             __global__
  #define __device__             __device__
  #define __host__               __host__
  #define __shared__             __shared__
  #define __syncthreads          __syncthreads
  #define blockIdx               hipBlockIdx_x
  #define threadIdx              hipThreadIdx_x
  #define blockDim               hipBlockDim_x
  #define gridDim                hipGridDim_x
  #define hipLaunchKernelGGL(F,GRID,BLOCK,SHMEM,STREAM,...) \
          hipLaunchKernelGGL(F, dim3(GRID), dim3(BLOCK), SHMEM, STREAM, __VA_ARGS__)
#endif