cuda Toolkit 10.2 + VS2015 C++ cuda GPU運算步驟

cuda Toolkit 10.2 + VS2015 C++ cuda

GPU運算步驟

 

包含頭文件

#include "cuda_runtime.h"
#include "device_launch_parameters.h"

 

1. GPU 內存申請

• cudaError_t cudaStatus = cudaMalloc(void **p, size_t s)

2. 內存拷貝 host memory -> Gpu Buffer

• cudaStatus = cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind)  //cudaMemcpyHostToDevice

  enum __evice_builtin__ cudaMemcpyKind
      {
          cudaMemcpyHostToHost          =   0,      /**< Host   -> Host */
          cudaMemcpyHostToDevice        =   1,      /**< Host   -> Device */
          cudaMemcpyDeviceToHost        =   2,      /**< Device -> Host */
          cudaMemcpyDeviceToDevice      =   3,      /**< Device -> Device */
          cudaMemcpyDefault             =   4       /**< Direction of the transfer is inferred from the pointer values. Requires unified virtual addressing */
      };

   

3. 調用 kernel 方法

• addKernel <<<1, size >>>(dev_c, dev_a, dev_b);

• __global__ void addKernel(int *c, const int *a, const int *b)
  {
  	int i = threadIdx.x + blockIdx.x;
  	printf("blockIdx.x %d threadIdx.x %d \n", blockIdx.x, threadIdx.x);
  	printf("i == %d \n", i);
  
  
  	c[i] = a[i] + b[i];
  }

   

4. 內存拷貝 Gpu Buffer->Host memory

• cudaMemcpy(void *dst, const void *src, size_t count, enum cudaMemcpyKind kind);  // cudaMemcpyDeviceToHost

5. 設備重置

• cudaStatus = cudaDeviceReset();