CMakeList 编译cuda程序

安装cuda  网上有很多

vim gpu.cu

#include "cuda_runtime.h"
#include "device_launch_parameters.h"
#include <sys/time.h>
#include <stdio.h>
#include <math.h>
#include <iostream>
#define Row  1024
#define Col  1024

__global__ void matrix_mul_gpu(int* M, int* N,int* P, int width)
{
   int i = threadIdx.x + blockDim.x * blockIdx.x;
   int j = threadIdx.y + blockDim.y * blockIdx.y;
  
   int sum = 0;
   for(int k = 0; k < width; ++k)
   {
     int a = M[j * width + k];
     int b = N[k * width + i];
     sum += a * b;
   }
  P[j * width + i] = sum;
}


int main()
{
  struct timeval start, end;
  gettimeofday(&start, NULL);

  int *A = (int*)malloc(sizeof(int) * Row * Col);
  int *B = (int*)malloc(sizeof(int) * Row * Col);
  int *C = (int*)malloc(sizeof(int) * Row * Col);

  int *d_dataA, *d_dataB, *d_dataC;
  cudaMalloc((void**)&d_dataA, sizeof(int) * Row * Col);
  cudaMalloc((void**)&d_dataB, sizeof(int) * Row * Col);
  cudaMalloc((void**)&d_dataC, sizeof(int) * Row * Col);

  for( int i = 0; i < Row * Col; ++i)
  { 
    A[i] = 90;
    B[i] = 10;
  }
 
  cudaMemcpy(d_dataA, A, sizeof(int) * Row * Col,cudaMemcpyHostToDevice);
  cudaMemcpy(d_dataB, B, sizeof(int) * Row * Col,cudaMemcpyHostToDevice);
  dim3 threadPerBlock(16, 16);
  dim3 blockNumber((Col + threadPerBlock.x-1)/threadPerBlock.x,
                   (Row + threadPerBlock.y-1)/threadPerBlock.y);
  matrix_mul_gpu << <blockNumber, threadPerBlock>> >(d_dataA,d_dataB,d_dataC,Col);
 cudaMemcpy(C, d_dataC,sizeof(int) * Row * Col,cudaMemcpyDeviceToHost);

 std::cout << C[ 512 * 1024 + 512] << std::endl;
 free(A);
 free(B);
 free(C);
 cudaFree(d_dataA);
 cudaFree(d_dataB);
 cudaFree(d_dataC);

 gettimeofday(&end, NULL);
 int timeuse = 1000000 * (end.tv_sec - start.tv_sec) + end.tv_usec - start.tv_usec;
 printf("totoal time is %f ms\r",timeuse / 1000.0);
  return 0;
}

保存(上代码仅供测试)

vim CMakeLists.txt

cmake_minimum_required(VERSION 2.6)
 
project(cudatest)

find_package(CUDA)
if( NOT CUDA_FOUND )
        message(STATUS "CUDA Not Found. Project will not be build.")
endif( CUDA_FOUND )

cuda_add_executable(${PROJECT_NAME} gpu.cu)

保存

cmake .
cmake --build .
./gputest