HELLO CUDA(VECTOR SUM内核编程）

设备查询与内核出错处理

实验1：设备查询与内核出错处理

1.1查询你机器上GPU设备的参数 l  新建.cu文件 l  调用cudaGetDeviceCount()得到GPU设备的数量 l  调用cudaGetDeviceProperties()函数得到GPU设备的属性结构体 l  解释关键属性的含义，至少包括设备名称、计算能力为多少、设备可用全局内存、每线程块最大线程数、设备可用全局内存容量、每线程块可用共享内存容量、每线程块可用寄存器数量、每线程块最大线程数、每个处理器簇最大驻留线程数、设备中的处理器簇数量等 l  可参考WILT 3.2节   1.2首先自己从头编写并运行VectorSum内核，然后尝试多种查看错误的方式 l  在内核函数内printf信息 l  组合调用cudaGetLastError()和cudaGetErrorString()，返回出错字符串 l  宏 l  在内核启动时，使用非法参数，检验是否成功处理 l  可参考WILT 附录A.3节   1.3 本实验 3学时完成
实验内容及结果：（写出完整操作过程）
1.1查询数据 int main() {         int numDevices;         cudaGetDeviceCount(&numDevices);         printf("Number of Devices:%d\n",numDevices);         for (int i =0;i<numDevices;++i)         {                  printf("----------------------\n");                  cudaDeviceProp cdp;                  cudaGetDeviceProperties(&cdp,i);                  printf("Device Number: %d\n",i);                  printf("Device Name: %s\n",cdp.name);                  printf("Compute capability: %d.%d\n",cdp.major,cdp.minor);                  printf("Maximum threads/block: %d\n",cdp.maxThreadsPerBlock);                  printf("Shared memory /block: %lu bytes \n",cdp.sharedMemPerBlock);                  printf("Total global memory : %lu bytes\n",cdp.totalGlobalMem);         }   } 1.2VectorSum的完成   #include "cuda_runtime.h" #include "device_launch_parameters.h"   #include <stdio.h> #include<cuda.h>       __global__ void addKernel(int * c, const int *a, const int *b) {         int i = threadIdx.x;         c[i] = a[i] + b[i];   }   int main() {         int numDevices;         cudaGetDeviceCount(&numDevices);         printf("Number of Devices:%d\n",numDevices);         for (int i =0;i<numDevices;++i)         {                  printf("----------------------\n");                  cudaDeviceProp cdp;                  cudaGetDeviceProperties(&cdp,i);                  printf("Device Number: %d\n",i);                  printf("Device Name: %s\n",cdp.name);                  printf("Compute capability: %d.%d\n",cdp.major,cdp.minor);                  printf("Maximum threads/block: %d\n",cdp.maxThreadsPerBlock);                  printf("Shared memory /block: %lu bytes \n",cdp.sharedMemPerBlock);                  printf("Total global memory : %lu bytes\n",cdp.totalGlobalMem);         }             const int arraySize = 5;     const int a[arraySize] = { 1, 2, 3, 4, 5 };     const int b[arraySize] = { 10, 20, 30, 40, 50 };     int c[arraySize] = { 0 };           CUresult cudaStatus;       int *dev_a, *dev_b, *dev_c;           //Allocate GPU buffere for three vectors (two inputs, one outputs)         cudaMalloc((void**)&dev_c,arraySize * sizeof(int));         cudaMalloc((void**)&dev_a,arraySize * sizeof(int));         cudaMalloc((void**)&dev_b,arraySize * sizeof(int));           //Copy input vectors from host memory to GPU buffers         cudaMemcpy(dev_a,a, arraySize * sizeof(int),cudaMemcpyHostToDevice);         cudaMemcpy(dev_b,b, arraySize * sizeof(int),cudaMemcpyHostToDevice);           //Launch a kernel on the GPU with one thread for each element         addKernel<<<1,arraySize>>>(dev_c,dev_a,dev_b);             //Check errors         cudaStatus = cudaGetLastError();         if(cudaStatus != cudaSuccess)         {                  fprintf(stderr,"addvector failed: %s\n",cudaGetErrorString(cudaStatus));                  fprintf(stderr,"%s\n",cudaGetErrorString(cudaStatus));                  goto Error;         }         //any errors return         cudaStatus = cudaDeviceSynchronize();         if(cudaStatus != cudaSuccess)         {                  fprintf(stderr,"cudaDeviceSynchronize returned %d\n",cudaStatus);                  fprintf(stderr,"%s\n",cudaGetErrorString(cudaStatus));                  goto Error;         }                 //Copy output vector from GPU buffer to host memory         cudaMemcpy(c,dev_c,arraySize *   sizeof(int),cudaMemcpyDeviceToHost);           cudaFree(dev_c);         cudaFree(dev_a);         cudaFree(dev_b);           printf("{ 1, 2, 3, 4, 5}\n + \n{10,20,30,40,50} \n = \n{%d,%d,%d,%d,%d}\n",c[0],c[1],c[2],c[3],c[4]);           getchar();           return 0; }
实验过程中遇到的问题如何解决的？本次实验的体会（详细）？
1.Memcpy函数使用的时候参数较多，第一次少写一个造成问题     2.调试的时候出现大量的warning，函数调用的时候<<<blocks,threads>>>经常出现警示。     3.代码完成后发现输出正常但c[i]数组输出有问题，均为[0]发现并没有将数组的值发生改变。