CUDA初探：CUDA的初始化

    首先介绍一下开发环境，Visual Studio 2008 + CUDA Wizard for Visual Studio. 确保显卡支持CUDA（GeForce 8系列之后，否则只能用模拟模式）并安装CUDA DDK及CUDA Toolkit。
    安装完CUDA Wizard for Visual Studio之后，如果编译运行时出现"无法找到cutil32D.dll"的错误，则需要将安装的CUDA的sdk的路径加到系统环境变量中：

例如C:\Program Files\NVIDIA Corporation\NVIDIA CUDA SDK\bin\win32\
下的
├─Debug
├─EmuDebug
├─EmuRelease
└─Release

几个目录都加入到系统环境变量Path中，这样才能在运行程序的时候找到相应的dll库。
     
     然后就可以在VS下直接编译CUDA程序了（*.cu)。以下是一个CUDA初始化的程序。

#include <stdio.h>
#include <cuda_runtime.h>

bool InitCUDA()
{
    int count;
    cudaGetDeviceCount(&count);
    if(count==0)
    {
        fprintf(stderr,"There is no device.\n");
        return false;
    }
    int i;
    for(i=0;i<count;i++)
    {
        cudaDeviceProp prop;
        if(cudaGetDeviceProperties(&prop,i) == cudaSuccess)
        {
            if(prop.major>=1)
            {
                //枚举详细信息
                printf("Identify: %s\n",prop.name);
                printf("Host Memory: %d\n",prop.canMapHostMemory);                
                printf("Clock Rate: %d khz\n",prop.clockRate);                
                printf("Compute Mode: %d\n",prop.computeMode);                
                printf("Device Overlap: %d\n",prop.deviceOverlap);                
                printf("Integrated: %d\n",prop.integrated);                
                printf("Kernel Exec Timeout Enabled: %d\n",prop.kernelExecTimeoutEnabled);                
                printf("Max Grid Size: %d * %d * %d\n",prop.maxGridSize[0],prop.maxGridSize[1],prop.maxGridSize[2]);
                printf("Max Threads Dim: %d * %d * %d\n",prop.maxThreadsDim[0],prop.maxThreadsDim[1],prop.maxThreadsDim[2]);
                printf("Max Threads per Block: %d\n",prop.maxThreadsPerBlock);
                printf("Maximum Pitch: %d bytes\n",prop.memPitch);
                printf("Minor Compute Capability: %d\n",prop.minor);
                printf("Number of Multiprocessors: %d\n",prop.multiProcessorCount);                
                printf("32bit Registers Availble per Block: %d\n",prop.regsPerBlock);
                printf("Shared Memory Available per Block: %d bytes\n",prop.sharedMemPerBlock);
                printf("Alignment Requirement for Textures: %d\n",prop.textureAlignment);
                printf("Constant Memory Available: %d bytes\n",prop.totalConstMem);
                printf("Global Memory Available: %d bytes\n",prop.totalGlobalMem);
                printf("Warp Size: %d threads\n",prop.warpSize);
                break;
            }
        }
    }
    if(i==count)
    {
        fprintf(stderr,"There is no device supporting CUDA.\n");
        return false;
    }
    cudaSetDevice(i);
    return true;
}

void main()
{
    if(!InitCUDA())
        {
            getchar();
            return;
        }
        printf("CUDA initialized.\n");
        getchar();        
}    
    
    

运行结果如下：