Today I write a simple test to call CUDA RuntimeAPI.

1 DeviceManagement RT

 1.1 cudaGetDeviceCount

 cudaError_t cudaGetDeviceCount(int *count)

 use this function to get the number of devices with compute capability

 1.2 cudaGetDevice

 cudaError_t cudaGetDeviceC(int *dev)

 use this function to get device number which is currently being used

 1.3 cudaGetDeviceProperties

 cudaError_t cudaGetDeviceProperties(struct cudaDeviceProp *prop, int dev)

 use this function to get the detail properties of dev device, you will know how large the GlobalMemory it has, how many register per block, etc.

 1.4 cudaChooseDevice

 cudaError_t cudaChooseDevice(int *dev, const struct cudaDeviceProp *prop)

 returns in *dev the device which properties best match *prop, if you have many devices, you can choose them by setting *prop, then get the device number.

 1.5 cudaSetDevice

 cudaError_t cudaSetDevice(int dev)

 after you choose the device you want to use, set the device number and call this function, then the specify one will work for you.

At last I post the code which can get all the CUDA device prop in you computer

/* * Copyright (c) 2009, SCUT Chenhu_lab * All rights reserved. * * filename：cudart_device.cu * content：test runtime api of cuda * * current version：1.0 * auther: Jiangfeng Peng * finish date：2009-6-4 * * old version: null * ori auther: null * finish daa: null */ #include <stdio.h> #include <stdlib.h> #include <cuda_runtime.h> bool InitCUDA() { int count; int i,j; cudaDeviceProp prop; /* cudaError_t cudaGetDeviceCount(int *count) -- returns the number of compute-capable devices */ cudaGetDeviceCount(&count); if (count == 0) { fprintf(stderr, "There is no device.\n"); return false; } else { printf("\ncudaGetdeviceCount: %d\n\n", count); } for (i=0,j=0; i < count; i++) { /* cudaError_t cudaGetDeviceProperties(struct cudaDeviceProp *prop, int dev) -- returns information on the compute-device */ if (cudaGetDeviceProperties(&prop, i) == cudaSuccess) { printf("Num %d cudaDeviceProp :\n", i); printf( " name %s \n totalGlobalMem %dB = %1.2fMB\n sharedMemPerBlock %dB = %dKB \n regsPerBlock %d = %dk\n" " warpSize %d \n memPitch %dB = %dKB\n maxThreadsPerBlock %d \n maxThreadsDim: x %d y %d z %d \n" " maxGridSize: x %d y %d z %d \n totalConstMem %dB = %dKB \n major %d \n minor %d \n clockRate %dKHZ = %1.2fGHZ \n" " textureAlignment %d \n deviceOverlap %d \n multiProcessorCount %d \n\n", prop.name, prop.totalGlobalMem, (double)prop.totalGlobalMem/1024/1024, prop.sharedMemPerBlock, prop.sharedMemPerBlock/1024, prop.regsPerBlock, prop.regsPerBlock/1024, prop.warpSize,prop.memPitch, prop.memPitch/1024, prop.maxThreadsPerBlock, prop.maxThreadsDim[0], prop.maxThreadsDim[1], prop.maxThreadsDim[2], prop.maxGridSize[0], prop.maxGridSize[1], prop.maxGridSize[2], prop.totalConstMem, prop.totalConstMem/1024, prop.major, prop.minor, prop.clockRate, (double)prop.clockRate/1024/1024, prop.textureAlignment, prop.deviceOverlap, prop.multiProcessorCount); if (prop.major >= 1) { j = 1; } } } if(j == 0) { fprintf(stderr, "There is no device supporting CUDA 1.x.\n"); return false; } cudaSetDevice(i); return true; } int main() { if(!InitCUDA()) { return 0; } printf("CUDA initialized.\n"); return 0; }

发表于 @ 2009年06月04日　21:06:00 | 评论( loading... ) | 编辑| 举报| 收藏

旧一篇:A Contributor's Guide to Launchpad.net - Part 2 - Code Management | 新一篇:assert() c++ 使用方法