一般有兩種方法可以在vs2013上添加運(yùn)行cuda8.0程序：

一、直接新建一個(gè)基于CUDA8.0的項(xiàng)目：如下圖所示，

點(diǎn)擊確定后即可生成test_cuda項(xiàng)目；默認(rèn)會(huì)自動(dòng)生成一個(gè)kernel.cu文件；默認(rèn)已經(jīng)配置好Debug/Release, Win32/x64環(huán)境，直接編譯運(yùn)行，結(jié)果如下圖所示：函數(shù)執(zhí)行的是兩個(gè)數(shù)組的加操作。移除kernel.cu文件，加入自己需要的cuda文件即可進(jìn)行實(shí)際操作了，非常方便。

二、實(shí)際情況下，多是在已有的項(xiàng)目中添加一些cuda文件，用于加速，下面說(shuō)下具體的操作步驟：

1、新建一個(gè)CUDA_Test x64控制臺(tái)空工程；

2、新建CUDA_Test.cpp文件；

3、選中CUDA_Test項(xiàng)目，右鍵單擊-->生成依賴項(xiàng)-->生成自定義，勾選CUDA8.0，點(diǎn)擊確定，如下圖所示：

4、完成第3步后，再次打開(kāi)工程的屬性配置，會(huì)多出兩項(xiàng)，CUDA C/C++和CUDA Linker，如下圖所示：

5、新建或添加幾個(gè)已有的文件，包括common.hpp、simple.hpp、simple.cpp、simple.cu，各個(gè)文件內(nèi)容如下：

common.hpp:

#ifndef FBC_CUDA_TEST_COMMON_HPP_
#define FBC_CUDA_TEST_COMMON_HPP_#define PRINT_ERROR_INFO(info) { \fprintf(stderr, "Error: %s, file: %s, func: %s, line: %d\n", #info, __FILE__, __FUNCTION__, __LINE__); \return -1; }#endif // FBC_CUDA_TEST_COMMON_HPP_

simple.hpp:

#ifndef FBC_CUDA_TEST_SIMPLE_HPP_
#define FBC_CUDA_TEST_SIMPLE_HPP_// reference: C:\ProgramData\NVIDIA Corporation\CUDA Samples\v8.0\0_Simple
int test_vectorAdd();int vectorAdd_cpu(const float *A, const float *B, float *C, int numElements);int vectorAdd_gpu(const float *A, const float *B, float *C, int numElements);#endif // FBC_CUDA_TEST_SIMPLE_HPP_

simple.cpp:

#include "simple.hpp"
#include <stdlib.h>
#include <iostream>
#include "common.hpp"// =========================== vector add =============================
int test_vectorAdd()
{// Vector addition: C = A + B, implements element by element vector additionconst int numElements{ 50000 };float* A = new float[numElements];float* B = new float[numElements];float* C1 = new float[numElements];float* C2 = new float[numElements];// Initialize vectorfor (int i = 0; i < numElements; ++i) {A[i] = rand() / (float)RAND_MAX;B[i] = rand() / (float)RAND_MAX;}int ret = vectorAdd_cpu(A, B, C1, numElements);if (ret != 0) PRINT_ERROR_INFO(vectorAdd_cpu);ret = vectorAdd_gpu(A, B, C2, numElements);if (ret != 0) PRINT_ERROR_INFO(vectorAdd_gpu);for (int i = 0; i < numElements; ++i) {if (fabs(C1[i] - C2[i]) > 1e-5) {fprintf(stderr, "Result verification failed at element %d!\n", i);return -1;}}delete[] A;delete[] B;delete[] C1;delete[] C2;return 0;
}int vectorAdd_cpu(const float *A, const float *B, float *C, int numElements)
{for (int i = 0; i < numElements; ++i) {C[i] = A[i] + B[i];}return 0;
}

simple.cu:

#include "simple.hpp"
#include <iostream>
#include <cuda_runtime.h> // For the CUDA runtime routines (prefixed with "cuda_")
#include <device_launch_parameters.h>// reference: C:\ProgramData\NVIDIA Corporation\CUDA Samples\v8.0\0_Simple// =========================== vector add =============================
__global__ void vectorAdd(const float *A, const float *B, float *C, int numElements)
{int i = blockDim.x * blockIdx.x + threadIdx.x;if (i < numElements) {C[i] = A[i] + B[i];}
}int vectorAdd_gpu(const float *A, const float *B, float *C, int numElements)
{// Error code to check return values for CUDA callscudaError_t err{ cudaSuccess };size_t length{ numElements * sizeof(float) };fprintf(stderr, "Length: %d\n", length);float* d_A{ nullptr };float* d_B{ nullptr };float* d_C{ nullptr };err = cudaMalloc(&d_A, length);if (err != cudaSuccess) {fprintf(stderr, "Failed to allocate device vector A (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaMalloc(&d_B, length);if (err != cudaSuccess) {fprintf(stderr, "Failed to allocate device vector B (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaMalloc(&d_C, length);if (err != cudaSuccess) {fprintf(stderr, "Failed to allocate device vector C (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaMemcpy(d_A, A, length, cudaMemcpyHostToDevice);if (err != cudaSuccess) {fprintf(stderr, "Failed to copy vector A from host to device (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaMemcpy(d_B, B, length, cudaMemcpyHostToDevice);if (err != cudaSuccess) {fprintf(stderr, "Failed to copy vector B from host to device (error code %s)!\n", cudaGetErrorString(err));return -1;}// Launch the Vector Add CUDA kernelint threadsPerBlock = 256;int blocksPerGrid = (numElements + threadsPerBlock - 1) / threadsPerBlock;fprintf(stderr, "CUDA kernel launch with %d blocks of %d threads\n", blocksPerGrid, threadsPerBlock);vectorAdd << <blocksPerGrid, threadsPerBlock >> >(d_A, d_B, d_C, numElements);err = cudaGetLastError();if (err != cudaSuccess) {fprintf(stderr, "Failed to launch vectorAdd kernel (error code %s)!\n", cudaGetErrorString(err));return -1;}// Copy the device result vector in device memory to the host result vector in host memory.err = cudaMemcpy(C, d_C, length, cudaMemcpyDeviceToHost);if (err != cudaSuccess) {fprintf(stderr, "Failed to copy vector C from device to host (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaFree(d_A);if (err != cudaSuccess) {fprintf(stderr, "Failed to free device vector A (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaFree(d_B);if (err != cudaSuccess) {fprintf(stderr, "Failed to free device vector B (error code %s)!\n", cudaGetErrorString(err));return -1;}err = cudaFree(d_C);if (err != cudaSuccess) {fprintf(stderr, "Failed to free device vector C (error code %s)!\n", cudaGetErrorString(err));return -1;}return err;
}

CUDA_Test.cpp:

#include <iostream>
#include "simple.hpp"int main()
{int ret = test_vectorAdd();if (ret == 0) fprintf(stderr, "***** test success *****\n");else fprintf(stderr, "===== test fail =====\n");return 0;
}

6、調(diào)整屬性配置項(xiàng)：

(1)、CUDA C/C++-->Common中Target Machine Platform中默認(rèn)是32-bit(--machine32)，因?yàn)槭莤64，所以將其調(diào)整為64-bit(--machine 64)；

(2)、添加附加庫(kù)：鏈接器-->輸入-->附加依賴項(xiàng)：cudart.lib；

(3)、消除nvcc warning: The 'compute_20', 'sm_20', and'sm_21' architectures are deprecated, and may be removed in a future release：CUDA C/C++-->Device: Code Generation:由compute_20,sm_20修改為compute_30,sm_30; compute_35,sm_35; compute_37,sm_37;compute_50,sm_50; compute_52,sm_52; compute_60,sm_60

以上code是參考NVIDIA Corporation\CUDA Samples\v8.0\0_Simple中vectorAdd例子進(jìn)行的改寫(xiě)，輸出結(jié)果如下：

GitHub：https://github.com/fengbingchun/CUDA_Test

總結(jié)

以上是生活随笔為你收集整理的windows10 vs2013控制台工程中添加并编译cuda8.0文件操作步骤的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。