{"id":112474,"date":"2026-10-03T20:01:11","date_gmt":"2026-10-03T12:01:11","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/112474.html"},"modified":"2026-10-03T20:01:11","modified_gmt":"2026-10-03T12:01:11","slug":"%e3%80%90cuda-%e5%85%a5%e9%97%a8%e7%b3%bb%e5%88%97%e3%80%91cuda-c-%e4%bb%a3%e7%a0%81%e7%bb%93%e6%9e%84%e3%80%81host-device-%e8%be%b9%e7%95%8c%e4%b8%8e-nvcc-%e7%bc%96%e8%af%91%e6%b5%81%e7%a8%8b","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/112474.html","title":{"rendered":"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218"},"content":{"rendered":"

🔥 \u672c\u6587\u4e13\u680f：CUDA 🌸\u4f5c\u8005\u4e3b\u9875：\u52aa\u529b\u52aa\u529b\u518d\u52aa\u529bwz<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

💪 \u4eca\u65e5\u535a\u5ba2\u52b1\u5fd7\u8bed\u5f55：\u7406\u8bba\u771f\u6b63\u5f00\u59cb\u53d8\u6210\u81ea\u5df1\u7684\u4e1c\u897f，\u5f80\u5f80\u4e0d\u662f\u5728\u201c\u542c\u61c2\u201d\u7684\u65f6\u5019，\u800c\u662f\u5728\u4eb2\u624b\u628a\u7b2c\u4e00\u6bb5\u4ee3\u7801\u7f16\u8bd1\u51fa\u6765\u7684\u65f6\u5019\u3002<\/p>\n

\n

\u601d\u7ef4\u5bfc\u56fe<\/h3>\n

CUDA \u7406\u8bba\u57fa\u7840
\n \u2193
\n\u4e0d\u80fd\u53ea\u505c\u7559\u5728 Grid \/ Block \/ Thread \/ \u663e\u5b58\u8fd9\u4e9b\u6982\u5ff5
\n \u2193
\n\u5f00\u59cb\u771f\u6b63\u7f16\u5199 CUDA C++ \u4ee3\u7801
\n \u2193
\n\u9996\u5148\u89e3\u51b3\u4e00\u4e2a\u95ee\u9898：CUDA \u7a0b\u5e8f\u5230\u5e95\u600e\u4e48\u88ab\u7f16\u8bd1\u51fa\u6765？
\n \u2193
\n\u666e\u901a C++
\n.cpp \u2192 g++ \u2192 \u53ef\u6267\u884c\u6587\u4ef6
\n \u2193
\nCUDA C++
\n.cu \u2192 nvcc
\n \u2193
\n\u4e00\u4e2a .cu \u4e2d\u53ef\u4ee5\u540c\u65f6\u5b58\u5728
\nHost Code（CPU） + Device Code（GPU）
\n \u2193
\nnvcc \u4f5c\u4e3a\u6574\u4e2a CUDA \u7f16\u8bd1\u6d41\u7a0b\u7684\u5165\u53e3
\n \u2193
\nHost Code \u2192 Host C++ Compiler
\nDevice Code \u2192 NVIDIA Device \u7f16\u8bd1\u94fe
\n \u2193
\n\u6700\u7ec8\u7ec4\u5408\u5e76\u94fe\u63a5\u6210\u53ef\u6267\u884c\u6587\u4ef6
\n \u2193
\n\u8fdb\u4e00\u6b65\u8ba4\u8bc6 CUDA \u6e90\u7801\u672c\u8eab
\n \u2193
\nCUDA \u8bed\u8a00\u6269\u5c55
\n__global__ \/ __device__ \/ <<< >>> \u7b49
\n \u2193
\n\u8fd9\u4e9b\u4e0d\u662f\u666e\u901a\u5e93\u51fd\u6570，\u800c\u662f\u7f16\u8bd1\u5668\u80fd\u591f\u8bc6\u522b\u7684 CUDA C++ \u8bed\u8a00\u6269\u5c55
\n \u2193
\nCUDA Runtime API
\ncudaMalloc \/ cudaMemcpy \/ cudaFree \u7b49
\n \u2193
\n\u5c5e\u4e8e\u8fd0\u884c\u5e93 API，\u4e0e\u8bed\u8a00\u6269\u5c55\u662f\u4e24\u4e2a\u5c42\u6b21
\n \u2193
\n\u7ee7\u7eed\u660e\u786e Host \/ Device \u7684\u6267\u884c\u8fb9\u754c
\n \u2193
\n\u666e\u901a Host \u51fd\u6570\u4e0d\u80fd\u76f4\u63a5\u5728 Device \u4e2d\u8c03\u7528
\nDevice \u4ee3\u7801\u53ea\u80fd\u8c03\u7528\u5177\u6709 Device \u4fa7\u5b9e\u73b0\u7684\u51fd\u6570
\n \u2193
\nprintf \u662f CUDA \u7279\u6b8a\u652f\u6301\u7684 Device-side \u51fd\u6570\u4e4b\u4e00
\nstd::cout \/ \u666e\u901a std::vector \u5219\u4e0d\u80fd\u76f4\u63a5\u653e\u8fdb Kernel \u4e2d\u4f7f\u7528
\n \u2193
\n\u7f16\u5199\u7b2c\u4e00\u4e2a hello.cu
\n \u2193
\nnvcc hello.cu -o hello
\n \u2193
\n\u6210\u529f\u751f\u6210\u53ef\u6267\u884c\u6587\u4ef6<\/p>\n

\n

\u5f15\u5165<\/h3>\n

\u5728\u6b64\u524d\u5b66\u4e60 CUDA \u7684\u8fc7\u7a0b\u4e2d，\u6211\u4eec\u5df2\u7ecf\u8ba4\u8bc6\u4e86\u5f88\u591a\u7406\u8bba\u4e0a\u7684\u6982\u5ff5，\u4f8b\u5982：<\/p>\n

CPU \/ GPU
\nHost \/ Device
\nGrid
\nBlock
\nThread
\nWarp
\nGlobal Memory
\nShared Memory
\nKernel<\/p>\n

\u8fd9\u4e9b\u6982\u5ff5\u80fd\u591f\u5e2e\u52a9\u6211\u4eec\u7406\u89e3 GPU \u5230\u5e95\u662f\u600e\u6837\u7ec4\u7ec7\u8ba1\u7b97\u7684，\u4f46\u662f\u5982\u679c\u4e00\u76f4\u505c\u7559\u5728\u7406\u8bba\u5c42\u9762，\u5f88\u5bb9\u6613\u51fa\u73b0\u4e00\u4e2a\u95ee\u9898：<\/p>\n

\u6bcf\u4e2a\u6982\u5ff5\u4f3c\u4e4e\u90fd\u8ba4\u8bc6，\u4f46\u662f\u4e00\u65e6\u771f\u6b63\u770b\u5230\u4e00\u4e2a .cu \u6587\u4ef6，\u5374\u4e0d\u77e5\u9053\u8fd9\u4e9b\u6982\u5ff5\u7a76\u7adf\u5bf9\u5e94\u4ee3\u7801\u4e2d\u7684\u54ea\u4e00\u90e8\u5206\u3002<\/p>\n

\u56e0\u6b64\u63a5\u4e0b\u6765\u9700\u8981\u5f00\u59cb\u628a\u7406\u8bba\u6620\u5c04\u5230\u771f\u6b63\u7684 CUDA C++ \u4ee3\u7801\u4e2d\u3002<\/p>\n

\u4e0d\u8fc7\u5728\u771f\u6b63\u7814\u7a76 cudaMalloc\u3001cudaMemcpy\u3001\u7ebf\u7a0b\u7d22\u5f15\u4ee5\u53ca\u77e9\u9635\u8ba1\u7b97\u4e4b\u524d，\u6211\u89c9\u5f97\u66f4\u5e94\u8be5\u5148\u628a\u6700\u5e95\u5c42\u7684\u4e00\u6761\u94fe\u8def\u7406\u987a：<\/p>\n

CUDA \u6e90\u4ee3\u7801
\n \u2193
\n\u7a76\u7adf\u7531\u8c01\u7f16\u8bd1？
\n \u2193
\nCPU \u4ee3\u7801\u548c GPU \u4ee3\u7801\u5982\u4f55\u5171\u5b58\u5728\u4e00\u4e2a\u7a0b\u5e8f\u4e2d？
\n \u2193
\n\u6700\u540e\u53c8\u600e\u6837\u751f\u6210\u4e00\u4e2a\u53ef\u6267\u884c\u6587\u4ef6？<\/p>\n

\u8fd9\u4e5f\u662f\u672c\u6587\u9996\u5148\u8981\u89e3\u51b3\u7684\u95ee\u9898\u3002<\/p>\n

\n

\u4e00\u3001\u666e\u901a C++ \u7a0b\u5e8f\u662f\u600e\u6837\u88ab\u7f16\u8bd1\u51fa\u6765\u7684<\/h2>\n

\u5728\u7406\u89e3 CUDA \u4e4b\u524d，\u53ef\u4ee5\u5148\u56de\u5230\u6211\u4eec\u6700\u719f\u6089\u7684\u666e\u901a C++\u3002<\/p>\n

\u5047\u8bbe\u5b58\u5728\u4e00\u4e2a：<\/p>\n

main.cpp<\/p>\n

\u5176\u4e2d\u53ea\u5305\u542b\u666e\u901a CPU \u4ee3\u7801，\u90a3\u4e48\u5728 Linux \u4e0b\u6700\u5e38\u89c1\u7684\u6d41\u7a0b\u5c31\u662f：<\/p>\n

g++ main.cpp -o<\/span> main<\/p>\n

\u6574\u4f53\u53ef\u4ee5\u7b80\u5355\u7406\u89e3\u4e3a：<\/p>\n

main.cpp
\n \u2193
\ng++
\n \u2193
\n\u9884\u5904\u7406 \/ \u7f16\u8bd1 \/ \u6c47\u7f16 \/ \u94fe\u63a5
\n \u2193
\nmain
\n \u2193
\n.\/main
\n \u2193
\nCPU \u6267\u884c<\/p>\n

\u4e5f\u5c31\u662f\u8bf4，\u5bf9\u4e8e\u666e\u901a C++ \u7a0b\u5e8f，\u6211\u4eec\u9762\u5bf9\u7684\u57fa\u672c\u5c31\u662f：<\/p>\n

C++ Source Code
\n \u2193
\nHost Compiler
\n \u2193
\nCPU \u53ef\u6267\u884c\u7a0b\u5e8f<\/p>\n

\u4f46\u662f CUDA \u7a0b\u5e8f\u4f1a\u591a\u51fa\u4e00\u4e2a\u975e\u5e38\u660e\u663e\u7684\u95ee\u9898：<\/p>\n

\u540c\u4e00\u4e2a\u7a0b\u5e8f\u4e2d，\u4e0d\u4ec5\u5b58\u5728 CPU \u8981\u6267\u884c\u7684\u4ee3\u7801，\u540c\u65f6\u8fd8\u5b58\u5728 GPU \u8981\u6267\u884c\u7684\u4ee3\u7801\u3002<\/p>\n

\u56e0\u6b64，\u5b83\u4e0d\u80fd\u518d\u7b80\u5355\u5730\u5168\u90e8\u4ea4\u7ed9\u666e\u901a g++ \u5904\u7406\u3002<\/p>\n

\n

\u4e8c\u3001CUDA \u7a0b\u5e8f\u672c\u8d28\u4e0a\u662f Host Code + Device Code<\/h2>\n

\u4e00\u4e2a CUDA C++ \u7a0b\u5e8f\u901a\u5e38\u4e0d\u662f\u201c\u53ea\u6709 GPU \u4ee3\u7801\u201d\u3002<\/p>\n

\u76f8\u53cd，\u4e00\u4e2a\u975e\u5e38\u5178\u578b\u7684 CUDA \u7a0b\u5e8f\u5b9e\u9645\u4e0a\u7531\u4e24\u90e8\u5206\u5171\u540c\u7ec4\u6210：<\/p>\n

CUDA Program
\n\u2502
\n\u251c\u2500\u2500 Host Code
\n\u2502 \u2514\u2500\u2500 CPU \u6267\u884c
\n\u2502
\n\u2514\u2500\u2500 Device Code
\n \u2514\u2500\u2500 GPU \u6267\u884c<\/p>\n

\u4f8b\u5982\u540e\u9762\u4f1a\u4f7f\u7528\u7684\u7b2c\u4e00\u6bb5\u4ee3\u7801：<\/p>\n

#<\/span>include<\/span> <cstdio><\/span><\/span><\/p>\n

__global__ void<\/span> helloFromGPU<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from GPU\\\\n"<\/span>)<\/span>;<\/span>
\n}<\/span><\/p>\n

int<\/span> main<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from CPU\\\\n"<\/span>)<\/span>;<\/span><\/p>\n

helloFromGPU<<<\/span><<\/span>1<\/span>,<\/span> 1<\/span>>><\/span>><\/span>(<\/span>)<\/span>;<\/span><\/p>\n

return<\/span> 0<\/span>;<\/span>
\n}<\/span><\/p>\n

\u5176\u4e2d：<\/p>\n

int<\/span> main<\/span>(<\/span>)<\/span><\/p>\n

\u5c5e\u4e8e\u6b63\u5e38\u7684 Host \u4ee3\u7801，\u4e5f\u5c31\u662f CPU \u4fa7\u7a0b\u5e8f\u5165\u53e3\u3002<\/p>\n

\u800c：<\/p>\n

__global__ void<\/span> helloFromGPU<\/span>(<\/span>)<\/span><\/p>\n

\u5219\u58f0\u660e\u4e86\u4e00\u4e2a CUDA Kernel，\u5b83\u7684\u51fd\u6570\u4f53\u6700\u7ec8\u5728 GPU \u4e0a\u6267\u884c\u3002<\/p>\n

\u56e0\u6b64\u540c\u4e00\u4e2a\u6e90\u7801\u6587\u4ef6\u4e2d\u5b9e\u9645\u4e0a\u540c\u65f6\u5b58\u5728：<\/p>\n

CPU \u4e16\u754c
\n+
\nGPU \u4e16\u754c<\/p>\n

\u8fd9\u4e5f\u662f CUDA \u7f16\u8bd1\u4e0e\u666e\u901a C++ \u7f16\u8bd1\u76f8\u6bd4\u6700\u6838\u5fc3\u7684\u533a\u522b\u4e4b\u4e00\u3002<\/p>\n

\n

\u4e09\u3001nvcc：CUDA \u7f16\u8bd1\u6d41\u7a0b\u7684\u603b\u5165\u53e3<\/h2>\n

\u65e2\u7136\u540c\u4e00\u4e2a CUDA \u6e90\u6587\u4ef6\u4e2d\u53ef\u80fd\u540c\u65f6\u5b58\u5728 Host Code \u548c Device Code，\u90a3\u4e48\u5c31\u9700\u8981\u4e00\u4e2a\u80fd\u591f\u534f\u8c03\u4e24\u5957\u7f16\u8bd1\u6d41\u7a0b\u7684\u5165\u53e3\u3002<\/p>\n

\u8fd9\u4e2a\u5165\u53e3\u5c31\u662f：<\/p>\n

nvcc<\/p>\n

nvcc \u5168\u79f0\u53ef\u4ee5\u7406\u89e3\u4e3a NVIDIA CUDA Compiler Driver\u3002<\/p>\n

\u8fd9\u91cc\u201cDriver\u201d\u8fd9\u4e2a\u8bcd\u5f88\u91cd\u8981，\u56e0\u4e3a\u6211\u4eec\u4e0d\u5e94\u8be5\u7b80\u5355\u5730\u628a\u5b83\u7406\u89e3\u6210：<\/p>\n

nvcc \u81ea\u5df1\u4e00\u4e2a\u7a0b\u5e8f\u628a CPU \u4ee3\u7801\u548c GPU \u4ee3\u7801\u5168\u90e8\u76f4\u63a5\u7f16\u8bd1\u5b8c\u3002<\/p>\n

\u66f4\u51c6\u786e\u7684\u5fc3\u667a\u6a21\u578b\u5e94\u8be5\u662f：<\/p>\n

hello.cu
\n \u2193
\n nvcc
\n \u2502
\n \u251c\u2500\u2500 Host Code
\n \u2502 \u2193
\n \u2502 Host C++ Compiler
\n \u2502 Linux \u4e0b\u901a\u5e38\u662f gcc \/ g++
\n \u2502 \u2193
\n \u2502 Host Object Code
\n \u2502
\n \u2514\u2500\u2500 Device Code
\n \u2193
\n NVIDIA CUDA Device \u7f16\u8bd1\u94fe
\n \u2193
\n PTX \/ GPU Binary \u7b49 Device Code<\/p>\n

\u968f\u540e Device \u4ee3\u7801\u4f1a\u88ab\u5d4c\u5165 Host \u4fa7\u76ee\u6807\u6587\u4ef6，\u5e76\u4e0e CUDA Runtime \u7b49\u6240\u9700\u5185\u5bb9\u5b8c\u6210\u540e\u7eed\u94fe\u63a5，\u6700\u7ec8\u5f97\u5230\u4e00\u4e2a\u666e\u901a\u7684\u53ef\u6267\u884c\u6587\u4ef6\u3002<\/p>\n

\u56e0\u6b64\u53ef\u4ee5\u5148\u628a nvcc \u7406\u89e3\u6210：<\/p>\n

CUDA \u6574\u4e2a\u7f16\u8bd1\u6d41\u7a0b\u7684\u201c\u603b\u8c03\u5ea6\u5165\u53e3\u201d\u3002<\/p>\n

NVIDIA \u5b98\u65b9\u6587\u6863\u5bf9\u8fd9\u4e00\u70b9\u7684\u63cf\u8ff0\u4e5f\u5f88\u660e\u786e：CUDA \u6e90\u6587\u4ef6\u53ef\u4ee5\u540c\u65f6\u5305\u542b\u4f20\u7edf C++ Host \u4ee3\u7801\u548c GPU Device \u51fd\u6570，nvcc \u4f1a\u534f\u8c03 Host \u7f16\u8bd1\u5668\u4ee5\u53ca NVIDIA \u81ea\u5df1\u7684 Device \u7f16\u8bd1\u5de5\u5177\u94fe\u5b8c\u6210\u5904\u7406\u3002<\/p>\n

\n

\u56db\u3001\u5148\u51c6\u5907 CUDA \u7f16\u8bd1\u73af\u5883<\/h2>\n

\u65e2\u7136\u73b0\u5728\u7684\u76ee\u6807\u9996\u5148\u662f：<\/p>\n

.cu
\n \u2193
\nnvcc
\n \u2193
\n\u53ef\u6267\u884c\u6587\u4ef6<\/p>\n

\u90a3\u4e48\u7b2c\u4e00\u4ef6\u4e8b\u5c31\u662f\u68c0\u67e5\u5f53\u524d\u673a\u5668\u6709\u6ca1\u6709 nvcc\u3002<\/p>\n

\u6267\u884c：<\/p>\n

nvcc –version<\/span><\/p>\n

\u4e00\u5f00\u59cb\u5f53\u524d Ubuntu \u4e91\u670d\u52a1\u5668\u4e2d\u5e76\u6ca1\u6709\u5b89\u88c5 nvcc：<\/p>\n

Command 'nvcc' not found<\/p>\n

\u56e0\u6b64\u5b89\u88c5 CUDA Toolkit \u540e，\u518d\u6b21\u68c0\u67e5：<\/p>\n

nvcc –version<\/span><\/p>\n

\u5f53\u524d\u73af\u5883\u8f93\u51fa\u4e3a：<\/p>\n

nvcc: NVIDIA (R) Cuda compiler driver
\nCuda compilation tools, release 12.0, V12.0.140<\/p>\n

\u4e5f\u5c31\u662f\u8bf4，\u5f53\u524d\u4f7f\u7528\u7684\u662f CUDA 12.0 \u7684\u7f16\u8bd1\u5de5\u5177\u94fe\u3002<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

\u81f3\u6b64，\u81f3\u5c11\u5df2\u7ecf\u5177\u5907\u4e86：<\/p>\n

CUDA Toolkit \u2705
\nnvcc \u2705
\nCUDA 12.0 \u2705<\/p>\n

\n

\u4e94\u3001\u6ca1\u6709 GPU，\u8fd8\u80fd\u4e0d\u80fd\u5b66\u4e60 CUDA \u7f16\u8bd1？<\/h2>\n

\u8fd9\u91cc\u53c8\u9047\u5230\u4e86\u53e6\u4e00\u4e2a\u95ee\u9898\u3002<\/p>\n

\u5f53\u524d\u4f7f\u7528\u7684\u662f\u4e00\u53f0\u4e91\u670d\u52a1\u5668，\u6240\u4ee5\u7ee7\u7eed\u68c0\u67e5：<\/p>\n

nvidia-smi<\/p>\n

\u7ed3\u679c\u53d1\u73b0\u547d\u4ee4\u672c\u8eab\u4e0d\u5b58\u5728\u3002<\/p>\n

\u968f\u540e\u8fdb\u4e00\u6b65\u6267\u884c：<\/p>\n

lspci |<\/span> grep<\/span> -i<\/span> nvidia<\/p>\n

\u540c\u6837\u6ca1\u6709\u4efb\u4f55 NVIDIA \u8bbe\u5907\u8f93\u51fa\u3002<\/p>\n

\u56e0\u6b64\u5bf9\u4e8e\u8fd9\u53f0\u865a\u62df\u673a，\u66f4\u51c6\u786e\u7684\u8bf4\u6cd5\u662f：<\/p>\n

\u5f53\u524d\u865a\u62df\u673a\u5e76\u6ca1\u6709\u5411\u7cfb\u7edf\u66b4\u9732\u53ef\u7528\u7684 NVIDIA GPU \u8bbe\u5907\u3002<\/p>\n

\u8fd9\u610f\u5473\u7740\u76ee\u524d\u4e0d\u80fd\u771f\u6b63\u5b8c\u6210：<\/p>\n

CPU \u53d1\u8d77 Kernel
\n \u2193
\nCUDA Runtime \/ Driver
\n \u2193
\nNVIDIA GPU
\n \u2193
\n\u771f\u6b63\u6267\u884c GPU \u6307\u4ee4<\/p>\n

\u4f46\u662f\u8fd9\u91cc\u9700\u8981\u628a\u7f16\u8bd1\u548c\u8fd0\u884c\u660e\u786e\u533a\u5206\u5f00\u3002<\/p>\n

1. \u7f16\u8bd1\u9636\u6bb5<\/h3>\n

hello.cu
\n \u2193
\nnvcc
\n \u2193
\nHost + Device Code \u7f16\u8bd1
\n \u2193
\n\u94fe\u63a5
\n \u2193
\nhello<\/p>\n

\u8fd9\u4e00\u9636\u6bb5\u7684\u91cd\u70b9\u662f\u628a CUDA \u6e90\u7801\u8f6c\u6362\u6210\u6700\u7ec8\u53ef\u6267\u884c\u6587\u4ef6\u3002<\/p>\n

2. \u8fd0\u884c\u9636\u6bb5<\/h3>\n

.\/hello
\n \u2193
\nCPU \u542f\u52a8\u8fdb\u7a0b
\n \u2193
\n\u7a0b\u5e8f\u53d1\u8d77 CUDA Kernel
\n \u2193
\nCUDA Driver \u627e\u5230 NVIDIA GPU
\n \u2193
\nGPU \u6267\u884c Kernel<\/p>\n

\u5f53\u524d\u673a\u5668\u7f3a\u5c11\u7684\u662f\u7b2c\u4e8c\u90e8\u5206\u771f\u6b63\u80fd\u591f\u6267\u884c Kernel \u7684 GPU \u73af\u5883\u3002<\/p>\n

\u56e0\u6b64\u672c\u6587\u5148\u5c06\u76ee\u6807\u6536\u7a84：<\/p>\n

\u5148\u8dd1\u901a CUDA \u6e90\u7801\u7684\u7f16\u8bd1\u94fe\u8def，\u6682\u65f6\u4e0d\u8ba8\u8bba\u771f\u6b63\u7684 GPU \u8fd0\u884c\u8fc7\u7a0b\u3002<\/p>\n

\u8fd9\u5e76\u4e0d\u59a8\u788d\u6211\u4eec\u8ba4\u8bc6 .cu\u3001nvcc\u3001Host\/Device Code \u4ee5\u53ca CUDA C++ \u7684\u57fa\u672c\u4ee3\u7801\u7ed3\u6784\u3002<\/p>\n

\n

\u516d\u3001\u4e3a\u4ec0\u4e48 CUDA \u6e90\u6587\u4ef6\u901a\u5e38\u4f7f\u7528 .cu<\/h2>\n

\u666e\u901a C++ \u6587\u4ef6\u901a\u5e38\u4f7f\u7528：<\/p>\n

.cpp
\n.cc
\n.cxx<\/p>\n

\u800c\u5f53\u4e00\u4e2a\u6587\u4ef6\u4e2d\u5305\u542b CUDA Device \u4ee3\u7801\u6216\u8005 Host + Device \u6df7\u5408\u4ee3\u7801\u65f6，\u901a\u5e38\u4f7f\u7528：<\/p>\n

.cu<\/p>\n

\u4f8b\u5982：<\/p>\n

hello.cu
\nvector_add.cu
\nmatrix_mul.cu<\/p>\n

\u53ef\u4ee5\u5148\u5efa\u7acb\u8fd9\u6837\u4e00\u4e2a\u6700\u76f4\u89c2\u7684\u533a\u5206：<\/p>\n

\u7eaf Host C++ \u4ee3\u7801
\n \u2193
\n.cpp
\n \u2193
\ng++ \/ clang++<\/p>\n

\u5305\u542b CUDA Device Code
\n \u2193
\n.cu
\n \u2193
\nnvcc<\/p>\n

\u9700\u8981\u6ce8\u610f\u7684\u662f：<\/p>\n

.cu \u5e76\u4e0d\u610f\u5473\u7740\u8fd9\u4e2a\u6587\u4ef6\u91cc\u9762\u5168\u90e8\u90fd\u662f GPU \u4ee3\u7801\u3002<\/p>\n

\u4e00\u4e2a .cu \u5b8c\u5168\u53ef\u4ee5\u540c\u65f6\u5305\u542b：<\/p>\n

\u666e\u901a C++ Host Code
\n+
\nCUDA Device Code<\/p>\n

\u8fd9\u4e5f\u662f\u5b9e\u9645 CUDA \u5de5\u7a0b\u4e2d\u975e\u5e38\u5e38\u89c1\u7684\u5199\u6cd5\u3002<\/p>\n

\u9664\u6b64\u4e4b\u5916，\u8fd8\u7ecf\u5e38\u4f1a\u770b\u5230：<\/p>\n

.cuh<\/p>\n

\u5b83\u901a\u5e38\u88ab\u4f5c\u4e3a CUDA \u76f8\u5173\u5934\u6587\u4ef6\u7684\u547d\u540d\u7ea6\u5b9a，\u4f8b\u5982：<\/p>\n

kernel.cuh
\noperator.cuh<\/p>\n

\u4e0d\u8fc7 .cuh \u66f4\u591a\u5c5e\u4e8e\u5de5\u7a0b\u4e0a\u7684\u5e38\u89c1\u7ea6\u5b9a，\u5e76\u4e0d\u662f\u8bf4\u6240\u6709 CUDA \u5934\u6587\u4ef6\u90fd\u5fc5\u987b\u4f7f\u7528\u8fd9\u4e2a\u540e\u7f00\u3002<\/p>\n

\n

\u4e03\u3001\u4e3a\u4ec0\u4e48\u6ca1\u6709 #include <cuda_runtime.h> \u4e5f\u80fd\u5199 __global__<\/h2>\n

\u7b2c\u4e00\u6b21\u770b\u5230\u4e0b\u9762\u8fd9\u6bb5\u4ee3\u7801\u65f6，\u5f88\u5bb9\u6613\u4ea7\u751f\u4e00\u4e2a\u7591\u95ee：<\/p>\n

#<\/span>include<\/span> <cstdio><\/span><\/span><\/p>\n

__global__ void<\/span> helloFromGPU<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from GPU\\\\n"<\/span>)<\/span>;<\/span>
\n}<\/span><\/p>\n

\u8fd9\u91cc\u53ea\u6709：<\/p>\n

#<\/span>include<\/span> <cstdio><\/span><\/span><\/p>\n

\u5374\u6ca1\u6709\u663e\u5f0f\u5f15\u5165：<\/p>\n

#<\/span>include<\/span> <cuda_runtime.h><\/span><\/span><\/p>\n

\u90a3\u4e48\u4e3a\u4ec0\u4e48：<\/p>\n

__global__<\/p>\n

\u4ecd\u7136\u80fd\u591f\u6b63\u5e38\u8bc6\u522b？<\/p>\n

\u6838\u5fc3\u539f\u56e0\u5728\u4e8e：<\/p>\n

CUDA C++ \u7684\u201c\u8bed\u8a00\u6269\u5c55\u201d\u548c CUDA Runtime \u63d0\u4f9b\u7684\u201c\u5e93 API\u201d\u4e0d\u662f\u540c\u4e00\u4e2a\u5c42\u6b21\u3002<\/p>\n

1. CUDA \u8bed\u8a00\u6269\u5c55<\/h3>\n

\u4f8b\u5982：<\/p>\n

__global__
\n__device__
\n__host__
\n<<<\/span><<\/span>grid,<\/span> block>><\/span>><\/span><\/p>\n

\u8fd9\u4e9b\u5185\u5bb9\u4e0d\u662f\u67d0\u4e2a\u666e\u901a\u51fd\u6570\u5e93\u4e2d\u63d0\u4f9b\u7684\u51fd\u6570\u3002<\/p>\n

\u5b83\u4eec\u5c5e\u4e8e CUDA C++ \u7684\u8bed\u8a00\u6269\u5c55，\u7531 CUDA \u7f16\u8bd1\u5668\u76f4\u63a5\u8bc6\u522b\u3002<\/p>\n

\u8fd9\u4e2a\u611f\u89c9\u548c\u666e\u901a C++ \u4e2d\u7684：<\/p>\n

sizeof<\/span>
\nnew<\/span>
\nauto<\/span><\/p>\n

\u6709\u4e9b\u7c7b\u4f3c\u3002<\/p>\n

\u4f8b\u5982\u6211\u4eec\u4f7f\u7528：<\/p>\n

sizeof<\/span>(<\/span>int<\/span>)<\/span><\/p>\n

\u5e76\u4e0d\u662f\u56e0\u4e3a\u67d0\u4e2a\u5934\u6587\u4ef6\u91cc\u9762\u58f0\u660e\u4e86\u4e00\u4e2a\u53eb sizeof() \u7684\u51fd\u6570，\u800c\u662f\u56e0\u4e3a\u7f16\u8bd1\u5668\u672c\u8eab\u5c31\u8ba4\u8bc6\u8fd9\u79cd\u8bed\u6cd5\u3002<\/p>\n

\u540c\u7406：<\/p>\n

__global__ void<\/span> kernel<\/span>(<\/span>)<\/span><\/p>\n

\u4ee5\u53ca：<\/p>\n

kernel<<<\/span><<\/span>1<\/span>,<\/span> 1<\/span>>><\/span>><\/span>(<\/span>)<\/span>;<\/span><\/p>\n

\u662f nvcc \u80fd\u591f\u8bc6\u522b\u7684 CUDA C++ \u8bed\u8a00\u7ed3\u6784\u3002<\/p>\n

\u6240\u4ee5\u8fd9\u91cc\u53ef\u4ee5\u5efa\u7acb：<\/p>\n

CUDA \u8bed\u8a00\u6269\u5c55
\n\u2502
\n\u251c\u2500\u2500 __global__
\n\u251c\u2500\u2500 __device__
\n\u251c\u2500\u2500 __host__
\n\u2514\u2500\u2500 <<< >>> Kernel Launch Syntax<\/p>\n

\u2193<\/p>\n

\u7531 CUDA \u7f16\u8bd1\u5668\u7406\u89e3
\n\u800c\u4e0d\u662f\u666e\u901a\u5e93\u51fd\u6570\u8c03\u7528<\/p>\n

2. CUDA Runtime API<\/h3>\n

\u53e6\u4e00\u7c7b\u5219\u662f：<\/p>\n

cudaMalloc<\/span>(<\/span>)<\/span>
\ncudaMemcpy<\/span>(<\/span>)<\/span>
\ncudaFree<\/span>(<\/span>)<\/span>
\ncudaDeviceSynchronize<\/span>(<\/span>)<\/span>
\ncudaGetLastError<\/span>(<\/span>)<\/span><\/p>\n

\u8fd9\u4e9b\u5c31\u4e0d\u662f\u8bed\u8a00\u8bed\u6cd5\u4e86，\u800c\u662f\u771f\u6b63\u7684 CUDA Runtime API\u3002<\/p>\n

\u6982\u5ff5\u4e0a\u53ef\u4ee5\u7406\u89e3\u4e3a：<\/p>\n

CUDA Runtime API
\n \u2193
\n\u5bf9\u5e94\u51fd\u6570\u58f0\u660e
\n \u2193
\ncuda_runtime.h
\n \u2193
\n\u6700\u7ec8\u8fd8\u9700\u8981\u94fe\u63a5 CUDA Runtime<\/p>\n

\u56e0\u6b64\u4ece\u4ee3\u7801\u9605\u8bfb\u7684\u89d2\u5ea6\u6765\u770b，\u6211\u4eec\u9700\u8981\u660e\u786e：<\/p>\n

__global__
\n \u2193
\n\u8bed\u8a00\u6269\u5c55<\/p>\n

cudaMalloc()
\n \u2193
\nRuntime API<\/p>\n

\u8fd9\u662f\u4e24\u4e2a\u5b8c\u5168\u4e0d\u540c\u7684\u5c42\u6b21\u3002<\/p>\n

\u4e00\u4e2a\u5bb9\u6613\u5ffd\u7565\u7684\u7ec6\u8282<\/h4>\n

\u4e25\u683c\u6765\u8bf4，\u4f7f\u7528 nvcc \u7f16\u8bd1 .cu \u6587\u4ef6\u65f6，nvcc \u9ed8\u8ba4\u4f1a\u5728\u7ffb\u8bd1\u5355\u5143\u9876\u90e8\u9690\u5f0f\u5305\u542b cuda_runtime.h\u3002<\/p>\n

\u56e0\u6b64\u67d0\u4e9b CUDA Runtime API \u5373\u4f7f\u6e90\u7801\u4e2d\u6ca1\u6709\u624b\u5199：<\/p>\n

#<\/span>include<\/span> <cuda_runtime.h><\/span><\/span><\/p>\n

\u4e5f\u53ef\u80fd\u4ecd\u7136\u80fd\u591f\u6210\u529f\u7f16\u8bd1\u3002<\/p>\n

\u4f46\u662f\u8fd9\u5e76\u4e0d\u4f1a\u6539\u53d8\u4e0a\u9762\u7684\u6838\u5fc3\u533a\u522b：<\/p>\n

__global__\u3001Kernel Launch \u8bed\u6cd5\u5c5e\u4e8e CUDA C++ \u8bed\u8a00\u5c42；cudaMalloc\u3001cudaMemcpy \u7b49\u5c5e\u4e8e CUDA Runtime API \u5c42\u3002<\/p>\n

\u5b9e\u9645\u5199\u4ee3\u7801\u65f6，\u5982\u679c\u76f4\u63a5\u4f7f\u7528 CUDA Runtime API，\u663e\u5f0f\u5199\u51fa：<\/p>\n

#<\/span>include<\/span> <cuda_runtime.h><\/span><\/span><\/p>\n

\u901a\u5e38\u4e5f\u66f4\u6709\u5229\u4e8e\u8868\u8fbe\u5f53\u524d\u6e90\u7801\u7684\u4f9d\u8d56\u5173\u7cfb\u3002<\/p>\n

\n

\u516b\u3001Host \u548c Device \u4e4b\u95f4\u5b58\u5728\u660e\u786e\u7684\u51fd\u6570\u8c03\u7528\u8fb9\u754c<\/h2>\n

\u8ba4\u8bc6\u4e86\u8bed\u8a00\u6269\u5c55\u4ee5\u540e，\u8fd8\u9700\u8981\u7ee7\u7eed\u89e3\u51b3\u4e00\u4e2a\u66f4\u91cd\u8981\u7684\u95ee\u9898：<\/p>\n

CPU \u4ee3\u7801\u548c GPU \u4ee3\u7801\u867d\u7136\u53ef\u4ee5\u5199\u5728\u540c\u4e00\u4e2a .cu \u6587\u4ef6\u4e2d，\u4f46\u662f\u5b83\u4eec\u5e76\u4e0d\u662f\u5904\u4e8e\u540c\u4e00\u4e2a\u6267\u884c\u4e16\u754c\u3002<\/p>\n

\u53ef\u4ee5\u5148\u628a\u51fd\u6570\u5206\u6210\u4e09\u7c7b\u3002<\/p>\n

1. \u666e\u901a\u51fd\u6570：\u9ed8\u8ba4\u5c5e\u4e8e Host<\/h3>\n

\u4f8b\u5982：<\/p>\n

void<\/span> func<\/span>(<\/span>)<\/span>
\n{<\/span>
\n}<\/span><\/p>\n

\u5982\u679c\u6ca1\u6709\u6dfb\u52a0\u4efb\u4f55 CUDA \u6267\u884c\u7a7a\u95f4\u4fee\u9970\u7b26，\u90a3\u4e48\u5b83\u9ed8\u8ba4\u5c31\u662f Host \u51fd\u6570\u3002<\/p>\n

\u4e5f\u53ef\u4ee5\u7c97\u7565\u770b\u6210：<\/p>\n

__host__ void<\/span> func<\/span>(<\/span>)<\/span>
\n{<\/span>
\n}<\/span><\/p>\n

\u5b83\u7531 CPU \u6267\u884c\u3002<\/p>\n

\n

2. __device__：GPU \u5185\u90e8\u51fd\u6570<\/h3>\n

\u4f8b\u5982：<\/p>\n

__device__ int<\/span> add<\/span>(<\/span>int<\/span> a,<\/span> int<\/span> b)<\/span>
\n{<\/span>
\n return<\/span> a +<\/span> b;<\/span>
\n}<\/span><\/p>\n

\u8fd9\u4e2a\u51fd\u6570：<\/p>\n

\u5728 GPU \u4e0a\u6267\u884c
\n\u5e76\u7531 Device Code \u8c03\u7528<\/p>\n

\u4f8b\u5982 Kernel \u5185\u90e8\u53ef\u4ee5\u8c03\u7528：<\/p>\n

__global__ void<\/span> kernel<\/span>(<\/span>)<\/span>
\n{<\/span>
\n int<\/span> x =<\/span> add<\/span>(<\/span>1<\/span>,<\/span> 2<\/span>)<\/span>;<\/span>
\n}<\/span><\/p>\n

\n

3. __global__：Kernel<\/h3>\n

\u4f8b\u5982：<\/p>\n

__global__ void<\/span> kernel<\/span>(<\/span>)<\/span>
\n{<\/span>
\n}<\/span><\/p>\n

\u6700\u57fa\u7840\u7684\u5fc3\u667a\u6a21\u578b\u5c31\u662f：<\/p>\n

Host \u4fa7\u53d1\u8d77 Kernel Launch
\n \u2193
\nkernel<<<grid, block>>>()
\n \u2193
\nDevice \/ GPU \u6267\u884c Kernel \u51fd\u6570\u4f53<\/p>\n

\u56e0\u6b64 __global__ \u51fd\u6570\u5c31\u662f\u8fde\u63a5 Host \u4e0e Device \u7684\u4e00\u4e2a\u975e\u5e38\u91cd\u8981\u7684\u5165\u53e3\u3002<\/p>\n

\n

\u4e5d\u3001\u4e3a\u4ec0\u4e48 Kernel \u91cc\u9762\u80fd\u591f\u8c03\u7528 printf<\/h2>\n

\u8fd9\u91cc\u53c8\u51fa\u73b0\u4e86\u4e00\u4e2a\u5f88\u6709\u610f\u601d\u7684\u95ee\u9898\u3002<\/p>\n

\u6211\u4eec\u7684 Kernel \u662f：<\/p>\n

__global__ void<\/span> helloFromGPU<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from GPU\\\\n"<\/span>)<\/span>;<\/span>
\n}<\/span><\/p>\n

\u4f46\u662f\u6211\u4eec\u77e5\u9053，\u666e\u901a\u7a0b\u5e8f\u4e2d\u7684 printf \u672c\u6765\u5c5e\u4e8e C\/C++ \u8fd0\u884c\u5e93\u4e2d\u7684\u51fd\u6570\u3002<\/p>\n

\u90a3\u4e48 GPU \u4ee3\u7801\u4e3a\u4ec0\u4e48\u80fd\u591f\u8c03\u7528\u5b83？<\/p>\n

\u5982\u679c\u7b80\u5355\u7406\u89e3\u6210：<\/p>\n

GPU Kernel
\n \u2193
\n\u76f4\u63a5\u8c03\u7528 CPU \u4e0a glibc \u7684 printf<\/p>\n

\u90a3\u80af\u5b9a\u662f\u4e0d\u5bf9\u7684\u3002<\/p>\n

\u56e0\u4e3a Host Code \u4e0e Device Code \u4e4b\u95f4\u5b58\u5728\u660e\u786e\u8fb9\u754c\u3002<\/p>\n

\u771f\u6b63\u7684\u539f\u56e0\u662f：<\/p>\n

CUDA \u4e13\u95e8\u4e3a Device Code \u63d0\u4f9b\u4e86 Device-side printf \u652f\u6301\u3002<\/p>\n

\u6240\u4ee5 Kernel \u4e2d：<\/p>\n

printf<\/span>(<\/span>"Hello from GPU\\\\n"<\/span>)<\/span>;<\/span><\/p>\n

\u8c03\u7528\u7684\u662f CUDA \u652f\u6301\u7684 Device \u4fa7\u8f93\u51fa\u673a\u5236，\u800c\u4e0d\u662f\u8ba9 GPU \u8dd1\u53bb\u76f4\u63a5\u6267\u884c CPU \u4e0a\u7684\u666e\u901a libc \u5b9e\u73b0\u3002<\/p>\n

\u53ef\u4ee5\u7b80\u5355\u7406\u89e3\u4e3a：<\/p>\n

GPU Thread
\n \u2193
\nDevice-side printf
\n \u2193
\n\u8bb0\u5f55\u8f93\u51fa\u53c2\u6570 \/ \u4fe1\u606f
\n \u2193
\n\u7531 CUDA \u8fd0\u884c\u673a\u5236\u6700\u7ec8\u628a\u683c\u5f0f\u5316\u8f93\u51fa\u5448\u73b0\u5728 Host \u8f93\u51fa\u6d41\u4e2d<\/p>\n

\u56e0\u6b64\u8fd9\u91cc\u771f\u6b63\u5e94\u8be5\u5efa\u7acb\u7684\u89c4\u5219\u662f：<\/p>\n

Device Code \u53ea\u80fd\u8c03\u7528\u5177\u6709 Device \u4fa7\u5b9e\u73b0\u6216\u8005\u88ab CUDA \u660e\u786e\u652f\u6301\u7684\u51fd\u6570\u3002<\/p>\n

\n

\u5341\u3001\u4e3a\u4ec0\u4e48\u4e0d\u80fd\u5728 Kernel \u4e2d\u968f\u4fbf\u4f7f\u7528 std::cout \u548c std::vector<\/h2>\n

\u6709\u4e86\u4e0a\u9762\u7684\u8fb9\u754c\u4ee5\u540e，\u4e0b\u9762\u8fd9\u79cd\u4ee3\u7801\u5c31\u5f88\u597d\u7406\u89e3\u4e86：<\/p>\n

#<\/span>include<\/span> <iostream><\/span><\/span><\/p>\n

__global__ void<\/span> kernel<\/span>(<\/span>)<\/span>
\n{<\/span>
\n std::<\/span>cout <<<\/span> "hello"<\/span>;<\/span>
\n}<\/span><\/p>\n

\u666e\u901a\u60c5\u51b5\u4e0b\u8fd9\u4e0d\u80fd\u76f4\u63a5\u6210\u7acb\u3002<\/p>\n

\u539f\u56e0\u4e0d\u662f：<\/p>\n

<iostream> \u6ca1\u6709 include \u6210\u529f<\/p>\n

\u800c\u662f：<\/p>\n

\u5f53\u524d\u6b63\u5728\u7f16\u8bd1 Device Code
\n \u2193
\nstd::cout \u5c5e\u4e8e\u666e\u901a Host C++ \u6807\u51c6\u5e93\u5b9e\u73b0
\n \u2193
\n\u6ca1\u6709\u53ef\u4f9b\u8fd9\u4e2a Kernel \u76f4\u63a5\u6267\u884c\u7684 Device \u7248\u672c
\n \u2193
\n\u65e0\u6cd5\u8fd9\u6837\u4f7f\u7528<\/p>\n

\u7c7b\u4f3c\u5730：<\/p>\n

#<\/span>include<\/span> <vector><\/span><\/span><\/p>\n

__global__ void<\/span> kernel<\/span>(<\/span>)<\/span>
\n{<\/span>
\n std::<\/span>vector<<\/span>int<\/span>><\/span> values;<\/span>
\n}<\/span><\/p>\n

\u666e\u901a\u7684 std::vector \u540c\u6837\u4e0d\u80fd\u76f4\u63a5\u88ab\u5f53\u6210 GPU Device \u5bb9\u5668\u5728 Kernel \u4e2d\u4f7f\u7528\u3002<\/p>\n

\u6240\u4ee5\u53ef\u4ee5\u5f62\u6210\u4e00\u4e2a\u975e\u5e38\u91cd\u8981\u7684\u5224\u65ad\u8fc7\u7a0b：<\/p>\n

\u5f53\u524d\u4ee3\u7801\u662f\u5426\u4f4d\u4e8e __global__ \/ __device__ \u51fd\u6570？
\n \u2193
\n \u662f
\n \u2193
\n\u51c6\u5907\u8c03\u7528\u67d0\u4e2a\u51fd\u6570 \/ \u4f7f\u7528\u67d0\u4e2a\u5e93\u80fd\u529b
\n \u2193
\n\u5b83\u662f\u5426\u5b58\u5728 Device \u4fa7\u53ef\u6267\u884c\u5b9e\u73b0？
\n \u2193
\n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n \u2193 \u2193
\n \u6709 \u6ca1\u6709
\n \u2193 \u2193
\n \u53ef\u4ee5 \u4e0d\u53ef\u4ee5<\/p>\n

printf \u4e4b\u6240\u4ee5\u80fd\u591f\u4f7f\u7528，\u4e0d\u662f\u56e0\u4e3a GPU \u53ef\u4ee5\u968f\u610f\u4f7f\u7528 CPU \u6807\u51c6\u5e93，\u800c\u662f\u56e0\u4e3a CUDA \u5bf9\u5b83\u8fdb\u884c\u4e86 Device \u4fa7\u652f\u6301\u3002<\/p>\n

\u540e\u9762\u7684 CUDA \u751f\u6001\u4e2d\u8fd8\u4f1a\u5b58\u5728 Thrust\u3001libcu++ \u7b49\u4e13\u95e8\u9762\u5411 CUDA \/ Device \u7684 C++ \u80fd\u529b，\u8fd9\u5c5e\u4e8e\u8fdb\u4e00\u6b65\u5b66\u4e60\u7684\u5185\u5bb9\u3002\u5f53\u524d\u9636\u6bb5\u5148\u628a\u6700\u57fa\u672c\u7684 Host\/Device \u8fb9\u754c\u5efa\u7acb\u8d77\u6765\u5373\u53ef\u3002<\/p>\n

\n

\u5341\u4e00\u3001\u7f16\u5199\u7b2c\u4e00\u4e2a CUDA C++ \u7a0b\u5e8f<\/h2>\n

\u5728\u8ba4\u8bc6\u5b8c\u524d\u9762\u7684\u4ee3\u7801\u7ed3\u6784\u4ee5\u540e，\u5c31\u53ef\u4ee5\u771f\u6b63\u521b\u5efa\u7b2c\u4e00\u4efd CUDA \u6e90\u6587\u4ef6：<\/p>\n

vim<\/span> hello.cu<\/p>\n

\u5185\u5bb9\u5982\u4e0b：<\/p>\n

\/\/ hello.cu<\/span><\/p>\n

#<\/span>include<\/span> <cstdio><\/span><\/span><\/p>\n

__global__ void<\/span> helloFromGPU<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from GPU\\\\n"<\/span>)<\/span>;<\/span>
\n}<\/span><\/p>\n

int<\/span> main<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from CPU\\\\n"<\/span>)<\/span>;<\/span><\/p>\n

helloFromGPU<<<\/span><<\/span>1<\/span>,<\/span> 1<\/span>>><\/span>><\/span>(<\/span>)<\/span>;<\/span><\/p>\n

return<\/span> 0<\/span>;<\/span>
\n}<\/span><\/p>\n

\u5b9e\u9645\u6e90\u7801\u5982\u4e0b\u56fe\u6240\u793a：<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

\u8fd9\u6bb5\u4ee3\u7801\u867d\u7136\u5f88\u77ed，\u4f46\u662f\u5df2\u7ecf\u5305\u542b\u4e86 CUDA \u7a0b\u5e8f\u6700\u6838\u5fc3\u7684\u4e24\u4e2a\u6267\u884c\u4e16\u754c\u3002<\/p>\n

Host \u4fa7<\/h3>\n

int<\/span> main<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from CPU\\\\n"<\/span>)<\/span>;<\/span><\/p>\n

helloFromGPU<<<\/span><<\/span>1<\/span>,<\/span> 1<\/span>>><\/span>><\/span>(<\/span>)<\/span>;<\/span><\/p>\n

return<\/span> 0<\/span>;<\/span>
\n}<\/span><\/p>\n

\u8fd9\u91cc\u7531 CPU \u6267\u884c\u3002<\/p>\n

\u5176\u4e2d：<\/p>\n

helloFromGPU<<<\/span><<\/span>1<\/span>,<\/span> 1<\/span>>><\/span>><\/span>(<\/span>)<\/span>;<\/span><\/p>\n

\u4e0d\u662f\u666e\u901a C++ \u51fd\u6570\u8c03\u7528，\u800c\u662f CUDA Kernel Launch\u3002<\/p>\n

\u76ee\u524d\u6682\u65f6\u53ea\u9700\u8981\u77e5\u9053：<\/p>\n

<<<1, 1>>><\/p>\n

\u63cf\u8ff0\u4e86 Kernel \u7684\u6267\u884c\u914d\u7f6e\u3002<\/p>\n

\u540e\u9762\u518d\u7ee7\u7eed\u628a\u5b83\u548c：<\/p>\n

Grid
\nBlock
\nThread<\/p>\n

\u6620\u5c04\u8d77\u6765\u3002<\/p>\n

Device \u4fa7<\/h3>\n

__global__ void<\/span> helloFromGPU<\/span>(<\/span>)<\/span>
\n{<\/span>
\n printf<\/span>(<\/span>"Hello from GPU\\\\n"<\/span>)<\/span>;<\/span>
\n}<\/span><\/p>\n

\u8fd9\u4e2a\u51fd\u6570\u4f53\u6700\u7ec8\u7531 GPU \u6267\u884c\u3002<\/p>\n

\u56e0\u6b64\u6574\u4e2a\u6e90\u7801\u53ef\u4ee5\u7b80\u5355\u753b\u6210：<\/p>\n

hello.cu
\n\u2502
\n\u251c\u2500\u2500 Host Code
\n\u2502 \u2514\u2500\u2500 main()
\n\u2502 \u2514\u2500\u2500 CPU \u6267\u884c
\n\u2502
\n\u2514\u2500\u2500 Device Code
\n \u2514\u2500\u2500 helloFromGPU()
\n \u2514\u2500\u2500 GPU \u6267\u884c<\/p>\n

\n

\u5341\u4e8c\u3001\u4f7f\u7528 nvcc \u5b8c\u6210\u7b2c\u4e00\u6b21 CUDA \u7f16\u8bd1<\/h2>\n

\u6e90\u7801\u5199\u5b8c\u4ee5\u540e，\u6267\u884c：<\/p>\n

nvcc hello.cu -o<\/span> hello<\/p>\n

\u8fd9\u6761\u547d\u4ee4\u548c\u6211\u4eec\u4ee5\u524d\u4f7f\u7528\u7684：<\/p>\n

g++ main.cpp -o<\/span> main<\/p>\n

\u5728\u4f7f\u7528\u4f53\u9a8c\u4e0a\u975e\u5e38\u76f8\u4f3c\u3002<\/p>\n

\u5176\u4e2d：<\/p>\n

hello.cu<\/p>\n

\u662f\u8f93\u5165 CUDA \u6e90\u6587\u4ef6\u3002<\/p>\n

\u800c：<\/p>\n

-o hello<\/p>\n

\u8868\u793a\u6700\u7ec8\u751f\u6210\u540d\u4e3a：<\/p>\n

hello<\/p>\n

\u7684\u53ef\u6267\u884c\u6587\u4ef6\u3002<\/p>\n

\u6267\u884c\u4ee5\u540e\u7ec8\u7aef\u6ca1\u6709\u6253\u5370\u4efb\u4f55\u9519\u8bef\u4fe1\u606f，\u7136\u540e\u7ee7\u7eed：<\/p>\n

ls<\/span> -l<\/span><\/p>\n

\u53ef\u4ee5\u770b\u5230：<\/p>\n

hello
\nhello.cu<\/p>\n

\u5176\u4e2d hello \u5df2\u7ecf\u5e26\u6709\u53ef\u6267\u884c\u6743\u9650\u3002<\/p>\n

\u5b9e\u9645\u7ed3\u679c\u5982\u4e0b：<\/p>\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n

\u8fd9\u610f\u5473\u7740\u7b2c\u4e00\u6761 CUDA \u7f16\u8bd1\u94fe\u5df2\u7ecf\u771f\u6b63\u8dd1\u901a：<\/p>\n

hello.cu
\n \u2193
\nnvcc hello.cu -o hello
\n \u2193
\n\u89e3\u6790 CUDA C++
\n \u2193
\nHost Code + Device Code \u5206\u522b\u8fdb\u5165\u5bf9\u5e94\u7f16\u8bd1\u6d41\u7a0b
\n \u2193
\nDevice Code \u88ab\u7ec4\u5408\u8fdb Host \u4fa7\u76ee\u6807\u6587\u4ef6
\n \u2193
\n\u5b8c\u6210\u94fe\u63a5
\n \u2193
\nhello<\/p>\n

\u8fd9\u91cc\u8fd8\u6709\u4e00\u4e2a\u548c g++ \u5f88\u76f8\u4f3c\u7684\u73b0\u8c61：<\/p>\n

\u7f16\u8bd1\u6210\u529f\u65f6，\u7f16\u8bd1\u5668\u901a\u5e38\u4e0d\u4f1a\u4e13\u95e8\u8f93\u51fa\u4e00\u6bb5\u201c\u7f16\u8bd1\u6210\u529f\u201d\u3002\u6ca1\u6709\u62a5\u9519，\u672c\u8eab\u5f80\u5f80\u5c31\u662f\u6210\u529f\u3002<\/p>\n

\u6700\u7ec8\u518d\u901a\u8fc7 ls -l \u6216\u8005\u76f4\u63a5\u68c0\u67e5\u76ee\u6807\u6587\u4ef6，\u5c31\u80fd\u591f\u786e\u8ba4\u53ef\u6267\u884c\u6587\u4ef6\u5df2\u7ecf\u751f\u6210\u3002<\/p>\n

\n

\u5341\u4e09\u3001\u73b0\u5728\u4e3a\u4ec0\u4e48\u6682\u65f6\u4e0d\u6267\u884c .\/hello<\/h2>\n

\u6b63\u5e38\u60c5\u51b5\u4e0b，\u7f16\u8bd1\u5b8c\u6210\u4ee5\u540e\u4e0b\u4e00\u6b65\u81ea\u7136\u4f1a\u6267\u884c：<\/p>\n

.\/hello<\/p>\n

\u4f46\u662f\u5f53\u524d\u4e91\u670d\u52a1\u5668\u6ca1\u6709\u5411\u865a\u62df\u673a\u66b4\u9732 NVIDIA GPU\u3002<\/p>\n

\u56e0\u6b64\u5982\u679c\u771f\u6b63\u8fdb\u5165\u8fd0\u884c\u9636\u6bb5：<\/p>\n

.\/hello
\n \u2193
\nCPU \u6267\u884c main()
\n \u2193
\n\u9047\u5230 helloFromGPU<<<1, 1>>>()
\n \u2193
\nCUDA Runtime \u5c1d\u8bd5\u542f\u52a8 Kernel
\n \u2193
\n\u9700\u8981\u53ef\u7528 NVIDIA Device<\/p>\n

\u5f53\u524d\u73af\u5883\u5c31\u5728\u6700\u540e\u8fd9\u91cc\u7f3a\u5c11\u771f\u5b9e GPU\u3002<\/p>\n

\u6240\u4ee5\u672c\u6587\u6682\u65f6\u53ea\u5b8c\u6210：<\/p>\n

CUDA Source Code
\n \u2193
\nCUDA Compilation
\n \u2193
\nExecutable<\/p>\n

\u81f3\u4e8e\u771f\u6b63：<\/p>\n

Host Launch Kernel
\n \u2193
\nGPU Execute Kernel<\/p>\n

\u540e\u9762\u53ef\u4ee5\u6362\u5230\u5e26 NVIDIA GPU \u7684\u73af\u5883，\u4f8b\u5982\u516c\u53f8\u7684 Jetson Orin \u6216\u8005 GPU \u4e91\u4e3b\u673a\u4ee5\u540e\u7ee7\u7eed\u9a8c\u8bc1\u3002<\/p>\n

\u8fd9\u4e5f\u518d\u6b21\u8bf4\u660e：<\/p>\n

CUDA Toolkit \/ nvcc
\n\u89e3\u51b3\u7684\u662f：CUDA \u7a0b\u5e8f\u80fd\u4e0d\u80fd\u88ab\u7f16\u8bd1<\/p>\n

NVIDIA GPU + Driver
\n\u89e3\u51b3\u7684\u662f：Device Code \u80fd\u4e0d\u80fd\u771f\u6b63\u88ab\u6267\u884c<\/p>\n

\u8fd9\u4e24\u4e2a\u95ee\u9898\u9700\u8981\u660e\u786e\u533a\u5206\u3002<\/p>\n

\n

\u5341\u56db\u3001\u91cd\u65b0\u6574\u7406\u6574\u4e2a CUDA C++ \u7f16\u8bd1\u5fc3\u667a\u6a21\u578b<\/h2>\n

\u5230\u8fd9\u91cc，\u53ef\u4ee5\u628a\u672c\u6587\u6240\u6709\u5185\u5bb9\u91cd\u65b0\u4e32\u6210\u4e00\u6761\u5b8c\u6574\u94fe\u8def\u3002<\/p>\n

\u9996\u5148，\u666e\u901a C++ \u7a0b\u5e8f\u662f：<\/p>\n

.cpp
\n \u2193
\ng++
\n \u2193
\nCPU \u53ef\u6267\u884c\u6587\u4ef6<\/p>\n

\u4f46\u662f CUDA \u7a0b\u5e8f\u4e2d\u540c\u65f6\u5b58\u5728：<\/p>\n

Host Code
\n+
\nDevice Code<\/p>\n

\u56e0\u6b64\u901a\u5e38\u4f7f\u7528：<\/p>\n

.cu<\/p>\n

\u5e76\u4ea4\u7ed9：<\/p>\n

nvcc<\/p>\n

\u4f5c\u4e3a\u6574\u4e2a CUDA \u7f16\u8bd1\u6d41\u7a0b\u7684\u5165\u53e3\u3002<\/p>\n

\u7136\u540e：<\/p>\n

hello.cu
\n \u2502
\n \u2193
\n nvcc
\n \u2502
\n \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2534\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510
\n \u2193 \u2193
\n Host Code Device Code
\n \u2193 \u2193
\n Host C++ Compiler CUDA Device Toolchain
\n \u2193 \u2193
\n CPU Object GPU Code \/ Fatbinary
\n \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u252c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518
\n \u2193
\n \u6700\u7ec8\u94fe\u63a5
\n \u2193
\n hello<\/p>\n

\u4e0e\u6b64\u540c\u65f6，\u5728\u6e90\u7801\u5c42\u9762\u8fd8\u9700\u8981\u533a\u5206：<\/p>\n

CUDA C++
\n\u2502
\n\u251c\u2500\u2500 \u8bed\u8a00\u6269\u5c55
\n\u2502 \u251c\u2500\u2500 __global__
\n\u2502 \u251c\u2500\u2500 __device__
\n\u2502 \u251c\u2500\u2500 __host__
\n\u2502 \u2514\u2500\u2500 <<< >>>
\n\u2502
\n\u2514\u2500\u2500 CUDA Runtime API
\n \u251c\u2500\u2500 cudaMalloc
\n \u251c\u2500\u2500 cudaMemcpy
\n \u251c\u2500\u2500 cudaFree
\n \u2514\u2500\u2500 cudaDeviceSynchronize<\/p>\n

\u6700\u540e\u8fd8\u5fc5\u987b\u4fdd\u6301 Host\/Device \u8fb9\u754c\u610f\u8bc6：<\/p>\n

Host Code
\n \u2193
\nCPU \u4e16\u754c<\/p>\n

Device Code
\n \u2193
\nGPU \u4e16\u754c<\/p>\n

GPU Kernel \u4e0d\u662f\u628a CPU \u7684\u6574\u4e2a C++ \u8fd0\u884c\u73af\u5883\u76f4\u63a5\u642c\u8fc7\u53bb\u3002<\/p>\n

Kernel \u5185\u90e8\u53ea\u80fd\u4f7f\u7528\u80fd\u591f\u88ab\u7f16\u8bd1\u4e3a Device Code \u7684\u51fd\u6570\u548c\u80fd\u529b\u3002<\/p>\n

\u56e0\u6b64：<\/p>\n

Device-side printf
\n \u2705 CUDA \u63d0\u4f9b\u652f\u6301<\/p>\n

\u666e\u901a std::cout
\n \u274c \u4e0d\u80fd\u76f4\u63a5\u5728 Kernel \u4e2d\u8fd9\u6837\u4f7f\u7528<\/p>\n

\u666e\u901a std::vector
\n \u274c \u4e0d\u80fd\u76f4\u63a5\u4f5c\u4e3a Device \u5bb9\u5668\u8fd9\u6837\u4f7f\u7528<\/p>\n

\u7406\u89e3\u4e86\u8fd9\u4e00\u5c42\u4ee5\u540e，\u540e\u9762\u518d\u7ee7\u7eed\u5b66\u4e60：<\/p>\n

cudaMalloc
\ncudaMemcpy
\ncudaFree
\nthreadIdx
\nblockIdx
\nblockDim
\nGrid \/ Block \/ Thread<\/p>\n

\u5c31\u4e0d\u518d\u53ea\u662f\u5355\u72ec\u8bb0\u51e0\u4e2a API \u6216\u51e0\u4e2a\u5173\u952e\u5b57，\u800c\u662f\u80fd\u591f\u628a\u5b83\u4eec\u653e\u56de\u5b8c\u6574\u7684 CUDA \u7a0b\u5e8f\u6267\u884c\u94fe\u8def\u4e2d\u7406\u89e3\u3002<\/p>\n

\n

\u603b\u7ed3<\/h2>\n

\u672c\u6587\u5e76\u6ca1\u6709\u6025\u7740\u8fdb\u5165 CUDA \u7684\u77e9\u9635\u4e58\u6cd5\u3001\u663e\u5b58\u4f18\u5316\u6216\u8005\u590d\u6742 Kernel，\u800c\u662f\u5148\u89e3\u51b3\u4e86\u6700\u57fa\u7840\u7684\u4ee3\u7801\u4e0e\u7f16\u8bd1\u95ee\u9898\u3002<\/p>\n

\u9996\u5148，CUDA C++ \u7a0b\u5e8f\u901a\u5e38\u540c\u65f6\u5305\u542b CPU \u4fa7\u7684 Host Code \u548c GPU \u4fa7\u7684 Device Code，\u56e0\u6b64\u5b83\u4e0e\u666e\u901a\u53ea\u9762\u5411 CPU \u7684 C++ \u7a0b\u5e8f\u5b58\u5728\u660e\u663e\u533a\u522b\u3002<\/p>\n

\u5176\u6b21，.cu \u662f\u5305\u542b CUDA Device Code \u6216 Host\/Device \u6df7\u5408\u4ee3\u7801\u65f6\u6700\u5e38\u89c1\u7684\u6e90\u6587\u4ef6\u540e\u7f00，\u800c nvcc \u5219\u8d1f\u8d23\u534f\u8c03\u6574\u4e2a CUDA \u7f16\u8bd1\u8fc7\u7a0b\u3002Host Code \u4f1a\u4ea4\u7ed9 Host C++ \u7f16\u8bd1\u5668\u5904\u7406，Device Code \u5219\u8fdb\u5165 NVIDIA \u7684 Device \u7f16\u8bd1\u94fe，\u6700\u540e\u518d\u7ec4\u5408\u6210\u6700\u7ec8\u7684\u53ef\u6267\u884c\u6587\u4ef6\u3002<\/p>\n

\u518d\u6b21，\u9700\u8981\u660e\u786e CUDA \u7684\u8bed\u8a00\u6269\u5c55\u548c CUDA Runtime API \u5e76\u4e0d\u662f\u540c\u4e00\u4e2a\u4e1c\u897f\u3002__global__\u3001__device__ \u4ee5\u53ca Kernel Launch \u8bed\u6cd5\u5c5e\u4e8e CUDA C++ \u8bed\u8a00\u5c42，\u800c cudaMalloc\u3001cudaMemcpy \u7b49\u5219\u5c5e\u4e8e Runtime API\u3002<\/p>\n

\u6b64\u5916，Host \u4e0e Device \u4e4b\u95f4\u5b58\u5728\u660e\u786e\u7684\u6267\u884c\u8fb9\u754c\u3002\u666e\u901a CPU \u51fd\u6570\u548c Host \u6807\u51c6\u5e93\u80fd\u529b\u4e0d\u80fd\u60f3\u5f53\u7136\u5730\u653e\u5230 Kernel \u5185\u90e8\u4f7f\u7528；Kernel \u4e2d\u80fd\u591f\u8c03\u7528\u7684\u51fd\u6570\u5fc5\u987b\u5177\u6709 Device \u4fa7\u53ef\u6267\u884c\u5b9e\u73b0\u3002printf \u662f CUDA \u4e13\u95e8\u652f\u6301\u7684 Device-side \u51fd\u6570，\u56e0\u6b64\u624d\u53ef\u4ee5\u51fa\u73b0\u5728 Kernel \u4e2d\u3002<\/p>\n

\u6700\u540e，\u6211\u4eec\u5df2\u7ecf\u771f\u6b63\u5b8c\u6210：<\/p>\n

hello.cu
\n \u2193
\nnvcc hello.cu -o hello
\n \u2193
\nhello<\/p>\n

\u7b2c\u4e00\u6b21 CUDA C++ \u7f16\u8bd1\u94fe\u8def\u5df2\u7ecf\u8dd1\u901a\u3002<\/p>\n

\u5f53\u524d\u4e91\u670d\u52a1\u5668\u867d\u7136\u6ca1\u6709 NVIDIA GPU，\u56e0\u6b64\u8fd8\u4e0d\u80fd\u771f\u6b63\u6267\u884c Kernel，\u4f46\u662f\u8fd9\u5e76\u4e0d\u5f71\u54cd\u6211\u4eec\u5148\u628a CUDA \u7684\u6e90\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4ee5\u53ca nvcc \u7f16\u8bd1\u6a21\u578b\u5efa\u7acb\u8d77\u6765\u3002<\/p>\n

\u4e0b\u4e00\u6b65\u5c31\u53ef\u4ee5\u5728\u8fd9\u4e2a\u57fa\u7840\u4e0a\u7ee7\u7eed\u5411\u771f\u6b63\u7684 CUDA Runtime \u7f16\u7a0b\u63a8\u8fdb，\u628a：<\/p>\n

Host Memory
\n \u2193
\ncudaMalloc
\n \u2193
\nDevice Memory
\n \u2193
\ncudaMemcpy
\n \u2193
\nKernel<\/p>\n

\u8fd9\u4e00\u6761\u6570\u636e\u4e0e\u6267\u884c\u94fe\u771f\u6b63\u6620\u5c04\u5230\u4ee3\u7801\u4e2d\u3002<\/p>\n

\n

\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\"<\/p>\n","protected":false},"excerpt":{"rendered":"

🔥 \u672c\u6587\u4e13\u680f：CUDA 🌸\u4f5c\u8005\u4e3b\u9875：\u52aa\u529b\u52aa\u529b\u518d\u52aa\u529bwz 💪 \u4eca\u65e5\u535a\u5ba2\u52b1\u5fd7\u8bed\u5f55：\u7406\u8bba\u771f\u6b63\u5f00\u59cb\u53d8\u6210\u81ea\u5df1\u7684\u4e1c\u897f，\u5f80\u5f80\u4e0d\u662f\u5728\u201c\u542c\u61c2\u201d\u7684\u65f6\u5019，\u800c\u662f\u5728\u4eb2\u624b\u628a\u7b2c\u4e00\u6bb5\u4ee3\u7801\u7f16\u8bd1\u51fa\u6765\u7684\u65f6\u5019\u3002 \u601d\u7ef4\u5bfc\u56fe
\nCUDA \u7406\u8bba\u57fa\u7840\u2193
\n\u4e0d\u80fd\u53ea\u505c\u7559\u5728 Grid \/ Block \/ Thread \/ \u663e\u5b58\u8fd9\u4e9b\u6982\u5ff5\u2193
\n\u5f00\u59cb\u771f\u6b63\u7f16\u5199 CUDA C \u4ee3\u7801\u2193
\n\u9996\u5148\u89e3\u51b3\u4e00\u4e2a\u95ee\u9898：CUDA <\/p>\n","protected":false},"author":2,"featured_media":112467,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[55,190],"topic":[],"class_list":["post-112474","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-c","tag-190"],"yoast_head":"\n\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/112474.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"🔥 \u672c\u6587\u4e13\u680f：CUDA 🌸\u4f5c\u8005\u4e3b\u9875：\u52aa\u529b\u52aa\u529b\u518d\u52aa\u529bwz 💪 \u4eca\u65e5\u535a\u5ba2\u52b1\u5fd7\u8bed\u5f55：\u7406\u8bba\u771f\u6b63\u5f00\u59cb\u53d8\u6210\u81ea\u5df1\u7684\u4e1c\u897f，\u5f80\u5f80\u4e0d\u662f\u5728\u201c\u542c\u61c2\u201d\u7684\u65f6\u5019，\u800c\u662f\u5728\u4eb2\u624b\u628a\u7b2c\u4e00\u6bb5\u4ee3\u7801\u7f16\u8bd1\u51fa\u6765\u7684\u65f6\u5019\u3002 \u601d\u7ef4\u5bfc\u56fe CUDA \u7406\u8bba\u57fa\u7840\u2193 \u4e0d\u80fd\u53ea\u505c\u7559\u5728 Grid \/ Block \/ Thread \/ \u663e\u5b58\u8fd9\u4e9b\u6982\u5ff5\u2193 \u5f00\u59cb\u771f\u6b63\u7f16\u5199 CUDA C \u4ee3\u7801\u2193 \u9996\u5148\u89e3\u51b3\u4e00\u4e2a\u95ee\u9898：CUDA\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/112474.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-03T12:01:11+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261003120107-6ac0ee83ef20f.gif\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/112474.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/112474.html\",\"name\":\"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-03T12:01:11+00:00\",\"dateModified\":\"2026-10-03T12:01:11+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/112474.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/112474.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/112474.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/112474.html","og_locale":"zh_CN","og_type":"article","og_title":"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"🔥 \u672c\u6587\u4e13\u680f：CUDA 🌸\u4f5c\u8005\u4e3b\u9875：\u52aa\u529b\u52aa\u529b\u518d\u52aa\u529bwz 💪 \u4eca\u65e5\u535a\u5ba2\u52b1\u5fd7\u8bed\u5f55：\u7406\u8bba\u771f\u6b63\u5f00\u59cb\u53d8\u6210\u81ea\u5df1\u7684\u4e1c\u897f，\u5f80\u5f80\u4e0d\u662f\u5728\u201c\u542c\u61c2\u201d\u7684\u65f6\u5019，\u800c\u662f\u5728\u4eb2\u624b\u628a\u7b2c\u4e00\u6bb5\u4ee3\u7801\u7f16\u8bd1\u51fa\u6765\u7684\u65f6\u5019\u3002 \u601d\u7ef4\u5bfc\u56fe CUDA \u7406\u8bba\u57fa\u7840\u2193 \u4e0d\u80fd\u53ea\u505c\u7559\u5728 Grid \/ Block \/ Thread \/ \u663e\u5b58\u8fd9\u4e9b\u6982\u5ff5\u2193 \u5f00\u59cb\u771f\u6b63\u7f16\u5199 CUDA C \u4ee3\u7801\u2193 \u9996\u5148\u89e3\u51b3\u4e00\u4e2a\u95ee\u9898：CUDA","og_url":"https:\/\/www.wsisp.com\/helps\/112474.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-03T12:01:11+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/10\/20261003120107-6ac0ee83ef20f.gif"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"7 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/112474.html","url":"https:\/\/www.wsisp.com\/helps\/112474.html","name":"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-03T12:01:11+00:00","dateModified":"2026-10-03T12:01:11+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/112474.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/112474.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/112474.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u3010CUDA \u5165\u95e8\u7cfb\u5217\u3011CUDA C++ \u4ee3\u7801\u7ed3\u6784\u3001Host\/Device \u8fb9\u754c\u4e0e nvcc \u7f16\u8bd1\u6d41\u7a0b\u4e0e\u7f16\u7a0b\u5b9e\u6218"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/112474","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=112474"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/112474\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/112467"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=112474"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=112474"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=112474"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=112474"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}