{"id":344,"date":"2024-09-03T13:37:17","date_gmt":"2024-09-03T04:37:17","guid":{"rendered":"https:\/\/kkaneko-lab.com\/?p=344"},"modified":"2024-09-03T18:17:58","modified_gmt":"2024-09-03T09:17:58","slug":"%e3%80%90cuda-net%e3%80%91c-cli%e3%82%92%e4%bd%bf%e3%81%a3%e3%81%a6c%e3%81%a7gpu%e4%b8%a6%e5%88%97%e8%a8%88%e7%ae%97","status":"publish","type":"post","link":"https:\/\/kkaneko-lab.com\/?p=344","title":{"rendered":"\u3010CUDA + .Net\u3011C++\/CLI\u3092\u4f7f\u3063\u3066C#\u3067CUDA"},"content":{"rendered":"\n

0.\u3000\u306f\u3058\u3081\u306b<\/h2>\n\n\n\n

\u3000CUDA\u306fC\/C++\u306e\u307f\u3092\u30b5\u30dd\u30fc\u30c8\u3057\u307e\u3059\u304c\uff0cC++\/CLI\u3092\u4f7f\u3046\u3053\u3068\u3067C\uff03\u3067\u3082CUDA\u306b\u3088\u308b\u8a08\u7b97\u3092\u7c21\u5358\u306b\u5b9f\u88c5\u3067\u304d\u307e\u3059\uff0eC++\/CLI\u3068\u306fC++\u306bC#\u306e\u30de\u30cd\u30fc\u30b8\u30c9\u306a\u6a5f\u80fd\u3092\u8ffd\u52a0\u3057\u305f\u8a00\u8a9e\u3067\uff0cC++\u3084CUDA\u3067\u8a18\u8ff0\u3057\u305f\u30b3\u30fc\u30c9\u3092\uff0cC#\u3067\u547c\u3073\u51fa\u305b\u308b\u5f62\u5f0f\u306eDLL\u3068\u3057\u3066\u30d3\u30eb\u30c9\u3067\u304d\u307e\u3059\uff0e\u305d\u306e\u305f\u3081\u4e0b\u56f3\u306e\u3088\u3046\u306b\u30bd\u30ea\u30e5\u30fc\u30b7\u30e7\u30f3\u5185\u3067\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u5206\u3051\u3066CUDA\u306b\u95a2\u3059\u308b\u51e6\u7406\u3092C++\/CLI\u3067\u8a18\u8ff0\u3059\u308b\u3053\u3068\u3067\uff0cC#\u3067\u3082GPU\u306b\u3088\u308b\u4e26\u5217\u8a08\u7b97\u51e6\u7406\u3092\u5b9f\u88c5\u3067\u304d\u307e\u3059\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u3000\u305d\u3053\u3067\u672c\u7a3f\u3067\u306f\uff0cC++\/CLI\u3092\u6d3b\u7528\u3057\u3066\uff0cC#\u306b\u304a\u3044\u3066CUDA\u3067\u30d9\u30af\u30c8\u30eb\u548c\u3092\u8a08\u7b97<\/mark><\/strong>\u3059\u308b\u30b5\u30f3\u30d7\u30eb\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u4f5c\u6210\u3057\u307e\u3059\uff0e<\/p>\n\n\n

\n
\"\"<\/figure>\n<\/div>\n\n\n

\u30bd\u30fc\u30b9\u30b3\u30fc\u30c9\uff1a<\/mark>https:\/\/github.com\/kkaneko1090\/CUDACLITest<\/a><\/mark><\/strong><\/p>\n\n\n\n

\uff11\uff0e\u4e8b\u524d\u6e96\u5099<\/h2>\n\n\n\n

\u2460Visual studio community\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n

\u2461CUDA\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\uff0e\u4eca\u56de\u306fCUDA Toolkit 12.3\u3067\u5b9f\u88c5\u3057\u307e\u3059\uff0ehttps:\/\/developer.nvidia.com\/cuda-toolkit<\/a><\/p>\n\n\n\n

\u2462\u30ef\u30fc\u30af\u30ed\u30fc\u30c9\u3067\u300cC++\u306b\u3088\u308b\u30c7\u30b9\u30af\u30c8\u30c3\u30d7\u958b\u767a\u300d\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u3000\u52a0\u3048\u3066\uff0cC++\/CLI\u3092\u4f7f\u7528\u3059\u308b\u305f\u3081\u306b\uff0cVisual studio Installer\u306e\u500b\u5225\u30b3\u30f3\u30dd\u30fc\u30cd\u30f3\u30c8\u3067\u30d3\u30eb\u30c9\u30c4\u30fc\u30eb\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\uff0eCUDA 12.3\u3067\u306f MSVC v14.3X\u307e\u3067\u3057\u304b\u30b5\u30dd\u30fc\u30c8\u3055\u308c\u3066\u3044\u306a\u3044\u306e\u4e0b\u56f3\u306e\u300cv143 \u30d3\u30eb\u30c9\u30c4\u30fc\u30eb\u306eC++\/CLI\u30b5\u30dd\u30fc\u30c8\uff0814.39-17.9\uff09\u300d\u3092\u8ffd\u52a0\u3057\u307e\u3057\u305f\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

2\uff0e\u30d7\u30ed\u30b8\u30a7\u30af\u30c81\uff08C++\/CLI\uff09\u306e\u4f5c\u6210<\/h2>\n\n\n\n

\u3000\u6700\u521d\u306b\uff0cCUDA\u95a2\u6570\u3092\u683c\u7d0d\u3059\u308bDLL\u3092\u4f5c\u6210\u3059\u308b\u305f\u3081\u306eC++\/CLI\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u4f5c\u6210\u3057\u307e\u3059\uff0e\u4e0b\u56f3\u306e\u3088\u3046\u306b\uff0cCUDA 12.3 Runtime\u3092\u9078\u629e\u3057\u3066\uff0c\u307e\u305a\u306fCUDA\u306b\u5bfe\u5fdc\u3057\u305fC++\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u4f5c\u6210\u3057\u307e\u3059\uff0e\u4eca\u56de\u306f\uff0c\u201dCUDACLITest\u201d\u3068\u3044\u3046\u540d\u524d\u3067\u4f5c\u6210\u3057\u307e\u3057\u305f\uff0e\u3053\u306eC++(CUDA)\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u306e\u8a2d\u5b9a\u3092\u5909\u66f4\u3059\u308b\u3053\u3068\u3067\uff0cC++\/CLI\u306e\u6a5f\u80fd\u3092\u4f7f\u3048\u308b\u3088\u3046\u306b\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u3000DLL\u3092\u4f5c\u6210\u3059\u308b\u305f\u3081\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u306a\u306e\u3067\uff0c\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u306e\u30d7\u30ed\u30d1\u30c6\u30a3\u3067\u300c\u69cb\u6210\u306e\u7a2e\u985e\u300d\u3092DLL\u306b\u5909\u66f4\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u3000\u6b21\u306bC++\/CLI\u30d7\u30ed\u30d1\u30c6\u30a3\u3092\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u8a2d\u5b9a\u3059\u308b\u3053\u3068\u3067\uff0c.Net 8.0\u3092\u30bf\u30fc\u30b2\u30c3\u30c8\u3068\u3057\u305fDLL\u306e\u4f5c\u6210\u304c\u53ef\u80fd\u3068\u306a\u308a\u307e\u3059\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

\u3000\u5ff5\u306e\u305f\u3081\uff0cMSVC\u30c4\u30fc\u30eb\u30bb\u30c3\u30c8\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u78ba\u8a8d\u3057\u307e\u3059\uff0e\u4e0a\u8a18\u306e\u901a\u308a\uff0cCUDA 12.3 \u3067\u306fMSVC\u306fv14.3X\u307e\u3067\u3057\u304b\u30b5\u30dd\u30fc\u30c8\u3055\u308c\u3066\u3044\u306a\u3044\u306e\u3067\uff0c\u3082\u3057\u3053\u308c\u304c14.4X\u3068\u306a\u3063\u3066\u3044\u308b\u5834\u5408\u306f\uff0c\u4ee5\u4e0b\u306e\u3088\u3046\u306b14.3X\u306b\u5909\u66f4\u3057\u306a\u3044\u3068\u30d3\u30eb\u30c9\u304c\u901a\u3089\u306a\u3044\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\uff0e<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

3. \u30d9\u30af\u30c8\u30eb\u548c\u3092\u8a08\u7b97\u3059\u308bCUDA\u30b3\u30fc\u30c9\u3092\u8a18\u8ff0<\/h2>\n\n\n\n

\u3000\u4ee5\u4e0b\u306e\u3088\u3046\u306b4\u3064\u306e\u30d5\u30a1\u30a4\u30eb\u3067DLL\u3092\u69cb\u7bc9\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n

\u2460Kernel.cu\u30fb\u30fb\u30fb\u30fb\u30fb\u30d9\u30af\u30c8\u30eb\u548c\u3092\u8a08\u7b97\u3059\u308b\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u3092\u5b9a\u7fa9<\/p>\n\n\n\n

\u2461Kernel.cuh \u30fb\u30fb\u30fb\u30fbKernel.cu\u306e\u30d8\u30c3\u30c0\u30fc\u30d5\u30a1\u30a4\u30eb<\/p>\n\n\n\n

\u2462Calculator.cpp\u30fb\u30fb\u30fb\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u3092\u547c\u3073\u51fa\u3057\uff0c\u8a08\u7b97\u3092\u5b9f\u884c\u3059\u308b\u95a2\u6570\u3092\u5b9a\u7fa9<\/p>\n\n\n\n

\u2463Calculator.h\u30fb\u30fb\u30fb\u30fbCalculator.cpp\u306e\u30d8\u30c3\u30c0\u30fc\u30d5\u30a1\u30a4\u30eb<\/p>\n\n\n\n

Kernel.cu\u3067\u306f\uff0c\u4ee5\u4e0b\u306e\u3088\u3046\u306b2\u30d9\u30af\u30c8\u30eb\u306e\u548c\u3092\u8a08\u7b97\u3059\u308b\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u3092\u8a18\u8ff0\u3057\u307e\u3057\u305f\uff0eCUDA\u3067\u306e\u4e26\u5217\u8a08\u7b97\u306e\u4ed5\u69d8\u306f\uff0c\u4ee5\u4e0b\u306e\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u30ac\u30a4\u30c9\u3092\u53c2\u7167\u3057\u3066\u304f\u3060\u3055\u3044\uff0ehttps:\/\/docs.nvidia.com\/cuda\/cuda-c-programming-guide\/index.html<\/p>\n\n\n\n

\/\/Kernel.cu\n\n#pragma once\n#include "Kernel.cuh"\n\n\/\/\/ <summary>\n\/\/\/ \u30d9\u30af\u30c8\u30eb\u548c\u3092\u8a08\u7b97\u3059\u308b\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\n\/\/\/ <\/summary>\n\/\/\/ <param name="vec_0">\u30d9\u30af\u30c8\u30eb0<\/param>\n\/\/\/ <param name="vec_1">\u30d9\u30af\u30c8\u30eb1<\/param>\n\/\/\/ <param name="result">\u8a08\u7b97\u7d50\u679c\u306e\u30d9\u30af\u30c8\u30eb<\/param>\n\/\/\/ <param name="length">\u30d9\u30af\u30c8\u30eb\u306e\u9577\u3055<\/param>\n\/\/\/ <returns><\/returns>\n__global__ void CudaAddKernel(float* vec_0, float* vec_1, float* result, int* length) {\n\t\/\/\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\n\tint index = (blockIdx.x * blockDim.x) + threadIdx.x;\n\n\t\/\/\u30a4\u30f3\u30c7\u30c3\u30af\u30b9\u304c\u7bc4\u56f2\u5185\u306e\u3068\u304d\n\tif (index < *length) {\n\t\t\/\/\u30d9\u30af\u30c8\u30eb\u306e\u8981\u7d20\u3069\u3046\u3057\u3092\u8db3\u3057\u5408\u308f\u305b\u308b\n\t\tresult[index] = vec_0[index] + vec_1[index];\n\t}\n}<\/code><\/pre><\/div>\n\n\n\n
Kernel.cuh\u306f\u30d8\u30c3\u30c0\u30fc\u30d5\u30a1\u30a4\u30eb\u3067\uff0c\u4ee5\u4e0b\u306e\u901a\u308a\u3067\u3059\uff0e<\/p>\n\n\n\n
\/\/Kernel.cuh\n\n#pragma\n#include "cuda_runtime.h"\n#include "device_launch_parameters.h"\n\n\/\/\/ <summary>\n\/\/\/ \u30d9\u30af\u30c8\u30eb\u548c\u3092\u8a08\u7b97\u3059\u308b\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\n\/\/\/ <\/summary>\n\/\/\/ <param name="vec_0">\u30d9\u30af\u30c8\u30eb0<\/param>\n\/\/\/ <param name="vec_1">\u30d9\u30af\u30c8\u30eb1<\/param>\n\/\/\/ <param name="result">\u8a08\u7b97\u7d50\u679c\u306e\u30d9\u30af\u30c8\u30eb<\/param>\n\/\/\/ <param name="length">\u30d9\u30af\u30c8\u30eb\u306e\u9577\u3055<\/param>\n\/\/\/ <returns><\/returns>\n__global__ void CudaAddKernel(float* vec_0, float* vec_1, float* result, int* length);<\/code><\/pre><\/div>\n\n\n\n
\u3000Calculator.cpp\u3067\u306f\uff0c\u4e0a\u8a18\u306e\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u201dCudaAddKernel\u201d\u3092\u547c\u3073\u51fa\u3057\u3066GPU\u3067\u8a08\u7b97\u3092\u884c\u3044\uff0c\u8a08\u7b97\u7d50\u679c\u3092\u8fd4\u3059\u30b3\u30fc\u30c9Calculator::Add(array<float>^ vec_0, array<float>^ vec_1)<\/mark><\/strong>\u3092\u8a18\u8ff0\u3057\u307e\u3059\uff0e\u4eca\u56de\u306f\uff0c\u30cd\u30fc\u30e0\u30b9\u30da\u30fc\u30b9\u3092CUDA\uff0c\u30af\u30e9\u30b9\u540d\u3092Calculator\u3068\u3057\u307e\u3057\u305f\uff0e\u30af\u30e9\u30b9\u306e\u5b9a\u7fa9\u306b\u304a\u3044\u3066\uff0cpublic ref <\/mark><\/strong>class \u3068\u306a\u3063\u3066\u304a\u308a\u307e\u3059\u304c\uff0cref\u3092\u4ed8\u3051\u308b\u3053\u3068\u3067C#\u3067\u547c\u3073\u51fa\u3057\u53ef\u80fd\u306a\u30af\u30e9\u30b9\u3068\u3057\u3066\u5b9a\u7fa9\u3067\u304d\u307e\u3059\uff0e<\/p>\n\n\n\n
\u3000\u4ee5\u4e0b\u306e\u3088\u3046\u306bCalculator.cpp\u3092\u8a18\u8ff0\u3057\u307e\u3057\u305f\uff0e\u3053\u306e\u95a2\u6570\u306e\u5f15\u6570\u306e\u578b\u306f\uff0carray<float><\/mark><\/strong>\u3067\u3059\u304c\u3053\u308c\u306f\uff0cC#\u306b\u304a\u3051\u308bfloat[]\u3068\u540c\u7fa9\u3067\u3059\uff0e\u3053\u306e\u3088\u3046\u306bC++\/CLI\u3067\u306fC#\u306e\u30de\u30cd\u30fc\u30b8\u30c9\u306a\u6a5f\u80fd\u3092\u4f7f\u7528\u3067\u304d\u307e\u3059\uff0e\u30de\u30cd\u30fc\u30b8\u30c9\u306a\u5909\u6570\u306b\u306f\u982d\u306b^\u304c\u4ed8\u304d\u307e\u3059\uff0e<\/p>\n\n\n\n
\u3000\u95a2\u6570 “Calculator::Add” \u3067\u306f\uff0c\u307e\u305aC#\u306e\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217 array<float> \u3092 C++ \u306e\u30a2\u30f3\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217 float* \u306b pin_ptr \u3092\u4f7f\u3063\u3066\u5909\u63db\u3057\u307e\u3059\uff0e\u6b21\u306f\u30c7\u30d0\u30a4\u30b9\u5909\u6570\u306e\u30e1\u30e2\u30ea\u78ba\u4fdd\u3068\u30c7\u30fc\u30bf\u8ee2\u9001\u3092\u884c\u3063\u3066\u3044\u307e\u3059\u304c\uff0c\u500b\u3005\u306e\u51e6\u7406\u306f\u901a\u5e38\u306eCUDA\u3068\u540c\u69d8\u3067\u3059\uff0e\u305d\u306e\u5f8c\uff0c\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u3092\u547c\u3073\u51fa\u3057\u3066GPU\u4e0a\u3067\u8a08\u7b97\u3092\u884c\u3044\u307e\u3059\u304c\uff0cC++\/CLI \u3067\u306f\u901a\u5e38\u306e CUDA \u3068\u7570\u306a\u308a \u201dcudaLaunchKernel\u201d \u306b\u3088\u308a\u30ab\u30fc\u30cd\u30eb\u3092\u5b9f\u884c\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n
\u3000\u6700\u5f8c\u306b\u30c7\u30d0\u30a4\u30b9\u3067\u306e\u8a08\u7b97\u7d50\u679c d_result \u3092\u30db\u30b9\u30c8\u5909\u6570 h_result \u306b\u8ee2\u9001\u3057\u307e\u3059\uff0e\u305d\u3057\u3066\u30de\u30cd\u30fc\u30b8\u30c9\u306a\u914d\u5217\u306b\u5909\u63db\u3057\u3066\u304b\u3089return \u3067\u8a08\u7b97\u7d50\u679c\u3092\u8fd4\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n
\/\/Calculator.cpp\n\n#pragma once\n#include "Calculator.h"\n#include "cuda_runtime.h"\n#include "device_launch_parameters.h"\n#include "Kernel.cuh"\n\nnamespace CUDA {\n\t\/\/\/ <summary>\n\t\/\/\/ \u540c\u3058\u9577\u3055\u306e\u30d9\u30af\u30c8\u30eb\u306e\u548c\n\t\/\/\/ <\/summary>\n\t\/\/\/ <param name="vec_0">\u30d9\u30af\u30c8\u30eb0<\/param>\n\t\/\/\/ <param name="vec_1">\u30d9\u30af\u30c8\u30eb1<\/param>\n\t\/\/\/ <returns><\/returns>\n\tarray<float>^ Calculator::Add(array<float>^ vec_0, array<float>^ vec_1) {\n\t\t\n#pragma region \u30db\u30b9\u30c8\u5909\u6570\u306e\u7528\u610f\n\t\t\/\/pin_ptr\u3067\u914d\u5217\u3092\u56fa\u5b9a\u3057\u3001\u30dd\u30a4\u30f3\u30bf\u30fc\u3092\u53d6\u5f97\n\t\tpin_ptr<float> vec_0_pin_ptr = &vec_0[0];\n\t\tpin_ptr<float> vec_1_pin_ptr = &vec_1[0];\n\t\t\/\/\u30a2\u30f3\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217\u306e\u30db\u30b9\u30c8\u5909\u6570\u3092\u7528\u610f\n\t\tfloat* h_vec_0 = vec_0_pin_ptr; \/\/\u30d9\u30af\u30c8\u30eb0\u306e\u30a2\u30f3\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217\n\t\tfloat* h_vec_1 = vec_1_pin_ptr; \/\/\u30d9\u30af\u30c8\u30eb1\u306e\u30a2\u30f3\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217\n\t\tfloat* h_result = new float[vec_0->Length]; \/\/\u8a08\u7b97\u7d50\u679c\u306e\u30a2\u30f3\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217\n\t\tint h_length = vec_0->Length; \/\/\u30d9\u30af\u30c8\u30eb\u306e\u9577\u3055\n#pragma endregion\n\n#pragma region \u30c7\u30d0\u30a4\u30b9\u5909\u6570\u306e\u7528\u610f\n\t\t\/\/\u30db\u30b9\u30c8\u5909\u6570\u306b\u5bfe\u5fdc\u3057\u305f\uff0c\u30c7\u30d0\u30a4\u30b9\u5909\u6570\u3092\u7528\u610f\n\t\tfloat* d_vec_0; \/\/\u30d9\u30af\u30c8\u30eb0\n\t\tfloat* d_vec_1; \/\/\u30d9\u30af\u30c8\u30eb0\n\t\tfloat* d_result; \/\/\u8a08\u7b97\u7d50\u679c\n\t\tint* d_length; \/\/\u30d9\u30af\u30c8\u30eb\u306e\u9577\u3055\n\t\t\/\/\u30c7\u30d0\u30a4\u30b9\u3092\u6307\u5b9a\n\t\tcudaError_t cuda_status = cudaSetDevice(0);\n\t\t\/\/\u30c7\u30d0\u30a4\u30b9\u5909\u6570\u306e\u30e1\u30e2\u30ea\u78ba\u4fdd\n\t\tcuda_status = cudaMalloc(&d_vec_0, h_length * sizeof(float));\n\t\tcuda_status = cudaMalloc(&d_vec_1, h_length * sizeof(float));\n\t\tcuda_status = cudaMalloc(&d_result, h_length * sizeof(float));\n\t\tcuda_status = cudaMalloc(&d_length, sizeof(int));\n\t\t\/\/\u30db\u30b9\u30c8\u5909\u6570\u306e\u5024\u3092\u30c7\u30d0\u30a4\u30b9\u5909\u6570\u306b\u30b3\u30d4\u30fc\n\t\tcuda_status = cudaMemcpy(d_vec_0, h_vec_0, h_length * sizeof(float), cudaMemcpyHostToDevice);\n\t\tcuda_status = cudaMemcpy(d_vec_1, h_vec_1, h_length * sizeof(float), cudaMemcpyHostToDevice);\n\t\tcuda_status = cudaMemcpy(d_length, &h_length, sizeof(int), cudaMemcpyHostToDevice);\n#pragma endregion\n\n#pragma region \u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u306e\u5b9f\u884c\n\t\t\/\/\u4e26\u5217\u8a08\u7b97\u6761\u4ef6\n\t\tint max_thread_num = 256; \/\/\u6700\u5927\u30b9\u30ec\u30c3\u30c9\u6570\n\t\tdim3 grid(h_length \/ max_thread_num + 1); \/\/\u30b0\u30ea\u30c3\u30c9\u306e\u6b21\u5143\n\t\tdim3 block(max_thread_num); \/\/\u30d6\u30ed\u30c3\u30af\u306e\u6b21\u5143\n\t\t\/\/\u5f15\u6570\n\t\tvoid* args[] = { &d_vec_0, &d_vec_1, &d_result, &d_length };\n\t\t\/\/\u30ab\u30fc\u30cd\u30eb\u5b9f\u884c\n\t\tcuda_status = cudaLaunchKernel((const void*)CudaAddKernel, grid, block, args);\n\t\t\/\/\u51e6\u7406\u5f85\u3061\n\t\tcuda_status = cudaDeviceSynchronize();\n#pragma endregion\n\n\t\t\/\/\u8a08\u7b97\u7d50\u679c\u306e\u30c7\u30d0\u30a4\u30b9\u5909\u6570\u306e\u5024\u3092\uff0c\u30db\u30b9\u30c8\u306b\u30b3\u30d4\u30fc\n\t\tcuda_status = cudaMemcpy(h_result, d_result, h_length * sizeof(float), cudaMemcpyDeviceToHost);\n\t\t\/\/\u30a2\u30f3\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217\u3092\uff0c\u30de\u30cd\u30fc\u30b8\u30c9\u914d\u5217\u306b\u5909\u63db\n\t\tarray<float>^ result_managed = gcnew array<float>(vec_0->Length);\n\t\tfor (int i = 0; i < result_managed->Length; i++) result_managed[i] = h_result[i];\n\n\t\t\/\/new\u3067\u78ba\u4fdd\u3057\u305f\u30e1\u30e2\u30ea\u306e\u958b\u653e\n\t\tdelete[] h_result;\n\t\t\/\/\u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u306e\u958b\u653e\n\t\tcuda_status = cudaFree(d_vec_0);\n\t\tcuda_status = cudaFree(d_vec_1);\n\t\tcuda_status = cudaFree(d_result);\n\t\tcuda_status = cudaFree(d_length);\n\n\t\treturn result_managed;\n\t}\n}<\/code><\/pre><\/div>\n\n\n\n
\u4ee5\u4e0a\u304cC++\/CLI\u306b\u3088\u308b\u95a2\u6570\u306e\u5b9f\u88c5\u306b\u306a\u308a\u307e\u3059\uff0e\u3053\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u30d3\u30eb\u30c9\u3059\u308b\u3053\u3068\u3067C#\u3067\u3082\u547c\u3073\u51fa\u3057\u53ef\u80fd\u306aDLL\u3092\u4f5c\u6210\u3067\u304d\u307e\u3059\uff0e<\/p>\n\n\n\n
4.\u3000\u30d7\u30ed\u30b8\u30a7\u30af\u30c82(C#)\u3092\u4f5c\u6210<\/h2>\n\n\n\n
\u3000\u4ee5\u4e0a\u306e\u3088\u3046\u306b\u5b9f\u88c5\u3057\u305fCalculator\u30af\u30e9\u30b9\u3092\u4f7f\u7528\u3059\u308bC#\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u4f5c\u6210\u3057\u307e\u3059\uff0e\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u4eca\u56de\u306f\u30b3\u30f3\u30bd\u30fc\u30eb\u30a2\u30d7\u30ea\u3092\u9078\u629e\u3057\u307e\u3057\u305f\uff0e<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
.Net\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u306f\uff0cC++\/CLI\u306e\u8a2d\u5b9a\u306b\u5408\u308f\u305b\u3066 8.0 \u3068\u8a2d\u5b9a\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u4eca\u56de\u306f\u300cTestConsole\u300d\u3068\u3044\u3046\u540d\u524d\u3067\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u4f5c\u6210\u3057\u307e\u3057\u305f\uff0e\u4f5c\u6210\u5f8c\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u53c2\u7167\u306e\u8ffd\u52a0\u304b\u3089\uff0c\u4e8b\u524d\u306b\u4f5c\u6210\u3057\u305fC++\/CLI\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3092\u53c2\u7167\u8ffd\u52a0\u3057\u307e\u3059\uff0e<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u3000Program.cs\u306b\u306f\u4ee5\u4e0b\u306e\u3088\u3046\u306b\uff0c\u81ea\u4f5c\u3057\u305fCalculator\u30af\u30e9\u30b9\u3092\u4f7f\u3063\u3066\u30d9\u30af\u30c8\u30eb\u548c\u3092GPU\u3067\u8a08\u7b97\u3059\u308b\u30b3\u30fc\u30c9\u3092\u8a18\u8ff0\u3057\u307e\u3057\u305f\uff0e\u6b63\u3057\u304f\u4f9d\u5b58\u95a2\u4fc2\u304c\u8a2d\u5b9a\u3067\u304d\u3066\u3044\u308c\u3070\uff0c \u201dusing CUDA\u201d\u306e\u3088\u3046\u306bC++\/CLI\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u3067\u5b9a\u7fa9\u3057\u305f\u540d\u524d\u7a7a\u9593\u304c\u8a8d\u8b58\u3055\u308c\u307e\u3059\uff0e<\/p>\n\n\n\n
\u3000\u3053\u306e\u30b3\u30fc\u30c9\u3067\u306f\uff0c\u9577\u30551000\u306e\u30d9\u30af\u30c8\u30eb\u3092\u7528\u610f\u3057\u3066\uff0c\u305d\u306e\u548c\u3092\u8a08\u7b97\u3057\u307e\u3059\uff0e\u305d\u306e\u5f8c\uff0c\u8a08\u7b97\u7d50\u679c\u306e\u30d9\u30af\u30c8\u30eb\u306e\u8981\u7d20\u3092\u9806\u756a\u306b\u30b3\u30f3\u30bd\u30fc\u30eb\u8868\u793a\u3057\u3066\u3044\u307e\u3059\uff0e<\/p>\n\n\n\n
\/\/Program.cs\n\nusing CUDA;\nusing System.Diagnostics;\n\n\/\/\u30d9\u30af\u30c8\u30eb\u306e\u9577\u3055\nint vectorLength = 1000;\n\n\/\/2\u3064\u306e\u30d9\u30af\u30c8\u30eb\u3092\u7528\u610f\nfloat[] vector0 = new float[vectorLength];\nfloat[] vector1 = new float[vectorLength];\n\n\/\/\u30d9\u30af\u30c8\u30eb\u306e\u4e2d\u8eab\u3092\u8a2d\u5b9a\nfor(int i = 0; i < vectorLength; i++)\n{\n    vector0[i] = i;\n    vector1[i] = i * 1000;\n}\n\n\/\/GPU\u3067\u30d9\u30af\u30c8\u30eb\u548c\u3092\u8a08\u7b97\nfloat[] sum = CUDA.Calculator.Add(vector0, vector1);\n\n\/\/\u30b3\u30f3\u30bd\u30fc\u30eb\u8868\u793a\nforeach(float value in sum) Console.WriteLine(value); \/\/\u8a08\u7b97\u7d50\u679c<\/code><\/pre><\/div>\n\n\n\n
\u3000\u4ee5\u4e0b\u306e\u3088\u3046\u306b\uff0cGPU\u3092\u4f7f\u3063\u3066\u6b63\u3057\u304f\u8db3\u3057\u7b97\u304c\u5b9f\u884c\u3067\u304d\u3066\u3044\u308b\u3053\u3068\u304c\u308f\u304b\u308a\u307e\u3059\uff0e<\/p>\n\n\n\n
\"\"<\/figure>\n\n\n\n
\u88dc\u8db3<\/h2>\n\n\n\n
C#\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u5074\u306e\u30d7\u30ed\u30d1\u30c6\u30a3\u3067\uff0c\u4e0b\u306e\u3088\u3046\u306b\u30cd\u30a4\u30c6\u30a3\u30d6\u30b3\u30fc\u30c9\u306e\u30c7\u30d0\u30c3\u30b0\u3092\u6709\u52b9\u306b\u3059\u308b\u3068\uff0cC++\/CLI\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u306b\u8a2d\u5b9a\u3057\u305f\u30d6\u30ec\u30fc\u30af\u30dd\u30a4\u30f3\u30c8\u3067\u505c\u6b62\u3059\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\uff0e<\/p>\n\n\n\n
\"\"<\/figure>\n","protected":false},"excerpt":{"rendered":"
0.\u3000\u306f\u3058\u3081\u306b \u3000CUDA\u306fC\/C++\u306e\u307f\u3092\u30b5\u30dd\u30fc\u30c8\u3057\u307e\u3059\u304c\uff0cC++\/CLI\u3092\u4f7f\u3046\u3053\u3068\u3067C\uff03\u3067\u3082CUDA\u306b\u3088\u308b\u8a08\u7b97\u3092\u7c21\u5358\u306b\u5b9f\u88c5\u3067\u304d\u307e\u3059\uff0eC++\/CLI\u3068\u306fC++\u306bC#\u306e\u30de\u30cd\u30fc\u30b8\u30c9\u306a\u6a5f\u80fd\u3092\u8ffd\u52a0\u3057\u305f\u8a00\u8a9e\u3067\uff0cC++\u3084CUDA\u3067 … <\/p>\n
“\u3010CUDA + .Net\u3011C++\/CLI\u3092\u4f7f\u3063\u3066C#\u3067CUDA” \u306e<\/span>\u7d9a\u304d\u3092\u8aad\u3080<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"_locale":"ja","_original_post":"https:\/\/kkaneko-lab.com\/?p=344","footnotes":""},"categories":[5],"tags":[13,19,18],"class_list":["post-344","post","type-post","status-publish","format-standard","hentry","category-5","tag-c","tag-c-cli","tag-cuda","ja"],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=\/wp\/v2\/posts\/344","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=344"}],"version-history":[{"count":24,"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=\/wp\/v2\/posts\/344\/revisions"}],"predecessor-version":[{"id":399,"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=\/wp\/v2\/posts\/344\/revisions\/399"}],"wp:attachment":[{"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=344"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=344"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kkaneko-lab.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=344"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}