{"id":521,"date":"2024-04-27T22:04:42","date_gmt":"2024-04-27T13:04:42","guid":{"rendered":"https:\/\/chocottopro.com\/?p=521"},"modified":"2024-04-27T22:06:44","modified_gmt":"2024-04-27T13:06:44","slug":"pycuda%e3%81%a7%e5%a7%8b%e3%82%81%e3%82%8bgpu%e3%83%97%e3%83%ad%e3%82%b0%e3%83%a9%e3%83%9f%e3%83%b3%e3%82%b0%e5%85%a5%e9%96%80%ef%bc%9a%e3%82%a4%e3%83%b3%e3%82%b9%e3%83%88%e3%83%bc%e3%83%ab%e3%81%8b","status":"publish","type":"post","link":"https:\/\/chocottopro.com\/?p=521","title":{"rendered":"PyCuda\u3067\u59cb\u3081\u308bGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u5165\u9580\uff1a\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u304b\u3089\u5b9f\u8df5\u7684\u306a\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u307e\u3067"},"content":{"rendered":"\n<p>GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306b\u8208\u5473\u304c\u3042\u308bPython\u30e6\u30fc\u30b6\u30fc\u306b\u3068\u3063\u3066\u3001PyCuda\u306f\u5f37\u529b\u306a\u30c4\u30fc\u30eb\u3067\u3059\u3002\u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001PyCuda\u306e\u57fa\u790e\u304b\u3089\u5fdc\u7528\u307e\u3067\u3001\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u3001\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3001\u5b9f\u8df5\u7684\u306a\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002PyCuda\u3092\u4f7f\u3063\u305fGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u4e16\u754c\u306b\u98db\u3073\u8fbc\u3093\u3067\u307f\u307e\u3057\u3087\u3046\uff01<\/p>\n\n\n\n<div class=\"wp-block-sgb-block-simple sgb-box-simple sgb-box-simple--title-normal sgb-box-simple--with-border\"><div style=\"background-color:var(--wp--preset--color--sango-main);color:#FFF\" class=\"sgb-box-simple__title\">\u3053\u306e\u8a18\u4e8b\u3092\u8aad\u3093\u3060\u3089\u308f\u304b\u308b\u3053\u3068<\/div><div class=\"sgb-box-simple__body\" style=\"border-color:var(--wp--preset--color--sango-main);background-color:#FFF\">\n<ul class=\"wp-block-list\">\n<li>PyCuda\u306e\u6982\u8981\u3068GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u5229\u70b9 <\/li>\n\n\n\n<li>PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u3068\u74b0\u5883\u8a2d\u5b9a <\/li>\n\n\n\n<li>PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u57fa\u672c\u7684\u306a\u66f8\u304d\u65b9 <\/li>\n\n\n\n<li>\u30d9\u30af\u30c8\u30eb\u52a0\u7b97\u3084\u884c\u5217\u7a4d\u306a\u3069\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9 <\/li>\n\n\n\n<li>\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3084\u79d1\u5b66\u8a08\u7b97\u3067\u306ePyCuda\u306e\u6d3b\u7528\u4f8b <\/li>\n\n\n\n<li>PyCuda\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u4e8b\u4f8b\u3068\u6210\u679c <\/li>\n\n\n\n<li>PyCuda\u306e\u5b66\u7fd2\u30ea\u30bd\u30fc\u30b9\u3068\u5c06\u6765\u306e\u5c55\u671b<\/li>\n<\/ul>\n<\/div><\/div>\n\n\n\n<div class=\"toc\"><br \/>\n<b>Warning<\/b>:  Undefined array key \"is_admin\" in <b>\/home\/c7479301\/public_html\/chocottopro.com\/wp-content\/themes\/sango-theme\/library\/gutenberg\/dist\/classes\/Toc.php<\/b> on line <b>116<\/b><br \/>\n<br \/>\n<b>Warning<\/b>:  Undefined array key \"is_category_top\" in <b>\/home\/c7479301\/public_html\/chocottopro.com\/wp-content\/themes\/sango-theme\/library\/gutenberg\/dist\/classes\/Toc.php<\/b> on line <b>121<\/b><br \/>\n<br \/>\n<b>Warning<\/b>:  Undefined array key \"is_top\" in <b>\/home\/c7479301\/public_html\/chocottopro.com\/wp-content\/themes\/sango-theme\/library\/gutenberg\/dist\/classes\/Toc.php<\/b> on line <b>128<\/b><br \/>\n    <div id=\"toc_container\" class=\"sgb-toc--bullets js-smooth-scroll\" data-dialog-title=\"Table of Contents\">\n      <p class=\"toc_title\">\u76ee\u6b21 <\/p>\n      <ul class=\"toc_list\">  <li class=\"first\">    <a href=\"#i-0\">PyCuda\u3068\u306f\uff1fGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u57fa\u790e\u77e5\u8b58<\/a>    <ul class=\"menu_level_1\">      <li class=\"first\">        <a href=\"#i-1\">GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u6982\u8981\u3068\u5229\u70b9<\/a>      <\/li>      <li class=\"last\">        <a href=\"#i-2\">PyCuda\u306e\u7279\u5fb4\u3068\u4ed6\u306eGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u306e\u6bd4\u8f03<\/a>      <\/li>    <\/ul>  <\/li>  <li>    <a href=\"#i-3\">PyCuda\u306e\u74b0\u5883\u8a2d\u5b9a\u3068\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5<\/a>    <ul class=\"menu_level_1\">      <li class=\"first\">        <a href=\"#i-4\">PyCuda\u3092\u4f7f\u3046\u305f\u3081\u306b\u5fc5\u8981\u306a\u74b0\u5883\u3068\u6e96\u5099\u3059\u308b\u3082\u306e<\/a>      <\/li>      <li>        <a href=\"#i-5\">PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u624b\u9806\uff08Windows, macOS, Linux\uff09<\/a>      <\/li>      <li class=\"last\">        <a href=\"#i-9\">\u3088\u304f\u3042\u308b\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u30a8\u30e9\u30fc\u3068\u5bfe\u51e6\u6cd5<\/a>      <\/li>    <\/ul>  <\/li>  <li>    <a href=\"#i-10\">PyCuda\u306e\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3068\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9<\/a>    <ul class=\"menu_level_1\">      <li class=\"first\">        <a href=\"#i-11\">PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u57fa\u672c\u69cb\u9020\u3068\u66f8\u304d\u65b9<\/a>      <\/li>      <li>        <a href=\"#i-12\">PyCuda\u3092\u4f7f\u3063\u305f\u7c21\u5358\u306a\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08\u30d9\u30af\u30c8\u30eb\u52a0\u7b97\u3001\u884c\u5217\u7a4d\u306a\u3069\uff09<\/a>      <\/li>      <li class=\"last\">        <a href=\"#i-13\">PyCuda\u306e\u30c7\u30d0\u30c3\u30b0\u65b9\u6cd5\u3068\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u6e2c\u5b9a<\/a>      <\/li>    <\/ul>  <\/li>  <li>    <a href=\"#i-14\">PyCuda\u306e\u5b9f\u8df5\u7684\u306a\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u3068\u5fdc\u7528\u4f8b<\/a>    <ul class=\"menu_level_1\">      <li class=\"first\">        <a href=\"#i-15\">\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306b\u304a\u3051\u308bPyCuda\u306e\u6d3b\u7528\u4f8b<\/a>      <\/li>      <li>        <a href=\"#i-16\">\u79d1\u5b66\u8a08\u7b97\u3084\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u3067\u306ePyCuda\u306e\u4f7f\u3044\u65b9<\/a>      <\/li>      <li class=\"last\">        <a href=\"#i-17\">PyCuda\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u4e8b\u4f8b\u3068\u6210\u679c<\/a>      <\/li>    <\/ul>  <\/li>  <li class=\"last\">    <a href=\"#i-18\">PyCuda\u306e\u3055\u3089\u306a\u308b\u5b66\u7fd2\u30ea\u30bd\u30fc\u30b9\u3068\u4eca\u5f8c\u306e\u5c55\u671b<\/a>    <ul class=\"menu_level_1\">      <li class=\"first\">        <a href=\"#i-19\">PyCuda\u306e\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3068\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb<\/a>      <\/li>      <li>        <a href=\"#i-20\">PyCuda\u306b\u95a2\u3059\u308b\u66f8\u7c4d\u3084\u60c5\u5831\u30b5\u30a4\u30c8<\/a>      <\/li>      <li class=\"last\">        <a href=\"#i-21\">PyCuda\u3068GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u5c06\u6765\u6027<\/a>      <\/li>    <\/ul>  <\/li><\/ul>\n      \n    <\/div><\/div><h2 class=\"wp-block-heading\" id=\"i-0\">PyCuda\u3068\u306f\uff1fGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u57fa\u790e\u77e5\u8b58<\/h2>\n\n\n\n<p>PyCuda\u306f\u3001NVIDIA\u304c\u63d0\u4f9b\u3059\u308bCUDA\u30d1\u30e9\u30ec\u30eb\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u30d7\u30e9\u30c3\u30c8\u30d5\u30a9\u30fc\u30e0\u3092Python\u304b\u3089\u5229\u7528\u3059\u308b\u305f\u3081\u306e\u30aa\u30fc\u30d7\u30f3\u30bd\u30fc\u30b9\u30e9\u30a4\u30d6\u30e9\u30ea\u3067\u3059\u3002CUDA\u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u3067\u3001GPU\u306e\u5f37\u529b\u306a\u4e26\u5217\u8a08\u7b97\u80fd\u529b\u3092\u6d3b\u7528\u3057\u3001CPU\u3088\u308a\u3082\u9ad8\u901f\u306b\u51e6\u7406\u3092\u5b9f\u884c\u3067\u304d\u307e\u3059\u3002PyCuda\u3092\u4f7f\u3048\u3070\u3001Python\u306e\u89aa\u3057\u307f\u3084\u3059\u3044\u6587\u6cd5\u3067GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3092\u884c\u3046\u3053\u3068\u304c\u3067\u304d\u3001NumPy\u30e9\u30a4\u30af\u306a\u69cb\u6587\u3067\u76f4\u611f\u7684\u306b\u30b3\u30fc\u30c9\u3092\u8a18\u8ff0\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-1\">GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u6982\u8981\u3068\u5229\u70b9<\/h3>\n\n\n\n<p>GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306f\u3001GPU\u306e\u5927\u898f\u6a21\u306a\u4e26\u5217\u8a08\u7b97\u80fd\u529b\u3092\u5229\u7528\u3057\u3066\u3001\u8a08\u7b97\u901f\u5ea6\u3092\u5927\u5e45\u306b\u5411\u4e0a\u3055\u305b\u308b\u624b\u6cd5\u3067\u3059\u3002GPU\u306f\u3001\u3082\u3068\u3082\u3068\u30b0\u30e9\u30d5\u30a3\u30c3\u30af\u30b9\u51e6\u7406\u3092\u9ad8\u901f\u5316\u3059\u308b\u305f\u3081\u306b\u958b\u767a\u3055\u308c\u307e\u3057\u305f\u304c\u3001\u305d\u306e\u4e26\u5217\u51e6\u7406\u80fd\u529b\u306f\u79d1\u5b66\u8a08\u7b97\u3084\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306a\u3069\u306e\u5206\u91ce\u3067\u3082\u6d3b\u7528\u3055\u308c\u3066\u3044\u307e\u3059\u3002GPU\u3092\u4f7f\u7528\u3059\u308b\u3053\u3068\u3067\u3001CPU\u3088\u308a\u3082\u9ad8\u901f\u306b\u5927\u91cf\u306e\u8a08\u7b97\u3092\u51e6\u7406\u3067\u304d\u308b\u305f\u3081\u3001\u8a08\u7b97\u30b3\u30b9\u30c8\u3092\u524a\u6e1b\u3057\u3064\u3064\u3001\u51e6\u7406\u6642\u9593\u3092\u77ed\u7e2e\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n<p>GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u4e3b\u306a\u5229\u70b9\u306f\u4ee5\u4e0b\u306e\u901a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5927\u898f\u6a21\u306a\u4e26\u5217\u8a08\u7b97\u306b\u3088\u308a\u3001CPU\u306b\u6bd4\u3079\u3066\u9ad8\u901f\u306b\u51e6\u7406\u3092\u5b9f\u884c\u3067\u304d\u308b\u3002<\/li>\n\n\n\n<li>\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3084\u79d1\u5b66\u8a08\u7b97\u306a\u3069\u3001\u8a08\u7b97\u8ca0\u8377\u306e\u9ad8\u3044\u30bf\u30b9\u30af\u306b\u9069\u3057\u3066\u3044\u308b\u3002<\/li>\n\n\n\n<li>\u30de\u30eb\u30c1\u30b3\u30a2CPU\u3088\u308a\u3082\u30b3\u30b9\u30c8\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u304c\u9ad8\u3044\u3002<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-2\">PyCuda\u306e\u7279\u5fb4\u3068\u4ed6\u306eGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068\u306e\u6bd4\u8f03<\/h3>\n\n\n\n<p>PyCuda\u306f\u3001CUDA\u30ab\u30fc\u30cd\u30eb\u3092Python\u304b\u3089\u76f4\u63a5\u547c\u3073\u51fa\u3059\u3053\u3068\u304c\u3067\u304d\u3001GPU\u30e1\u30e2\u30ea\u306e\u5272\u308a\u5f53\u3066\u3084\u30c7\u30fc\u30bf\u8ee2\u9001\u3092\u67d4\u8edf\u306b\u7ba1\u7406\u3067\u304d\u307e\u3059\u3002\u307e\u305f\u3001Elementwise\u30ab\u30fc\u30cd\u30eb\u3084Reduction\u30ab\u30fc\u30cd\u30eb\u306a\u3069\u306e\u9ad8\u30ec\u30d9\u30ebAPI\u3092\u63d0\u4f9b\u3057\u3066\u304a\u308a\u3001GPUArray\u3001GPUMatrix\u3001Scan\u306a\u3069\u306e\u4fbf\u5229\u306a\u30c7\u30fc\u30bf\u69cb\u9020\u3082\u7528\u610f\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4ed6\u306eGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3068PyCuda\u3092\u6bd4\u8f03\u3059\u308b\u3068\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u7279\u5fb4\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>NVIDIA CUDA: PyCuda\u306fCUDA\u3092\u76f4\u63a5\u5229\u7528\u3059\u308b\u305f\u3081\u3001\u4f4e\u30ec\u30d9\u30eb\u3067\u67d4\u8edf\u6027\u304c\u9ad8\u3044\u3002<\/li>\n\n\n\n<li>OpenCL: \u30aa\u30fc\u30d7\u30f3\u30b9\u30bf\u30f3\u30c0\u30fc\u30c9\u306eGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3060\u304c\u3001PyCuda\u307b\u3069\u5e83\u304f\u306f\u4f7f\u308f\u308c\u3066\u3044\u306a\u3044\u3002<\/li>\n\n\n\n<li>Numba: Python\u30b3\u30fc\u30c9\u304b\u3089GPU\u30b3\u30fc\u30c9\u3092\u81ea\u52d5\u751f\u6210\u3059\u308b\u30b3\u30f3\u30d1\u30a4\u30e9\u3002PyCuda\u3088\u308a\u624b\u8efd\u306b\u4f7f\u3048\u308b\u304c\u3001\u67d4\u8edf\u6027\u306f\u4f4e\u3044\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4ee5\u4e0b\u306f\u3001PyCuda\u3092\u4f7f\u3063\u305f\u7c21\u5358\u306a\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3067\u3059\u3002\u3053\u306e\u30b3\u30fc\u30c9\u3067\u306f\u3001GPU\u3092\u4f7f\u3063\u30662\u3064\u306e\u914d\u5217\u3092\u8db3\u3057\u5408\u308f\u305b\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">import numpy as np\nimport pycuda.autoinit\nimport pycuda.driver as cuda\nfrom pycuda.compiler import SourceModule\n\n# CUDA\u30ab\u30fc\u30cd\u30eb\u3092\u5b9a\u7fa9\nmod = SourceModule(\"\"\"\n__global__ void add_arrays(float *a, float *b, float *c, int n) {\n    int i = blockIdx.x * blockDim.x + threadIdx.x;\n    if (i &lt; n) {\n        c[i] = a[i] + b[i];\n    }\n}\n\"\"\")\n\n# \u30db\u30b9\u30c8\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099\na = np.random.randn(1000).astype(np.float32)\nb = np.random.randn(1000).astype(np.float32)\nc = np.zeros_like(a)\n\n# \u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\na_gpu = cuda.mem_alloc(a.nbytes)\nb_gpu = cuda.mem_alloc(b.nbytes)\nc_gpu = cuda.mem_alloc(c.nbytes)\n\n# \u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\ncuda.memcpy_htod(a_gpu, a)\ncuda.memcpy_htod(b_gpu, b)\n\n# \u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\nadd_arrays = mod.get_function(\"add_arrays\")\nadd_arrays(a_gpu, b_gpu, c_gpu, np.int32(a.size), block=(1024, 1, 1), grid=(a.size \/\/ 1024 + 1, 1))\n\n# \u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\ncuda.memcpy_dtoh(c, c_gpu)\n\nprint(c[:10])  # \u7d50\u679c\u3092\u8868\u793a<\/pre>\n\n\n\n<p>\u3053\u306e\u3088\u3046\u306b\u3001PyCuda\u3092\u4f7f\u3046\u3053\u3068\u3067\u3001Python\u306e\u89aa\u3057\u307f\u3084\u3059\u3044\u6587\u6cd5\u3067GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3092\u884c\u3046\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002PyCuda\u306f\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3084\u79d1\u5b66\u8a08\u7b97\u306a\u3069\u306e\u5206\u91ce\u3067\u6d3b\u7528\u3055\u308c\u3066\u304a\u308a\u3001GPU\u306e\u4e26\u5217\u51e6\u7406\u80fd\u529b\u3092\u5f15\u304d\u51fa\u3059\u3053\u3068\u3067\u3001\u9ad8\u901f\u306a\u8a08\u7b97\u3092\u5b9f\u73fe\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"i-3\">PyCuda\u306e\u74b0\u5883\u8a2d\u5b9a\u3068\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5<\/h2>\n\n\n\n<p>PyCuda\u3092\u4f7f\u7528\u3059\u308b\u306b\u306f\u3001NVIDIA\u88fd\u306eGPU\u3068CUDA Toolkit\u304c\u5fc5\u8981\u3067\u3059\u3002\u307e\u305f\u3001Python (\u30d0\u30fc\u30b8\u30e7\u30f33.6\u4ee5\u4e0a) \u3068NumPy (\u30d0\u30fc\u30b8\u30e7\u30f31.18\u4ee5\u4e0a) \u3082\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u3066\u304a\u304f\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\u3002\u3053\u3053\u3067\u306f\u3001\u5404OS (Windows\u3001macOS\u3001Linux) \u3054\u3068\u306ePyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u624b\u9806\u3092\u8aac\u660e\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-4\">PyCuda\u3092\u4f7f\u3046\u305f\u3081\u306b\u5fc5\u8981\u306a\u74b0\u5883\u3068\u6e96\u5099\u3059\u308b\u3082\u306e<\/h3>\n\n\n\n<p>PyCuda\u3092\u4f7f\u7528\u3059\u308b\u305f\u3081\u306e\u5fc5\u8981\u74b0\u5883\u306f\u4ee5\u4e0b\u306e\u901a\u308a\u3067\u3059\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>NVIDIA\u88fd\u306eGPU (CUDA Compute Capability 3.0\u4ee5\u4e0a)<\/li>\n\n\n\n<li>CUDA Toolkit (\u30d0\u30fc\u30b8\u30e7\u30f38.0\u4ee5\u4e0a)<\/li>\n\n\n\n<li>Python (\u30d0\u30fc\u30b8\u30e7\u30f33.6\u4ee5\u4e0a)<\/li>\n\n\n\n<li>NumPy (\u30d0\u30fc\u30b8\u30e7\u30f31.18\u4ee5\u4e0a)<\/li>\n<\/ul>\n\n\n\n<p>\u307e\u305a\u3001NVIDIA\u88fd\u306eGPU\u3092\u642d\u8f09\u3057\u305f\u30b3\u30f3\u30d4\u30e5\u30fc\u30bf\u3092\u7528\u610f\u3057\u3066\u304f\u3060\u3055\u3044\u3002\u6b21\u306b\u3001CUDA Toolkit\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\u3002CUDA Toolkit\u306fNVIDIA\u306e\u516c\u5f0f\u30a6\u30a7\u30d6\u30b5\u30a4\u30c8\u304b\u3089\u30c0\u30a6\u30f3\u30ed\u30fc\u30c9\u3067\u304d\u307e\u3059\u3002\u307e\u305f\u3001Python\u3068NumPy\u3082\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u3066\u304a\u304d\u307e\u3057\u3087\u3046\u3002\u3053\u308c\u3089\u306f\u3001PyCuda\u3092\u4f7f\u7528\u3059\u308b\u305f\u3081\u306b\u5fc5\u8981\u306a\u74b0\u5883\u3067\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-5\">PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u624b\u9806\uff08Windows, macOS, Linux\uff09<\/h3>\n\n\n\n<p>PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u624b\u9806\u306f\u3001OS\u306b\u3088\u3063\u3066\u5c11\u3057\u7570\u306a\u308a\u307e\u3059\u3002\u4ee5\u4e0b\u306b\u3001\u5404OS\u3054\u3068\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u624b\u9806\u3092\u793a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<h4 class=\"wp-block-sgb-headings sgb-heading\" id=\"i-6\"><span class=\"sgb-heading__inner sgb-heading--type1 with-icon\" style=\"background-color:transparent;border-color:transparent;font-size:1.2em\"><i class=\"fa fa-check\" style=\"color:var(--wp--preset--color--sango-main)\"><\/i><span class=\"sgb-heading__text\" style=\"color:#333\">a. Windows:<\/span><\/span><\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li>CUDA Toolkit\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\u3002<\/li>\n\n\n\n<li>\u30b3\u30de\u30f3\u30c9\u30d7\u30ed\u30f3\u30d7\u30c8\u307e\u305f\u306fPowerShell\u3092\u958b\u304d\u3001\u4ee5\u4e0b\u306e\u30b3\u30de\u30f3\u30c9\u3092\u5b9f\u884c\u3057\u307e\u3059\u3002<\/li>\n<\/ol>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">   pip install pycuda<\/pre>\n\n\n\n<h4 class=\"wp-block-sgb-headings sgb-heading\" id=\"i-7\"><span class=\"sgb-heading__inner sgb-heading--type1 with-icon\" style=\"background-color:transparent;border-color:transparent;font-size:1.2em\"><i class=\"fa fa-check\" style=\"color:var(--wp--preset--color--sango-main)\"><\/i><span class=\"sgb-heading__text\" style=\"color:#333\">b. macOS:<\/span><\/span><\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li>CUDA Toolkit\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\u3002<\/li>\n\n\n\n<li>\u30bf\u30fc\u30df\u30ca\u30eb\u3092\u958b\u304d\u3001\u4ee5\u4e0b\u306e\u30b3\u30de\u30f3\u30c9\u3092\u5b9f\u884c\u3057\u307e\u3059\u3002<\/li>\n<\/ol>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">   brew install pycuda<\/pre>\n\n\n\n<h4 class=\"wp-block-sgb-headings sgb-heading\" id=\"i-8\"><span class=\"sgb-heading__inner sgb-heading--type1 with-icon\" style=\"background-color:transparent;border-color:transparent;font-size:1.2em\"><i class=\"fa fa-check\" style=\"color:var(--wp--preset--color--sango-main)\"><\/i><span class=\"sgb-heading__text\" style=\"color:#333\">c. Linux:<\/span><\/span><\/h4>\n\n\n\n<ol class=\"wp-block-list\">\n<li>CUDA Toolkit\u3092\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u307e\u3059\u3002<\/li>\n\n\n\n<li>\u30bf\u30fc\u30df\u30ca\u30eb\u3092\u958b\u304d\u3001\u4ee5\u4e0b\u306e\u30b3\u30de\u30f3\u30c9\u3092\u5b9f\u884c\u3057\u307e\u3059\u3002<\/li>\n<\/ol>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">   pip install pycuda<\/pre>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-9\">\u3088\u304f\u3042\u308b\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u30a8\u30e9\u30fc\u3068\u5bfe\u51e6\u6cd5<\/h3>\n\n\n\n<p>PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u4e2d\u306b\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306a\u30a8\u30e9\u30fc\u304c\u767a\u751f\u3059\u308b\u3053\u3068\u304c\u3042\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>nvcc not found: CUDA Toolkit\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u304c\u6b63\u3057\u304f\u884c\u308f\u308c\u3066\u3044\u306a\u3044\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002\u74b0\u5883\u5909\u6570\u306e\u8a2d\u5b9a\u3092\u78ba\u8a8d\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/li>\n\n\n\n<li>No module named \u2018pycuda\u2019: PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u306b\u5931\u6557\u3057\u3066\u3044\u307e\u3059\u3002pip\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3084\u3001CUDA Toolkit\u3068\u306e\u4e92\u63db\u6027\u3092\u78ba\u8a8d\u3057\u3066\u304f\u3060\u3055\u3044\u3002<\/li>\n\n\n\n<li>ImportError: libcudart.so.X.X: cannot open shared object file: PyCuda\u3068CUDA Toolkit\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u304c\u4e00\u81f4\u3057\u3066\u3044\u306a\u3044\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u63c3\u3048\u3066\u304f\u3060\u3055\u3044\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u3053\u308c\u3089\u306e\u30a8\u30e9\u30fc\u304c\u767a\u751f\u3057\u305f\u5834\u5408\u306f\u3001\u30a8\u30e9\u30fc\u30e1\u30c3\u30bb\u30fc\u30b8\u3092\u78ba\u8a8d\u3057\u3001\u9069\u5207\u306a\u5bfe\u51e6\u3092\u884c\u3063\u3066\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n<p>\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u304c\u6b63\u5e38\u306b\u5b8c\u4e86\u3057\u305f\u304b\u3069\u3046\u304b\u306f\u3001\u4ee5\u4e0b\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3092\u5b9f\u884c\u3059\u308b\u3053\u3068\u3067\u78ba\u8a8d\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">import pycuda.driver as cuda\nimport pycuda.autoinit\nfrom pycuda.compiler import SourceModule\n\nmod = SourceModule(\"\"\"\n__global__ void myfunc(void)\n{\n  printf(\"Hello PyCUDA!!\\\\n\");\n}\n\"\"\")\n\nfunc = mod.get_function(\"myfunc\")\nfunc(block=(1,1,1))<\/pre>\n\n\n\n<p>\u4e0a\u8a18\u306e\u30b3\u30fc\u30c9\u3092\u5b9f\u884c\u3057\u3066\u3001\u201dHello PyCUDA!!\u201d\u3068\u8868\u793a\u3055\u308c\u308c\u3070\u3001PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u306f\u6210\u529f\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4ee5\u4e0a\u304c\u3001PyCuda\u306e\u74b0\u5883\u8a2d\u5b9a\u3068\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u306e\u89e3\u8aac\u3067\u3059\u3002OS\u3054\u3068\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u624b\u9806\u3092\u8e0f\u307e\u3048\u3001\u6b63\u3057\u304f\u74b0\u5883\u3092\u8a2d\u5b9a\u3059\u308b\u3053\u3068\u3067\u3001PyCuda\u3092\u4f7f\u3063\u305fGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3092\u59cb\u3081\u308b\u3053\u3068\u304c\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"i-10\">PyCuda\u306e\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3068\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9<\/h2>\n\n\n\n<p>PyCuda\u3092\u4f7f\u3063\u305f\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3067\u306f\u3001\u307e\u305a\u30db\u30b9\u30c8(CPU)\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099\u3057\u3001\u30c7\u30d0\u30a4\u30b9(GPU)\u306e\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\u307e\u3059\u3002\u6b21\u306b\u3001\u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\u3057\u3001\u30ab\u30fc\u30cd\u30eb(GPU\u3067\u5b9f\u884c\u3055\u308c\u308b\u95a2\u6570)\u3092\u547c\u3073\u51fa\u3057\u307e\u3059\u3002\u6700\u5f8c\u306b\u3001\u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\u3057\u307e\u3059\u3002\u3053\u306e\u4e00\u9023\u306e\u6d41\u308c\u304c\u3001PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u57fa\u672c\u7684\u306a\u69cb\u9020\u3067\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-11\">PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u57fa\u672c\u69cb\u9020\u3068\u66f8\u304d\u65b9<\/h3>\n\n\n\n<p>PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u57fa\u672c\u69cb\u9020\u306f\u3001\u4ee5\u4e0b\u306e\u3088\u3046\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u30db\u30b9\u30c8(CPU)\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099<\/li>\n\n\n\n<li>\u30c7\u30d0\u30a4\u30b9(GPU)\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066<\/li>\n\n\n\n<li>\u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001<\/li>\n\n\n\n<li>\u30ab\u30fc\u30cd\u30eb(GPU\u3067\u5b9f\u884c\u3055\u308c\u308b\u95a2\u6570)\u3092\u547c\u3073\u51fa\u3057<\/li>\n\n\n\n<li>\u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001<\/li>\n<\/ol>\n\n\n\n<p>\u3053\u306e\u69cb\u9020\u306b\u6cbf\u3063\u3066\u3001PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u66f8\u3044\u3066\u3044\u304d\u307e\u3059\u3002\u307e\u305a\u3001<code>pycuda.autoinit<\/code>\u3068<code>pycuda.driver<\/code>\u3092\u30a4\u30f3\u30dd\u30fc\u30c8\u3057\u307e\u3059\u3002\u6b21\u306b\u3001<code>SourceModule<\/code>\u3092\u4f7f\u3063\u3066CUDA\u30ab\u30fc\u30cd\u30eb\u3092\u5b9a\u7fa9\u3057\u307e\u3059\u3002\u3053\u306e\u30ab\u30fc\u30cd\u30eb\u306f\u3001GPU\u3067\u5b9f\u884c\u3055\u308c\u308b\u95a2\u6570\u3067\u3059\u3002<\/p>\n\n\n\n<p>\u30db\u30b9\u30c8\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099\u3057\u305f\u3089\u3001<code>cuda.mem_alloc()<\/code>\u3092\u4f7f\u3063\u3066\u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\u307e\u3059\u3002\u305d\u3057\u3066\u3001<code>cuda.memcpy_htod()<\/code>\u3092\u4f7f\u3063\u3066\u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\u306b\u306f\u3001<code>mod.get_function()<\/code>\u3092\u4f7f\u3063\u3066\u30ab\u30fc\u30cd\u30eb\u95a2\u6570\u3092\u53d6\u5f97\u3057\u3001\u9069\u5207\u306a\u30d6\u30ed\u30c3\u30af\u30b5\u30a4\u30ba\u3068\u30b0\u30ea\u30c3\u30c9\u30b5\u30a4\u30ba\u3092\u6307\u5b9a\u3057\u3066\u5b9f\u884c\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u6700\u5f8c\u306b\u3001<code>cuda.memcpy_dtoh()<\/code>\u3092\u4f7f\u3063\u3066\u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-12\">PyCuda\u3092\u4f7f\u3063\u305f\u7c21\u5358\u306a\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\uff08\u30d9\u30af\u30c8\u30eb\u52a0\u7b97\u3001\u884c\u5217\u7a4d\u306a\u3069\uff09<\/h3>\n\n\n\n<p>\u4ee5\u4e0b\u306f\u3001PyCuda\u3092\u4f7f\u3063\u305f\u30d9\u30af\u30c8\u30eb\u52a0\u7b97\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3067\u3059\u3002<\/p>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">import numpy as np\nimport pycuda.autoinit\nimport pycuda.driver as cuda\nfrom pycuda.compiler import SourceModule\n\n# CUDA\u30ab\u30fc\u30cd\u30eb\u3092\u5b9a\u7fa9\nmod = SourceModule(\"\"\"\n__global__ void add_arrays(float *a, float *b, float *c, int n) {\n    int i = blockIdx.x * blockDim.x + threadIdx.x;\n    if (i &lt; n) {\n        c[i] = a[i] + b[i];\n    }\n}\n\"\"\")\n\n# \u30db\u30b9\u30c8\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099\na = np.random.randn(1000).astype(np.float32)\nb = np.random.randn(1000).astype(np.float32)\nc = np.zeros_like(a)\n\n# \u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\na_gpu = cuda.mem_alloc(a.nbytes)\nb_gpu = cuda.mem_alloc(b.nbytes)\nc_gpu = cuda.mem_alloc(c.nbytes)\n\n# \u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\ncuda.memcpy_htod(a_gpu, a)\ncuda.memcpy_htod(b_gpu, b)\n\n# \u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\nadd_arrays = mod.get_function(\"add_arrays\")\nadd_arrays(a_gpu, b_gpu, c_gpu, np.int32(a.size), block=(1024, 1, 1), grid=(a.size \/\/ 1024 + 1, 1))\n\n# \u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\ncuda.memcpy_dtoh(c, c_gpu)\n\nprint(c[:10])  # \u7d50\u679c\u3092\u8868\u793a<\/pre>\n\n\n\n<p>\u3053\u306e\u30b3\u30fc\u30c9\u3067\u306f\u3001\u307e\u305aCUDA\u30ab\u30fc\u30cd\u30eb<code>add_arrays<\/code>\u3092\u5b9a\u7fa9\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u306e\u30ab\u30fc\u30cd\u30eb\u306f\u30012\u3064\u306e\u914d\u5217<code>a<\/code>\u3068<code>b<\/code>\u306e\u8981\u7d20\u3092\u8db3\u3057\u5408\u308f\u305b\u3001\u7d50\u679c\u3092\u914d\u5217<code>c<\/code>\u306b\u683c\u7d0d\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u6b21\u306b\u3001\u30db\u30b9\u30c8\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099\u3057\u3001\u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\u307e\u3059\u3002\u305d\u3057\u3066\u3001<code>cuda.memcpy_htod()<\/code>\u3092\u4f7f\u3063\u3066\u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\u969b\u306f\u3001<code>add_arrays<\/code>\u95a2\u6570\u3092\u53d6\u5f97\u3057\u3001\u9069\u5207\u306a\u30d6\u30ed\u30c3\u30af\u30b5\u30a4\u30ba\u3068\u30b0\u30ea\u30c3\u30c9\u30b5\u30a4\u30ba\u3092\u6307\u5b9a\u3057\u3066\u5b9f\u884c\u3057\u307e\u3059\u3002\u6700\u5f8c\u306b\u3001<code>cuda.memcpy_dtoh()<\/code>\u3092\u4f7f\u3063\u3066\u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\u3057\u3001\u7d50\u679c\u3092\u8868\u793a\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4ee5\u4e0b\u306f\u3001PyCuda\u3092\u4f7f\u3063\u305f\u884c\u5217\u7a4d\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3067\u3059\u3002<\/p>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">import numpy as np\nimport pycuda.autoinit\nimport pycuda.driver as cuda\nfrom pycuda.compiler import SourceModule\n\n# CUDA\u30ab\u30fc\u30cd\u30eb\u3092\u5b9a\u7fa9\nmod = SourceModule(\"\"\"\n__global__ void matmul(float *a, float *b, float *c, int N) {\n    int row = blockIdx.y * blockDim.y + threadIdx.y;\n    int col = blockIdx.x * blockDim.x + threadIdx.x;\n\n    if (row &lt; N &amp;&amp; col &lt; N) {\n        float sum = 0;\n        for (int i = 0; i &lt; N; i++) {\n            sum += a[row * N + i] * b[i * N + col];\n        }\n        c[row * N + col] = sum;\n    }\n}\n\"\"\")\n\n# \u30db\u30b9\u30c8\u5074\u3067\u30c7\u30fc\u30bf\u3092\u6e96\u5099\nN = 1024\na = np.random.randn(N, N).astype(np.float32)\nb = np.random.randn(N, N).astype(np.float32)\nc = np.zeros((N, N), dtype=np.float32)\n\n# \u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\na_gpu = cuda.mem_alloc(a.nbytes)\nb_gpu = cuda.mem_alloc(b.nbytes)\nc_gpu = cuda.mem_alloc(c.nbytes)\n\n# \u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\ncuda.memcpy_htod(a_gpu, a)\ncuda.memcpy_htod(b_gpu, b)\n\n# \u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\nblock_size = 32\ngrid_size = (N \/\/ block_size + 1, N \/\/ block_size + 1)\nmatmul = mod.get_function(\"matmul\")\nmatmul(a_gpu, b_gpu, c_gpu, np.int32(N), block=(block_size, block_size, 1), grid=grid_size)\n\n# \u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\ncuda.memcpy_dtoh(c, c_gpu)\n\nprint(np.mean(np.abs(c - np.dot(a, b))))  # \u7d50\u679c\u3092\u691c\u8a3c<\/pre>\n\n\n\n<p>\u3053\u306e\u30b3\u30fc\u30c9\u3067\u306f\u3001CUDA\u30ab\u30fc\u30cd\u30eb<code>matmul<\/code>\u3092\u5b9a\u7fa9\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u306e\u30ab\u30fc\u30cd\u30eb\u306f\u30012\u3064\u306e\u884c\u5217<code>a<\/code>\u3068<code>b<\/code>\u306e\u7a4d\u3092\u8a08\u7b97\u3057\u3001\u7d50\u679c\u3092\u884c\u5217<code>c<\/code>\u306b\u683c\u7d0d\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\u969b\u306f\u3001<code>matmul<\/code>\u95a2\u6570\u3092\u53d6\u5f97\u3057\u3001\u9069\u5207\u306a\u30d6\u30ed\u30c3\u30af\u30b5\u30a4\u30ba\u3068\u30b0\u30ea\u30c3\u30c9\u30b5\u30a4\u30ba\u3092\u6307\u5b9a\u3057\u3066\u5b9f\u884c\u3057\u307e\u3059\u3002\u6700\u5f8c\u306b\u3001<code>cuda.memcpy_dtoh()<\/code>\u3092\u4f7f\u3063\u3066\u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\u3057\u3001<code>np.dot()<\/code>\u3092\u4f7f\u3063\u3066CPU\u3067\u8a08\u7b97\u3057\u305f\u7d50\u679c\u3068\u6bd4\u8f03\u3059\u308b\u3053\u3068\u3067\u3001\u7d50\u679c\u3092\u691c\u8a3c\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-13\">PyCuda\u306e\u30c7\u30d0\u30c3\u30b0\u65b9\u6cd5\u3068\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u6e2c\u5b9a<\/h3>\n\n\n\n<p>PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u30c7\u30d0\u30c3\u30b0\u306b\u306f\u3001\u4ee5\u4e0b\u306e\u65b9\u6cd5\u304c\u5f79\u7acb\u3061\u307e\u3059\u3002<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><code>printf()<\/code>\u3092\u4f7f\u3063\u305f\u30c7\u30d0\u30c3\u30b0\u51fa\u529b\n<ul class=\"wp-block-list\">\n<li>\u30ab\u30fc\u30cd\u30eb\u5185\u3067<code>printf()<\/code>\u3092\u4f7f\u3063\u3066\u3001\u5909\u6570\u306e\u5024\u3084\u4e2d\u9593\u7d50\u679c\u3092\u51fa\u529b\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><code>cuda.mem_get_info()<\/code>\u3092\u4f7f\u3063\u305f\u30e1\u30e2\u30ea\u4f7f\u7528\u91cf\u306e\u78ba\u8a8d\n<ul class=\"wp-block-list\">\n<li><code>cuda.mem_get_info()<\/code>\u3092\u4f7f\u3063\u3066\u3001\u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u306e\u7dcf\u5bb9\u91cf\u3068\u7a7a\u304d\u5bb9\u91cf\u3092\u78ba\u8a8d\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><code>pycuda.tools<\/code>\u30e2\u30b8\u30e5\u30fc\u30eb\u3092\u4f7f\u3063\u305f\u30a8\u30e9\u30fc\u30c1\u30a7\u30c3\u30af\n<ul class=\"wp-block-list\">\n<li><code>pycuda.tools<\/code>\u30e2\u30b8\u30e5\u30fc\u30eb\u306b\u306f\u3001<code>clear_context_caches()<\/code>\u3084<code>mark_cuda_test()<\/code>\u306a\u3069\u306e\u4fbf\u5229\u306a\u95a2\u6570\u304c\u7528\u610f\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n\n\n\n<p>PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u3092\u6e2c\u5b9a\u3059\u308b\u306b\u306f\u3001\u4ee5\u4e0b\u306e\u65b9\u6cd5\u304c\u5f79\u7acb\u3061\u307e\u3059\u3002<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li><code>time<\/code>\u30e2\u30b8\u30e5\u30fc\u30eb\u3092\u4f7f\u3063\u305f\u5b9f\u884c\u6642\u9593\u306e\u6e2c\u5b9a\n<ul class=\"wp-block-list\">\n<li><code>time.time()<\/code>\u3092\u4f7f\u3063\u3066\u3001\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u5b9f\u884c\u6642\u9593\u3092\u6e2c\u5b9a\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><code>nvprof<\/code>\u3092\u4f7f\u3063\u305f\u30d7\u30ed\u30d5\u30a1\u30a4\u30ea\u30f3\u30b0\n<ul class=\"wp-block-list\">\n<li><code>nvprof<\/code>\u306f\u3001CUDA\u30d7\u30ed\u30b0\u30e9\u30e0\u306e\u30d7\u30ed\u30d5\u30a1\u30a4\u30ea\u30f3\u30b0\u30c4\u30fc\u30eb\u3067\u3059\u3002\u30ab\u30fc\u30cd\u30eb\u306e\u5b9f\u884c\u6642\u9593\u3084\u30e1\u30e2\u30ea\u8ee2\u9001\u306e\u6642\u9593\u3092\u8a73\u7d30\u306b\u5206\u6790\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ul>\n<\/li>\n\n\n\n<li><code>pycuda.driver.Event<\/code>\u3092\u4f7f\u3063\u305f\u6642\u9593\u6e2c\u5b9a\n<ul class=\"wp-block-list\">\n<li><code>pycuda.driver.Event<\/code>\u3092\u4f7f\u3063\u3066\u3001\u30ab\u30fc\u30cd\u30eb\u306e\u5b9f\u884c\u6642\u9593\u3084\u30e1\u30e2\u30ea\u8ee2\u9001\u306e\u6642\u9593\u3092\u6e2c\u5b9a\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n\n\n\n<p>\u4ee5\u4e0a\u304c\u3001PyCuda\u306e\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3068\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u306e\u89e3\u8aac\u3067\u3059\u3002PyCuda\u3092\u4f7f\u3046\u3053\u3068\u3067\u3001Python\u304b\u3089\u7c21\u5358\u306bGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3092\u59cb\u3081\u3089\u308c\u307e\u3059\u3002\u9069\u5207\u306a\u30c7\u30d0\u30c3\u30b0\u65b9\u6cd5\u3068\u30d1\u30d5\u30a9\u30fc\u30de\u30f3\u30b9\u6e2c\u5b9a\u65b9\u6cd5\u3092\u6d3b\u7528\u3057\u306a\u304c\u3089\u3001PyCuda\u30d7\u30ed\u30b0\u30e9\u30e0\u3092\u958b\u767a\u3057\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"i-14\">PyCuda\u306e\u5b9f\u8df5\u7684\u306a\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u3068\u5fdc\u7528\u4f8b<\/h2>\n\n\n\n<p>PyCuda\u306f\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3084\u79d1\u5b66\u8a08\u7b97\u3001\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306a\u3069\u3001\u5e45\u5e83\u3044\u5206\u91ce\u3067\u6d3b\u7528\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u3053\u3053\u3067\u306f\u3001PyCuda\u306e\u5b9f\u8df5\u7684\u306a\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u3068\u5fdc\u7528\u4f8b\u3092\u3044\u304f\u3064\u304b\u7d39\u4ecb\u3057\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-15\">\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306b\u304a\u3051\u308bPyCuda\u306e\u6d3b\u7528\u4f8b<\/h3>\n\n\n\n<p>\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3067\u306f\u3001\u5927\u91cf\u306e\u30c7\u30fc\u30bf\u3092\u51e6\u7406\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\u305f\u3081\u3001GPU\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u304c\u4e0d\u53ef\u6b20\u3067\u3059\u3002PyCuda\u306f\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u3068\u3057\u3066\u4f7f\u7528\u3055\u308c\u3001\u8a08\u7b97\u306e\u9ad8\u901f\u5316\u306b\u8ca2\u732e\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4f8b\u3048\u3070\u3001Chainer\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u306f\u3001PyCuda\u3092\u4f7f\u3063\u3066\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af\u306e\u8a08\u7b97\u3092\u9ad8\u901f\u5316\u3057\u3066\u3044\u307e\u3059\u3002\u307e\u305f\u3001TensorFlow\u3084PyTorch\u3067\u3082\u3001PyCuda\u3092\u30d0\u30c3\u30af\u30a8\u30f3\u30c9\u3068\u3057\u3066\u4f7f\u7528\u3059\u308b\u3053\u3068\u3067\u3001GPU\u3092\u6d3b\u7528\u3057\u305f\u9ad8\u901f\u306a\u5b66\u7fd2\u3068\u63a8\u8ad6\u304c\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u7279\u306b\u3001\u7573\u307f\u8fbc\u307f\u30cb\u30e5\u30fc\u30e9\u30eb\u30cd\u30c3\u30c8\u30ef\u30fc\u30af(CNN)\u306e\u5b66\u7fd2\u3068\u63a8\u8ad6\u3067\u306f\u3001PyCuda\u306b\u3088\u308b\u9ad8\u901f\u5316\u306e\u52b9\u679c\u304c\u9855\u8457\u3067\u3059\u3002CNN\u306f\u3001\u753b\u50cf\u8a8d\u8b58\u3084\u7269\u4f53\u691c\u51fa\u306a\u3069\u306e\u30bf\u30b9\u30af\u3067\u5e83\u304f\u4f7f\u308f\u308c\u3066\u3044\u307e\u3059\u304c\u3001\u5927\u91cf\u306e\u753b\u50cf\u30c7\u30fc\u30bf\u3092\u51e6\u7406\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\u305f\u3081\u3001GPU\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u304c\u91cd\u8981\u306b\u306a\u308a\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-16\">\u79d1\u5b66\u8a08\u7b97\u3084\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u3067\u306ePyCuda\u306e\u4f7f\u3044\u65b9<\/h3>\n\n\n\n<p>\u79d1\u5b66\u8a08\u7b97\u3084\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306e\u5206\u91ce\u3067\u3082\u3001PyCuda\u306f\u5927\u304d\u306a\u5f79\u5272\u3092\u679c\u305f\u3057\u3066\u3044\u307e\u3059\u3002\u4ee5\u4e0b\u306f\u3001PyCuda\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u306e\u4e8b\u4f8b\u3067\u3059\u3002<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u6d41\u4f53\u529b\u5b66\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306e\u9ad8\u901f\u5316<\/li>\n\n\n\n<li>Navier-Stokes\u65b9\u7a0b\u5f0f\u306e\u6570\u5024\u89e3\u6cd5\u3084Lattice Boltzmann\u6cd5\u306e\u8a08\u7b97\u3092PyCuda\u3067\u9ad8\u901f\u5316\u3059\u308b\u3053\u3068\u3067\u3001\u5927\u898f\u6a21\u306a\u6d41\u4f53\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u3092\u52b9\u7387\u7684\u306b\u5b9f\u884c\u3067\u304d\u307e\u3059\u3002<\/li>\n\n\n\n<li>\u5206\u5b50\u52d5\u529b\u5b66\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306e\u9ad8\u901f\u5316<\/li>\n\n\n\n<li>\u5206\u5b50\u9593\u306e\u76f8\u4e92\u4f5c\u7528\u3092\u8a08\u7b97\u3059\u308b\u969b\u306b\u3001PyCuda\u3092\u4f7f\u3063\u3066\u30ec\u30ca\u30fc\u30c9\u30fb\u30b8\u30e7\u30fc\u30f3\u30ba\u30dd\u30c6\u30f3\u30b7\u30e3\u30eb\u3084\u7c92\u5b50\u9593\u529b\u306e\u8a08\u7b97\u3092\u9ad8\u901f\u5316\u3067\u304d\u307e\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u5927\u898f\u6a21\u306a\u5206\u5b50\u30b7\u30b9\u30c6\u30e0\u306e\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u304c\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\u3002<\/li>\n\n\n\n<li>\u6709\u9650\u8981\u7d20\u6cd5(FEM)\u306e\u9ad8\u901f\u5316<\/li>\n\n\n\n<li>\u69cb\u9020\u89e3\u6790\u3084\u96fb\u78c1\u754c\u89e3\u6790\u306a\u3069\u3067\u4f7f\u308f\u308c\u308bFEM\u3067\u306f\u3001\u5927\u898f\u6a21\u306a\u9023\u7acb\u65b9\u7a0b\u5f0f\u3092\u89e3\u304f\u5fc5\u8981\u304c\u3042\u308a\u307e\u3059\u3002PyCuda\u3092\u4f7f\u3063\u3066\u525b\u6027\u65b9\u7a0b\u5f0f\u306e\u7d44\u307f\u7acb\u3066\u3068\u89e3\u6cd5\u3092\u9ad8\u901f\u5316\u3059\u308b\u3053\u3068\u3067\u3001\u5927\u898f\u6a21\u306aFEM\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u3092\u52b9\u7387\u7684\u306b\u5b9f\u884c\u3067\u304d\u307e\u3059\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u4ee5\u4e0b\u306f\u3001\u5206\u5b50\u52d5\u529b\u5b66\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306b\u304a\u3051\u308b\u7c92\u5b50\u9593\u529b\u306e\u8a08\u7b97\u3092\u9ad8\u901f\u5316\u3059\u308bPyCuda\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3067\u3059\u3002<\/p>\n\n\n\n<pre class=\"EnlighterJSRAW\" data-enlighter-language=\"generic\" data-enlighter-theme=\"\" data-enlighter-highlight=\"\" data-enlighter-linenumbers=\"\" data-enlighter-lineoffset=\"\" data-enlighter-title=\"\" data-enlighter-group=\"\">import numpy as np\nimport pycuda.autoinit\nimport pycuda.driver as cuda\nfrom pycuda.compiler import SourceModule\n\n# \u30ec\u30ca\u30fc\u30c9\u30fb\u30b8\u30e7\u30fc\u30f3\u30ba\u30dd\u30c6\u30f3\u30b7\u30e3\u30eb\u3092\u8a08\u7b97\u3059\u308bCUDA\u30ab\u30fc\u30cd\u30eb\nmod = SourceModule(\"\"\"\n#define NUM_ATOMS 1024\n#define BLOCK_SIZE 256\n\n__device__ float calc_distance(float* positions, int i, int j) {\n    float dx = positions[i] - positions[j];\n    float dy = positions[i + NUM_ATOMS] - positions[j + NUM_ATOMS];\n    float dz = positions[i + 2 * NUM_ATOMS] - positions[j + 2 * NUM_ATOMS];\n    return sqrtf(dx * dx + dy * dy + dz * dz);\n}\n\n__global__ void lennard_jones(float* positions, float* forces, float epsilon, float sigma) {\n    int i = blockIdx.x * blockDim.x + threadIdx.x;\n    if (i &lt; NUM_ATOMS) {\n        float fx = 0.0f, fy = 0.0f, fz = 0.0f;\n        for (int j = 0; j &lt; NUM_ATOMS; j++) {\n            if (i != j) {\n                float r = calc_distance(positions, i, j);\n                float r2inv = 1.0f \/ (r * r);\n                float r6inv = r2inv * r2inv * r2inv;\n                float force = 24.0f * epsilon * r2inv * r6inv * (2.0f * r6inv - 1.0f);\n                fx += force * (positions[i] - positions[j]) \/ r;\n                fy += force * (positions[i + NUM_ATOMS] - positions[j + NUM_ATOMS]) \/ r;\n                fz += force * (positions[i + 2 * NUM_ATOMS] - positions[j + 2 * NUM_ATOMS]) \/ r;\n            }\n        }\n        forces[i] = fx;\n        forces[i + NUM_ATOMS] = fy;\n        forces[i + 2 * NUM_ATOMS] = fz;\n    }\n}\n\"\"\")\n\n# \u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306e\u30d1\u30e9\u30e1\u30fc\u30bf\nnum_atoms = 1024\nepsilon = 1.0\nsigma = 1.0\ndt = 0.01\n\n# \u521d\u671f\u4f4d\u7f6e\u3068\u529b\u306e\u914d\u5217\u3092\u521d\u671f\u5316\npositions = np.random.rand(num_atoms * 3).astype(np.float32)\nforces = np.zeros_like(positions)\n\n# \u30c7\u30d0\u30a4\u30b9\u30e1\u30e2\u30ea\u3092\u5272\u308a\u5f53\u3066\npositions_gpu = cuda.mem_alloc(positions.nbytes)\nforces_gpu = cuda.mem_alloc(forces.nbytes)\n\n# \u30db\u30b9\u30c8\u304b\u3089\u30c7\u30d0\u30a4\u30b9\u3078\u30c7\u30fc\u30bf\u3092\u8ee2\u9001\ncuda.memcpy_htod(positions_gpu, positions)\n\n# \u30ab\u30fc\u30cd\u30eb\u3092\u547c\u3073\u51fa\u3059\nlennard_jones = mod.get_function(\"lennard_jones\")\nblock = (BLOCK_SIZE, 1, 1)\ngrid = (num_atoms \/\/ BLOCK_SIZE + 1, 1)\nlennard_jones(positions_gpu, forces_gpu, np.float32(epsilon), np.float32(sigma), block=block, grid=grid)\n\n# \u30c7\u30d0\u30a4\u30b9\u304b\u3089\u30db\u30b9\u30c8\u3078\u7d50\u679c\u3092\u8ee2\u9001\ncuda.memcpy_dtoh(forces, forces_gpu)\n\nprint(forces[:10])  # \u7d50\u679c\u3092\u8868\u793a<\/pre>\n\n\n\n<p>\u3053\u306e\u30b3\u30fc\u30c9\u3067\u306f\u3001\u30ec\u30ca\u30fc\u30c9\u30fb\u30b8\u30e7\u30fc\u30f3\u30ba\u30dd\u30c6\u30f3\u30b7\u30e3\u30eb\u306b\u57fa\u3065\u3044\u3066\u7c92\u5b50\u9593\u529b\u3092\u8a08\u7b97\u3057\u3066\u3044\u307e\u3059\u3002<code>calc_distance<\/code>\u95a2\u6570\u3067\u30c7\u30d0\u30a4\u30b9\u4e0a\u306e2\u7c92\u5b50\u9593\u306e\u8ddd\u96e2\u3092\u8a08\u7b97\u3057\u3001<code>lennard_jones<\/code>\u30ab\u30fc\u30cd\u30eb\u3067\u5168\u7c92\u5b50\u306b\u50cd\u304f\u529b\u3092\u8a08\u7b97\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u306e\u3088\u3046\u306bPyCuda\u3092\u4f7f\u3046\u3053\u3068\u3067\u3001\u5927\u898f\u6a21\u306a\u5206\u5b50\u52d5\u529b\u5b66\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u3092\u9ad8\u901f\u306b\u5b9f\u884c\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-17\">PyCuda\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u4e8b\u4f8b\u3068\u6210\u679c<\/h3>\n\n\n\n<p>PyCuda\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u306e\u6210\u679c\u306f\u3001\u591a\u304f\u306e\u4e8b\u4f8b\u3067\u5831\u544a\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u4f8b\u3048\u3070\u3001\u884c\u5217\u7a4d\u306e\u8a08\u7b97\u3067\u306f\u3001CPU\u306b\u6bd4\u3079\u3066\u6570\u5341\u500d\u306e\u9ad8\u901f\u5316\u3092\u9054\u6210\u3067\u304d\u307e\u3059\u3002\u3042\u308b\u4e8b\u4f8b\u3067\u306f\u300110000\u00d710000\u306e\u884c\u5217\u7a4d\u30920.5\u79d2\u3067\u8a08\u7b97\u3059\u308b\u3053\u3068\u306b\u6210\u529f\u3057\u305f\u305d\u3046\u3067\uff53\u3002<\/p>\n\n\n\n<p>\u753b\u50cf\u51e6\u7406\u306e\u5206\u91ce\u3067\u3082\u3001PyCuda\u3092\u4f7f\u3063\u305f\u9ad8\u901f\u5316\u306e\u52b9\u679c\u304c\u793a\u3055\u308c\u3066\u3044\u307e\u3059\u30021000\u679a\u306e\u753b\u50cf\u306b\u5bfe\u3059\u308b\u30a8\u30c3\u30b8\u691c\u51fa\u3092\u3001\u308f\u305a\u304b1\u79d2\u3067\u5b9f\u884c\u3067\u304d\u305f\u3068\u3044\u3046\u4e8b\u4f8b\u3082\u3042\u308b\u305d\u3046\u3067\u3059\u3002<\/p>\n\n\n\n<p>\u91d1\u878d\u5de5\u5b66\u306e\u5206\u91ce\u3067\u306f\u3001\u30aa\u30d7\u30b7\u30e7\u30f3\u4fa1\u683c\u306e\u30e2\u30f3\u30c6\u30ab\u30eb\u30ed\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306bPyCuda\u304c\u6d3b\u7528\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u3042\u308b\u4e8b\u4f8b\u3067\u306f\u3001100\u4e07\u30d1\u30b9\u306e\u30e2\u30f3\u30c6\u30ab\u30eb\u30ed\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u3092\u77ed\u6642\u9593\u3067\u5b9f\u884c\u3059\u308b\u3053\u3068\u306b\u6210\u529f\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n<p>\u3053\u306e\u3088\u3046\u306b\u3001PyCuda\u3092\u4f7f\u3046\u3053\u3068\u3067\u3001\u69d8\u3005\u306a\u5206\u91ce\u3067\u5927\u5e45\u306a\u9ad8\u901f\u5316\u3092\u9054\u6210\u3067\u304d\u307e\u3059\u3002\u7279\u306b\u3001\u5927\u898f\u6a21\u306a\u30c7\u30fc\u30bf\u3092\u51e6\u7406\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\u30bf\u30b9\u30af\u3084\u3001\u8907\u96d1\u306a\u8a08\u7b97\u3092\u7e70\u308a\u8fd4\u3057\u5b9f\u884c\u3059\u308b\u5fc5\u8981\u304c\u3042\u308b\u30bf\u30b9\u30af\u3067\u306f\u3001PyCuda\u306b\u3088\u308b\u9ad8\u901f\u5316\u306e\u52b9\u679c\u304c\u9855\u8457\u3067\u3059\u3002<\/p>\n\n\n\n<p>PyCuda\u306f\u3001\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u3084\u79d1\u5b66\u8a08\u7b97\u3001\u30b7\u30df\u30e5\u30ec\u30fc\u30b7\u30e7\u30f3\u306a\u3069\u306e\u5206\u91ce\u3067\u3001\u5fc5\u8981\u4e0d\u53ef\u6b20\u306a\u30c4\u30fc\u30eb\u3068\u306a\u3063\u3066\u3044\u307e\u3059\u3002\u4eca\u5f8c\u3082\u3001PyCuda\u3092\u6d3b\u7528\u3057\u305f\u9ad8\u901f\u5316\u306e\u4e8b\u4f8b\u304c\u5897\u3048\u3066\u3044\u304f\u3053\u3068\u304c\u671f\u5f85\u3055\u308c\u307e\u3059\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"i-18\">PyCuda\u306e\u3055\u3089\u306a\u308b\u5b66\u7fd2\u30ea\u30bd\u30fc\u30b9\u3068\u4eca\u5f8c\u306e\u5c55\u671b<\/h2>\n\n\n\n<p>\u3053\u3053\u307e\u3067PyCuda\u306e\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3084\u5fdc\u7528\u4f8b\u3092\u898b\u3066\u304d\u307e\u3057\u305f\u304c\u3001\u3055\u3089\u306bPyCuda\u3092\u6df1\u304f\u5b66\u3073\u305f\u3044\u65b9\u306b\u304a\u3059\u3059\u3081\u306e\u5b66\u7fd2\u30ea\u30bd\u30fc\u30b9\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002\u307e\u305f\u3001PyCuda\u3068GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u4eca\u5f8c\u306e\u5c55\u671b\u306b\u3064\u3044\u3066\u3082\u8003\u3048\u3066\u307f\u307e\u3057\u3087\u3046\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-19\">PyCuda\u306e\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3068\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb<\/h3>\n\n\n\n<p>PyCuda\u3092\u5b66\u3076\u4e0a\u3067\u6700\u3082\u91cd\u8981\u306a\u30ea\u30bd\u30fc\u30b9\u306f\u3001\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3068\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u3059\u3002<\/p>\n\n\n\n<p>PyCuda\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\uff08<a href=\"https:\/\/documen.tician.de\/pycuda\/\">https:\/\/documen.tician.de\/pycuda\/<\/a>\uff09\u3067\u306f\u3001PyCuda\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u3001\u57fa\u672c\u7684\u306a\u4f7f\u3044\u65b9\u3001API\u30ea\u30d5\u30a1\u30ec\u30f3\u30b9\u306a\u3069\u304c\u8a73\u3057\u304f\u8aac\u660e\u3055\u308c\u3066\u3044\u307e\u3059\u3002PyCuda\u3092\u4f7f\u3044\u59cb\u3081\u308b\u969b\u306e\u53c2\u8003\u306b\u306a\u308b\u60c5\u5831\u304c\u8c4a\u5bcc\u306b\u63d0\u4f9b\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u307e\u305f\u3001PyCuda Tutorial\uff08<a href=\"https:\/\/documen.tician.de\/pycuda\/tutorial.html\">https:\/\/github.com\/inducer\/pycuda-tutorial<\/a>\uff09\u306f\u3001PyCuda\u306e\u57fa\u790e\u304b\u3089\u5fdc\u7528\u307e\u3067\u3092\u5b66\u3079\u308b\u512a\u308c\u305f\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3067\u3059\u3002\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3068\u4e01\u5be7\u306a\u8aac\u660e\u304c\u7528\u610f\u3055\u308c\u3066\u304a\u308a\u3001PyCuda\u306e\u69d8\u3005\u306a\u6a5f\u80fd\u3092\u6bb5\u968e\u7684\u306b\u7406\u89e3\u3067\u304d\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u3055\u3089\u306b\u3001PyCuda\u306e\u516c\u5f0f\u30ea\u30dd\u30b8\u30c8\u30ea\uff08<a href=\"https:\/\/github.com\/inducer\/pycuda\/tree\/main\/examples\">https:\/\/github.com\/inducer\/pycuda\/tree\/main\/examples<\/a>\uff09\u306b\u306f\u3001\u6570\u591a\u304f\u306e\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u304c\u7528\u610f\u3055\u308c\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u3089\u306e\u30b3\u30fc\u30c9\u3092\u8aad\u3080\u3053\u3068\u3067\u3001PyCuda\u306e\u5b9f\u8df5\u7684\u306a\u4f7f\u3044\u65b9\u3092\u5b66\u3076\u3053\u3068\u304c\u3067\u304d\u308b\u3067\u3057\u3087\u3046\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-20\">PyCuda\u306b\u95a2\u3059\u308b\u66f8\u7c4d\u3084\u60c5\u5831\u30b5\u30a4\u30c8<\/h3>\n\n\n\n<p>PyCuda\u306b\u3064\u3044\u3066\u3001\u3088\u308a\u4f53\u7cfb\u7684\u306b\u5b66\u3073\u305f\u3044\u65b9\u306b\u306f\u3001\u66f8\u7c4d\u304c\u304a\u3059\u3059\u3081\u3067\u3059\u3002<\/p>\n\n\n\n<p>\u201cPython\u3067\u59cb\u3081\u308bCUDA\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u201d by Dr. t-tetsuya\u306f\u3001Python\u3092\u4f7f\u3063\u3066\u3044\u308b\u30a8\u30f3\u30b8\u30cb\u30a2\u3001\u7814\u7a76\u8005\u3001\u5b66\u751f\u306e\u65b9\u306b\u5411\u3051\u305fCUDA\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u5165\u9580\u66f8\u3067\u3059\u3002<\/p>\n\n\n\n<p>\u201cHands-On GPU Programming with Python and CUDA\u201d by Dr. Brian Tuomanen\u306f\u3001PyCuda\u3092\u542b\u3080GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u5165\u9580\u66f8\u3067\u3059\u3002CUDA\u306e\u57fa\u790e\u304b\u3089\u3001\u5b9f\u969b\u306e\u30d7\u30ed\u30b8\u30a7\u30af\u30c8\u307e\u3067\u5e45\u5e83\u304f\u6271\u3063\u3066\u304a\u308a\u3001GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u5168\u4f53\u50cf\u3092\u63b4\u3080\u306e\u306b\u9069\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>PyCuda\u306b\u95a2\u3059\u308b\u6700\u65b0\u60c5\u5831\u3084\u6280\u8853\u8a18\u4e8b\u3092\u63a2\u3059\u306a\u3089\u3001NVIDIA Developer Blog\uff08<a href=\"https:\/\/developer.nvidia.com\/blog\">https:\/\/developer.nvidia.com\/blog<\/a>\uff09\u304c\u304a\u3059\u3059\u3081\u3067\u3059\u3002NVIDIA\u306e\u516c\u5f0f\u30d6\u30ed\u30b0\u3067\u3042\u308a\u3001GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306b\u95a2\u3059\u308b\u6709\u76ca\u306a\u60c5\u5831\u304c\u5b9a\u671f\u7684\u306b\u66f4\u65b0\u3055\u308c\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u307e\u305f\u3001Stack Overflow\uff08<a href=\"https:\/\/stackoverflow.com\/questions\/tagged\/pycuda\">https:\/\/stackoverflow.com\/questions\/tagged\/pycuda<\/a>\uff09\u3067\u306f\u3001PyCuda\u306b\u95a2\u3059\u308b\u8cea\u554f\u3068\u56de\u7b54\u304c\u6d3b\u767a\u306b\u884c\u308f\u308c\u3066\u3044\u307e\u3059\u3002\u5b9f\u969b\u306e\u30b3\u30fc\u30c7\u30a3\u30f3\u30b0\u3067\u906d\u9047\u3057\u305f\u554f\u984c\u306e\u89e3\u6c7a\u7b56\u3092\u898b\u3064\u3051\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u3067\u3057\u3087\u3046\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"i-21\">PyCuda\u3068GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u5c06\u6765\u6027<\/h3>\n\n\n\n<p>\u8fd1\u5e74\u3001\u6a5f\u68b0\u5b66\u7fd2\u3084\u30c7\u30a3\u30fc\u30d7\u30e9\u30fc\u30cb\u30f3\u30b0\u306e\u767a\u5c55\u306b\u4f34\u3044\u3001GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u9700\u8981\u304c\u6025\u901f\u306b\u9ad8\u307e\u3063\u3066\u3044\u307e\u3059\u3002\u5927\u898f\u6a21\u306a\u30c7\u30fc\u30bf\u51e6\u7406\u3084\u8907\u96d1\u306a\u8a08\u7b97\u3092\u9ad8\u901f\u306b\u884c\u3046\u305f\u3081\u306b\u3001GPU\u306e\u4e26\u5217\u51e6\u7406\u80fd\u529b\u304c\u6b20\u304b\u305b\u306a\u304f\u306a\u3063\u3066\u3044\u308b\u306e\u3067\u3059\u3002<\/p>\n\n\n\n<p>\u307e\u305f\u3001\u30a8\u30c3\u30b8\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u3084IoT\u306e\u666e\u53ca\u306b\u3088\u308a\u3001\u7d44\u307f\u8fbc\u307f\u6a5f\u5668\u3067\u306eGPU\u6d3b\u7528\u3082\u9032\u3093\u3067\u3044\u304f\u3068\u4e88\u60f3\u3055\u308c\u307e\u3059\u3002\u30bb\u30f3\u30b5\u30fc\u30c7\u30fc\u30bf\u306e\u30ea\u30a2\u30eb\u30bf\u30a4\u30e0\u51e6\u7406\u3084\u3001AI\u6a5f\u80fd\u306e\u5b9f\u88c5\u306a\u3069\u3001GPU\u3092\u4f7f\u3063\u305f\u9ad8\u5ea6\u306a\u51e6\u7406\u304c\u3088\u308a\u8eab\u8fd1\u306b\u306a\u3063\u3066\u3044\u304f\u3067\u3057\u3087\u3046\u3002<\/p>\n\n\n\n<p>\u30de\u30eb\u30c1GPU\u74b0\u5883\u3084\u30af\u30e9\u30a6\u30c9GPU\u306e\u5229\u7528\u3082\u5897\u52a0\u3057\u3066\u304a\u308a\u3001\u3088\u308a\u5927\u898f\u6a21\u306a\u4e26\u5217\u51e6\u7406\u304c\u53ef\u80fd\u306b\u306a\u308a\u307e\u3059\u3002PyCuda\u3092\u4f7f\u3048\u3070\u3001\u3053\u308c\u3089\u306e\u74b0\u5883\u3092\u52b9\u679c\u7684\u306b\u6d3b\u7528\u3057\u3001\u3055\u3089\u306a\u308b\u9ad8\u901f\u5316\u3092\u5b9f\u73fe\u3067\u304d\u308b\u306f\u305a\u3067\u3059\u3002<\/p>\n\n\n\n<p>CUDA\u306b\u52a0\u3048\u3066\u3001OpenCL\u3084Vulkan\u306a\u3069\u306e\u30aa\u30fc\u30d7\u30f3\u30b9\u30bf\u30f3\u30c0\u30fc\u30c9\u3082\u767a\u5c55\u3057\u3066\u3044\u307e\u3059\u3002\u3053\u308c\u306b\u3088\u308a\u3001GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u30a8\u30b3\u30b7\u30b9\u30c6\u30e0\u304c\u3088\u308a\u8c4a\u304b\u306b\u306a\u308a\u3001\u69d8\u3005\u306a\u74b0\u5883\u3067GPU\u3092\u6d3b\u7528\u3057\u3084\u3059\u304f\u306a\u308b\u3068\u671f\u5f85\u3055\u308c\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4ee5\u4e0a\u306e\u3088\u3046\u306b\u3001GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306f\u3053\u308c\u304b\u3089\u307e\u3059\u307e\u3059\u91cd\u8981\u306b\u306a\u3063\u3066\u3044\u304f\u3068\u8003\u3048\u3089\u308c\u307e\u3059\u3002PyCuda\u306f\u3001\u305d\u306e\u3088\u3046\u306a\u672a\u6765\u306b\u304a\u3044\u3066\u3082\u3001GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u3092\u652f\u3048\u308b\u91cd\u8981\u306a\u30c4\u30fc\u30eb\u3067\u3042\u308a\u7d9a\u3051\u308b\u3067\u3057\u3087\u3046\u3002<\/p>\n\n\n\n<p>PyCuda\u3092\u5b66\u3076\u3053\u3068\u306f\u3001\u3053\u308c\u304b\u3089\u306e\u30b3\u30f3\u30d4\u30e5\u30fc\u30c6\u30a3\u30f3\u30b0\u306e\u4e16\u754c\u3067\u5927\u3044\u306b\u5f79\u7acb\u3064\u306f\u305a\u3067\u3059\u3002\u516c\u5f0f\u30c9\u30ad\u30e5\u30e1\u30f3\u30c8\u3084\u30c1\u30e5\u30fc\u30c8\u30ea\u30a2\u30eb\u3001\u66f8\u7c4d\u306a\u3069\u3092\u6d3b\u7528\u3057\u3066\u3001PyCuda\u3068GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u30b9\u30ad\u30eb\u3092\u78e8\u3044\u3066\u3044\u304d\u307e\u3057\u3087\u3046\u3002\u7686\u3055\u3093\u306e\u4eca\u5f8c\u306e\u6d3b\u8e8d\u3092\u671f\u5f85\u3057\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n<p>\u4ee5\u4e0a\u3067\u3001\u300cPyCuda\u3067\u59cb\u3081\u308bGPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u5165\u9580\u300d\u306e\u8a18\u4e8b\u306f\u7d42\u4e86\u3067\u3059\u3002PyCuda\u306e\u57fa\u790e\u304b\u3089\u5fdc\u7528\u307e\u3067\u3001\u5e45\u5e83\u3044\u30c8\u30d4\u30c3\u30af\u3092\u6271\u3044\u307e\u3057\u305f\u3002\u3053\u306e\u8a18\u4e8b\u304c\u3001\u8aad\u8005\u306e\u7686\u3055\u3093\u306b\u3068\u3063\u3066\u3001PyCuda\u3068GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306e\u7406\u89e3\u3092\u6df1\u3081\u308b\u4e00\u52a9\u3068\u306a\u308c\u3070\u5e78\u3044\u3067\u3059\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>GPU\u30d7\u30ed\u30b0\u30e9\u30df\u30f3\u30b0\u306b\u8208\u5473\u304c\u3042\u308bPython\u30e6\u30fc\u30b6\u30fc\u306b\u3068\u3063\u3066\u3001PyCuda\u306f\u5f37\u529b\u306a\u30c4\u30fc\u30eb\u3067\u3059\u3002\u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001PyCuda\u306e\u57fa\u790e\u304b\u3089\u5fdc\u7528\u307e\u3067\u3001\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u65b9\u6cd5\u3001\u30b5\u30f3\u30d7\u30eb\u30b3\u30fc\u30c9\u3001\u5b9f\u8df5\u7684\u306a\u30e6\u30fc\u30b9\u30b1\u30fc\u30b9\u3092\u7d39\u4ecb\u3057\u307e\u3059\u3002PyCuda &#8230; <\/p>\n","protected":false},"author":1,"featured_media":523,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[],"class_list":{"0":"post-521","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-python"},"_links":{"self":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts\/521","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=521"}],"version-history":[{"count":2,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts\/521\/revisions"}],"predecessor-version":[{"id":524,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/posts\/521\/revisions\/524"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=\/wp\/v2\/media\/523"}],"wp:attachment":[{"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=521"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=521"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/chocottopro.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=521"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}