{"id":54,"date":"2012-01-29T18:38:12","date_gmt":"2012-01-29T09:38:12","guid":{"rendered":"http:\/\/staka.jp\/wordpress\/?p=54"},"modified":"2012-01-29T18:38:12","modified_gmt":"2012-01-29T09:38:12","slug":"gpgpu%e3%81%a7%e7%b2%be%e5%ba%a6%e4%bf%9d%e8%a8%bc%e4%bb%98%e3%81%8d%e8%a8%88%e7%ae%97","status":"publish","type":"post","link":"https:\/\/staka.jp\/wordpress\/?p=54","title":{"rendered":"GPGPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u8a08\u7b97"},"content":{"rendered":"<p>\u91d1\u878d\u306e\u8a08\u7b97\u306b\u304a\u3044\u3066\u3001\u8a08\u7b97\u7cbe\u5ea6\u306f\u8d85\u91cd\u8981\u3067\u3042\u308b\u3002\u7cbe\u5ea6\u304c\u4f4e\u3051\u308c\u3070\u4fa1\u683c\u4ed8\u3051\u306b\u5931\u6557\u3057\u5927\u640d\u3059\u308b\u304b\u3082\u3057\u308c\u306a\u3044\u3002<br \/>\nCUDA\u306b\u306f\u3001__fadd_rn(x, y)\u3001__fadd_rz(x, y)\u3001__dadd_rd(x, y)\u3001__dadd_ru(x, y)\u3001\u306e\u3088\u3046\u306b\u56db\u5247\u6f14\u7b97\u3067\u4e38\u3081\u306e\u65b9\u5411\u3092\u6307\u5b9a\u53ef\u80fd\u306a\u95a2\u6570\u304c\u7528\u610f\u3055\u308c\u3066\u3044\u308b\u3002\u6570\u5b66\u95a2\u6570\u306e\u8aa4\u5dee\u306b\u3064\u3044\u3066\u306e\u60c5\u5831\u3082\u4e0e\u3048\u3089\u308c\u3066\u304a\u308aGPGPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u8a08\u7b97\u3092\u884c\u3046\u3053\u3068\u304c\u53ef\u80fd\u3067\u3042\u308b\u3002<br \/>\n\u524d\u56de\u540c\u69d8\u3001\u30d6\u30e9\u30c3\u30af\u30b7\u30e7\u30fc\u30eb\u30ba\u30e2\u30c7\u30eb\u3067\u306e\u30aa\u30d7\u30b7\u30e7\u30f3\u30d7\u30e9\u30a4\u30b7\u30f3\u30b0\u3092\u4f8b\u306b\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u8a08\u7b97\u306e\u5b9f\u9a13\u3092\u884c\u3063\u305f\u3002\u6761\u4ef6\u306f\u4e0b\u8a18\u306e\u901a\u308a\u3002<\/p>\n<ul>\n<li>\u8a08\u6e2c\u5bfe\u8c61\u306e\u30aa\u30d7\u30b7\u30e7\u30f3\u660e\u7d30\u306f1024\u4e07\u4ef6\u3002\u30d1\u30e9\u30e1\u30fc\u30bf\u306f\u4e71\u6570\u306b\u3088\u308a\u767a\u751f\u3055\u305b\u308b\u3002<\/li>\n<li>\u30d7\u30ec\u30df\u30a2\u30e0\u3092\u7b97\u51fa\u3002\uff08\u30c7\u30eb\u30bf\u3001\u30ac\u30f3\u30de\u306f\u3081\u3093\u3069\u30fc\u306a\u306e\u3067\u7701\u7565\u3057\u305f\u3002\uff09<\/li>\n<li>\u8a08\u7b97\u306f\u3059\u3079\u3066double\u578b\u3067\u884c\u3046\u3002\uff08\u5358\u7cbe\u5ea6\u7248\u306f\u5f8c\u534a\uff09<\/li>\n<li>Amazon EC2\u306e\u30af\u30e9\u30b9\u30bf GPU \u30af\u30a2\u30c9\u30e9\u30d7\u30eb \u30a8\u30af\u30b9\u30c8\u30e9\u30e9\u30fc\u30b8\u3092\u4f7f\u7528\u3002GPU\u306fFermi M2050\u3092<strong>1\u3064<\/strong>\u4f7f\u7528\u3002CPU\u306f<strong>1\u30b3\u30a2\u306e\u307f<\/strong>\u4f7f\u7528\u3002\u30b3\u30f3\u30d1\u30a4\u30eb\u306f\u5168\u90e8CUDA\u3067\u3084\u3063\u3066\u3044\u308b\u3002\uff08CPU\u306b\u306f\u4e0d\u5229\u306a\u6761\u4ef6\u3060\u304c\u3001CPU\u3068\u306e\u8a08\u7b97\u6642\u9593\u306f\u6bd4\u8f03\u3057\u306a\u3044\u3002\uff09<\/li>\n<li>\u56db\u5247\u6f14\u7b97\u306b\u3064\u3044\u3066\u306f\u4e38\u3081\u306e\u65b9\u5411\u3092\u6307\u5b9a\u3057\u3064\u3064\u533a\u9593\u6f14\u7b97\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u3092\u884c\u3046\u3002<\/li>\n<li>\u6570\u5b66\u95a2\u6570\u306b\u3064\u3044\u3066\u306f\u8aa4\u7b97\u60c5\u5831\u3092\u52a0\u5473\u3057\u3064\u3064\u533a\u9593\u3092\u8a2d\u5b9a\u3057\u7cbe\u5ea6\u4fdd\u8a3c\u3092\u884c\u3046\u3002<strong>\uff08\u308f\u308a\u3068\u3066\u304d\u30fc\u306b\u3084\u3063\u3066\u3044\u308b\u3002\uff09<\/strong><\/li>\n<\/ul>\n<p>\u7d50\u679c\u306f\u4e0b\u8a18\u306e\u901a\u308a\u3002\u6570\u5b57\u306e\u9806\u306fGPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u8a08\u7b97\u3092\u3057\u305f\u30d7\u30ec\u30df\u30a2\u30e0\u306e\u4e0b\u754c\u3001CPU\u3067\u306e\u8a08\u7b97\u7d50\u679c\u3001GPU\u3067\u306e\u8a08\u7b97\u7d50\u679c\u3001GPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u8a08\u7b97\u3092\u3057\u305f\u30d7\u30ec\u30df\u30a2\u30e0\u306e\u4e0a\u754c\u3068\u306a\u3063\u3066\u3044\u308b\u3002\u6700\u5f8c\u306e\u6642\u9593\u306f\u3001CPU\u3067\u306e\u8a08\u7b97\u6642\u9593\u3001GPU\u3078\u306e\u30c7\u30fc\u30bf\u8ee2\u9001\u3068\u521d\u671f\u5316\u306b\u304b\u304b\u3063\u305f\u6642\u9593\u3001GPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u8a08\u7b97\u3092\u3057\u305f\u3068\u304d\u306b\u304b\u304b\u3063\u305f\u6642\u9593\uff08\u7d50\u679c\u306e\u8ee2\u9001\u6642\u9593\u8fbc\u307f\uff09\u3001GPU\u3067\u3075\u3064\u30fc\u306b\u8a08\u7b97\u3057\u305f\u3068\u304d\u306b\u304b\u304b\u3063\u305f\u6642\u9593\uff08\u7d50\u679c\u306e\u8ee2\u9001\u6642\u9593\u8fbc\u307f\uff09\u3002<\/p>\n<blockquote>\n<pre>GPU lower &lt; CPU | GPU &lt; GPU upper\nPREMIUM[0]: 0.405499973123749 &lt; 0.405499973123820 | 0.405499973123821 &lt; 0.405499973123893\nPREMIUM[1024000]: 4.029060162124779 &lt; 4.029060162124900 | 4.029060162124893 &lt; 4.029060162125027\nPREMIUM[2048000]: 0.733045301526717 &lt; 0.733045301526774 | 0.733045301526773 &lt; 0.733045301526825\nPREMIUM[3072000]: 1.977617064175732 &lt; 1.977617064175853 | 1.977617064175839 &lt; 1.977617064175931\nPREMIUM[4096000]: 0.162532300560706 &lt; 0.162532300560737 | 0.162532300560734 &lt; 0.162532300560762\nPREMIUM[5120000]: 0.767333091428844 &lt; 0.767333091428927 | 0.767333091428925 &lt; 0.767333091428991\nPREMIUM[6144000]: 8.282037131422769 &lt; 8.282037131422967 | 8.282037131422962 &lt; 8.282037131423138\nPREMIUM[7168000]: 0.020357967912149 &lt; 0.020357967912157 | 0.020357967912157 &lt; 0.020357967912164\nPREMIUM[8192000]: 0.442812778153888 &lt; 0.442812778153947 | 0.442812778153947 &lt; 0.442812778154016\nPREMIUM[9216000]: 0.727449524196905 &lt; 0.727449524196972 | 0.727449524196975 &lt; 0.727449524197043\nCPU time: 2779.697021 (ms)\nGPU Initial time: 2082.956055 (ms)\nGPU Processing time 1: 442.318848 (ms)\nGPU Processing time 2: 89.383057 (ms)<\/pre>\n<\/blockquote>\n<p>GPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u306e\u8a08\u7b97\u3092\u884c\u3046\u305f\u3081\u306b\u666e\u901a\u306e5\u500d\u304f\u3089\u3044\u306e\u6642\u9593\u304c\u304b\u304b\u308b\u3002\u5358\u7cbe\u5ea6\u3067\u8a08\u7b97\u3059\u308b\u3068\u4e0b\u8a18\u306e\u3088\u3046\u306b\u306a\u308b\u3002<\/p>\n<blockquote>\n<pre>GPU lower &lt; CPU | GPU &lt; GPU upper\nPREMIUM[0]: 0.39829 &lt; 0.39859 | 0.39859 &lt; 0.39882\nPREMIUM[1024000]: 4.04288 &lt; 4.04349 | 4.04350 &lt; 4.04403\nPREMIUM[2048000]: 0.72358 &lt; 0.72385 | 0.72385 &lt; 0.72404\nPREMIUM[3072000]: 1.98723 &lt; 1.98767 | 1.98767 &lt; 1.98803\nPREMIUM[4096000]: 0.16034 &lt; 0.16050 | 0.16050 &lt; 0.16059\nPREMIUM[5120000]: 0.77226 &lt; 0.77262 | 0.77262 &lt; 0.77288\nPREMIUM[6144000]: 8.28720 &lt; 8.28772 | 8.28772 &lt; 8.28814\nPREMIUM[7168000]: 0.02022 &lt; 0.02033 | 0.02033 &lt; 0.02036\nPREMIUM[8192000]: 0.44575 &lt; 0.44603 | 0.44603 &lt; 0.44622\nPREMIUM[9216000]: 0.72577 &lt; 0.72608 | 0.72608 &lt; 0.72630\nCPU time: 8085.477051 (ms)\nGPU Initial time: 2080.006104 (ms)\nGPU Processing time 1: 200.956787 (ms)\nGPU Processing time 2: 45.681152 (ms)<\/pre>\n<\/blockquote>\n<p>\u3053\u306e\u4f8b\u306b\u3064\u3044\u3066\u306fCPU\u306b\u3064\u3044\u3066\u306f\u5358\u7cbe\u5ea6\u306e\u65b9\u304c\u500d\u7cbe\u5ea6\u3088\u308a\u9045\u3044\u3002\u7406\u7531\u3068\u3057\u3066\u306f\u6700\u9069\u5316\uff08\u7279\u306bSSE\u306a\u306e\u304bFPU\u306a\u306e\u304b\uff09\u3084\u3089\u30b3\u30f3\u30d1\u30a4\u30e9\u3084\u3089CPU\u306e\u7279\u6027\u3084\u3089\u30e1\u30e2\u30ea\u3084\u3089\u306e\u8a71\u984c\u304c\u3042\u308b\u304c\u3053\u3053\u3067\u306f\u7701\u7565\u3059\u308b\u3002<br \/>\n\u5358\u7cbe\u5ea6\u306e\u5834\u5408\u3082\u666e\u901a\u306e\u8a08\u7b97\u3068\u304f\u3089\u3079\u30665\u500d\u304f\u3089\u3044\u306e\u6642\u9593\u304c\u304b\u304b\u3063\u3066\u3044\u308b\u3002\u8a08\u7b97\u91cf\u3082\u305d\u3046\u3060\u304c\u5206\u5c90\u547d\u4ee4\u304c\u8ffd\u52a0\u3055\u308c\u305f\u30da\u30ca\u30eb\u30c6\u30a3\u304c\u5927\u304d\u3044\u3082\u306e\u3068\u601d\u308f\u308c\u308b\u3002<br \/>\n\u5358\u7cbe\u5ea6\u306e\u65b9\u306f\u7cbe\u5ea6\u4fdd\u8a3c\u6709\u308a\u7121\u3057\u306e\u6642\u9593\u306e\u6bd4\u7387\u304c\u5909\u308f\u308b\u304b\u3092\u8abf\u3079\u308b\u4e8b\u304c\u4e3b\u76ee\u7684\u3067\u3042\u308b\u305f\u3081\u3001\u56db\u5247\u6f14\u7b97\u4ee5\u5916\u306e\u95a2\u6570\u306b\u95a2\u308f\u308b\u7cbe\u5ea6\u4fdd\u8a3c\u3092\u975e\u5e38\u306b\u8352\u304f\u884c\u3063\u3066\u3044\u308b\u3002\u4e0a\u754c\u3001\u4e0b\u754c\u306e\u5dee\u304c\u5e83\u3044\u306e\u306f\u305d\u306e\u305f\u3081\u3067\u3042\u308b\u3002<br \/>\n\u4ee5\u4e0a\u306e\u7d50\u679c\u304b\u3089\u3001\u30d6\u30e9\u30c3\u30af\u30b7\u30e7\u30fc\u30eb\u30ba\u30e2\u30c7\u30eb\u3067\u306e\u30aa\u30d7\u30b7\u30e7\u30f3\u30d7\u30e9\u30a4\u30b7\u30f3\u30b0\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u8a08\u7b97\u3092\u884c\u3046\u305f\u3081\u306b\u306f\u901a\u5e38\u306e5\u500d\u7a0b\u5ea6\u306e\u8a08\u7b97\u6642\u9593\u304c\u5fc5\u8981\u3067\u3042\u308b\u3053\u3068\u304c\u308f\u304b\u308b\u3002<\/p>\n<h4>\u88dc\u8a18<\/h4>\n<p>\u6700\u521d\u306b\u8a08\u7b97\u7cbe\u5ea6\u304c\u8d85\u91cd\u8981\u3068\u304b\u66f8\u3044\u305f\u3051\u308c\u3069\u3082\u3001\u305d\u3082\u305d\u3082\u30d6\u30e9\u30c3\u30af\u30b7\u30e7\u30fc\u30eb\u30ba\u30e2\u30c7\u30eb\u3067\u306f\u30dc\u30e9\u30c6\u30a3\u30ea\u30c6\u30a3\u30b9\u30de\u30a4\u30eb\u3084\u91d1\u5229\u306a\u3093\u304b\u306e\u671f\u9593\u69cb\u9020\u3092\u7121\u8996\u3057\u3066\u3057\u307e\u3063\u3066\u3044\u308b\u306e\u3067\u3001\u305d\u306e\u30e2\u30c7\u30eb\u306b\u8d77\u56e0\u3059\u308b\u5dee\u7570\u306e\u307b\u3046\u304c\u6570\u5024\u8a08\u7b97\u4e0a\u306e\u8aa4\u5dee\u3088\u308a\u5927\u304d\u3044\u3002\u306a\u306e\u3067\u3001\u7cbe\u5ea6\u4fdd\u8a3c\u3092\u3084\u308b\u610f\u5473\u306f\u3042\u3093\u307e\u308a\u7121\uff08\uff52\uff59\u30fb\u30fb\u30fb\u3044\u3084\u3059\u3053\u3057\u306f\u3042\u308b\u304b\uff4d\u30fb\u30fb\u30fb\u307e\u3041\u3061\u3087\u3063\u3068\u306f\u30fb\u30fb\u30fb<br \/>\n\u73fe\u5b9f\u554f\u984c\u3067\u306e\u4e0a\u754c\u4e0b\u754c\u3092\u6c42\u3081\u305f\u3051\u308c\u3070\u6b63\u306e\u30d0\u30a4\u30a2\u30b9\u304c\u304b\u304b\u308b\u5f0f\u3068\u8ca0\u306e\u30d0\u30a4\u30a2\u30b9\u304c\u304b\u304b\u308b\u5f0f\u3067\u631f\u307f\u6483\u3061\u30fc\u3068\u3044\u3046\u611f\u3058\u3067\u3084\u308b\u5fc5\u8981\u304c\u3042\u308b\u3002\uff08\u30a2\u30e1\u30ea\u30ab\u30f3\u30aa\u30d7\u30b7\u30e7\u30f3\u306e\u30d7\u30e9\u30a4\u30b7\u30f3\u30b0\u3067\u898b\u304b\u3051\u308b\u3002\u3068\u306f\u3044\u3048\u3001\u305d\u3053\u307e\u3067\u3084\u3063\u3066\u3082\u5b9f\u969b\u554f\u984c\u30e2\u30c7\u30eb\u8aa4\u5dee\u306f\u6b8b\u308b\u3002\u3002\u3002\uff09<br \/>\n\u305d\u308c\u306f\u304a\u3044\u3068\u3044\u3066\u8a08\u7b97\u6642\u9593\u304c5\u500d\u3068\u3044\u3046\u306e\u306f\u305d\u3093\u306a\u3082\u3093\u304b\u306a\u3068\u3044\u3046\u5370\u8c61\u3002\u4e0a\u754c\u3001\u4e0b\u754c\u306e\u305f\u3081\u306b\u6700\u4f4e2\u500d\u306e\u8a08\u7b97\u3068\u30e1\u30e2\u30ea\u304c\u5fc5\u8981\u3067\u3001\u304b\u3064\u3001\u5272\u308a\u7b97\u306e\u5834\u5408\u306f\u6b63\u8ca0\u306e\u5834\u5408\u5206\u3051\u304c\u5fc5\u8981\u306a\u305f\u3081\u5206\u5c90\u304c\u5165\u3063\u3066\u3057\u307e\u3046\u3002<br \/>\n\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u6570\u5024\u8a08\u7b97\u306f\u9762\u767d\u3044\u8003\u3048\u65b9\u3060\u3068\u601d\u3046\u304c\u3001\u30a4\u30de\u30a4\u30c1\u6d41\u884c\u3063\u3066\u3044\u306a\u3044\u6c17\u304c\u3057\u306a\u304f\u3082\u7121\u3044\u3002\u30b3\u30f3\u30d1\u30a4\u30e9\u30aa\u30d7\u30b7\u30e7\u30f3\u306a\u308aVM\u306e\u8a2d\u5b9a\u306a\u308a\u3067\u6c17\u8efd\u306b\u4f7f\u3048\u308b\u6642\u4ee3\u304c\u6765\u305f\u3089\u5229\u7528\u3059\u308b\u6a5f\u4f1a\u306f\u3042\u308a\u305d\u3046\u3060\u304c\u3002\u3002\u3002<br \/>\n&nbsp;<br \/>\n&nbsp;<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u91d1\u878d\u306e\u8a08\u7b97\u306b\u304a\u3044\u3066\u3001\u8a08\u7b97\u7cbe\u5ea6\u306f\u8d85\u91cd\u8981\u3067\u3042\u308b\u3002\u7cbe\u5ea6\u304c\u4f4e\u3051\u308c\u3070\u4fa1\u683c\u4ed8\u3051\u306b\u5931\u6557\u3057\u5927\u640d\u3059\u308b\u304b\u3082\u3057\u308c\u306a\u3044\u3002 CUDA\u306b\u306f\u3001__fadd_rn(x, y)\u3001__fadd_rz(x, y)\u3001__dadd_rd(x, y)\u3001__dadd &hellip; <a href=\"https:\/\/staka.jp\/wordpress\/?p=54\" class=\"more-link\"><span class=\"screen-reader-text\">&#8220;GPGPU\u3067\u7cbe\u5ea6\u4fdd\u8a3c\u4ed8\u304d\u8a08\u7b97&#8221; \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":{"footnotes":""},"categories":[3,5],"tags":[18,39,82,86],"class_list":["post-54","post","type-post","status-publish","format-standard","hentry","category-gpgpu","category-linux","tag-amazon-ec2","tag-gpu","tag-82","tag-86"],"_links":{"self":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/54","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=54"}],"version-history":[{"count":0,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/54\/revisions"}],"wp:attachment":[{"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=54"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=54"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/staka.jp\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=54"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}