ãã®ææ¸ã¯ Paralle Reduction ã®é
ã«ãã Whitepaper ã®æ訳ã§ãã CUDAã§ããã°ã©ã ãæ¸ãã«ããã£ã¦ã®ãã¯ããã¯ãå¤æ°è©°ã¾ã£ã¦ãã¾ãã 並åãªãã¯ã·ã§ã³ (Parallel Reduction) ãªãã¯ã·ã§ã³ã¯ä¸¦åãã¼ã¿å¦çãæ§æããä¸è¬çãã¤éè¦ãªé¨åã§ããã CUDAã§ã®å®è£
ã¯å®¹æã ãç解ããã®ã¯å®è£
ãããå°é£ã§ããã æé©åã®é常ã«æçãªäºä¾ã§ããã ãã®ææ¸ã§ã¯7種é¡ã®ç°ãªããã¼ã¸ã§ã³ãä¸ã¤ãã¤æ¤è¨¼ããã ããããã® thread block ã§ã¯(äºå)æ¨æ§é ãã¼ã¹ã®ã¢ããã¼ããç¨ããã (å®è£
ã«ããã£ã¦ã¯)ãä¸ã¤ã®å¯¾è±¡ã«ã¤ãã¦è¤æ°ã® thread block ãç¨ã㦠reduction ãå®è¡ã§ããããã«ããå¿
è¦ãããã 巨大ãªé
åã«å¯¾ãã¦å¦çãè¡ãå ´åã«å
¨ã¦ã® multiprocessor ã稼åããã ããã«ã¯ãå thread block
{{#tags}}- {{label}}
{{/tags}}