並åå¦çæ代ã®æ¼ç®ã®ä¸ææ§
ã足ãç®ã®çããä¸æã«æ±ºã¾ããªãã¦æ話ã«ãªããããã¨ããã話ã
ã確èªããç¯å²ã§ã¯ãMATLAB 7.8 (2009a, 64bit) ã«è¶³ãç®ããããï¼ãã¨ãã°ãµã¤ãºã500,000x1ã®è¡åã®åè¦ç´ ãå ¨é¨è¶³ãï¼ã¨ããã®çãã¯ä¸æã«æ±ºã¾ããªããå ¨ãåä¸ã®ãã¼ã¿ã«ã¤ãã¦åãPCã»ç°å¢ã§ç¹°ãè¿ãè¨ç®ããã¦ãã»ã¼æ¯åéãçããè¿ã£ã¦ããã
ãããã«ã¤ãã¦MathWorks社ï¼ã®å°ãªãã¨ãä¸ç¤¾å¡ï¼ã®è¦è§£ãæ¸ããã¦ãããã¼ã¸ãè¦ã¤ããã
Comparing Single-threaded vs. Multithreaded Floating Point Calculations
Loren on the Art of MATLAB
December 4th, 2009
ã彼女ã®è¦è§£ãè¦ç´ããã¨ã
- æµ®åå°æ°ç¹æ°ãç¨ããæ¼ç®ã«ããã¦ã足ãç®ã¯çµåæ³åãæºãããªãï¼ a+(b+c) â (a+b)+c ï¼ã
- æ¼ç®ã®é åºã¯å¦ç系任ãã§ããã
- ç¹ã«ãã«ãã³ã¢ã§ä¸¦åæ¼ç®ãããå ´åãæ¼ç®é åºã®ä»»ææ§ã¯é«ã¾ãã
ãã£ã¦è¶³ãç®ã®çããé決å®è«çã«ãªãã®ã¯ãã°ã§ã¯ãªãã¦ä»æ§ã§ãã¨ãï¼ã»ã»ã»ã¨ãè¨ããªããææ°ãã¼ã¸ã§ã³ï¼7.9, 2009bï¼ã¯ããããæåã示ããªãï¼è¶³ãç®ã®çããä¸æã«æ±ºã¾ãï¼ã®ã§ãMathWorks社ããã£ã±ãããã¯ã¾ããã¨æã£ã¦ããã®ãããããªããï¼
ãå人çã«ã¯ãå¦çç³»ãç°ãªãã°æ¼ç®çµæãéã£ã¦ããå¯è½æ§ã¯ããã ãããªãã¨ã¯æã£ã¦ãããã§ãåãå¦çç³»ã«åãè¨ç®ããããã¦ãä¸ç§åã®çãã¨ä»ã®çããç°ãªãã¨ã¯æã£ã¦ããªãã£ããããã¦ãã£ããæ¼ç®ã®ä¸ææ§ãè¦è¾¼ãã ï¼æ æ°ãªï¼ã³ã¼ããä¸é¨æ¸ãã¦ããã®ã§ããã°ããä½ãèµ·ãã¦ããã®ãå¤ããªãã£ãã
ã彼女ã®è¨ããã¨ããç解ã¯åºæ¥ããå¹çãæåªå
ãããå ´åã«ã¯é åºã«ä»»ææ§ãæãããæ¹ãè¯ãå ´é¢ã¨ããã®ã¯ããã ãããããããããããå°æ¥çã«ã¯ãæ¼ç®çµæã¯ä¸æã«æ±ºã¾ããªãããã©é«éãããæ¼ç®çµæã¯ä¸æã«æ±ºã¾ããã©ä½éããªæ¼ç®ã®ã©ã¡ãããé¸ã¶ã¨ããæ代ãæ¥ãã®ãããããªãããããã¯ããæ°å¤è¨ç®ã®çµæã¯ç¢ºççã«æ±ºã¾ãã¨ããã®ã常èã«ãªãã足ãç®ã®çããä¸æã«æ±ºã¾ããªãã¦21ä¸ç´åé ã¾ã§ã®ç§æçãªèãæ¹ã¨è¦ãªãããããã«ãªãã®ããã
-
-
- -
-
ããã¼ã¿ã®çµ±è¨çãªæ§è³ªã«ãã£ã¦ã¯è¶³ãç®çµæã®ã確çåå¸ããçµæ§å¤ãªå½¢ã«ãªããããªæ°ããããã©ããã®è¾ºã¯ã©ãæ±ãããããã«ãªããã ãããã10åè¨ç®ãã¦å¹³åãåãã¨ãã«ãªã£ããçµæ§ã·ã¥ã¼ã«ã ã
-
-
- -
-
ã並åæ¼ç®è£ ç½®ã®å¹çæ§ã¨é決å®è«çãªæ§è³ªã¯ä¸å¯åãªã®ã ï¼ããã¨ãã°è³ã¿ï¼ä»¥ä¸ç¥ï¼