Á¤º¸°úÇÐȸ ³í¹®Áö A : ½Ã½ºÅÛ ¹× ÀÌ·Ð
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
´ÙÁß½º·¹µù ½ÇÇà ¹æ½Ä¿¡ ÀÇÇÑ Id ÇÔ¼ö ¾ð¾îÀÇ ¹Ýº¹ ±¸¹® ½ÇÇà ±â¹ý ºñ±³ |
¿µ¹®Á¦¸ñ(English Title) |
Comparison of the Loop Implementation of Id by Multithreading |
ÀúÀÚ(Author) |
ÇÏ»óÈ£
±èÈïÇÑ
ÇÑ»ó¿µ
Snagho Ha
Heunghwan Kim
Sangyong Han
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 23 NO. 05 PP. 0519 ~ 0528 (1996. 05) |
Çѱ۳»¿ë (Korean Abstract) |
´ÙÁß½º·¹µù(multithreading)Àº ±ä ¸Þ¸ð¸® Áöü ½Ã°£°ú µ¿±âÈÀÇ ¹®Á¦Á¡À» ÇØ°áÇÒ ¼ö ÀÖ´Ù´Â Á¡¿¡¼ ´ë±Ô¸ð º´·Ä ±â°è ±¸Á¶¿¡ ¸Å¿ì ÀûÇÕÇÏ´Ù. ÇÔ¼ö ¾ð¾î Id¸¦ ´ÙÁß½º·¹µù ¹æ½ÄÀ¸·Î ½ÇÇà½ÃÅ°±â À§ÇÑ ¿©·¯ ¿¬±¸°¡ ÀÌ·ç¾îÁ® ¿ÔÀ¸³ª, ¹Ýº¹ ±¸¹®¿¡ °üÇÑ ¿¬±¸´Â ¾ÆÁ÷ ¹ÌÈíÇÏ´Ù. º» ³í¹®¿¡¼´Â ´ÙÁß½º·¹µå ±¸Á¶»ó¿¡¼ ¹Ýº¹ ±¸¹®ÀÇ ¼¼ °¡Áö ½ÇÇà ¹æ½Ä, Áï Àç±Í ÇÔ¼ö, K-Æîħ ·çÇÁ, ·çÇÁ ¼øÂ÷ȸ¦ ºñ±³ÇÑ´Ù. ƯÈ÷, ³í¹®¿¡¼ Á¦¾ÈµÈ ·çÇÁ ¼øÂ÷È ¹æ½Ä¿¡¼´Â ´Ù¸¥ µÎ ¹æ½Ä°ú´Â ´Þ¸®, ·çÇÁÀÇ ¸ðµç ¹Ýº¹À» ¼øÂ÷ÀûÀ¸·Î ½ÇÇà½ÃŲ´Ù. ½Ã¹Ä·¹À̼ÇÀÇ ºÐ¼® °á°ú´Â º´·Ä¼ºÀÌ Àû°í, º¹ÀâÇÑ ÀÚ·á Á¾¼ÓÀ» °®´Â ·çÇÁ¿¡ ´ëÇؼ ·çÇÁ ¼øÂ÷ÈÀÇ ¹æ½ÄÀÌ ´Ù¸¥ µÎ ¹æ½Äº¸´Ù ¸Å¿ì È¿°úÀûÀÓÀ» º¸¿©ÁØ´Ù. |
¿µ¹®³»¿ë (English Abstract) |
Multithreading is attractive for massively parallel architectures in that it can tolerate long memory latencies and synchronization overheads. Although several methods to execute Id programs by multithreading have been studied, the efficient implementation method of its iteration constructs has been nearly ignored. In this paper, we compare three schemes of loop execution : recursion, k-unfolding loop, and loop sequentialization. Especially, loop sequentialization presented in this paper executes all iterations sequentially unlike other this schemes that unfold loops. The simulation analysis shows that loop sequentialization is very effective for loops with small computation and complicated data dependencies in their bodies. |
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|