Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
ÄÚµå »ðÀÔ µµ±¸ È°¿ë GPU ¼º´É ºÐ¼® µµ±¸ ±¸Çö |
¿µ¹®Á¦¸ñ(English Title) |
Implementing GPU Performance Analysis Tool Using Code Instrumentation Tool |
ÀúÀÚ(Author) |
±èÇöÁØ
È«¼ºÀÎ
ÇÑȯ¼ö
Hyunjun Kim
Sungin Hong
Hwansoo Han
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 24 NO. 07 PP. 0352 ~ 0357 (2018. 07) |
Çѱ۳»¿ë (Korean Abstract) |
´ë·®ÀÇ ¾²·¹µå°¡ º´·Ä·Î µ¿ÀÛÇÏ´Â GPU¿¡¼ ÃÖÀûÈµÈ Äڵ带 ÀÛ¼ºÇÏ´Â °ÍÀº ½±Áö ¾Ê´Ù. º» ³í¹®¿¡¼´Â ¼º´É ºÐ¼®À» À§ÇØ GPU¸¦ ´ë»óÀ¸·Î ÄÚµå »ðÀÔÀÌ °¡´ÉÇÑ SASSI¸¦ ÀÌ¿ëÇÏ¿© ÀûÀº ¿À¹öÇìµå·Î Ä¿³Î ÇÔ¼öÀÇ ¼º´É Á¤º¸¸¦ µ¿ÀûÀ¸·Î ÃßÃâÇÏ´Â µµ±¸¸¦ ±¸ÇöÇÏ¿´´Ù. ¶ÇÇÑ, ÃßÃâÇÑ GPU ¼º´É Á¤º¸¸¦ ºÐ¼®ÇÏ¿© ¼º´É ºÐ¼® °¡´ÉÇÑ µµ±¸¸¦ °³¹ßÇÏ¿´´Ù. µµ¸ÞÀÎ-ƯÁ¤ °³¹ßÀڵ鵵 °³¹ßÇÑ ¼º´É ºÐ¼® µµ±¸¸¦ È°¿ëÇÏ¿© ¼º´ÉÃÖÀûÈµÈ GPU Ä¿³Î ÇÔ¼ö¸¦ ÀÛ¼ºÇϴµ¥ µµ¿òÀÌ µÉ °ÍÀ̶ó ±â´ëÇÑ´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Domain-specific programmers find it difficult to understand and apply optimizations for programs in which massive threads run in parallel. In this paper, we implemented a tool to dynamically generate the profiling information of GPU kernels with less overhead. We used SASSI to instrument a profiling code into the GPU kernels to generate a performance trace. We also developed a profiling tool to analyze the extracted traces to investigate the performance overhead. We expect domainspecific developers to benefit from the performance-optimized GPU kernel using the performance analysis tools.
|
Å°¿öµå(Keyword) |
¼º´É ºÐ¼® µµ±¸
ÄÚµå »ðÀÔ µµ±¸
GPU ¸Þ¸ð¸®
¾²·¹µå ½ºÄÉÁÙ¸µ
performance analysis tool
code instrumentation tool
GPU memory
thread scheduling
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|