CUDAãåå¼·ãããã¨ãã¦ï¼ã¾ãã¤ã¾ããã®ãï¼
ThreadãBlock, Gridãªã©ã®ã½ããã¦ã§ã¢ä¸ã®æ¦å¿µã¨ï¼
Streaming Multiprocessor(SM)ï¼CUDA Coreï¼Warp, Occupancyãªã©ã®ãã¼ãã¦ã§ã¢ã®æ¦å¿µããã£ã¡ãã«ãªãç¹ã ï¼
ä»åã¯ããã¦ç解ã§ããï¼æ°ãããï¼ã®ã§ï¼å¿ããªããã¡ã«æ¸ãæ®ãã¦ããï¼
ã½ããçï¼ã¨æãããï¼æ¦å¿µ
- Thread
- ããã¤ã¹ä¸ã®ããã°ã©ã ãåãã¨ãã®æå°åä½
- éåæã«åã
- Block
- Threadãã¾ã¨ãããã®
- 3次å 表ç¾
- Grid
- Blockãã¾ã¨ãããã®
ãã¼ãçï¼ã¨æãããï¼æ¦å¿µ
- CUDA Core
- SM
- ä¸ã®Blockã¯ãã¼ãã§ã¯SMã«ããã
- 1ã¤ã®SMã§å®è¡ãããBlockã¯1ã¤
- 1ã¤ã®SMã«è¤æ°ã®Blockãåæã«åå¨ã§ãã
- Registerã¯å ¨Threadã§åå²
- Shared memoryã¯Blockã§åå²
- Warp
- ä¸ã§SMãã¨ã¨æ¸ãããï¼ãã詳細ã«ã¯ï¼ãã®Warpã®ã¹ã¬ãããã¨ã«åä½ãã
- Occupancy
å®éã«æ¸ãã¦ã¿ããï¼ããã§ã¯CUDAã®å
¥éæ¸ã®å
容ã¨åãã§ï¼
åå¦è
ã«ã¯çµ¶å¯¾ã«ç解ã§ããªãï¼
ãã®åç»ã«ã§ã¦ãããã®å³ãããããããã£ã
GPGPUと流体計算(@ponkotuy) @ 関東GPGPU勉強会 - YouTube