[B! sse] dannã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

dann id:dann

sseã«é–¢ã™ã‚‹dannã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (25)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

HTTPãƒ‘ãƒ¼ã‚µã«ãŠã‘ã‚‹SSE4.2æœ€é©åŒ–ã®å¨åŠ›ã¨æ³¨æ„ç‚¹ - Cybozu Inside Out | ã‚µã‚¤ãƒœã‚¦ã‚ºã‚¨ãƒ³ã‚¸ãƒ‹ã‚¢ã®ãƒ–ãƒã‚°
ã“ã‚“ã«ã¡ã¯ã€ã‚µã‚¤ãƒœã‚¦ã‚ºãƒ»ãƒ©ãƒœã®å…‰æˆã§ã™ã€‚ PicoHTTPParserã¯@kazuhoã•ã‚“ãŸã¡ãŒé–‹ç™ºã—ã¦ã„ã‚‹é«˜é€ŸãªHTTPãƒ‘ãƒ¼ã‚µã§ã™ã€‚ åŒã˜ä½œè€…ã«ã‚ˆã‚‹HTTPã‚µãƒ¼ãƒH2Oã«ã‚‚ä½¿ã‚ã‚Œã¦ã„ã¾ã™ã€‚ 11æœˆ4æ—¥ã®é–‹ç™ºãƒ–ãƒã‚°ã«ã‚ˆã‚‹ã¨ã€ãã®æ™‚ç‚¹ã§Node.jsãªã©ã«ä½¿ã‚ã‚Œã¦ã„ã‚‹http-parserã®10å€ç¨‹åº¦ã®é€Ÿåº¦ã‚’èª‡ã‚‹ãã†ã§ã™ï¼ˆç¾åœ¨ã¯http-parserã‚‚é€Ÿåº¦å‘ä¸Šã—ãã®å·®ã¯ç¸®ã¾ã‚Šã¾ã—ãŸã€‚ãã‚Œã§ã‚‚4å€ä»¥ä¸Šã®å·®ãŒã‚ã‚‹ã‚ˆã†ã§ã™ï¼‰ã€‚ è©²å½“ãƒ–ãƒã‚°ã«ã¯ãã®é«˜é€ŸåŒ–ã®ãŸã‚ã®ãƒŽã‚¦ãƒã‚¦ãŒæ›¸ã‹ã‚Œã¦ã„ã¦å¤§å¤‰èˆˆå‘³æ·±ã„ã§ã™ã€‚ãŸã Intelç³»CPUã«æè¼‰ã•ã‚Œã¦ã„ã‚‹SIMDå‘½ä»¤ã¯ç”¨ã„ã‚‰ã‚Œã¦ã„ã¾ã›ã‚“ã§ã—ãŸã€‚ä»Šå›žã€@kazuhoã•ã‚“ã¨ä¸€ç·’ã«æ–‡å—åˆ—å‡¦ç†å°‚ç”¨ã®SSE4.2ã‚’ç”¨ã„ã‚‹ã“ã¨ã§1.7ï½ž1.9å€ã®é«˜é€ŸåŒ–ã‚’é”æˆã—ã¾ã—ãŸï¼ˆImproving Parser Performance using SSE Instructions (in
dann 2014/12/04
sse
ãƒªãƒ³ã‚¯
Improving Parser Performance using SSE Instructions (in case of PicoHTTPParser)
PCMPxSTRx is a SIMD instruction that can be used for parsing text. In _SIDD_CMP_RANGES mode, it checks at most 16 bytes at once, if each byte is within given set of ranges. Herumi and I have created a wrapper function for the instruction named findchar_fast that iterates though every 16 bytes of a given buffer to find the first occurrence of a byte within a set of given ranges. And the function is
dann 2014/12/02
ãƒªãƒ³ã‚¯
marisa-trie ã® PopCount ã‚’æ”¹è‰¯ã—ã¾ã—ãŸ - ã‚„ãŸï¼ ã¯ã¦ãªæ—¥è¨˜
ç°¡æ½”ãƒ“ãƒƒãƒˆãƒ™ã‚¯ãƒˆãƒ«ã®åŸºæœ¬æ“ä½œã§ã‚ã‚‹ rank/select ã«ãŠã„ã¦é‡è¦ãªå½¹å‰²ã‚’æžœãŸã™ PopCount ã‚’æ”¹è‰¯ã™ã‚‹ã“ã¨ã«ã‚ˆã‚Šï¼Œmarisa-trie ï¼ˆGoogle Code Archive - Long-term storage for Google Code Project Hosting.ï¼‰ ã‚’å°‘ã—ã ã‘é«˜é€ŸåŒ–ã™ã‚‹ã“ã¨ãŒã§ãã¾ã—ãŸï¼Žæˆæžœã¯ Subversion ã‹ã‚‰ã‚½ãƒ¼ã‚¹ã‚³ãƒ¼ãƒ‰ã‚’ãƒ€ã‚¦ãƒ³ãƒãƒ¼ãƒ‰ã—ã¦ãƒ“ãƒ«ãƒ‰ã™ã‚‹ã“ã¨ã«ã‚ˆã‚Šç¢ºèªã§ãã¾ã™ï¼Ž ä¸€ã¤ç›®ã®æ”¹è‰¯ç‚¹ã¯ SSE4.2 ã§è¿½åŠ ã•ã‚ŒãŸ popcnt å‘½ä»¤ã§ã™ï¼Žãã®åã®é€šã‚Šï¼ŒPopCount å°‚ç”¨ã® CPU å‘½ä»¤ã§ã‚ã‚Šï¼ŒPopCount ã‚’é«˜é€Ÿã«å®Ÿç¾ã§ãã¾ã™ï¼ŽãŸã ã—ï¼Œç¾çŠ¶ã§ã¯ä½¿ãˆã‚‹ CPU ãŒé™ã‚‰ã‚Œã¦ã„ã‚‹ãŸã‚ï¼Œconfigure ã« --enable-popcnt ã‚’æŒ‡å®šã—ã¦ãƒ“ãƒ«ãƒ‰ã—ãŸã¨ãã ã‘æœ‰åŠ¹ã«ãªã‚‹ã‚ˆã†ã«ã—ã¦ã„ã¾ã™ï¼Ž äºŒã¤ç›®ã®æ”¹è‰¯ç‚¹ã¯ pop
dann 2012/06/11
popcnt

gcc

sse
ãƒªãƒ³ã‚¯
SSEï¼’å‘½ä»¤ï¼ˆmovdquï¼‰ã‚’ä½¿ç”¨ã—ãŸï¼‘ï¼’ï¼˜ãƒ“ãƒƒãƒˆå˜ä½ã®ãƒ¡ãƒ¢ãƒªé–“è»¢é€ãƒ—ãƒã‚°ãƒ©ãƒ - ã¡ãƒ¼ã¡ãƒ¼ã®å°ãƒã‚¿éƒ¨å±‹
SSEï¼’ã§ã¯ã€ï¼‘ï¼’ï¼˜ãƒ“ãƒƒãƒˆæ•´æ•°æ¼”ç®—ã®SIMDå‘½ä»¤ã®æ‹¡å¼µãŒè¡Œã‚ã‚Œã¦ãŠã‚Šã€æ¼”ç®—ã ã‘ã§ãªãã€movdqaãƒ»movdquã¨ã„ã†å‘½ä»¤ã‚’ä½¿ã†ã“ã¨ã§ï¼‘ï¼’ï¼˜ãƒ“ãƒƒãƒˆå˜ä½ã§ãƒ¡ãƒ¢ãƒªé–“ã®ãƒ‡ãƒ¼ã‚¿ã‚³ãƒ”ãƒ¼ã‚’é«˜é€Ÿã«è¡Œã†ã“ã¨ãŒã§ãã¾ã™ã€‚movdqaã¯è»¢é€å¯¾è±¡ã®ãƒ¡ãƒ¢ãƒªã®ã‚¢ãƒ‰ãƒ¬ã‚¹ãŒï¼‘ï¼–ãƒã‚¤ãƒˆã®å¢ƒç•Œï¼ˆã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ ãƒˆï¼‰ã«æƒã£ã¦ã‚‹å¿…è¦ãŒã‚ã‚Šã¾ã™ï¼ˆã‚¢ãƒ‰ãƒ¬ã‚¹å€¤ãŒï¼‘ï¼–ã§å‰²ã‚Šåˆ‡ã‚Œã‚‹ï¼‰ã€‚ã“ã‚Œã«å¯¾ã—ã¦ã€movdquã§ã¯ï¼‘ï¼–ãƒã‚¤ãƒˆã®ã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ãƒˆåˆ¶ç´„ã¯ãªãã€ã©ã®ã‚¢ãƒ‰ãƒ¬ã‚¹å€¤ã‹ã‚‰ã§ã‚‚èªã¿æ›¸ãã§ãã¾ã™ã€‚ãŸã ã€ã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ãƒˆãŒæƒã£ã¦ã„ã‚‹åˆ†ã€CPUãŒå‡¦ç†ã—ã‚„ã™ã„ãŸã‚ã€movdqaå‘½ä»¤ã®æ–¹ãŒå®Ÿè¡Œé€Ÿåº¦ãŒé€Ÿã„ã§ã™ã€‚é€Ÿã•ã‚’è¿½æ±‚ã™ã‚‹ãªã‚‰ã°ã€mallocã‚„é…åˆ—ã®ç¢ºä¿æ™‚ã«ï¼‘ï¼–ãƒã‚¤ãƒˆã®ã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ãƒˆã‚’æ„è˜ã™ã‚‹å¿…è¦ãŒã‚ã‚Šã¾ã™ã€‚å„å‘½ä»¤ã®åŸºæœ¬çš„ãªå®Ÿè¡Œé€Ÿåº¦ã¯Intelã‹ã‚‰ä»¥ä¸‹ã®è³‡æ–™ãŒå‡ºã¦ã„ã¾ã™ãŒã€ã“ã®è³‡æ–™ä¸Šã§ã¯movdqaãƒ» movdquã®ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆã¨ãƒ¬ã‚¤ãƒ†ãƒ³ã‚·ã¯åŒã˜ã«ãªã£ã¦ã¾ã™ããƒ»
dann 2011/11/02
sse

simd
ãƒªãƒ³ã‚¯
æ¥“ software: x86 SIMD Technique ã‚¢ãƒ¼ã‚«ã‚¤ãƒ–
2008å¹´02æœˆ23æ—¥ ã¯ã˜ã‚ã« ã‚¤ãƒ³ãƒˆãƒªãƒ³ã‚·ãƒƒã‚¯ã§ MMX ã‚„ SSE ãŒä½¿ãˆã‚‹ã‚ˆã†ã«ãªã‚Šã€æ‰‹è»½ã« SIMD ãŒåˆ©ç”¨ã§ãã‚‹ã‚ˆã†ã«ãªã£ãŸã€‚ ã§ã‚‚ã€ã„ã–ä½¿ãŠã†ã¨ã™ã‚‹ã¨ã©ã†æ›¸ã‘ã°ã„ã„ã®ã‹ã‚ã‹ã‚‰ãªã„å ´é¢ã«å‡ºãã‚ã™ã“ã¨ãŒå¤šã„ã€‚ ã ã‹ã‚‰ã¨è¨€ã£ã¦ã€ä¸€åº¦ã°ã‚‰ã—ã¦æ™®é€šã«å‡¦ç†ã™ã‚‹ã¨é…ããªã£ã¦ã—ã¾ã†ã€‚ ã“ã“ã§ã¯ SIMD åŒ–ã™ã‚‹ã«å½“ãŸã£ã¦ç›´é¢ã—ãŸå•é¡Œã‚’è§£æ±ºã—ãŸæ–¹æ³•ã‚„ä»–ã®ã‚½ãƒ¼ã‚¹ã‚³ãƒ¼ãƒ‰ã‹ã‚‰å¦ã‚“ã ãƒ†ã‚¯ãƒ‹ãƒƒã‚¯ã€æ¤œç´¢ã—ã¦è¦‹ã¤ã‘ãŸã‚‚ã®ãªã©ã‚’ã¾ã¨ã‚ã¦è¡Œã“ã†ã¨æ€ã†ã€‚ æŠ•ç¨¿è€… Takenori : 20:48 | ãƒˆãƒ©ãƒƒã‚¯ãƒãƒƒã‚¯ æ¯”è¼ƒ SIMD ã®æ¯”è¼ƒã¯ã€å˜ã«ãƒžã‚¹ã‚¯ã‚’ç”Ÿæˆã—ã¦ãã‚Œã‚‹ã ã‘ã ã€‚ æ¯”è¼ƒã—ã¦åˆ†å²ã™ã‚‹ãªã©ã¯å‡ºæ¥ãªã„ã€‚ ç”Ÿæˆã•ã‚ŒãŸãƒžã‚¹ã‚¯ã‚’ä½¿ã£ã¦ä½•ã¨ã‹ã™ã‚‹ã—ã‹ãªã„ã€‚ ã“ã‚Œã¯ã„ã¤ã‚‚åˆ†å²ã§æ›¸ã„ã¦ã„ã‚‹ã¨ã€ã©ã†æ›¸ã‘ã°ã„ã„ã®ã‹ã•ã£ã±ã‚Šã‚ã‹ã‚‰ãªã„å•é¡Œã ã€‚ ã§ã‚‚ã€ã“ã‚ŒãŒä½¿ãˆã‚‹ã¨åˆ†å²ã‚’æ¶ˆã—ã¦ä¸¦åˆ—ã§å‡¦ç†ã§ãã‚‹ã‚ˆã†ã«ãªã‚‹ã€‚ æ¯”è¼ƒçš„ã‚ã‹ã‚Šã‚„ã™ã„ã§
dann 2011/11/02
simd

sse
ãƒªãƒ³ã‚¯
æ±äº¬å·¥æ¥å¤§å¦ æƒ…å ±ç†å·¥å¦é™¢ æ•°ç†ãƒ»è¨ˆç®—ç§‘å¦ç³»
dann 2011/11/01
simd

sse
ãƒªãƒ³ã‚¯
[PDF] ã„ã¾ã©ãã˜ã‚ƒãªã„ ã‚¢ã‚»ãƒ³ãƒ–ãƒ©å…¥é–€ / ä¹…ä¿ç”°å±•è¡Œ
ã„ã¾ã©ãã˜ã‚ƒãªã„ ã‚¢ã‚»ãƒ³ãƒ–ãƒ©ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚° x86ã‚¢ã‚»ãƒ³ãƒ–ãƒªè¨€èªžã®åŸºç¤Žã‹ã‚‰SSEã¾ã§ 1 è‡ªå·±ç´¹ä»‹ ï‚— ä¹…ä¿ç”°å±•è¡Œ â—¦ @nobu_k, id:nobu-q ï‚— æ¤œç´¢ã‚¨ãƒ³ã‚¸ãƒ³Sedueã‚’ä½œã£ã¦ã¾ã™ 2 æœ¬æ—¥ã®å†…å®¹ ï‚— x86ã‚¢ã‚»ãƒ³ãƒ–ãƒªè¨€èªžã®åŸºç¤Ž ï‚— ã‚¤ãƒ³ãƒ©ã‚¤ãƒ³ã‚¢ã‚»ãƒ³ãƒ–ãƒ©ã§éŠã¶ ï‚— SSEã‚’ä½¿ã£ã¦ã¿ã‚‹ 3 ã‚¢ã‚»ãƒ³ãƒ–ãƒªè¨€èªžã¨ã¯ ï‚— æ©Ÿæ¢°èªžã«è¿‘ã„ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°è¨€èªž â—¦ æ©Ÿæ¢°èªžã¨ä¸€å¯¾ä¸€ã§å¯¾å¿œ 4 if (x < 0) { x = -x; } mov eax, [ebp + 8] cmp eax, 0 jge L1 neg eax L1: 8B 45 08 3D 00 00 00 00 7D 02 F7 D8 Cè¨€èªž ã‚¢ã‚»ãƒ³ãƒ–ãƒªè¨€èªž æ©Ÿæ¢°èªžèªã‚€ã®ã¯åŽ³ã—ã„ ã¾ã ãƒžã‚· ã‚¢ã‚»ãƒ³ãƒ–ãƒªè¨€èªžã®ç‰¹å¾´ ï‚— ç§»æ¤æ€§ãŒä½Žã„ â—¦ CPUã‚„å‡¦ç†ç³»ã«ã‚ˆã£ã¦ã™ã¹ã¦ãŒå¤‰ã‚ã‚‹ ï‚— èªã¿æ›¸ããŒå¤§å¤‰ â—¦ å¯èªæ€§ãŒã‚‚ã®ã™ã”ãä½Žã„ ï‚—
dann 2011/10/28
gcc

sse
ãƒªãƒ³ã‚¯
x86ã‚„x64ã«ãŠã‘ã‚‹æœ€é©åŒ–ã¾ã¨ã‚ - ãšmemo
dann 2011/10/08
sse
ãƒªãƒ³ã‚¯
SIMDæ¼”ç®— - MUGI COM
Cè¨€èªžã§SSEã‚„SSE2ã‚’ä½¿ã†æ–¹æ³•ã«ã¤ã„ã¦ã€‚ãƒã‚¤ãƒ³ãƒˆãŒã„ãã¤ã‹ã‚ã‚Šã¾ã™ã€‚ xmmintrin.hã‚„emmintrin.hã‚’ã‚¤ãƒ³ã‚¯ãƒ«ãƒ¼ãƒ‰ã™ã‚‹ SSEå‘½ä»¤ã®ã¿ãªã‚‰xmmintrin.hã€SSE2å‘½ä»¤ã‚‚ä½¿ã†ãªã‚‰emmintrin.h SIMDå‘½ä»¤ã§ã‚¢ã‚¯ã‚»ã‚¹ã™ã‚‹ãƒ¡ãƒ¢ãƒªã¯16ãƒã‚¤ãƒˆã‚¢ãƒ©ã‚¤ãƒ³ãƒ¡ãƒ³ãƒˆã«ã™ã‚‹ gccã®å ´åˆã€_mm_mallocé–¢æ•°ã‚’ä½¿ãˆã°ã‚ˆã„ã€‚ MMXå‘½ä»¤ã¯AMDã®CPUã§ã¯ä½¿ãˆãªã„ Pentium4ä»¥é™ã«é™å®šã•ã‚Œã¾ã™ãŒã€SSEã‚„SSE2ã‚’ä½¿ã†ã®ãŒè‰¯ã„ã§ã—ã‚‡ã†ã€‚ ã“ã‚Œã‚‰ã‚’è¸ã¾ãˆã¦ã€ç”»åƒã®å„ãƒ”ã‚¯ã‚»ãƒ«ã®RGBå€¤ã‚’åè»¢ã•ã›ã‚‹ä¾‹ã‚’ç¤ºã—ã¾ã™ã€‚ å„ãƒ”ã‚¯ã‚»ãƒ«ã¯8bitã®RGBå€¤ãŒä¸¦ã‚“ã§ã„ã‚‹ã‚‚ã®ã¨ã—ã¾ã™ã€‚8bitã®æ•´æ•°æ¼”ç®—ã‚’è¡Œã†ã®ã§ã€8bitè¨ˆç®—ã‚’16å€‹ä¸¦åˆ—ã§è¡Œã†SSE2å‘½ä»¤ã‚’ä½¿ç”¨ã—ã¾ã™ã€‚ æœ€åˆã¯SSE2ã®ã‚¤ãƒ³ã‚¯ãƒ«ãƒ¼ãƒ‰ãƒ•ã‚¡ã‚¤ãƒ«ã€‚ #include <emmintrin.h> ã¤ã¥ã„ã¦16ãƒã‚¤ãƒˆã‚¢ãƒ©ã‚¤ãƒ³ãƒ¡ãƒ³
dann 2011/10/08
sse
ãƒªãƒ³ã‚¯
é«˜é€Ÿãªå€ç²¾åº¦æŒ‡æ•°é–¢æ•°expã®å®Ÿè£…
4. æœ€çµ‚å®Ÿè£…ã‚³ãƒ¼ãƒ‰ ï®ä»Šå›žã®ç›®æ¨™ã¯ã“ã®ã‚³ãƒ¼ãƒ‰ã®æ„å‘³ã‚’ç†è§£ã™ã‚‹ã“ã¨ 2011/8/6 /204 union di { uint64_t i; double d; }; double expd(double x) { di di; di.d = x * 2954.639443740597 + 6755399441055744ULL; uint64_t iax = tbl[di.i & 2047]; double t = (di.d - 6755399441055744ULL) * 0.0003384507717577858 - x; uint64_t u = ((di.i + 2095104) >> 11) << 52; double y = (3.0000000027955394 - t) * (t * t) * 0.16666666685227835 - t + 1; di.i = u
dann 2011/10/02
sse

x86-64

cpu
ãƒªãƒ³ã‚¯
SSE4.2ã®æ–‡å—åˆ—å‡¦ç†å‘½ä»¤ã®ç´¹ä»‹
1. SSE4.2ã®æ–‡å—åˆ—å‡¦ç†å‘½ä»¤ã®ç´¹ä»‹ Cybozu Labs 2011/8/6 å…‰æˆæ»‹ç”Ÿï¼ˆ8/23åŠ ç†ä¿®æ£p3, p.25ï¼‰ x86/x64æœ€é©åŒ–å‹‰å¼·ä¼š#1 2011/8/6 /41 1 3. èª¬æ˜Žã¨ã‚³ãƒ¼ãƒ‰ ï®Intelãƒ—ãƒã‚»ãƒƒã‚µæœ€é©åŒ–ãƒžãƒ‹ãƒ¥ã‚¢ãƒ«ã‚’èªã‚‚ã† ï®http://homepage1.nifty.com/herumi/prog/intel-opt.html ï®ã‚³ãƒ¼ãƒ‰ç‰‡ ï®https://github.com/herumi/opti/ ï®ã‚¢ãƒ©ã‚¤ãƒ¡ãƒ³ãƒˆã¨ãƒšãƒ¼ã‚¸å¢ƒç•Œã«é–¢ã™ã‚‹è£œè¶³ï¼ˆå¿…èªï¼‰ ï®http://homepage1.nifty.com/herumi/diary/1108.html#8 2011/8/6 3 /41 4. SIMDå‘ã/ä¸å‘ã ï®SIMDã¯4å€‹ï¼ˆor 8/16ï¼‰å€‹ã®ãƒ‡ãƒ¼ã‚¿ã‚’åŒæ™‚ã«åŒã˜ã‚ˆã† ã«å‡¦ç†ã™ã‚‹ãŸã‚ã®ã‚‚ã® ï®SIMDå‘ã ï®å…ˆç¨‹ã®æœ€å¤§å€¤ã‚’æ±‚ã‚ã‚‹å ´åˆï¼Œ4å€‹ãšã¤ã‚„ã£
dann 2011/08/07
sse
ãƒªãƒ³ã‚¯
SSE4.2ã®æ–‡å—åˆ—å‡¦ç†å‘½ä»¤ã®ç´¹ä»‹ã«é–¢ã™ã‚‹è£œè¶³ï¼ˆã‚¢ã‚¤ãƒ©ãƒ³ã¨ãƒšãƒ¼ã‚¸å¢ƒç•Œï¼‰
x86/x64æœ€é©åŒ–å‹‰å¼·ä¼š1 ãªã‚“ã¨ã‹ç„¡äº‹çµ‚äº†. ustãŒã†ã¾ãã„ã£ãŸã‚Šã„ã‹ãªã‹ã£ãŸã‚Š, é‹å–¶ãŒæ‰‹é–“å–ã£ãŸã‚Šã¨ç”³ã—è¨³ãªã„. ä¸»å‚¬ã™ã‚‹ã¨ã„ã†ã®ã¯ãŸã„ã¸ã‚“ã ãªã‚. åå‰ã ã‘ã¯webãªã‚Šã§ã‚ˆãè¦‹ã‹ã‘ã¦ã„ãŸã‘ã©ä¼šã£ãŸã“ã¨ãªã‹ã£ãŸæ–¹ã€…ã«ä¼šãˆãŸã®ã§æº€è¶³. ã§ã‚‚ã“ã‚Œã¾ãŸåˆã‚ã¦ãŠä¼šã„ã—ãŸ@takehiro_tã•ã‚“ã¨ã¯æ®†ã©ã—ã‚ƒã¹ã‚Œãªã‹ã£ãŸ. æ®‹å¿µ. å€‹äººçš„ã«ã¯w_oã•ã‚“ã®ãƒ™ãƒ³ãƒãƒžãƒ¼ã‚¯ã®çµæžœã‹ã‚‰ç†ç”±ã‚’æŽ¢ã—ã¦ã„ãéƒ¨åˆ†ãŒèˆˆå‘³æ·±ã‹ã£ãŸ. åçœç‚¹ : ç™ºè¡¨ã®é–“ã¯5åˆ†ãƒžãƒ¼ã‚¸ãƒ³ã‚’å…¥ã‚Œã¦ãŠã. ã‚¿ã‚¤ãƒžãƒ¼ã‚ã‚‹ã¨ã‚ˆã•ã’. ã¨ã‚Šã‚ãˆãšã„ãã¤ã‹ã®è³‡æ–™ã¸ã®ãƒªãƒ³ã‚¯(uploadã•ã‚Œã‚Œã°éšæ™‚æ›´æ–°). herumi : æ¡ä»¶åˆ†å²ã¨cmovã¨maxps m_asama : IA32/Intel64ã«ãŠã‘ã‚‹ã‚ãƒ£ãƒƒã‚·ãƒ¥åˆ©ç”¨æœ€é©åŒ– sinya8282 : é–‹ç™ºä¸ã®JITç‰ˆgrepã®è‹¦åŠ´è©± TAKESAKO : ãƒ“ãƒƒãƒˆã‚’æ•°ãˆã‚‹ herumi
dann 2011/08/07
sse

x64

cpu
ãƒªãƒ³ã‚¯
advanced optimization SSE
floatåž‹æµ®å‹•å°æ•°ã‚’4ã¤ä¸¦åˆ—ã«æ¼”ç®—ã™ã‚‹æ©Ÿæ§‹ã§ã™. MMXã‚„3D Now!ã®æ§˜ã«FPUãƒ¬ã‚¸ã‚¹ã‚¿ã‚’åˆ©ç”¨ã—ã¦ã„ã‚‹ã®ã§ã¯ãªãå…¨ãæ–°è¦ã«ä½œã‚‰ã‚ŒãŸãƒ¬ã‚¸ã‚¹ã‚¿ã§ã™. ãã®ãŸã‚, MMXã‚„3D Now!ã¨é•ã£ã¦FPUãƒ¬ã‚¸ã‚¹ã‚¿ã¨ã®åˆ‡æ›¿ãˆ[emms, femms]ã¯ä¸è¦ã§ã™ãŒ, ä»£ã‚ã‚Šã«OSãŒSSEãƒ¬ã‚¸ã‚¹ã‚¿ã‚’ã‚µãƒãƒ¼ãƒˆã—ã¦ã„ã‚‹å¿…è¦ãŒã‚ã‚Šã¾ã™. åŸºæœ¬çš„ã«æœ€ä¸‹ä½ãƒ‡ãƒ¼ã‚¿ã®ã¿ã‚’æ“ä½œã™ã‚‹å‘½ä»¤ã«ã¯ss, 4ã¤ã®(ãƒ‘ãƒƒã‚¯ã•ã‚ŒãŸ)ãƒ‡ãƒ¼ã‚¿ã‚’æ“ä½œã™ã‚‹å‘½ä»¤ã«ã¯psãŒã¤ãã¾ã™. å¤§æŠµã®å‘½ä»¤ã¯ssã‚ˆã‚Šã¯psã®æ–¹ãŒãƒ¬ã‚¤ãƒ†ãƒ³ã‚·, ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆã¨ã‚‚ã«1ã‚¯ãƒãƒƒã‚¯ãšã¤ä½™è¨ˆã«ã‹ã‹ã‚Šã¾ã™. ä»¥ä¸‹, è¡¨ã®ä¸ã®`F'ã¯floatåž‹ãƒ‡ãƒ¼ã‚¿1ã¤ã‚’è¡¨ã—ã¾ã™. `a-bã‚µã‚¤ã‚¯ãƒ«'ã¯ãƒ¬ã‚¤ãƒ†ãƒ³ã‚·a, ã‚¹ãƒ«ãƒ¼ãƒ—ãƒƒãƒˆbã‚’è¡¨ã—ã¾ã™. xmm?ã¨æ›¸ãã®ã¯é¢å€’ãªã®ã§xm?ã¨æ›¸ãã¾ã™. 4.1 èªã¿è¾¼ã¿ / æ›¸ãè¾¼ã¿
dann 2011/08/07
sse
ãƒªãƒ³ã‚¯
ãƒ›ãƒ¼ãƒ ï¼ˆçŽ„ç•Œï¼‰ | RESEARCH INSTITUTE FOR INFORMATION TECHNOLOGY, KYUSHU UNIVERSITY
dann 2011/08/07
gcc

sse
ãƒªãƒ³ã‚¯
SIMD and other techniques for fast numerical programming with gcc
SIMD and other techniques for fast numerical programming with gcc
dann 2011/08/07
sse

gcc
ãƒªãƒ³ã‚¯
CDWFS: Optimizing for SSE: A Case Study
Optimizing for SSE: A Case Study Originally appeared in Hugi Issue #25 Table of Contents Introduction Techniques Naive C++ Basic SSE Batch Processing 16-byte Memory Alignment Instruction Pairing Prefetching Improving Locality Application-Specific Specializations Conclusion and Future Possibilities Download Sample Code Links and Resources Revision History Introduction In this article I'll be descri
dann 2011/08/07
sse
ãƒªãƒ³ã‚¯
Using SSE to speed up computation - store, load and alignment
dann 2011/08/07
sse
ãƒªãƒ³ã‚¯
SSE intrinsics optimizations in popular compilers â€“
dann 2011/08/07
sse

gcc

icc
ãƒªãƒ³ã‚¯
SSE2ã§ãƒžãƒ³ãƒ‡ãƒ«ãƒ–ãƒé›†åˆ - merom686's blog
ä¸‹ã®ã‚³ãƒ¼ãƒ‰ã¯ã€SSE2ã§ãƒžãƒ³ãƒ‡ãƒ«ãƒ–ãƒé›†åˆã®è¨ˆç®—ã‚’ã™ã‚‹ã‚‚ã®ã€‚ã‹ãªã‚Šé«˜é€Ÿã«å‹•ä½œã™ã‚‹ã¨æ€ã†ã€‚ ã“ã®ã‚¨ãƒ³ãƒˆãƒªã§ã¯ã€ã“ã®ã‚³ãƒ¼ãƒ‰ã®è§£èª¬ã‚’ã™ã‚‹ã«ã‚‡ã€‚ ï¼ƒã†ã‚‹ã‚Šã®ç‰©ç½®ãã®Mandel100.zipã«ã‚ã‚‹SSEã®ã‚³ãƒ¼ãƒ‰ãŒå…ƒã«ãªã£ã¦ã„ã‚‹ã€‚ ; MASM 8.0 ã‚’ä½¿ç”¨ .686 .xmm .model flat, c .data align 16 D4 real8 4.0, 4.0 L1 oword -1 .code mandel_sse2 proc x:ptr real8, y:ptr real8, m:dword, count:ptr dword mov eax, x movapd xmm0, [eax] mov eax, y movapd xmm1, [eax] mov ecx, m movapd xmm6, xmm0 movapd xmm7, xmm1 pxor xmm5, xmm5 jmp brot
dann 2011/08/07
sse
ãƒªãƒ³ã‚¯
SSE Instruction Set
dann 2011/08/07
sse

intel
ãƒªãƒ³ã‚¯
1 2 æ¬¡ã®ãƒšãƒ¼ã‚¸