[B! SIMD] tekimenã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

tekimen id:tekimen

SIMDã«é–¢ã™ã‚‹tekimenã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (5)

${{author_name}}$

{{{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}}

x86-64æ©Ÿæ¢°èªžå…¥é–€ AVX/AVX-512ç·¨
ã“ã®è¨˜äº‹ã¯x86-64æ©Ÿæ¢°èªžå…¥é–€ã®ç¶šãã§ã™ã€‚SSE/AVX/AVX-512ã®ç´¹ä»‹ã¨ã€ãã‚Œã«ä¼´ã†VEXãƒ—ãƒªãƒ•ã‚£ãƒƒã‚¯ã‚¹ã€EVEXãƒ—ãƒªãƒ•ã‚£ãƒƒã‚¯ã‚¹ã®ç´¹ä»‹ã‚’è¡Œã„ã¾ã™ã€‚ SSE x86-64ã§å˜ç²¾åº¦ãƒ»å€ç²¾åº¦ã®æµ®å‹•å°æ•°ç‚¹æ•°ã‚’æ‰±ã†ã«ã¯ã€SSE (Streaming SIMD Extensions) ã¨å‘¼ã°ã‚Œã‚‹æ©Ÿèƒ½ï¼ˆå‘½ä»¤ã‚»ãƒƒãƒˆã¨ãƒ¬ã‚¸ã‚¹ã‚¿ãƒ¼ï¼‰ã‚’ä½¿ã„ã¾ã™ã€‚å…ƒã€…ã¯å‘½ä»¤ã‚»ãƒƒãƒˆæ‹¡å¼µã§ã—ãŸãŒã€x86-64ã§ã¯SSE2ãŒå¸¸ã«åˆ©ç”¨å¯èƒ½ãªã®ã§ã€æ¨™æº–ã§ä½¿ãˆã‚‹æ©Ÿèƒ½ã ã¨æ€ã£ã¦æ§‹ã„ã¾ã›ã‚“ã€‚ SSEã§ã¯128ãƒ“ãƒƒãƒˆå¹…ã®ãƒ¬ã‚¸ã‚¹ã‚¿ãƒ¼ã‚’16æœ¬ï¼ˆ32ãƒ“ãƒƒãƒˆãƒ¢ãƒ¼ãƒ‰ã§ã¯8æœ¬ï¼‰åˆ©ç”¨ã§ãã¾ã™ã€‚ãƒ‡ãƒ¼ã‚¿åž‹ã¯ç„¡å°SSEã§ã¯å˜ç²¾åº¦æµ®å‹•å°æ•°ç‚¹æ•°ï¼ˆ32ãƒ“ãƒƒãƒˆå¹…ï¼‰ãŒã€SSE2ä»¥é™ã§ã¯å€ç²¾åº¦æµ®å‹•å°æ•°ç‚¹æ•°ï¼ˆ64ãƒ“ãƒƒãƒˆå¹…ï¼‰ã‚„æ•´æ•°ãŒä½¿ãˆã¾ã™ã€‚ SSEã®128ãƒ“ãƒƒãƒˆå¹…ã®ãƒ¬ã‚¸ã‚¹ã‚¿ãƒ¼ã¯XMMãƒ¬ã‚¸ã‚¹ã‚¿ãƒ¼ã¨å‘¼ã°ã‚Œã€xmm0, xmm1, ..., xmm15ã®ã‚ˆã†ã«ç•ªå·ã‚’ã¤ã‘ã¦å‘¼ã°
tekimen 2024/01/03
SIMD

CPU

AVX
ãƒªãƒ³ã‚¯
SIMDã®åŸºç¤Žã¨é–¢æ•°ã®å®Ÿè£…ä¾‹
é…ä¿¡è¬›ç¾©ã€€è¨ˆç®—ç§‘å¦æŠ€è¡“ç‰¹è«–Aï¼ˆ2023ï¼‰ï¼ˆ2023å¹´4æœˆ13æ—¥ï½ž7æœˆ27æ—¥ãƒ»ã‚ªãƒ³ãƒ©ã‚¤ãƒ³ï¼‰ ç¬¬11å›ž 6æœˆ29æ—¥ SIMDã®åŸºç¤Žã¨é–¢æ•°ã®å®Ÿè£…ä¾‹ https://www.r-ccs.riken.jp/outreach/schools/20230413-1/
tekimen 2023/09/20
SIMD

programming
ãƒªãƒ³ã‚¯
Armã®å‘½ä»¤ã‚»ãƒƒãƒˆã¨NEONæ‹¡å¼µ - Qiita
ã“ã®ã‚¢ãƒ‰ãƒ™ãƒ³ãƒˆã‚«ãƒ¬ãƒ³ãƒ€ãƒ¼ã®ã‚¹ã‚³ãƒ¼ãƒ—ã€å¯¾è±¡èªè€…ã«ã¤ã„ã¦ ã“ã®è¨˜äº‹ã¯ã²ã¨ã‚ŠNEONã‚¢ãƒ‰ãƒ™ãƒ³ãƒˆã‚«ãƒ¬ãƒ³ãƒ€ãƒ¼2020 1æ—¥ç›®ã®è¨˜äº‹ã§ã™ã€‚ ç›®æ¬¡ã«ã‚‚æ›¸ãã¾ã—ãŸãŒã€ã‚ãã¾ã§æ‰‹å³¶å€‹äººã®çŸ¥è˜ã®ç¯„å›²å†…ã§æ›¸ãã®ã§ã€ã™ã¹ã¦å…¬é–‹æƒ…å ±ãŒã‚‚ã¨ã«ãªã£ã¦ã„ã¾ã™ã€‚ å€‹äººã§è¶£å‘³ã®ç¯„å›²ã§æ›¸ãã®ã§ã€ç†è€…ã®å‹¤å‹™å…ˆã¨ã¯ç„¡é–¢ä¿‚ã§ã™ ä¸€äººadvent calendarãªã®ã§ã€çœã‚¨ãƒãƒ¢ãƒ¼ãƒ‰ãŒãƒ¢ãƒƒãƒˆãƒ¼ã«ãªã£ã¦ã¾ã™ã€‚æ‚ªã—ã‹ã‚‰ãšã€‚ Arm v8.4ã€SVEã¨è¡Œã£ãŸArmã®ç¾åœ¨é€²è¡Œå½¢ã®æ‹¡å¼µã«é–¢ã—ã¦ã¯å®Ÿæ©ŸãŒæ‰‹å…ƒã«ç„¡ã„ã“ã¨ã‚‚ã‚ã£ã¦ã€ã‚ã¾ã‚Šæ·±å…¥ã‚Šã¯ã—ãªã„äºˆå®š å®Ÿéš›ã®ã‚³ãƒ¼ãƒ‰ã¯ã‚¢ã‚»ãƒ³ãƒ–ãƒ©ã§ã¯ãªãã€GCCã®intrinsicã‚’ã‚‚ã¨ã«è§£èª¬ã™ã‚‹ C/C++ã®ã‚³ãƒ¼ãƒ‰ã‹ã‚‰NEONå‘½ä»¤ã‚’ä½¿ã†ã‚³ãƒ¼ãƒ‰ã‚’è§£èª¬ã™ã‚‹ã®ã§ã€ãã‚ŒãŒèªã‚ã‚‹ãã‚‰ã„ã®èªè€…ã‚’æƒ³å®šã€‚ æœ¬æ—¥ã®ãƒˆãƒ”ãƒƒã‚¯ ã„ããªã‚Šã€ŒNEONã¨ã¯ã€ã€ã¨é£›ã³è¾¼ã‚“ã§ã‚‚ã‚ã‹ã‚‰ãªã„äººãŒå±…ã‚‹ã¨æ€ã†ã®ã§ã€ç¬¬1æ—¥ç›®ã¯æ¦‚è¦ã‚’èª¬æ˜Žã—ã‚ˆã†ã¨æ€ã„ã¾ã™ã€‚
tekimen 2023/04/25
SIMD

cpu

ARM
ãƒªãƒ³ã‚¯
AVX/AVX2ã«ã‚ˆã‚‹æ•´æ•°ã®Shuffleã¨Permute - Qiita
Zen2ï¼ŒZen3ã®ãƒ¡ãƒ¢ãƒªã‹ã‚‰ã®shuffleã¯UopsãŒ1ã§ã™ï¼Žä»–ã¯2ã«ãªã‚Šã¾ã™ï¼Ž èª¬æ˜Ž floatã¨ã¯ç•°ãªã‚Šï¼Œ1ã¤ã®ãƒ¬ã‚¸ã‚¹ã‚¿ã®å€¤ã‚’ï¼Œ128ãƒ“ãƒƒãƒˆã®å£ã‚’è¶…ãˆãšã«intã®æ•´æ•°ã‚’ä¸¦ã¹æ›¿ãˆã¾ã™ï¼Ž floatã®å ´åˆã®permute_psã¨åŒã˜æŒ™å‹•ã‚’ã™ã‚‹ãŸã‚ï¼Œimmã«ã‚ˆã‚‹æŒ‡å®šã«ã‚ˆã£ã¦XMMã¨YMMã®ä¸¦ã³ã¯åŒã˜ã«ãªã‚Šã¾ã™ï¼Ž åŒæ§˜ã«_MM_SHUFFLEã§ã®æŒ‡å®šãŒä¾¿åˆ©ã§ã™ï¼Ž ä»¥ä¸‹ã«ã‚µãƒ³ãƒ—ãƒ«ã‚³ãƒ¼ãƒ‰ã‚’ç¤ºã—ã¾ã™ï¼Ž void test_shuffle_epi32() { __m256i a, d; a = _mm256_setr_epi32(0, 1, 2, 3, 4, 5, 6, 7); print_m256i_int(a); printf("\n"); d = _mm256_shuffle_epi32(a, _MM_SHUFFLE(3, 2, 1, 0)); print_m256i_int(d); d =
tekimen 2023/02/09
ãƒ—ãƒã‚°ãƒ©ãƒŸãƒ³ã‚°

SIMD
ãƒªãƒ³ã‚¯
é«˜é€ŸUTF-8ãƒãƒªãƒ‡ãƒ¼ã‚·ãƒ§ãƒ³ã®ä¸–ç•Œ - Qiita
å‚ç…§: http://www.unicode.org/versions/Unicode6.0.0/ch03.pdf ã® "Table 3-7. Well-Formed UTF-8 Byte Sequences" ã‚¢ãƒ«ã‚´ãƒªã‚ºãƒ ã‚’ç†è§£ã™ã‚‹ä¸Šã§é‡è¦ãªUTF-8ã®ç‰¹å¾´ã«ã¤ã„ã¦è¿°ã¹ã¾ã™ã€‚ 1ã‚³ãƒ¼ãƒ‰ãƒã‚¤ãƒ³ãƒˆã¯1-4ãƒã‚¤ãƒˆã®ã‚·ãƒ¼ã‚±ãƒ³ã‚¹ã§è¡¨ç¾ã•ã‚Œã‚‹ ä¸Šä½ãƒ‹ãƒ–ãƒ«ï¼ˆ1ãƒã‚¤ãƒˆ8ãƒ“ãƒƒãƒˆã®ã†ã¡ã€ä¸Šä½4ãƒ“ãƒƒãƒˆï¼‰ã‚’ç¢ºèªã™ã‚‹ã“ã¨ã§ã‚·ãƒ¼ã‚±ãƒ³ã‚¹ã®æƒ…å ±ãŒå¾—ã‚‰ã‚Œã‚‹ ãã®ãƒã‚¤ãƒˆãŒã‚·ãƒ¼ã‚±ãƒ³ã‚¹å…ˆé ãƒã‚¤ãƒˆã‹ã©ã†ã‹ã‚ã‹ã‚‹ ã‚‚ã—ãã‚ŒãŒã‚·ãƒ¼ã‚±ãƒ³ã‚¹å…ˆé ãƒã‚¤ãƒˆã ã£ãŸãªã‚‰ã€ä½•ãƒã‚¤ãƒˆã®ã‚·ãƒ¼ã‚±ãƒ³ã‚¹ã‹ã‚ã‹ã‚‹ å…ˆé ã§ãªã„ãƒã‚¤ãƒˆã¯åŸºæœ¬çš„ã«0x80..0xBFã®ç¯„å›²ãŒè¨±å®¹ã•ã‚Œã¦ã„ã‚‹ãŒã€ä½•ç®‡æ‰€ã‹ä¾‹å¤–ãŒã‚ã‚‹ã®ã§ãã‚Œã‚‚ãƒãƒªãƒ‡ãƒ¼ã‚·ãƒ§ãƒ³ã—ãªã‘ã‚Œã°ãªã‚‰ãªã„ ä¾‹å¤–ã®ç®‡æ‰€ã¯è¡¨ã§ã¯å¤ªå—ã§ç¤ºã—ãŸ ä¾‹ãˆã°ã€è¡¨ã‚’è¦‹ã¦åˆ†ã‹ã‚‹ã‚ˆã†ã«ã€å…ˆé ãƒã‚¤ãƒˆãŒ0xE0ã®ã¨ã2ãƒã‚¤ãƒˆç›®ã¯0xA0..0x
tekimen 2023/02/07
ã‚¢ãƒ«ã‚´ãƒªã‚ºãƒ

Unicode
ãƒªãƒ³ã‚¯
1