{"id":115974,"date":"2024-05-21T14:59:30","date_gmt":"2024-05-21T12:59:30","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/proekt-bend-razvivaet-vysokourovnevyj-yazyk-dlya-parallelnyh-vychislenij-na-gpu"},"modified":"2024-05-21T14:59:30","modified_gmt":"2024-05-21T12:59:30","slug":"proekt-bend-razvivaet-vysokourovnevyj-yazyk-dlya-parallelnyh-vychislenij-na-gpu","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/proekt-bend-razvivaet-vysokourovnevyj-yazyk-dlya-parallelnyh-vychislenij-na-gpu","title":{"rendered":"Projekt Bend rozv\u00edj\u00ed vysoce \u00farov\u0148ov\u00fd jazyk pro paraleln\u00ed v\u00fdpo\u010dty na GPU","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Byly zve\u0159ejn\u011bny prvn\u00ed verze projektu Bend, kter\u00fd vyv\u00edj\u00ed vysoce \u00farov\u0148ov\u00fd programovac\u00ed jazyk pro paraleln\u00ed v\u00fdpo\u010dty. Ten se prezentuje jako alternativa k n\u00edzko\u00farov\u0148ov\u00fdm jazyk\u016fm, jako jsou CUDA a Metal, a nab\u00edz\u00ed expresivn\u00ed syntaxi a mo\u017enosti zvy\u0161uj\u00edc\u00ed pohodl\u00ed v\u00fdvoje, typick\u00e9 pro jazyky jako Python a Haskell. K\u00f3d projektu je naps\u00e1n v jazyce Rust a distribuuje se pod licenc\u00ed Apache 2.0. <\/p>\n
Mezi mo\u017enostmi jazyka Bend se vyzna\u010duje rychl\u00e9 rozd\u011blen\u00ed objekt\u016f, mo\u017enost pou\u017eit\u00ed vy\u0161\u0161\u00edch funkc\u00ed, uz\u00e1v\u011br\u016f, pokra\u010dov\u00e1n\u00ed, neomezen\u00e9 rekurze, vzorov\u00e9ho porovn\u00e1v\u00e1n\u00ed, rekurzivn\u00edho porovn\u00e1v\u00e1n\u00ed (fold) a cykl\u016f (bend), celo\u010d\u00edseln\u00fdch, \u0159et\u011bzcov\u00fdch a seznamov\u00fdch typ\u016f. Jsou podporov\u00e1ny dva varianty syntaxe \u2014 v Python stylu a v Haskell stylu. V programech nen\u00ed nutn\u00e9 uv\u00e1d\u011bt \u0159\u00edzen\u00ed paralelizace pomoc\u00ed anotac\u00ed, explicitn\u011b vytv\u00e1\u0159et vl\u00e1kna a nastavovat z\u00e1mky. Paralelizace prob\u00edh\u00e1 automaticky, nap\u0159\u00edklad p\u0159i v\u00fdpo\u010dtu v\u00fdrazu \u201e((1 + 2) + (3 + 4))\u201c budou operace \u201e1 + 2\u201c a \u201e3 + 4\u201c provedeny paraleln\u011b. <\/p>\n
Programy napsan\u00e9 v jazyce Bend mohou b\u011b\u017eet na hardwaru, kter\u00fd poskytuje masov\u00fd paralelismus, nap\u0159\u00edklad na GPU, p\u0159i\u010dem\u017e vykazuj\u00ed prakticky line\u00e1rn\u00ed n\u00e1r\u016fst v\u00fdkonu v z\u00e1vislosti na po\u010dtu v\u00fdpo\u010detn\u00edch jader. K\u00f3d v jazyce Bend je kompilov\u00e1n do n\u00edzko\u00farov\u0148ov\u00e9ho mezilehl\u00e9ho k\u00f3du HVM2 (Higher-order Virtual Machine 2), kter\u00fd je pak kompilov\u00e1n do podoby pro C a CUDA. V sou\u010dasn\u00e9 dob\u011b projekt podporuje pouze b\u011bh na GPU NVIDIA. <\/p>\n
Pokud jde o v\u00fdkon, testovac\u00ed aplikace s implementac\u00ed bitonn\u00ed t\u0159\u00edd\u011bn\u00ed byla na CPU Apple M3 Max provedena za 12,15 sekundy v jednom vl\u00e1kn\u011b, p\u0159i vyu\u017eit\u00ed 16 vl\u00e1ken \u2014 za 0,96 sekundy, a p\u0159i zapojen\u00ed GPU NVIDIA RTX 4090 se 16k vl\u00e1kny \u2014 za 0,21 sekundy. def sort(d, s, tree): switch d: case 0: return tree case _: (x,y) = tree lft = sort(d-1, 0, x) rgt = sort(d-1, 1, y) return rots(d, s, lft, rgt) def rots(d, s, tree): switch d: case 0: return tree case _: (x,y) = tree return down(d, s, warp(d-1, s, x, y)) \u2026 \n Zdroj: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"