{"id":4200,"date":"2013-09-06T22:06:10","date_gmt":"2013-09-06T13:06:10","guid":{"rendered":"http:\/\/umezawa.dyndns.info\/wordpress\/?p=4200"},"modified":"2014-06-24T17:15:08","modified_gmt":"2014-06-24T08:15:08","slug":"avx-512-%e3%81%8c%e3%82%a2%e3%83%8a%e3%82%a6%e3%83%b3%e3%82%b9%e3%81%95%e3%82%8c%e3%81%a6%e3%81%84%e3%81%9f","status":"publish","type":"post","link":"http:\/\/umezawa.dyndns.info\/wordpress\/?p=4200","title":{"rendered":"AVX-512 \u304c\u30a2\u30ca\u30a6\u30f3\u30b9\u3055\u308c\u3066\u3044\u305f"},"content":{"rendered":"<p>x86\/x64 \u6700\u9069\u5316\u52c9\u5f37\u4f1a #6 \u3067\u805e\u3044\u3066\u304d\u307e\u3057\u305f\u304c\u3001<a href=\"http:\/\/software.intel.com\/en-us\/blogs\/2013\/avx-512-instructions\">AVX-512 \u306a\u308b\u3082\u306e<\/a>\u304c\u30a2\u30ca\u30a6\u30f3\u30b9\u3055\u308c\u3066\u307e\u3057\u305f\u3002\u540d\u524d\u304b\u3089\u4e88\u60f3\u3055\u308c\u308b\u901a\u308a\u3001SIMD \u30ec\u30b8\u30b9\u30bf\u304c 512bit \u306b\u306a\u308a\u307e\u3059\u3002SIMD \u30ec\u30b8\u30b9\u30bf\u304c 512bit \u3068\u3044\u3046\u3068 Xeon Phi \u306f\u6700\u521d\u304b\u3089\u305d\u3046\u306a\u3063\u3066\u3044\u308b\u306e\u3067\u3059\u304c\u3001\u3053\u308c\u304c\u666e\u901a\u306e Xeon \u306b\u964d\u308a\u3066\u304f\u308b\u30a4\u30e1\u30fc\u30b8\u3067\u3059\u3002<\/p>\n<p><!--more--><\/p>\n<p>AVX \u304c\u300c\u5358\u306a\u308b legacy SSE \u306e 256bit \u7248\u300d\u3067\u306f\u306a\u3044\u306e\u3068\u540c\u69d8\u3001AVX-512 \u3082\u300c\u5358\u306a\u308b AVX \u306e 512bit \u7248\u300d\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002<\/p>\n<ul>\n<li>SIMD \u30ec\u30b8\u30b9\u30bf\u304c zmm \u3068\u3044\u3046\u540d\u524d\u3067 512bit \u306b\u306a\u308b\u3002\u4e0b\u4f4d 256bit \u306f ymm \u3068\u3057\u3066\u30a2\u30af\u30bb\u30b9\u3067\u304d\u308b\uff08xmm \u306b\u5bfe\u3059\u308b ymm \u3068\u540c\u3058\uff09<\/li>\n<li>64bit \u30e2\u30fc\u30c9\u306e\u5834\u5408\u3001SIMD \u30ec\u30b8\u30b9\u30bf\u304c 32\u672c\u4f7f\u3048\u308b\u3002\u3053\u308c\u306f\u3001zmm \u3060\u3051\u3067\u306a\u304f ymm \u3084 xmm \u3067\u3082\u540c\u69d8\u3067\u3042\u308b\u300232bit \u30e2\u30fc\u30c9\u3067\u306f8\u672c\u306e\u307e\u307e\u3002<\/li>\n<li>8\u672c\u306e\u30de\u30b9\u30af\u30ec\u30b8\u30b9\u30bf\u304c\u8ffd\u52a0\u3055\u308c\u308b\u3002\u30de\u30b9\u30af\u30ec\u30b8\u30b9\u30bf\u306f\u30e9\u30a4\u30c8\u30de\u30b9\u30af\u306e\u6307\u5b9a\u306b\u4f7f\u308f\u308c\u308b\u3002\u3064\u307e\u308a\u30011\u547d\u4ee4\u3054\u3068\u306b merging(blend)\/zeroing \u64cd\u4f5c\u304c\u53ef\u80fd\u306b\u306a\u308b\u3002<\/li>\n<li>\u547d\u4ee4\u57cb\u3081\u8fbc\u307f\u3067 2\u3064\u76ee\u306e\u30bd\u30fc\u30b9\u30aa\u30da\u30e9\u30f3\u30c9\u306e\u30d6\u30ed\u30fc\u30c9\u30ad\u30e3\u30b9\u30c8\u3001\u4e38\u3081\u5236\u5fa1\u3001\u4f8b\u5916\u7121\u8996 \u304c\u53ef\u80fd\uff08\u547d\u4ee4\u306b\u3088\u308b\uff09<\/li>\n<li>\u4ee5\u4e0a\u3092\u5b9f\u73fe\u3059\u308b\u305f\u3081\u306b\u3001VEX \u30d7\u30ec\u30d5\u30a3\u30c3\u30af\u30b9\u306b\u7d9a\u3044\u3066 EVEX \u30d7\u30ec\u30d5\u30a3\u30c3\u30af\u30b9\u3092\u5c0e\u5165\u3059\u308b\u3002VEX \u3084 MVEX \u3068\u30d3\u30c3\u30c8\u30d1\u30bf\u30fc\u30f3\u306f\u91cd\u306a\u3089\u306a\u3044\u306e\u3067\u540c\u5c45\u53ef\u80fd\u3002<\/li>\n<li>\u3082\u3061\u308d\u3093\u666e\u901a\u306e\u610f\u5473\u3067\u306e\u65b0\u547d\u4ee4\u3082\u8ffd\u52a0<\/li>\n<\/ul>\n<p>\u65b0\u6a5f\u80fd\u3066\u3093\u3053\u76db\u308a\u3067\u3059\u3002\u6b69\u7559\u307e\u308a\u5927\u4e08\u592b\u306a\u3093\u3067\u3057\u3087\u3046\u304b\u3002<\/p>\n<p>\u6bd4\u8f03\u3059\u308b\u3068\u3053\u3093\u306a\u611f\u3058\u306b\u306a\u308a\u307e\u3059\u3002LNI \u3068\u6bd4\u8f03\u3059\u308b\u3068\u6a5f\u80fd\u7684\u306b\u306f\u4e00\u9577\u4e00\u77ed\u3068\u306a\u308a\u307e\u3059\u3002<\/p>\n<table border=\"border\">\n<tr>\n<th><\/th>\n<th>AVX<\/th>\n<th>LNI<br \/>(Larrabee New Instructions)<\/th>\n<th>AVX-512<\/th>\n<\/tr>\n<tr>\n<td>\u5b9f\u88c5\u30d7\u30ed\u30bb\u30c3\u30b5<\/td>\n<td>Sandy Bridge \u4ee5\u964d<\/td>\n<td>Xeon Phi (Knights Corner)<\/td>\n<td>\u307e\u3060\u4e0d\u660e<\/td>\n<\/tr>\n<tr>\n<td>\u4f7f\u308f\u308c\u308b\u30d7\u30ec\u30d5\u30a3\u30c3\u30af\u30b9<\/td>\n<td>VEX<\/td>\n<td>MVEX<\/td>\n<td>EVEX<\/td>\n<\/tr>\n<tr>\n<td>SIMD\u30ec\u30b8\u30b9\u30bf\u9577<\/td>\n<td>128bit (xmm)<br \/>256bit (ymm)<\/td>\n<td>512bit (zmm)<\/td>\n<td>128bit (xmm)<br \/>256bit (ymm)<br \/>512bit (zmm)<\/td>\n<\/tr>\n<tr>\n<td>SIMD\u30ec\u30b8\u30b9\u30bf\u6570<\/td>\n<td>8 (x86)<br \/>16 (x64)<\/td>\n<td>32<\/td>\n<td>8 (x86)<br \/>32 (x64)<\/td>\n<\/tr>\n<tr>\n<td>\u30de\u30b9\u30af\u30ec\u30b8\u30b9\u30bf<\/td>\n<td>\u306a\u3057<\/td>\n<td>\u3042\u308a<\/td>\n<td>\u3042\u308a<\/td>\n<\/tr>\n<tr>\n<td>\u30de\u30b9\u30af\u306b\u3088\u308b zeroing<\/td>\n<td>&#8211;<\/td>\n<td>\u306a\u3057<\/td>\n<td>\u3042\u308a<\/td>\n<\/tr>\n<tr>\n<td>\u30d6\u30ed\u30fc\u30c9\u30ad\u30e3\u30b9\u30c8<\/td>\n<td>\u306a\u3057<\/td>\n<td>\u3042\u308a<\/td>\n<td>\u30d5\u30eb\u30d6\u30ed\u30fc\u30c9\u30ad\u30e3\u30b9\u30c8\u306e\u307f<\/td>\n<\/tr>\n<tr>\n<td>swizzle<\/td>\n<td>\u306a\u3057<\/td>\n<td>\u3042\u308a<\/td>\n<td>\u306a\u3057<\/td>\n<\/tr>\n<tr>\n<td>\u4e38\u3081\u5236\u5fa1\/\u4f8b\u5916\u7121\u8996<\/td>\n<td>\u306a\u3057<\/td>\n<td>\u3042\u308a<\/td>\n<td>\u3042\u308a<\/td>\n<\/tr>\n<\/table>\n<p>\u3053\u308c\u304c\u3044\u3064\u51fa\u3066\u304f\u308b\u304b\u3067\u3059\u304c\u3001\u307e\u305a\u6b21\u306e Xeon Phi \u3067\u3042\u308b Knights Landing \u306b\u5165\u3063\u3066\u304f\u308b\u306e\u306f\u78ba\u5b9a\u3067\u3001\u305d\u306e\u5f8c\u666e\u901a\u306e Xeon \u306b\u6765\u308b\u3053\u3068\u306b\u306a\u3063\u3066\u3044\u307e\u3059\u3002\u554f\u984c\u306f Xeon Phi \u3067\u306f\u306a\u3044\u666e\u901a\u306e\u30d7\u30ed\u30bb\u30c3\u30b5\u306e\u65b9\u306a\u3093\u3067\u3059\u304c\u3001&#8221;and will also be supported by some future Xeon processors scheduled to be introduced after Knights Landing&#8221; \u3068\u306a\u3063\u3066\u304a\u308a\u3001Xeon \u9650\u5b9a\u2026\u3064\u307e\u308a Core i \u30b7\u30ea\u30fc\u30ba\u306b\u306f\u5b9f\u88c5\u3055\u308c\u306a\u3044\u3088\u3046\u306b\u3082\u8aad\u3081\u307e\u3059\u3002\u3060\u3068\u3059\u308b\u3068\u500b\u4eba\u7684\u306b\u306f\u975e\u5e38\u306b\u30c4\u30de\u30e9\u30ca\u30a4\u3053\u3068\u306b\u306a\u308a\u305d\u3046\u3067\u3059\u3002\u300c\u3055\u3089\u306b\u5c06\u6765\u300d\u306b Core i \u30b7\u30ea\u30fc\u30ba\u306b\u3082\u964d\u308a\u3066\u304d\u3066\u304f\u308c\u308c\u3070\u3044\u3044\u306e\u3067\u3059\u304c\u3002<\/p>\n<p>\u3068\u3082\u3042\u308c\u3001AVX2 \u304c Sandy Bridge \u304c\u51fa\u305f\u3053\u308d\u306b\u30a2\u30ca\u30a6\u30f3\u30b9\u3055\u308c\u3066 Haswell \u3067\u5b9f\u88c5\u3055\u308c\u305f\u3053\u3068\u3092\u8003\u3048\u308b\u3068\u3001Haswell \u304c\u51fa\u305f\u3053\u308d\u306b\u30a2\u30ca\u30a6\u30f3\u30b9\u3055\u308c\u305f AVX-512 \u306f Skylake \uff08\u306e Xeon \u7248\uff1f\uff09\u3042\u305f\u308a\u3067\u5b9f\u88c5\u3055\u308c\u3066\u3082\u3088\u3055\u305d\u3046\u306a\u3082\u306e\u3067\u3059\u3002AVX2 \u3088\u308a\u305a\u3063\u3068\u5927\u898f\u6a21\u306a\u306e\u3067\u9593\u306b\u5408\u3046\u306e\u304b\uff1f\u3068\u3082\u601d\u3044\u307e\u3059\u304c\u3001AVX-512 \u3092\u30d1\u30b9\u3057\u305f\u5834\u5408 Skylake \u306e\u65b0\u6a5f\u80fd\u306f\u4f55\u306b\u306a\u308b\u306e\u304b\u3068\u3044\u3046\u554f\u984c\u304c\u767a\u751f\u3057\u307e\u3059\u3002SHA Extensions \u3084 ADCX, ADOX \uff08AVX-512 \u306e\u30de\u30cb\u30e5\u30a2\u30eb\u306e\u5f8c\u308d\u306e\u65b9\u306b\u3042\u308b\uff09\u3060\u3051\u8ffd\u52a0\u3057\u3066\u7d42\u308f\u308a\u3060\u3063\u305f\u308a\u3057\u3066\u2026<\/p>\n<p>\u79cb\u306e Intel Developers Forum (IDF) \u306f9\u670810\u65e5\u304b\u3089\u59cb\u307e\u308b\u306e\u3067\u3001<del>\u305d\u3053\u3067\u4f55\u304b\u767a\u8868\u3055\u308c\u308b\u3067\u3057\u3087\u3046\u3002\u3055\u308c\u308b\u3068\u3044\u3044\u306a\u3002<\/del> \u7279\u306b\u3055\u308c\u306a\u304b\u3063\u305f\u3088\u3046\u3067\u3059\u3002\u3061\u3047\u3002<\/p>\n<p>\u3068\u3053\u308d\u3067\u3001AVX-512 \u3060\u3051\u3067 feature flags \u30924\u3064\u3082\u6301\u3063\u3066\u308b\u306e\u306f\u3084\u3081\u3066\u3082\u3089\u3048\u307e\u305b\u3093\u304b\u306d\u3047\u2026<\/p>\n","protected":false},"excerpt":{"rendered":"<p>x86\/x64 \u6700\u9069\u5316\u52c9\u5f37\u4f1a #6 \u3067\u805e\u3044\u3066\u304d\u307e\u3057\u305f\u304c\u3001AVX-512 \u306a\u308b\u3082\u306e\u304c\u30a2\u30ca\u30a6\u30f3\u30b9\u3055\u308c\u3066\u307e\u3057\u305f\u3002\u540d\u524d\u304b\u3089\u4e88\u60f3\u3055\u308c\u308b\u901a\u308a\u3001SIMD \u30ec\u30b8\u30b9\u30bf\u304c 512bit \u306b\u306a\u308a\u307e\u3059\u3002SIMD \u30ec\u30b8\u30b9\u30bf\u304c 512bit \u3068\u3044\u3046 [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[27],"tags":[],"class_list":["post-4200","post","type-post","status-publish","format-standard","hentry","category-technology"],"_links":{"self":[{"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4200","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=4200"}],"version-history":[{"count":23,"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4200\/revisions"}],"predecessor-version":[{"id":4858,"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=\/wp\/v2\/posts\/4200\/revisions\/4858"}],"wp:attachment":[{"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4200"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4200"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/umezawa.dyndns.info\/wordpress\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4200"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}