Go and CPU Caches
Credits: unsplash.comAccording to Jackie Stewart, a three-time world champion F1 driver, having an understanding of how a car works made him a better pilot. “You don’t have to be an engineer to be a racing driver, but you do have to have Mechanical Sympathy†Martin Thompson (the designer of the LMAX Disruptor) applied the concept of mechanical sympathy to programming. In a nutshell, understanding
OSã®ãƒ‡ãƒãƒƒã‚°ï¼šãƒ¡ãƒ¢ãƒªã‚¢ãƒã‚±ãƒ¼ã‚·ãƒ§ãƒ³è¬›åº§ | POSTD
追記：やã‚ã‚„ã‚Hacker Newsã•ã‚“ï¼ãŠã²ã•ã—ã¶ã‚Šï¼ãƒ¡ãƒ¢ãƒªç®¡ç†ã‚’æ·±ã掘り下ã’ãŸç§ã®æŠ•ç¨¿ã‚’èªã‚€é‚ªé”ã¯ã—ãŸããªã„ã—ã€ç§ã®æŠ•ç¨¿ã®ã‚ら探ã—ã‚’ã™ã‚‹äººãŸã¡ã‚’邪é”ã™ã‚‹ã¤ã‚‚ã‚Šã‚‚ã‚ã‚Šã¾ã›ã‚“。技術的ãªãƒžãƒ«ç§˜ãƒ†ã‚¯ãƒ‹ãƒƒã‚¯ã«æ³¨ç›®ã™ã‚‹ã®ã‚‚ã„ã„ã§ã—ょã†ï¼ˆã„ã‚„ã€ã‚ã‹ã‚Šã¾ã™ã‚ˆã€‚楽ã—ã„ã§ã™ã—ãï¼ï¼‰ã€‚ã§ã‚‚ã€ç§ãŸã¡ã¯ã²ã¨ã‚Šã¼ã£ã¡ã§ã‚½ãƒ•ãƒˆã‚¦ã‚§ã‚¢ã‚’書ã„ã¦ã„ã‚‹ã‚ã‘ã§ã¯ãªã„ã®ã§ã™ã€‚ã ã‹ã‚‰ã€ãŸã‚ã«ãªã‚‹æŠ€è¡“çš„ãªã‚³ãƒ³ãƒ†ãƒ³ãƒ„（ç§ã®è¨˜äº‹ã‚‚ãã†ã§ã‚ã‚ŠãŸã„ã‚‚ã®ã§ã™ï¼‰ã‚’æœã™ã ã‘ã˜ã‚ƒãªãã¦ã€æ”¿æ²»çš„ãªè©±é¡Œã«ã‚‚目をå‘ã‘ã‚‹ã“ã¨ã‚’å¼·ããŠã™ã™ã‚ã—ã¾ã™ã€‚ソフトウェア開発者ã§ã‚ã‚‹ç§ãŸã¡ã¯ã€ä»Šå¾Œæ•°åå¹´ã«ã‚ãŸã£ã¦ç¤¾ä¼šã‚’変ãˆã‚‹æœ€å‰ç·šã«ç«‹ã¤ç‰¹æ¨©ã‚’与ãˆã‚‰ã‚Œã¦ã„ã¾ã™ã€‚ç§ãŸã¡ã¯ã€è‡ªåˆ†ãŸã¡ã®ä»•äº‹ã‚’社会全体ã«å½¹ç«‹ã¦ã‚‰ã‚Œã‚‹ã‚ˆã†ã«ã™ã‚‹ãŸã‚ã®æƒ…å ±ã‚„çŸ¥è˜ã‚’å¿…è¦ã¨ã—ã¦ã„ã¾ã™ã€‚ ãã†ã„ã†æ„味ã§ã¿ãªã•ã‚“ã«ã¯ã€ Hacker NewsãŒã€Œæ”¿æ²»çš„ã€ãªã‚³ãƒ³ãƒ†ãƒ³ãƒ„を検閲ã™ã‚‹ã¨è¨€ã„出ã—ãŸã“㨠ã«å対ã—ã¦ã‚‚
Latency Numbers Every Programmer Should Know
latency.txt :�]�U Latency Comparison Numbers (~2012) ---------------------------------- L1 cache reference 0.5 ns Branch mispredict 5 ns L2 cache reference 7 ns 14x L1 cache Mutex lock/unlock 25 ns Main memory reference 100 ns 20x L2 cache, 200x L1 cache Compress 1K bytes with Zippy 3,000 ns 3 us Send 1K bytes over 1 Gbps network 10,000 ns 10 us Read 4K randomly from SSD* 150,000 ns 150 us ~1GB/se
コンピュータアーã‚テクãƒãƒ£ã®è©±(368) GCNアーã‚テクãƒãƒ£ã«ãŠã‘ã‚‹1次データã‚ャッシュã®æ§‹é€
L1データã‚ャッシュ AMDã®GCNアーã‚テクãƒãƒ£ã®GPUã¯èªã¿æ›¸ãå¯èƒ½ãª16KBã®å®¹é‡ã®1次データã‚ャッシュをæŒã£ã¦ã„る。NVIDIAã®GPUã¯1次データã‚ャッシュã®ä½ç½®ã¥ã‘ãŒä¸–代ã”ã¨ã«å¤‰ã‚Šã€æœ€é©åŒ–ãŒã‚„り難ã„ãŒã€GCNアーã‚テクãƒãƒ£ã®1次データã‚ャッシュã¯èªã¿æ›¸ãå¯èƒ½ãªã‚ャッシュã¨ã„ã†ä½ç½®ã¥ã‘ãŒä¸€è²«ã—ã¦ã„ã‚‹ã®ã¯ç«‹æ´¾ã§ã‚る。 GCNアーã‚テクãƒãƒ£ã®GPUã®1次データã‚ャッシュã®æ§‹é€ を図8.8ã«ç¤ºã™ã€‚1次データã‚ャッシュã¯ã€å®¹é‡ãŒ16KBã§ã€ãƒ©ã‚¤ãƒ³ã‚µã‚¤ã‚ºã¯64Bã€4wayã®ã‚»ãƒƒãƒˆã‚¢ã‚½ã‚·ã‚¢ãƒ†ã‚£ãƒ–ã§ã€LRU入れ替ãˆã‚’è¡Œã†ã€‚ 1次データã‚ャッシュã¸ã®æ›¸ãè¾¼ã¿ã‚’è¡Œã†å ´åˆã€CPUã§ã¯ã€ã‚ャッシュコヒーレンシプãƒãƒˆã‚³ãƒ«ã‚’使ã£ã¦ã€å„コアã®1次データã‚ャッシュ間ã®å†…容ã«çŸ›ç›¾ãŒç”Ÿã˜ãªã„よã†ã«ã™ã‚‹ã®ã§ã‚ã‚‹ãŒã€GPUã®å ´åˆã¯Wavefrontã”ã¨ã«64個ã®ç•°ãªã‚‹ã‚¢ãƒ‰ãƒ¬ã‚¹ã«æ›¸ãè¾¼ã¿ãŒè¡Œã‚れるå¯èƒ½æ€§ãŒã‚る。ã“ã®ãŸã‚ã€
SSDã‚ャッシュã®æ´å²
ç§ã¯2012å¹´é ƒã‹ã‚‰dm-writeboostを開発ã—ã¦ã„ã‚‹ã‚ã‘ã ãŒã€ç§ãŒdm-writeboostを開発ã—ã¯ã˜ã‚ã¦ã‹ã‚‰æ–°ã—ãSSDã‚ャッシュã¯é–‹ç™ºã•ã‚Œã¦ã„ãªã„。少ãªãã¨ã‚‚OSSã§ã¯ã€‚一応ã€Oh Yongseokã¨ã„ã†éŸ“国人ã®å¦ç”ŸãŒdm-writeboostã‚’å…ƒã«ã—ã¦dm-srcã¨ã„ã†ç ”究プãƒã‚¸ã‚§ã‚¯ãƒˆã‚’ã‚„ã£ãŸãŒ(https://ysoh.files.wordpress.com/2009/05/dm-src-ibm.pdf)ã€ãã‚Œã¯ãƒ—ãƒãƒˆã‚¿ã‚¤ãƒ—ã®ãƒ¬ãƒ™ãƒ«ã§ã‚りプãƒãƒ€ã‚¯ã‚·ãƒ§ãƒ³ã§ä½¿ã†ãƒ¬ãƒ™ãƒ«ã«ã¯è‡³ã‚‰ãªã‹ã£ãŸã€‚ ç§ãŒdm-writeboostを開発ã—ã¯ã˜ã‚ãŸã“ã‚ã¯ã¡ã‚‡ã†ã©ã€RHã®dm-cacheãŒã‚¢ãƒƒãƒ—ストリームã«ç¾ã‚ŒãŸé ƒã§ã‚ã‚Šã€ãã®é ƒã«ã¯bcacheã¯ã™ã§ã«å…¥ã£ã¦ã„ãŸã€‚今ã€ã‚¢ãƒƒãƒ—ストリームã«å…¥ã£ã¦ã„ã‚‹ã®ã¯bcacheã¨dm-cacheã®2ã¤ã ãŒã€bcacheãŒ3.9ã€dm-cacheãŒ3.10ã«ãƒžãƒ¼
ã‚ãªãŸã®ãŠå®¶ã«å±Šãã¾ã§ CPUã‚ャッシュアーã‚テクãƒãƒ£å…¥é–€ Dora.Panda
3. ãƒãƒ¼ãƒ‰ã‚¦ã‚§ã‚¢æ§‹æˆ 左図ã®æ§‹æˆã§èª¬æ˜Žã‚’進ã‚ã¾ã™ CPUã«ã‚ˆã£ã¦å®Ÿè£…ãŒç•°ãªã‚‹ã®ã§ã€æ©Ÿç¨® 固有ã®èª¬æ˜Žã¯çœãã¾ã™ã€‚ ç§ãªã‚Šã«è€ƒãˆãŸã€Œã‚ャッシュã®åŸºç¤Žã‚’ ç†è§£ã™ã‚‹ã€ãŸã‚ã®ä»®æƒ³CPUã§ã™ã€‚ 「ãã“ãã“ã€ãƒ¯ãƒ¼ã‚¹ãƒˆã‚±ãƒ¼ã‚¹ã§è€ƒãˆã¾ã™ã€‚ TLBヒットã€L1ã€L2ã‚ャッシュã¯ãƒŸã‚¹ ヒット。 ç´°ã‹ã„用語ã¯å¾Œã»ã©ã€‚ Processor Chip CPU Core L2 Cache Mem Ctrl DRAM HW… PCIe MMU L1 Cache
Memory Layouts for Binary Search
The first round of experiments is over, but you can still help out with the second round. Binary search is a wonderful algorithm: theoretically optimal, fast in practice, and discoverable by school children playing guessing games. Usually we think of binary search as being applied to a sorted array, and we teach it this way. It turns out that, because of cache effects, sorted order is usually not
Haskell and State
Haskell is a pure functional programming language. My question is: What are the advantages and disadvantages of using Haskell to solve problems involving lots of state, for example GUI programming or game programming? Also a secondary question: what methods are there to handle state in a functional way? Thanks in advance.
Haskell ã® State モナド (1) - 状態を模倣ã™ã‚‹
昔ã€ãƒ¢ãƒŠãƒ‰ãŒã‚ˆãã‚ã‹ã‚‰ãªã‹ã£ãŸã®ã§ã€ã•ã¾ã‚ˆã£ã¦ã„ãŸã‚‰ã€ … ãƒãƒƒãƒˆã§è¦‹ãŸMonadã®èª¬æ˜Žã§ä¸€ç•ªç§ãŒã‚ã‹ã‚Šã‚„ã™ã„ã¨æ€ã£ãŸã®ã¯ã€Wikibooksã®èª¬æ˜Žã€‚Hello World!ãŒãƒ–ラックボックスãªäººã¯ã€æ˜¯éžä¸€èªã‚’。 (404 Blog Not Found:Haskellã§ä¸€ç•ªé›£ã—ã„ã®ã¯ より) 最åˆã«ã“ã® Wikibooksã®èª¬æ˜Ž ã‚’èªã‚“ã ã®ã¯åŽ»å¹´ã® 11 æœˆé ƒã€‚ãã®ã¨ãã®æ–‡æ›¸ã®ãƒãƒ¼ã‚¸ãƒ§ãƒ³ã¯Â 05:13, 27 October 2008 ã§ã€ä»Šã¯å†…容ãŒéšåˆ†å¢—ãˆã¦ã„る。å‰ã®æ–‡æ›¸ã¯ã€ç¾åœ¨ã® Haskell/Understanding monads/State ã«ç›¸å½“ã™ã‚‹ã‚ˆã†ã 。 ã¨ã“ã‚ã§ã€ä¸Šè¨˜ã®è§£èª¬ã‚’最åˆèªã‚“ã ã¨ãå…¨ãæ„味ãŒã‚ã‹ã‚‰ãªã‹ã£ãŸã€‚ (@_@;) 「3 Stateful Computations〠ã§ã¯ã€ã€Œãƒ©ãƒ³ãƒ€ãƒ ãªæ•°å—を生æˆã™ã‚‹é–¢æ•°ã€ã‚’例ã«æŒ™ã’ã¦ãƒ¢ãƒŠãƒ‰ã®èª¬æ˜ŽãŒã•ã‚Œã¦ã„ãŸã‘ã‚Œã©ã€ä½•ãŒè¨€
Memoise
Memoi[sz]eã€Memoi[sz]ationã€ãƒ¡ãƒ¢åŒ–ã®è©±é¡Œ メモ化ã£ã¦ãªãã«ï¼Ÿé–¢æ•°ã®ãƒ¡ãƒ¢åŒ–memoise ã¯ç‰¹æ®Šãª ($) ã‹ã‚‚？Memo モジュール実装を共有ã™ã‚‹é”法 メモ化ã£ã¦ãªãã«ï¼Ÿ フィボナッãƒé–¢æ•°ã‚’考ãˆã¦ã¿ã‚ˆã†ã€å®šç¾©ã¯ fib 0 = 0 fib 1 = 1 fib n = fib (n-1) + fib (n-2) ã“れを使ã£ã¦ã€fib 7 を計算ã™ã‚‹ã¨ fib 7 -- fib 6 -- fib 5 -- fib 4 -- fib 3 -- fib 2 -- fib 1 -- 1 | | | | | | | | | | | fib 0 -- 0 | | | | | | | | | fib 1 -- 1 | | | | | | | fib 2 -- fib 1 -- 1 | | | | | | | fib 0 -- 0 | | | | | fib 3
"state monad" cache - Google 検索
williamyaoh.com › posts › 2020-07-12-deriving-state-... 2020/07/12 · Deriving the State monad from first principles ... In this series of posts, we're going to talk about some specific monad instances, the Reader- ...
2008-05-25
sampou.orgã®http://www.sampou.org/cgi-bin/haskell.cgi?Memoiseã‚’ã»ã¨ã‚“ã©ä¸¸å†™ã—ã—ã¦ï¼ŒStateモナドを使ã£ã¦ãƒ•ã‚£ãƒœãƒŠãƒƒãƒã‚’書ã„ã¦ã¿ã¾ã—ãŸï¼Ž import Control.Monad.State rfib :: Integer -> Integer rfib 0 = 0 rfib 1 = 1 rfib n = rfib (n-1) + rfib (n-2) type Cache = [(Integer,Integer)] -- ã‚ャッシュ cfib :: Integer -> State Cache Integer cfib 0 = return 0 cfib 1 = return 1 cfib n = do cache <- get case (lookup n cache) of Just k -> return k Nothin
ランã‚ング
ランã‚ング
ランã‚ング
リリースã€éšœå®³æƒ…å ±ãªã©ã®ã‚µãƒ¼ãƒ“スã®ãŠçŸ¥ã‚‰ã›
最新ã®äººæ°—エントリーã®é…ä¿¡
j次ã®ãƒ–ックマーク
kå‰ã®ãƒ–ックマーク
lã‚ã¨ã§èªã‚€
eコメント一覧を開ã
oページを開ã
Line size of L1 and L2 caches
GitHub - harishankarsuresh/Cache-Simulator: Cache simulator using C++ Standard Template Library
GitHub - IAIK/CJAG: CJAG is an open-source implementation of our cache-based jamming agreement.
cache oblivious - Google 検索
Optimizing longest Collatz chain in Haskell
Memoization in Haskell?
awkward monad transformer stack
Better pattern for caching results
{{#tags}}- {{label}}
{{/tags}}