cacheline 填充的意义 - V2EX
V2EX = way to explore
V2EX 是一个关于分享和探索的地方
现在注册
已注册用户请  登录
The Go Programming Language
http://golang.org/
Go Playground
Go Projects
Revel Web Framework
docxs
V2EX    Go 编程语言

cacheline 填充的意义

  •  
  •   docxs 2022-09-02 21:20:28 +08:00 1892 次点击
    这是一个创建于 1188 天前的主题,其中的信息可能已经有所发展或是发生改变。

    这种填充方式能解决 false sharing 吗,

    type PaddedStruct struct { _ CacheLinePad n int } type CacheLinePad struct { _ [CacheLinePadSize]byte } const CacheLinePadSize = 64 

    CacheLinePad 后面的字段,是怎么避免和其他结构体不共享一个 Cache 块了,把 CacheLinePad 放在第一个字段,只能避免和它前面的数据不在一个 cache 块吧,是不是应该这样

    type PaddedStruct struct { _ CacheLinePad n int _ CacheLinePad } 
    10 条回复    2022-09-04 19:26:17 +08:00
    slowman
        1
    slowman  
       2022-09-02 23:19:51 +08:00
    On ARM, 386, and 32-bit MIPS, ........ The first word in an allocated struct, array, or slice; in a global variable; or in a local variable can be relied upon to be 64-bit aligned.

    https://pkg.go.dev/sync/atomic
    slowman
        2
    slowman  
       2022-09-02 23:22:44 +08:00
    啊,请忽略上条
    qwe678
        3
    qwe678  
       2022-09-03 12:38:11 +08:00
    struct PaddedStruct {
    int a;
    char cache_line_pad[64];
    int b;
    };
    docxs
        4
    docxs  
    OP
       2022-09-03 16:38:46 +08:00 via iPhone
    @qwe678 这个结构体的 a 会不会和另一个结构体的尾部合成一个 cache 块了
    lysS
        5
    lysS  
       2022-09-04 11:15:30 +08:00
    @docxs #4 当然会,和 struct 无关,只关心是否两个变量是否同在一个 Cache Block 里面
    lysS
        6
    lysS  
       2022-09-04 11:30:28 +08:00
    我测出来了,性能差了 3 倍以上,https://go.dev/play/p/WF-CoUIe4bd

    goos: windows
    goarch: amd64
    pkg: btest
    cpu: 11th Gen Intel(R) Core(TM) i5-11320H @ 3.20GHz
    BenchmarkD1
    BenchmarkD1-8
    582 1844166 ns/op 66 B/op 3 allocs/op
    BenchmarkD2
    BenchmarkD2-8
    2710 387049 ns/op 59 B/op 3 allocs/op
    BenchmarkC1
    BenchmarkC1-8
    583 1974648 ns/op 53 B/op 2 allocs/op
    BenchmarkC2
    BenchmarkC2-8
    2938 396749 ns/op 42 B/op 2 allocs/op
    PASS
    ok btest 4.985s
    lysS
        7
    lysS  
       2022-09-04 11:35:29 +08:00
    docxs
        8
    docxs  
    OP
       2022-09-04 15:54:08 +08:00 via iPhone
    @lysS 嗯,我的意思是,要解决一个 struct 中的变量和其他 struct 中的变量造成 false sharing ,是不是应该在那个变量的前后都加上 cachelinepad ,只在前或后一端加都没法避免,我看很多的文章讲避免 false sharing ,都拿一端填充的例子讲,所以比较疑惑
    lysS
        9
    lysS  
       2022-09-04 18:31:54 +08:00   1
    @docxs #8 理论上是的,但是这种优化不是绝对的,把高频操作的变量分开就可以了
    docxs
        10
    docxs  
    OP
       2022-09-04 19:26:17 +08:00 via iPhone
    @lysS 好的,多谢
    关于     帮助文档     自助推广系统     博客     API     FAQ     Solana     916 人在线   最高记录 6679       Select Language
    创意工作者们的社区
    World is powered by solitude
    VERSION: 3.9.8.5 25ms UTC 20:19 PVG 04:19 LAX 12:19 JFK 15:19
    Do have faith in what you're doing.
    ubao msn snddm index pchome yahoo rakuten mypaper meadowduck bidyahoo youbao zxmzxm asda bnvcg cvbfg dfscv mmhjk xxddc yybgb zznbn ccubao uaitu acv GXCV ET GDG YH FG BCVB FJFH CBRE CBC GDG ET54 WRWR RWER WREW WRWER RWER SDG EW SF DSFSF fbbs ubao fhd dfg ewr dg df ewwr ewwr et ruyut utut dfg fgd gdfgt etg dfgt dfgd ert4 gd fgg wr 235 wer3 we vsdf sdf gdf ert xcv sdf rwer hfd dfg cvb rwf afb dfh jgh bmn lgh rty gfds cxv xcv xcs vdas fdf fgd cv sdf tert sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf sdf shasha9178 shasha9178 shasha9178 shasha9178 shasha9178 liflif2 liflif2 liflif2 liflif2 liflif2 liblib3 liblib3 liblib3 liblib3 liblib3 zhazha444 zhazha444 zhazha444 zhazha444 zhazha444 dende5 dende denden denden2 denden21 fenfen9 fenf619 fen619 fenfe9 fe619 sdf sdf sdf sdf sdf zhazh90 zhazh0 zhaa50 zha90 zh590 zho zhoz zhozh zhozho zhozho2 lislis lls95 lili95 lils5 liss9 sdf0ty987 sdft876 sdft9876 sdf09876 sd0t9876 sdf0ty98 sdf0976 sdf0ty986 sdf0ty96 sdf0t76 sdf0876 df0ty98 sf0t876 sd0ty76 sdy76 sdf76 sdf0t76 sdf0ty9 sdf0ty98 sdf0ty987 sdf0ty98 sdf6676 sdf876 sd876 sd876 sdf6 sdf6 sdf9876 sdf0t sdf06 sdf0ty9776 sdf0ty9776 sdf0ty76 sdf8876 sdf0t sd6 sdf06 s688876 sd688 sdf86