〈[Kernel] Linux Kernel 內的 Macro likely() 和 unlikely()〉的留言 https://blog.richliu.com/2007/02/01/428/kernel-linux-kernel-%e5%85%a7%e7%9a%84-macro-likely-%e5%92%8c-unlik/ Linux, 工作, 生活, 家人 Tue, 05 Jun 2012 03:51:12 +0000 hourly 1 https://wordpress.org/?v=6.6.2 留言者: 匿名訪客 https://blog.richliu.com/2007/02/01/428/kernel-linux-kernel-%e5%85%a7%e7%9a%84-macro-likely-%e5%92%8c-unlik/#comment-45236 Tue, 05 Jun 2012 03:51:12 +0000 http://blog.richliu.com/2007/02/01/428/#comment-45236 long __builtin_expect (long EXP, long C)
#define unlikely(x) __builtin_expect(!!(x), 0)
the built-in are that it is expected that EXP == C
==> unlikely(x) 表示期望 x == 0

上面的觀念…………有點問題

]]> 留言者: 匿名訪客 https://blog.richliu.com/2007/02/01/428/kernel-linux-kernel-%e5%85%a7%e7%9a%84-macro-likely-%e5%92%8c-unlik/#comment-30663 Wed, 01 Aug 2007 08:32:43 +0000 http://blog.richliu.com/2007/02/01/428/#comment-30663 最后结论说反了吧

]]> 留言者: jsli https://blog.richliu.com/2007/02/01/428/kernel-linux-kernel-%e5%85%a7%e7%9a%84-macro-likely-%e5%92%8c-unlik/#comment-19831 Thu, 01 Feb 2007 14:40:29 +0000 http://blog.richliu.com/2007/02/01/428/#comment-19831 現在手邊沒有 cross-compiler 可以試，不過 ARM 的 instruction set 有 conditional execution 的 flag, 可以取代部份的 branch, 再加上 ARM9 5-stage pipeline 和相對較小的 cache, Branch prediction 就沒那麼重要了。Branch prediction 通常是為了避免 pipeline stall 和 cache miss penalty, 這部份 x86 要嚴重多了。

]]>