Re: [HACKERS] Do we need the gcc feature "__builtin_expect" topromote the branches prediction? - Mailing list pgsql-hackers

From Craig Ringer
Subject Re: [HACKERS] Do we need the gcc feature "__builtin_expect" topromote the branches prediction?
Date
Msg-id CAMsr+YGW0UZyoOP+fCHb43h=DD8WCXNoLJGnua4ADZStt44jpA@mail.gmail.com
Whole thread Raw
In response to [HACKERS] Do we need the gcc feature "__builtin_expect" to promote the branches prediction?  (Hao Lee <mixtrue@gmail.com>)
Responses Re: [HACKERS] Do we need the gcc feature "__builtin_expect" topromote the branches prediction?  (Julien Rouhaud <julien.rouhaud@dalibo.com>)
List pgsql-hackers


On 2 Jun. 2017 16:42, "Hao Lee" <mixtrue@gmail.com> wrote:
Hi all, 
       There is a lot of "if statement" in system, and GCC provides a feature,"__builtin_expect", which  let compilers know which branch is mostly run.

Compilers and CPUs are really good at guessing this.

Humans are wrong about it more than we'd like too.

It's surprisingly rarely s good idea to use branch prediction hints.

See the vsrious Linux kernel discussions about this.

If you find concrete sites of frequent or costly branch mis-prediction please point them out, with benchmarks.


. as we known, miss-prediction will lead the performance lost(because the CPU will thrown away some instructions, and re-fetch some new instructions). so that we can tell GCC how produce more efficient code. for example as following. 
It will gain performance promotion i think. As i know, the in Linux kernel, this feature is also applied already. 

 #define likely(cond)     __builtin_expect(cond,true)
#define unlikely(cond)  __builtin_expect(cond,false)

if (likely(cond)) {
//most likely run.
   xxxx
} else //otherwise.
{
   xxxx
}


Best Regards.

Hom.

pgsql-hackers by date:

Previous
From: Marko Tiikkaja
Date:
Subject: [HACKERS] PG10 transition tables, wCTEs and multiple operations on the same table
Next
From: Dmitriy Sarafannikov
Date:
Subject: Re: [HACKERS] Broken hint bits (freeze)