Re: [PATCH 0/5] branch hint tweaks

[Brian Gerst <brgerst@gmail.com>]

On Tue, Nov 24, 2009 at 1:17 PM, Aaron Cohen <remleduff@gmail.com> wrote:
> On Tue, Nov 24, 2009 at 12:50 PM, Brian Gerst <brgerst@gmail.com> wrote:
>> On Tue, Nov 24, 2009 at 5:54 AM, Tim Blechmann <tim@klingt.org> wrote:
>>> hi all,
>>>
>>> doing some branch hint profiling of 2.6.31 on my nehalem machine showed
>>> a few
>>> incorrect branch hints.
>>> these patches remove or change the branch hints
>>
>> I'm not sure what version of GCC added this, but I wonder if this
>> option will fix the problem:
>>
>> -fno-guess-branch-probability
>>    Do not guess branch probabilities using heuristics.
>>
>>    GCC will use heuristics to guess branch probabilities if they are
>> not provided by profiling feedback (-fprofile-arcs). These heuristics
>> are based on the control flow graph. If some branch probabilities are
>
> Is the kernel's use of unlikely typically intended to mean "I consider
> the following unlikely so please optimize with that expectation" or is
> meant to indicate "the following is a slow-path so please pessimize
> it" or both?
>

Both.  It is meant to optimize static branch prediction (on x86,
forward branches are by default predicted not taken), and to reduce
icache footprint of infrequently used code in the main execution path.
 If a section of code is marked unlikely(), GCC should put the code at
the end of the function.

--
Brian Gerst