This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH 1/25] Remove nested functions: crypt/md5-crypt.c


On Tue, May 20, 2014 at 05:20:31PM +0200, OndÅej BÃlka wrote:
> Yes I wanted to write that benchmarking this change is probably
> pointless as gcc should optimize them in same way (if there is
> consistently difference in one way we could report that as gcc regression.)
> 
> Also perf data show that its a cold function so you should improve
> readability not performance (measuring difference is possible but you
> need to run that for few days and wait until standard deviation drops
> sufficiently to get statistically significant result). 
> 
> About only way this could make difference is if when it gets inlined in
> one way and does not in other.
> That is out of scope of microbenchmarks, instruction cache misses easily
> could make a function slower in wild but you would measure that inlined
> function is faster when you would call it in tight loop.

Fair enough.  I'll push the patch if Konstantin shows that the
generated code is not drastically different.

Siddhesh

Attachment: pgpDIlbssAw2a.pgp
Description: PGP signature


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]