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 2/2][BZ #16640] Remove strtok assembly implementation.


On Thu, Mar 06, 2014 at 12:51:07PM -0500, Carlos O'Donell wrote:
> On 03/06/2014 11:32 AM, OndÅej BÃlka wrote:
> > On Thu, Mar 06, 2014 at 11:20:13AM -0500, Carlos O'Donell wrote:
> >> On 02/27/2014 07:42 AM, OndÅej BÃlka wrote:
> >>> As followup this patch removes strtok assembly implementation as it is
> >>> around 2-4 times slower on sse4_2 capable machines which is a majority,
> >>> as previous benchtest demonstrated.
> >>>
> >>> You could try to gain some cycles on older machines by using ifuncs but
> >>> is a strtok important enough to complicate matters?
> >>>
> >>> OK to commit?
> >>>
> >>> 	* sysdeps/i386/i686/strtok.S: Remove.
> >>> 	* sysdeps/i386/i686/strtok_r.S: Likewise.
> >>> 	* sysdeps/i386/strtok.S: Likewise.
> >>> 	* sysdeps/i386/strtok_r.S: Likewise.
> >>> 	* sysdeps/x86_64/strtok.S: Likewise.
> >>> 	* sysdeps/x86_64/strtok_r.S: Likewise.
> >>
> >> Is this still the case with the oldest compiler we support?
> >>
> > I do not understand what are you reffering to.
> 
> If you remove the strtok assembly implemetnation, the build will
> fall back on use the C implementation e.g. string/strtok.c.
> That C implementation will have been compiled by whichever
> compiler you're using for your build. In glibc we support a wide
> range of compilers from gcc 4.4 and newer. Do the older compilers,
> specifically gcc 4.4, still generate good code for strtok.c which
> is faster than strtok.S?
> 
There is nothing to optimize, strtok code consist from two function
calls and some ifs and I do not see optimization oppurtinities here.


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