On Thu, 10 Dec 2015 22:07:50 +0530 Amitesh Singh <singh.amit...@gmail.com> said:

> On Dec 10, 2015 4:32 AM, "Carsten Haitzler" <ras...@rasterman.com> wrote:
> >
> > On Wed, 9 Dec 2015 20:39:30 +0530 Amitesh Singh <singh.amit...@gmail.com>
> said:
> >
> > > On Dec 9, 2015 6:20 PM, "Tom Hacohen" <t...@osg.samsung.com> wrote:
> > > >
> > > > On 09/12/15 12:40, Mike Blumenkrantz wrote:
> > > > > I can confirm that Vyacheslav is correct, and even if he did not
> mention
> > > > > stringshare, this change is still wrong on a fundamental level.
> > > > >
> > > > > Please revert it.
> > > >
> > > > What really stands out in this change, is that I don't get the reason
> > > > for it. Why was it even done? At worse, it's 0.00000000001% slower
> than
> > > > after the change, but it's definitely not "more correct".
> > > >
> > > > Ami, how did you even get to do this fix? as it doesn't, and can't fix
> > > > anything.
> > > I was profiling tween _ state function (elm label slide ) and this check
> > > was getting called every time and it was always false . So I thought to
> > > reduce one more check. I think it might not be the case in other
> scenarios.
> > >
> > > But you guys are right.. I would revert it in few mins.
> >
> > did it speed things up? :) what did your profiling say where most of the
> cpu
> > time is going? did you try perf and narrow down lines? or callgrind? get
> some
> > figures per line of code?
> Yes. I used callgrind.

and what did it say?

> >
> > as such the check for the ptr first would assume stringshared strings are
> there
> > and thus a ptr cmp is a fast and quick way of checking before doing a full
> > string check. :) both ptrs will be in registers already there regardless
> so its
> > a cmp + jmp (and on some archs it might use conditional commands after
> this
> > point too so even avoid the jmp)
> >
> > > Thanks
> > > ami
> > > >
> > > > --
> > > > Tom.
> > > >
> > > > >
> > > > > On Wed, Dec 9, 2015 at 7:08 AM Vyacheslav Reutskiy <
> > > reutskiy....@gmail.com>
> > > > > wrote:
> > > > >
> > > > >> Hello,
> > > > >>
> > > > >> I'm not sure that this changes is correct. The 'state_name' can be
> > > > >> pointer to eina_stringshare and comparing the two pointers faster
> > > > >> than strcmp. This fix looks doubtful.
> > > > >>
> > > > >> --
> > > > >> Viacheslav Reutskiy (rimmed)
> > > > >>
> > > > >> On Wed, Dec 9, 2015 at 12:20 PM, Amitesh Singh <
> amitesh...@samsung.com>
> > > > >> wrote:
> > > > >>
> > > > >>> ami pushed a commit to branch master.
> > > > >>>
> > > > >>>
> > > > >>>
> > > > >>
> > >
> http://git.enlightenment.org/core/efl.git/commit/?id=c892a1cb714fed496cbf5568c4d43880b6fb67b2
> > > > >>>
> > > > >>> commit c892a1cb714fed496cbf5568c4d43880b6fb67b2
> > > > >>> Author: Amitesh Singh <amitesh...@samsung.com>
> > > > >>> Date:   Wed Dec 9 15:46:41 2015 +0530
> > > > >>>
> > > > >>>      edje: calc - remove pointer comparison while finding part
> desc
> > > > >>>
> > > > >>>      Only strcmp comparision is realiable.
> > > > >>>      @fix
> > > > >>> ---
> > > > >>>   src/lib/edje/edje_calc.c | 3 +--
> > > > >>>   1 file changed, 1 insertion(+), 2 deletions(-)
> > > > >>>
> > > > >>> diff --git a/src/lib/edje/edje_calc.c b/src/lib/edje/edje_calc.c
> > > > >>> index b0742cf..c06e3ac 100644
> > > > >>> --- a/src/lib/edje/edje_calc.c
> > > > >>> +++ b/src/lib/edje/edje_calc.c
> > > > >>> @@ -460,8 +460,7 @@ _edje_part_description_find(Edje *ed,
> > > Edje_Real_Part
> > > > >>> *rp, const char *state_name
> > > > >>>        {
> > > > >>>           d = ep->other.desc[i];
> > > > >>>
> > > > >>> -        if (d->state.name && (d->state.name == state_name ||
> > > > >>> -                              !strcmp(d->state.name,
> state_name)))
> > > > >>> +        if (d->state.name && (!strcmp(d->state.name,
> state_name)))
> > > > >>>             {
> > > > >>>                if (!approximate)
> > > > >>>                  {
> > > > >>>
> > > > >>> --
> > > > >>>
> > > > >>>
> > > > >>>
> > > > >>
> > > > >>
> > >
> ------------------------------------------------------------------------------
> > > > >> _______________________________________________
> > > > >> enlightenment-devel mailing list
> > > > >> enlightenment-devel@lists.sourceforge.net
> > > > >> https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
> > > > >>
> > > > >
> > >
> ------------------------------------------------------------------------------
> > > > > _______________________________________________
> > > > > enlightenment-devel mailing list
> > > > > enlightenment-devel@lists.sourceforge.net
> > > > > https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
> > > > >
> > > >
> > > >
> > > >
> > >
> ------------------------------------------------------------------------------
> > > > _______________________________________________
> > > > enlightenment-devel mailing list
> > > > enlightenment-devel@lists.sourceforge.net
> > > > https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
> > >
> ------------------------------------------------------------------------------
> > > _______________________________________________
> > > enlightenment-devel mailing list
> > > enlightenment-devel@lists.sourceforge.net
> > > https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
> > >
> >
> >
> > --
> > ------------- Codito, ergo sum - "I code, therefore I am" --------------
> > The Rasterman (Carsten Haitzler)    ras...@rasterman.com
> >
> ------------------------------------------------------------------------------
> _______________________________________________
> enlightenment-devel mailing list
> enlightenment-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/enlightenment-devel
> 


-- 
------------- Codito, ergo sum - "I code, therefore I am" --------------
The Rasterman (Carsten Haitzler)    ras...@rasterman.com


------------------------------------------------------------------------------
_______________________________________________
enlightenment-devel mailing list
enlightenment-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/enlightenment-devel

Reply via email to