Re: Scintilla GTK2 profiling results

[Biswapesh Chattopadhyay <biswapesh_chatterjee tcscal co in>]

I updated Pango to latest pango-1-2 branch at Soren's sugegstion, but
the figures remain roughly the same. Is there any way to check whether
I'm using the latest library or not ?

I'm a bit hesitant about migrating to HEAD since GNOME 2.2 (jhbuild) is
my work desktop.

About the latest run, KCachegrind says this:

pango_fc_font_map_load_fontset: 20.82 (cumulative)
	-> pango_fc_font_map_get_patterns (99%)
		-> FcFontSort (41.2%)
		-> pango_fc_default_substitute (32%)
		-> FcFontRenderPrepare (14.5%)
		-> uniquify_pattern (5.7%)

Also,

pango_font_get_glyph_extents: 13.5% (cumulative)
	-> pango_xft_font_get_glyph_extents (95%)
		-> XftGlyphExtents (70%)


Another (unrelated) point of botheration seems to be:
g_object_get_qdata: 12.6% (cumulative)
	-> g_type_check_instance_is_a (42%)
	-> g_datalist_id_get_data (31%)

BTW, Soren just pointed out (on #anjuta) that the teh profile looks
quite different when loading multiple files - I'll check that out too.

The sample, incidentally, is a C++ file with 3K LOC

On Wed, 2003-04-16 at 06:20, Owen Taylor wrote:
> On Wed, 2003-04-16 at 06:33, Biswapesh Chattopadhyay wrote:
> > Sorry for the cross-post, but I though it might be helpful is the issues
> > are discussed with feedback from both Neil as well as GTK2 developers.
> >
> > Scintilla GTK2 version with Pango seems quite a bit slower than the
> > plain GTK1 counterpart. So I did a quick profiling using KCachegrind to
> > identify the bottlenecks. The test case was to fire SciTE with a C++
> > source file as parameter and ensure that the file is fully loaded and
> > displayed with styling. The application was then closed immediately.
> >
> > Platform: RedHat 8.0 + XFree86 4.3 (build from CVS source) + GNOME 2.2
> > (built from jhbuild) + CVS scintilla + scite
> 
> Could you check that you are using the current CVS version of Pango
> (either pango-1-2 or HEAD branches) the numbers below look more like
> what I'd expect from pango-1.2.1.
> 
> pango_fc_font_map_get_patterns() shouldn't have much cumulative time
> it with the current code, unless you are testing with a very small
> sample of text, or possibly exceeding Pango's cache size.
> 
> [kcachegrind is indeed a nice tool. I spent quite a bit of time looking
> at pango-1.2.1 with it a month or two ago]
> 
> Regards,
>                                            Owen
> 
-- 
Biswa.