Re: [directfb-dev] GDK-DirectFB Patches

[Attilio Fiandrotti <attilio fiandrotti gmail com>]

I can take care of committing all the gtk patches along with the 
previous patch for gdk events which i already tested and seems to be ok.
One question about the code inside

#ifndef NO_EXPERIMENTS
...
#endif

shall it be committed and activated, committed but not activated or not 
committed at all?

sincerely

Attilio

Denis Oliver Kropp wrote:
> Hi,
>
> here are a bunch of patches which improve the overall speed of GDK-DirectFB a lot
> and also contain a few fixes. In our special test case the load was reduced by 70%,
> speeding up tab switching 3.3 times!
>
> The patches should be applied in the following order (probably skipping the two HACKs):
>
> window_flip_group.patch
> no_background_pixmap_fix.patch
> blit_after_cairo_fix.patch
> rect_clip_fix.patch
> rgb16_default.patch
> HACK_no_clear.patch
> fast_blend.patch
> opt_clip_region_and_fill_rects.patch
> no_state_resets.patch
> opt_temp_region_etc.patch
> opt_temp_region2.patch
> HACK_no_clear_area.patch
>
>
> Another patch is for Cairo:
>
> cairo_show_glyphs.patch
>
>
> Each patch header has a description, here's an overview:
>
> - window_flip_group.patch
> GDK-DIRECTFB: During update processing, don't flip immediately, but accumulate a flipping region using the DirectFB
> update manager and flush all Flip()s at the end of all painting. This results in a UI update In One Go [tm]!
>
>
> - no_background_pixmap_fix.patch
> GDK-DIRECTFB: Fix setting of NULL pixmaps resulting in a GDK_NO_BG setting. Probably GDK had an
> API change at some point in time.
>
>
> - blit_after_cairo_fix.patch
> GDK-DIRECTFB: Fix wrong blitting flags being used by GDK after Cairo changed the state.
>
>
> - rect_clip_fix.patch
> GDK-DIRECTFB: Fix clipping rectangle for filling rectangles after Cairo has modified it.
>
>
> - rgb16_default.patch
> GDK-DIRECTFB: Use RGB16 format by default.
>
>
> - HACK_no_clear.patch
> GDK-DIRECTFB: Experimentally comment out the clear in gdk_window_impl_directfb_begin_paint_region().
>
> This speeds up the test app without visual impact, but it might harm other (traditional?) apps.
>
>
> - fast_blend.patch
> GDK-DIRECTFB: Start with our own implementation of gdk_drawable_draw_pixbuf() based on the original.
>
> First modifications are:
> - No scratch image used, saving two copies.
> - Replaced composite_565() implementation by an optimized version doing 32 bit wise I/O.
>
> Clipping might need to be fixed.
>
>
> - opt_clip_region_and_fill_rects.patch
> GDK-DIRECTFB: Another small improvement by using FillRectangles() instead of FillRectangle().
>
> Avoid a temporary memory allocation in gdk_directfb_clip_region().
>
>
> - no_state_resets.patch
> GDK-DIRECTFB: Don't set back states after drawing/blitting, just make sure they're correct before being used.
>
>
> - opt_temp_region_etc.patch
> GDK-DIRECTFB: Introduce temp_region functions and avoid temporary memory allocation in a lot of places.
>
>
> - opt_temp_region2.patch
> GDK-DIRECTFB: Changed remaining places where temp_region can be used. Added caching of the clip region during painting.
>
>
> - HACK_no_clear_area.patch
> GDK-DIRECTFB: Experimentally comment out the clear in _gdk_windowing_window_clear_area[_e] and don't invalidate in
> show_window_internal().
>
> This speeds up the test app without visual impact, but it might harm other (traditional?) apps.
>
>
> - cairo_show_glyphs.patch
> CAIRO-DIRECTFB: Use DirectFB for show_glyphs() even if it is unaccelerated.
>
> The software fallback in DirectFB is well optimized.
>
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> directfb-dev mailing list
> directfb-dev directfb org
> http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-dev