Keith Whitwell wrote:
>
> Brian Paul wrote:
> >
> > The following TrueColor (and DirectColor) visuals use optimized code
> > which directly pokes pixels into the XImage instead of using XPutPixesl:
> >
> > 32bpp, Rmask=0x0000ff, Gmask=0x00ff00, Bmask=0xff0000
> > 32bpp, Rmask=0xff0000, Gmask=0x00ff00, Bmask=0x0000ff
> > 24bpp, Rmask=0xff0000, Gmask=0x00ff00, Bmask=0x0000ff
> > 16bpp, Rmask=0xf800, Gmask=0x7e0, Bmask=0x001f (5-6-5 mode)
> > 8bpp, Rmask=0xe0, Gmask=0x1c, Bmask=0x3 (HP Color Recovery visual only)
> >
> > Perhaps you're using a 16bpp, depth=15 (5-5-5) visual?
> >
>
> Hmm... Looks like something is lost in the translation:
>
> xdpyinfo puts me in (5-6-5) mode with the same masks as above:
>
> default visual id: 0x23
> visual:
> visual id: 0x23
> class: TrueColor
> depth: 16 planes
> available colormap entries: 64 per subfield
> red, green, blue masks: 0xf800, 0x7e0, 0x1f
> significant bits in color specification: 6 bits
>
> Profiling isosurf gives:
>
> 83.33 3.15 3.15 smooth_TRUEDITHER_z_triangle
> 9.79 3.52 0.37 shade_fast_rgba_one_sided
> 1.59 3.58 0.06 glVertex3fv
> 1.32 3.63 0.05 gl_x86_cliptest_points4
> 0.79 3.66 0.03 glNormal3fv
>
> Examining smooth_TRUEDITHER_z_triangle shows:
>
> if (z < zRow[i])
> unsigned long p;
> PACK_TRUEDITHER( p, xx, yy, FixedToInt(ffr),
> FixedToInt(ffg), FixedToInt(ffb) );
> XMesaPutPixel( img, xx, yy, p );
> zRow[i] = z;
> }
>
> I guess that dithering hasn't been brought up to speed with the ability
> to write directly into the XImage. Turning off dither gives a speedup,
> but it's not huge. Isosurf is probably dominated by the setup rather
> than fill of these triangles.
Yeah, I just checked, 16bpp w/ dithering does use XPutPixel. When
dithering is disabled XPutPixel is avoided. At one time, dithering
wasn't supported in 16bpp mode. I guess I never optimized it when
I enabled it.
A better benchmark would be the demos/trispd program with a large
triangle size (-size option).
I'll put this optimization on my to-do list.
-Brian
----------------------------------------------------------------------
Brian Paul Avid Technology / Softimage [EMAIL PROTECTED]
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
