Re: Improving glXSwapBuffers performace

[Simon T. <sim...@gm...>]

On Apr 11, 2005 4:34 PM, Brian Paul <bri...@tu...> wrote:
> Simon Toedt wrote:
> > Has anyone yet looked into ways to improve rendering performance? I
> > have noticed that glXSwapBuffers() is slow (30secs +/-5secs on a
> > P3/600Mhz) and mainly spends its time in PsOutImageBytes() which calls
> > sprintf() in a tight loop:
> > (gdb) where
> > #0  PsOut_OutImageBytes (self=0x8c91ed0, nBytes=3, bytes=0xbfffec7d
> > "") at psout.c:1488
> > #1  0x080d63af in PsPutScaledImage (pDrawable=0x8a69eb0,
> > pGC=0x8c6f640, depth=24, x=0, y=0, w=9321, h=13431, leftPad=0,
> > format=2, imageRes=0,
> >     pImage=0x4066e008 "") at PsArea.c:157
> > #2  0x080d6a79 in PsPutImage (pDrawable=0x8a69eb0, pGC=0x8c6f640,
> > depth=24, x=0, y=0, w=9321, h=13431, leftPad=0, format=2,
> >     pImage=0x4066e008 "") at PsArea.c:332
> > #3  0x08291717 in XMesaSwapBuffers (b=0x4063d008) at xm_api.c:2391
> > #4  0x0828dca9 in __MESA_swapBuffers (glxPriv=0x8c6f178) at xf86glx.c:736
> > #5  0x0825a00f in __glXSwapBuffers (cl=0x8a69c60, pc=0x8c90a40
> > "\220\v\003") at glxcmds.c:1340
> > #6  0x082300ff in __glXDispatch (client=0x8a647b0) at glxext.c:435
> > #7  0x08083b69 in Dispatch () at dispatch.c:449
> > #8  0x0806a790 in main (argc=8, argv=0xbffff2a4, envp=0xbffff2c8) at main.c:439
> > #9  0x400e48ae in __libc_start_main () from /lib/libc.so.6
> 
> Well, at line 2391 of xm_api.c, XMesaXPutImage() is being called.
> That's a macro:
> 
> #define XMesaPutImage(__d,__b,__gc,__i,__sx,__sy,__x,__y,__w,__h) \
> do { \
>      /* Assumes: Images are always in ZPixmap format */ \
>      (void) __d; \
>      if (__sx || __sy) /* The non-trivial case */ \
>         XMesaPutImageHelper(__d,__b,__gc,__i,__sx,__sy,__x,__y,__w,__h); \
>      ValidateGC(__b, __gc); \
>      (*__gc->ops->PutImage)(__b, __gc, ((XMesaDrawable)(__b))->depth, \
>                            __x, __y, __w, __h, 0, ZPixmap, \
>                            ((XMesaImage *)(__i))->data); \
> } while (0)
> 
> The PsPutImage() function is an internal server function in the Xprint
> module.  I have no idea why that would be getting called.

I've just had a look at the code and did some modifications which
improve the performance a lot - its now almost twice as fast. Can
anyone pls look at the patch, make sure its correct and then apply it?
Thanks!
-- 
Simon