On Sun, 2002-07-07 at 22:19, Rogério Brito wrote:
> On Jul 07 2002, Jason E. Stewart wrote:
> > X now works beautifully on my pismo thanks to m3mirror, so I'm kinda
> > loathe to mess with it (why I'm still using X4.1 out of sid instead of
> > michel's 4.2 binaries).
>
> His binaries are quite non-intrusive. And, at least for my
> ibook2, they are far faster than the experimental 4.2 debs.
>
> Michel, did you compile your binaries with anything special
> that is not included in stock 4.2? What would explain the
> speed difference when, say, I am using xine?
Only the attached hack, which Peter Surda posted to the dri-devel list
some time ago. The problem with it has been that nobody knew why it has
any effect, let alone a positive one. :) I just had an idea though,
maybe without this, the X server spends a lot of time busy looping
waiting for enough DMA buffers to become available, stealing CPU cycles
from other processes for little gain. If this analysis is correct, a
better solution could be to emit HOSTDATA_BLT CCE packets directly with
BEGIN_RING() and friends, similar to
R128CCEScanlineCPUToScreenColorExpandFillPacket() (only in DRI CVS yet).
--
Earthling Michel Dänzer (MrCooper)/ Debian GNU/Linux (powerpc) developer
XFree86 and DRI project member / CS student, Free Software enthusiast
Index: r128_video.c
===================================================================
RCS file: /cvs/xc/programs/Xserver/hw/xfree86/drivers/ati/r128_video.c,v
retrieving revision 1.20
diff -p -u -r1.20 r128_video.c
--- r128_video.c 2001/10/02 11:44:16 1.20
+++ r128_video.c 2002/07/10 21:31:14
@@ -510,6 +510,8 @@ R128DMA(
int sizes[MAXPASSES], list[MAXPASSES];
drmDMAReq req;
+ usleep(1);
+
/* Verify conditions and bail out as early as possible */
if (!info->directRenderingEnabled)
return FALSE;
