[sisyphus] HyperZ on RV100

Konstantin A. Lepikhov =?iso-8859-1?q?lakostis_=CE=C1_anti-leasure=2Eru?=
Вт Ноя 9 01:23:11 MSK 2004


Hi!

JFYI. Также автору патча требуются тестеры с RV100.

----- Forwarded message from Dieter N?tzel <Dieter.Nuetzel на hamburg.de> -----

Date: Wed, 3 Nov 2004 22:42:02 +0100
From: Dieter N?tzel <Dieter.Nuetzel на hamburg.de>
To: Stephane Marchesin <marchesin на icps.u-strasbg.fr>,
        Keith Whitwell <keith на tungstengraphics.com>
Cc: dri-devel на lists.sourceforge.net,
        Jacek Pop??awski <jacekpoplawski на wp.pl>
Subject: Re: HyperZ on RV100
Organization: DN
User-Agent: KMail/1.7.1
In-Reply-To: <4188E46B.7000000 на icps.u-strasbg.fr>

Am Mittwoch, 3. November 2004 15:00 schrieb Stephane Marchesin:
> Dieter N?tzel wrote:
> >Am Dienstag, 2. November 2004 02:11 schrieb Stephane Marchesin:
> >>  Hi,
> >>
> >>The attached patches enable HyperZ on radeon (RV100 only for now).

OK, do you 'really" want some r200 numbers...;-)))

I've applied all three patches on Mesa / XFree86 DRI (the dri.freedesktop.org 
copy) CVS and fixed xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h.

First:
I had NO crash during all of my tests.

Second:
I see same (?) black horizontal line corruption as Jacek. (pitch?)
But only with texture enabled.
gears, ipers, gloss, quake3-smp, UT2k3/4

Vertical/horizontal black stippeled lines (blocks) in isosurf.

But texcyl is fine.

dual Athlon MP 1900+
1 GB DDR266, CL2
r200, 64 MB

(--) RADEON(0): Chipset: "ATI Radeon 8500 QL (AGP)" (ChipID = 0x514c)
(--) RADEON(0): Linear framebuffer at 0xe8000000
(--) RADEON(0): VideoRAM: 65536 kByte (128 bit DDR SDRAM)
(II) RADEON(0): AGP card detected

> >>People always ask me for an HyperZ benchmark, so :
> >>HyperZ gets me 45% more fps (48 fps -> 70 fps) in Quake3 four.dm68
> >>(Radeon 7000, athlon XP 2600).

Desktop: 1280x1024x24 (32)
All with AMD PM.

			window		fullscreen

640x480		198			199

1024x768		123			123

1280x1024	88			83


1280x1024 window mode was a little bit clipped (KDE window frame).
640x480 window mode _was_ ~164 fps before!!!

> >>It also speeds up glxgears from 450 fps to 1300 fps, which shows how
> >>useless glxears is as a benchmark :)

Even somewhat slower compared with my numbers from yesterday.

progs/demos> 7237 frames in  5.001 seconds = 1447.111 FPS
7780 frames in  5.000 seconds = 1556.000 FPS
7822 frames in  5.001 seconds = 1564.087 FPS
7790 frames in  5.000 seconds = 1558.000 FPS

without PM:

progs/demos> ./gears
9790 frames in  5.000 seconds = 1958.000 FPS
11608 frames in  5.000 seconds = 2321.600 FPS
11605 frames in  5.000 seconds = 2321.000 FPS
11603 frames in  5.000 seconds = 2320.600 FPS

> >>My case is probably a bit extreme though, since my video card is slow
> >>compared to my CPU.
> >
> >Why is glxgears so slow (without HyperZ) on your system?
> >rv100?
>
> Probably... glxgears is almost a benchmark for clearing speed IMO, and
> it seems RV100 clears are slow.

I see.

> >Do you have time for running viewperf-7.1.1/8.0.1?
> >I see depth (?) corruption (wrapped around a vertical axis) and text
> > fragments from KDE Konsole windows in the result images of 'ugs' when
> > alpha isn't used even without your patches.

gloss went from ~404 fps (with PM)

progs/demos> ./gloss
2321 frames in 5.001 seconds = 464.107 FPS
2597 frames in 5 seconds = 519.4 FPS
2605 frames in 5.001 seconds = 520.896 FPS
2591 frames in 5.001 seconds = 518.096 FPS
2610 frames in 5 seconds = 522 FPS

without PM:
progs/demos> ./gloss
2777 frames in 5 seconds = 555.4 FPS
3152 frames in 5 seconds = 630.4 FPS
3149 frames in 5 seconds = 629.8 FPS
3151 frames in 5 seconds = 630.2 FPS


doom3-demo
_playable_
640x480 window on my desktop (see above)
~27 fps


UT2k3
Not that much faster. (Not measurable.)
But was playable.


UT2k4
Much faster.
Now, playable.


Viewperf-7.1.1
~1 fps faster in most tests

But texture corruption in all tests.
Even worse when the 3D window is huge (!)
Full vertical dimension.
Maybe a pattern?
To few memory?

I have test images (ask me).

> Maybe when I have hyperz +
> color tiling working, it might be worth doing some benchmarks (color
> tiling brings another 10% increase in rendering speed).

We will see;-)

GREAT stuff!

-Dieter


-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_idU88&alloc_id065&op?ick
-- 
_______________________________________________
Dri-devel mailing list
Dri-devel на lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dri-devel

----- End forwarded message -----

-- 
WBR, Konstantin	      chat with ==>ICQ: 109916175
     Lepikhov,	      speak  to ==>JID: lakostis на jabber.org
aka L.A. Kostis       write  to ==>mailto:lakostis на pisem.net.nospam

...The information is like the bank... 			  (c) EC8OR



Подробная информация о списке рассылки Sisyphus