Zou, Nanhai wrote:
>> -----Original Message-----
>> From: Keith Whitwell [mailto:[EMAIL PROTECTED]
>> Sent: 2007?8?31? 18:27
>> To: Nan hai Zou
>> Cc: MESA-Devel
>> Subject: Re: mesa: Changes to 'master'
>>
>> Nan hai Zou wrote:
>>>  src/mesa/drivers/dri/i965/brw_clip_line.c  |    7 +
>>>  src/mesa/drivers/dri/i965/brw_clip_state.c |    2
>>>  src/mesa/drivers/dri/i965/brw_clip_tri.c   |  105
>> ++++++++++++++++++++++++++++-
>>>  src/mesa/drivers/dri/i965/brw_clip_util.c  |   12 ---
>>>  4 files changed, 112 insertions(+), 14 deletions(-)
>>>
>>> commit diffs at http://gitweb.freedesktop.org/?p=mesa/mesa.git;a=summary
>>>
>>> New commits:
>>> commit b47c9f8c915ae4ca8c7fa5ee3b6b64f17c38b569
>>> Author: Zou Nan hai <[EMAIL PROTECTED]>
>>> Date:   Fri Aug 31 13:42:20 2007 +0800
>>>
>>>       optimize 965 clip
>>>       1. increase clip thread number to 2
>>>       2. do cliptest for -rhw
>> Nan hai Zou,
>>
>> Isn't it the case that the hardware bug that requires the -rhw
>> workaround is fixed in later versions of the hardware?  It would be good
>> to turn the workaround off in those cases - both in the clipper and the
>> vertex shader.
>>
>> Likewise, the various workarounds for send destination dependency
>> checking (eg. in brw_vs_emit.c emit_math1(), etc.) can be turned off in
>> later revisions and steppings of the hardware.  This is probably quite
>> important as I believe the workaround defeats built in latency-hiding
>> characteristics of the hardware.
>>
>> It's also worth noting that the hardware support for guard band clipping
>> isn't turned on in the current i965 driver.  Making use of this could
>> give a good boost to clipping performance...
>>
>> Otherwise, looking good.  It's nice to see the driver continuing to evolve.
>>
>> Keith
> 
> Hi Keith, 
>       Yes it is for the -rhw workaround. It just do a fast clip test before 
> the relatively slower full clip when -rhw workaround is needed.
>       I am testing the 965 driver performance with some 3D engines,
> I see clipper take a lot of  GPU time with Doom III demo and becoming a 
> bottleneck.
>       I guess it is because the heavily used stencil shadow algorithm in 
> Doom3 III engine, because it is very likely shadow volumes get clipped than 
> normal scenes.
>     For other game engines, I don't see clipper becoming a bottleneck maybe 
> because they are not using stencil shadow. Although the stencil shadow does 
> not look work quite correct on DOOM 3 demo, I think it is still worth to 
> optimize the clipper..., later I will fix the 3D effect issue.
>       Actually this change for -rhw does not optimize the clip performance in 
> DOOM III too much. I have only got around 20% speed-up with cliptest on -rhw. 
> Another change clip thread number to 2 in this patch gives another more than 
> 50% speed up.
>     I have tried guard band method, not seeing any performance gain with Doom 
> III...
> I  will continue checking the 3D pipeline to try to speed up the driver 
> performance.


Nan hai Zou,

Interesting.  It seems like the -rhw path is important for Doom 3, in 
which case it would be worthwhile to turn off all workarounds on 
hardware where they aren't required.  I think that around the C stepping 
of crestline and possibly the latest stepping of broadwater the bug is 
fixed??

Keith



-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >>  http://get.splunk.com/
_______________________________________________
Mesa3d-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev

Reply via email to