Re: [Mesa3d-dev] Re: [Mesa3d-cvs] CVS Update: Mesa (branch: trunk)

[Keith W. <ke...@tu...>]

Eric Anholt wrote:
> On Tue, 2005-05-31 at 07:45 +0100, Keith Whitwell wrote:
> 
>>Eric Anholt wrote:
>>
>>>CVSROOT:	/cvs/mesa
>>>Module name:	Mesa
>>>Repository:	Mesa/src/mesa/drivers/dri/radeon/
>>>Changes by:	anholt@gabe.	05/05/30 21:04:24
>>>
>>>Log message:
>>>  Bugzilla #2195: Convert the radeon driver to the t_vertex interface.  This cuts
>>>  about 200 lines from the code and 25k from the binary, while matching other
>>>  drivers more closely.  In the worst case (tcl_mode=0) it appears to have
>>>  a performance cost of 4.4% +/- 0.3% on quake3 (800x600 demofours, 1ghz p3,
>>>  rv200).  Tested on ut2004, ut, q3, projtex.
>>
>>Does this performance cost still apply since the recent change to 
>>runtime-generate x86 code for the t_vertex.c mechanism?
>>
>>Keith
> 
> 
> Yes, this was tested after the runtime generation, and those new sse
> functions did show up in the oprofile.
> 
> Would the loss of HAVE_ELTS matter?
> 

For q3, yes that would be an issue and would explain the above.  I can't 
remember why you would lose HAVE_ELTS going to t_vertex.c though - I'll 
have a quick look and see if it can be restored.

If the sse functions (like sse_movups etc) are showing up in the 
significant percentages, that's a little odd as they are the generators 
only and you'd expect q3 to only generate a few vertex functions...

Keith