[Mesa-dev] Fixed function fragment shader to GLSL

[Kenneth Graunke]

On 10/18/2011 11:39 AM, Jakob Bornecrantz wrote:
> ----- Original Message -----
>> Here's this patch series again, part of the kill-Mesa-IR goal we have
>> for next release.  It's been no regressions for me for a while, I was
>> just trying to track down a 2% performance regression on gen6.  It
>> turns out that on my gen4 system, it's no performance regression, and
>> on gen7 (at least according to testing with the followon patches to
>> use the new coegen), it's a ~4% performance win.
>>
>> While many things were learned in the process, the cause of the 2%
>> loss appears to come down to some GPU instruction pipelining issue
>> with lots of handwaving on my part.  When using the new backend, we
>> generate like 1/3 fewer instructions (after accounting for the 8/16
>> compile).
>>
>> I'm not including the patches for using the new backend quite yet, as
>> there's a functional regression on gm45 in the new backend that I
>> need to track down first.
> 
> Hi, I know I told you to wait till Monday but I didn't get around to
> take a look at the changes, sorry about that, tho I would have
> appreciated that you at least pinged me before you pushed the changes.
> 
> While you say tested the changes I assume you didn't test any Gallium
> driver (which you probably should since the only real driver in use
> today that isn't Gallium is i965).
> 
> I'm also assuming this patch series will have the previous set of
> regressions as I outlined the last time this was up for discussion.
> 
> http://article.gmane.org/gmane.comp.video.mesa3d.devel/24837
> 
> Also I would appreciated these kind of change the whole world changes
> to have a larger Acked-by/Reviewed-by base the just one other guy
> using the same test set as the author.
> 
> Would it be possible to do the same resolution as before until we can
> actually discuss this change.
> 
> Cheers, Jakob.

We can't stall this forever...it's critical for our future work.

You had two weeks to comment (10/06 to 10/18); that should have been
plenty of time.  Not to mention the fact that this was originally
proposed clear back in January...this should come as no surprise.

Also, the issues you raised 9 months ago about code generation most
likely no longer apply: Gallium now uses Bryan Cain's new
st_glsl_to_tgsi translator, which includes its own optimization passes.
 At any rate, a new investigation is needed.

I'm strongly opposed to reverting this again, especially for unbounded
amounts of time like before.  This is the future.  If there are
remaining issues, let's just go ahead and fix them...