[Bug 89580] Implement a NIR -> vec4 pass

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=89580

--- Comment #36 from Jason Ekstrand <jason at jlekstrand.net> ---
(In reply to Eduardo Lima Mitev from comment #35)
> At this point, we have produced a new NIR pass which we called
> 'nir_lower_vec_and_coalesce', that will take the NIR shader right after SSA,
> and will try to coalesce registers by propagating vecN destination
> components to the instructions that define its sources. The pass will
> produce new reduced vecN instructions, containing the channels that were not
> propagated. Hence, this pass is compatible with lower_vec_to_movs(), and if
> in the future we decide to disable it, we can still use
> lower_vec_and_coalesce transparently.
> 
> Right now, the pass is very conservative about the conditions in which it
> will propagate register components. The reason is that at this point we are
> very interested in feedback about the whole idea, before we add more
> complexity to the pass.
> 
> The branch is here:
> https://github.com/Igalia/mesa/commits/elima/nir-vec4-quality
> 
> There are 2 patches only, one that adds the pass to NIR; and a second that
> just activates it on i965 for non-scalar shaders. Only the first one is
> interesting.
> 
> With this pass enabled, we get these shader-db VS results against
> vec4_visitor:
> 
> total instructions in shared programs: 1853747 -> 1762126 (-4.94%)
> instructions in affected programs:     1681255 -> 1589634 (-5.45%)
> helped:                                7751
> HURT:                                  9344
> GAINED:                                0
> LOST:                                  0
> 
> And these against NIR-vec4 as in current master:
> 
> total instructions in shared programs: 1801527 -> 1762126 (-2.19%)
> instructions in affected programs:     1156923 -> 1117522 (-3.41%)
> helped:                                10283
> HURT:                                  1281
> GAINED:                                0
> LOST:                                  0
>  
> It is not very impressive but there are a lot of clear opportunities left
> out, that would add some complexity that is maybe not worth at this point.
> 
> There are no piglit or dEQP regressions observed. 
> 
> What do you think?

I'm sorry I haven't had a chance to get back to you.  I've had a lot on my
plate the last few weeks.  I did take a brief look at it today but didn't
really give a full review.  Could you please send it to the list as at least an
RFC so I can more easily make comments inline?  Thanks!

I also took a stab at helping things out:

http://lists.freedesktop.org/archives/mesa-dev/2015-September/093737.html

I'm not 100% sure that I like it, but it seems to be effective:

> total instructions in shared programs: 1801527 -> 1761720 (-2.21%)
> instructions in affected programs:     978662 -> 938855 (-4.07%)
> helped:                                7924
> HURT:                                  497

With the total GLSL IR vs. NIR delta being

> total instructions in shared programs: 1853737 -> 1761720 (-4.96%)
> instructions in affected programs:     1686064 -> 1594047 (-5.46%)
> helped:                                7717
> HURT:                                  9942

Not sure how the two passes interact, I put your branch on top of mine and the
final GLSL IR vs. NIR results with both of them are

> total instructions in shared programs: 1853737 -> 1697708 (-8.42%)
> instructions in affected programs:     1695435 -> 1539406 (-9.20%)
> helped:                                11648
> HURT:                                  5540

That seems pretty decent to me.  Eventually, I do think we want to make
register coalescing in the back-end non-terrible, but these seem to help at the
very least.  If you get it sent to the list, I'll try and read your pass again
and wrap my head around it.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20150909/a314f5ef/attachment.html>