[Mesa-dev] [PATCH 0/9] Skip automatic execsize for instructions with a width of 4

[Pohjolainen, Topi]

On Tue, Mar 15, 2016 at 07:44:43AM +0100, Iago Toral wrote:
> On Mon, 2016-03-14 at 11:15 -0700, Mark Janes wrote:
> > Iago Toral <itoral at igalia.com> writes:
> > 
> > > On Wed, 2016-03-09 at 09:54 +0200, Pohjolainen, Topi wrote:
> > >> On Mon, Mar 07, 2016 at 10:48:49AM +0100, Samuel Iglesias Gons?lvez wrote:
> > >> > Hello,
> > >> > 
> > >> > There is only one patch from this series that has been reviewed (patch
> > >> > 1).
> > >> > 
> > >> > Our plans is to start sending patches for adding fp64 support to i965
> > >> > driver in the coming weeks but they depend on these patches.
> > >> > 
> > >> > Can someone take a look at them? ;)
> > >> > 
> > >> > Sam
> > >> > 
> > >> > 
> > >> > On Thu, 2015-12-17 at 14:44 +0100, Samuel Iglesias Gonsálvez wrote:
> > >> > > Hello,
> > >> > > 
> > >> > > This patch series is a updated version of the one Iago sent last
> > >> > > week [0] that includes patches for gen6 too, as suggested by Jason.
> > >> > > 
> > >> > > We checked the gen9 code paths that work with a horizontal width of 4
> > >> > > and we think there won't be any regression on gen9... but we don't
> > >> > > have any gen9 machine to run piglit with these patches. Can someone
> > >> > > check it?
> > >> 
> > >> I rebased it and ran it through the test system, gen9 seems to be fine, I
> > >> only got one regression, and that was on old g965:
> > >> 
> > >> /tmp/build_root/m64/lib/piglit/bin/ext_framebuffer_multisample-accuracy all_samples srgb depthstencil -auto -fbo
> > >> Pixels that should be unlit
> > >>   count = 236444
> > >>   RMS error = 0.025355
> > >> Pixels that should be totally lit
> > >>   count = 13308
> > >>   Perfect output
> > >> The error threshold for unlit and totally lit pixels test is 0.016650
> > >> Pixels that should be partially lit
> > >>   count = 12392
> > >>   RMS error = 0.273876
> > >> The error threshold for partially lit pixels is 0.333000
> > >> Samples = 0, Result = fail
> > >
> > > I managed to borrow gen4 hardware to test this. According to glxinfo:
> > > Mesa DRI Mobile Intel GM45 Express Chipset
> > >
> > > but it does not fail for me... never. I left this test running in a loop
> > > for 5 minutes and it never failed. Also, looking at the problem
> > > described in the output you pasted above, it seems that the problem was
> > > related to precision calculations and it looks rather odd that our
> > > execsize patches could've compromised the precision of anything...
> > >
> > > Mark, would it be possible for you or someone else to run the piglit
> > > test mentioned by Topi above on gen4 hardware against this Mesa branch?:
> > > https://github.com/Igalia/mesa/tree/i965-fix-execsize
> > 
> > I ran this branch through our CI, and it passed all tests.  My
> > recollection is that ext_framebuffer_multisample-accuracy occasionally
> > fails on g965.  I would not attribute any failure of that test to this
> > branch.
> 
> Great, thanks Matt!
> 
> Topi: in that case I guess we can say there are no regressions in any
> gen, right? Do you have any other comments to the branch? We have merged
> the comments you made so far in case you want to double-check that we
> did not miss anything.

Right. We identified two piglit failures, one on hsw and this one on g965.
The former is intermittent and reproducible with current mesa master. Now
that we have answer for the latter also, I concur, your series should be
regression free.

I can take a look if you have a branch somewhere. I only had a few minor
comments in the first place and the series looks good to me. But I'll double
check.