[Mesa-dev] [PATCH 1/2] nir/dead_cf: Don't crash on unreachable after-loop blocks

[Jason Ekstrand]

On Fri, Sep 2, 2016 at 10:39 AM, Kenneth Graunke <kenneth at whitecape.org>
wrote:

> On Thursday, September 1, 2016 7:00:01 PM PDT Connor Abbott wrote:
> > On Thu, Sep 1, 2016 at 6:19 PM, Jason Ekstrand <jason at jlekstrand.net>
> wrote:
> > > On Thu, Sep 1, 2016 at 12:41 PM, Connor Abbott <cwabbott0 at gmail.com>
> wrote:
> > >>
> > >> As-is, this change will make us delete trivially infinite loops (i.e.
> > >> loops with no break statement). The most likely scenario where we get
> > >> one of those (besides for silly Piglit tests) is a developer with an
> > >> accidental bug in their shader. In that case, it seems kinda mean to
> > >> then delete the entire loop and pretend like it doesn't exist. Maybe
> > >> add something like
> > >>
> > >> if (!after->imm_dom)
> > >>    return false;
> > >>
> > >> with a comment explaining what's going on? This would also help us
> > >> catch other places where we don't handle infinite loops correctly that
> > >> might've been hidden by this.
> > >
> > >
> > > I wrote that patch and kicked it off to Jenkins and it passed just
> fine.
> > > However, I don't know if I like it.  Matt and I were talking about this
> > > yesterday and his suggestion was to, if they have an infinite loop,
> delete
> > > the entire shader and draw pink.  That's more useful to developers than
> > > hanging their GPU.
> >
> > Hmm... I guess if you did that it would be wise to also print a
> > message somewhere the developer could see, since "uhh now the entire
> > thing is pink" probably won't help clarify what the problem is, even
> > if it shows that there is in fact a problem :)
> >
> > But I guess Matt is right that the danger is more theoretical than
> > practical. So go ahead and add my r-b to both patches with or without
> > the suggestion. Fake edges begone!
>

Thanks!  I went ahead and pushed the original version with your R-B.


> It would be great to be able to emit a warning to the shader info log,
> or a KHR debug message.  We've never hooked that up to NIR, but if we
> did, we could issue a warning and then basically do whatever we want
> with impunity.
>

I'm looking into this.  It shouldn't be terrible...
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160902/cb461eed/attachment.html>