[Mesa-dev] [PATCH v3 24/25] panfrost: Support batch pipelining

[Boris Brezillon]

On Fri, 6 Sep 2019 08:10:55 -0400
Alyssa Rosenzweig <alyssa at rosenzweig.io> wrote:

> > Now, if we go for the dep graph solution, that's probably a non issue,
> > since deps can be added at any point as long as they are described
> > before the flush happens.
> >
> > [snip]
> >
> > Thanks for the detailed explanation. I'll look into that. This being
> > said, I was wondering if we shouldn't merge this patch (after I
> > addressed your first comment maybe) before getting involved in a more
> > advanced solution (which I agree is what we should aim for).  
> 
> If it's alright, I would prefer to focus on patches 1-23; most of it
> looks wonderful so the few comments I had should be easily addressed for
> the v2.
> 
> Once all of that initial work is merged (and your revision queue and my
> review queue are cleared), we can circle back to this change.
> 
> I would prefer to go straight to a dep graph approach; this patch is a
> good intermediate step for testing the earlier patches in the series but
> given the extra complexity added for the draw flushing (which you
> mention is only needed with the non-graph solution), I don't know if we
> should merge.
> 
> Thoughts?

I'm definitely biased :-), but I do find the changes at hand not that
invasive: most of the logic is placed in helpers that are called in one
or 2 places. I mean, removing those explicit flushes when the time
comes shouldn't be too hard, and I do think it's one step in the right
direction even though it's not the perfect solution yet.

Anyway, I guess having patch 1 to 23 merged would already
significantly reduce my patch queue, and I'm definitely interested in
working on the dep graph solution, so I'm not strongly opposed to the
idea of dropping this patch.