[Mesa-dev] [RFC] Concrete proposal to split classic

[Kenneth Graunke]

On Monday, March 22, 2021 3:15:30 PM PDT Dylan Baker wrote:
> Hi list,
> 
> We've talked about it a number of times, but I think it's time time to
> discuss splitting the classic drivers off of the main development branch
> again, although this time I have a concrete plan for how this would
> work.
> 
> First, why? Basically, all of the classic drivers are in maintanence
> mode (even i965). Second, many of them rely on code that no one works
> on, and very few people still understand. There is no CI for most of
> them, and the Intel CI is not integrated with gitlab, so it's easy to
> unintentionally break them, and this breakage usually isn't noticed
> until just before or just after a release. 21.0 was held up (in small
> part, also me just getting behind) because of such breakages.
> 
> I konw there is some interest in getting i915g in good enough shape that
> it could replace i915c, at least for the common case. I also am aware
> that Dave, Ilia, and Eric (with some pointers from Ken) have been
> working on a gallium driver to replace i965. Neither of those things are
> ready yet, but I've taken them into account.
> 
> Here's the plan:
> 
> 1) 21.1 release happens
> 2) we remove classic from master
> 3) 21.1 reaches EOL because of 21.2
> 4) we fork the 21.1 branch into a "classic-lts"¹ branch
> 5) we disable all vulkan and gallium drivers in said branch, at least at
>    the Meson level
> 6) We change the name and precidence of the glvnd loader file
> 7) apply any build fixups (turn of intel generators for versions >= 7.5,
>    for example
> 8) maintain that branch with build and critical bug fixes only
> 
> This gives ditros and end users two options.
> 1) then can build *only* the legacy branch in the a normal Mesa provides
>    libGL interfaces fashion
> 2) They can use glvnd and install current mesa and the legacy branch in
>    parallel
> 
> Because of glvnd, we can control which driver will get loaded first, and
> thus if we decide i915g or the i965 replacement is ready and turn it on
> by default it will be loaded by default. An end user who doesn't like
> this can add a new glvnd loader file that makes the classic drivers
> higher precident and continue to use them.
> 
> Why fork from 21.1 instead of master?
> 
> First, it allows us to delete classic immediately, which will allow
> refactoring to happen earlier in the cycle, and for any fallout to be
> caught and hopefully fixed before the release. Second, it means that
> when a user is switched from 21.1 to the new classic-lts branch, there
> will be no regressions, and no one has to spend time figuring out what
> broke and fixing the lts branch.
> 
> When you say "build and critical bug fixes", what do you mean?
> 
> I mean update Meson if we rely on something that in the future is
> deprecated and removed, and would prevent building the branch or an
> relying on some compiler behavior that changes, gaping exploitable
> security holes, that kind of thing.
> 
> footnotes
> ¹Or whatever color you like your bikeshed

Hi Dylan,

I largely like this plan.  i915 and r200 and novueau_vieux definitely
need to be forked off, to keep them working as we do core refactors.

In the past, we weren't working on core much, and so they largely kept
working with some pain, but not -too- much effort.  But these days,
we're seeing a lot of work from Marek and others to clean up and rework
a lot of core drawing code, state upload code, and so on.  I remember
all the work Tim did to rework uniform handling and ancient classic was
a pain point for him as well.  I had to track down like 5 overlapping
Piglit regressions on i915 this release cycle, just to get the driver
working...no worse than it was before.  And i915 is tested regularly
in the Intel CI.  r200 isn't tested regularly.  Doubtful about vieux.

I had thought about also forking other old drivers like i915g or r300g.
But it sounds like there's still some interest in i915g, and those
tend to get fixed up as the Gallium infrastructure is still actively
being maintained (unlike things like swrast and tnl).  So I guess we
can leave those in mainline.

I'm hesitant about i965.  One thing I will say is that, if i965 is
included in the plan, there would probably be more interest in working
on that branch than mere "critical bug fixes" as you defined.  We might
discover new games that don't work; people might write bug fixes for
i965, at which point we should merge and ship them.  But we could still
do on-demand releases when enough interesting bug fixes have piled up,
rather than doing them on a schedule.  That should hopefully not be
too burdensome.

While a part of me hates the idea of forking i965 off, I think it may
actually be the best call.  We haven't done any new interesting feature
development on those platforms in ages, and they're complete in terms of
OpenGL spec support.  We aren't spending any time optimizing performance
on those platforms.  They're pretty much in bug-fix-only mode already.
Whether that bug fixing happens on master or on a "classic-lts" branch
doesn't matter; it should be basically the same for our users.

I think the best thing for users of those platforms would be for us to
finish the new Gallium driver and re-add support to mainline.  The new
Gallium driver would have massively better CPU overhead, gain all the
new core infrastructure improvements we're making, and also add OpenGL
compatibility profile support.  i965...isn't ever going to get most of
those, even if it stays in tree.  Your glvnd plan already handles this
nicely - when crocus is ready, it can use that instead.

Removing classic and i965 would be great for the other active driver
developers, as radeonsi/nouveau/panfrost/freedreno/etnaviv/lima/...
could receive new optimizations as we remove extra layers that add
overhead.  That's great for users of those newer GPUs.

...

One extra thought: can we also fork off anv Gen7.x support at the same
time?  If distros are already going to be building i965 for Gen7.x from
that branch, building Vulkan from there should be easy as well.

Jason and I have been talking about requiring softpin on Gen8+ in MR
!4431.  I have absolutely no regrets about making iris softpin-only;
it's been so much simpler and easier to work with.  anv still has a ton
of leftover complexity from supporting relocations for Gen7.x - and IMO
it's actually more complex than the support in i965 ever was.  We could
clean up a lot of cruft---and reduce the CPU overhead of the driver---if
we split it off.  The timeframes align---we were going to drop relocs
on Gen8+ after 21.1 forked.  If we forked Gen 7.x, we could just drop
them altogether.  It would be so nice.

--Ken
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: This is a digitally signed message part.
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20210322/1e29d6d8/attachment-0001.sig>