[Mesa-dev] megadrivers series

[Matt Turner]

On Mon, Sep 30, 2013 at 1:44 PM, Eric Anholt <eric at anholt.net> wrote:
> Here are the megadrivers changes, after the prep series I posted earlier.
> A few tiny updates to the prep series are available in my tree as
> "megadriver-prep" and this series is available as "megadrivers-5"
>
> FPS improvement on GLB2.7 with INTEL_NO_HW=1: 2.61061% +/- 1.16957% (n=50)
>
> One question I have is whether the hardlinks are going to cause problems
> for packaging.  I noticed that when I went and stripped the binaries
> trying to do a space comparison, I of course got brand new inodes each
> taking up their own set of disk space.  I do really like how hardlinks end
> up for installing on my test systems -- a single binary I can move around
> however I need.
>
> video from the talk I gave at XDC:
> http://www.youtube.com/watch?v=0fJq-2haT3Y
>
> I think Emil has been looking at doing the gallium side of things, so I
> haven't pushed forward with that.
>
> Note that the megadriver build does require an updated loader. I've done
> EGL and GLX, but the xserver still needs updating.  If I get some acks on
> the ABI I chose, I'll go do that.

In general I have some reservations about megadrivers. Mostly about
accepting the potential maintenance burden for something that is
arguably a distribution problem.

Aside: while I agree it's kind of rude to ship the same 50 MB
libmesa.a linked statically in multiple drivers, I have a hard time
buying the argument about needing to fit onto CDs (now 1 GB USB
drives), in part because I literally cannot buy 1 GB drives most
places. Anyway.

On the other hand, I love getting rid of libdricore and its
performance impact. Also the bit about not having to build src/glsl
and src/mesa/main twice.

If megadrivers is what we need to do to get rid of libdricore, okay.

I sent a few small comments, and with those fixed all but patch 10 are

Reviewed-by: Matt Turner <mattst88 at gmail.com>

and patch 10 is

Acked-by: Matt Turner <mattst88 at gmail.com>