[Intel-gfx] [PATCH gen4asm 1/3] build: Don't use automake's maintainer modey

Damien Lespiau damien.lespiau at intel.com
Thu Jan 17 13:02:11 CET 2013


On Thu, Jan 17, 2013 at 10:56:28AM +0000, Beauchesne, Gwenole wrote:
> > On Fri, Jan 11, 2013 at 09:41:05AM +0100, Daniel Vetter wrote:
> > I'm not against doing that, it'll shed more light/review/etc on the assembler (I
> > assume patches would have to hit this list first).
> > 
> > Xiang, Homer, Gwenole, what do you think of that idea?
> 
> I think distributions would prefer smaller self-contained projects,
> that don't need too many dependencies. If we make intel-gen4asm into
> i-g-t, we'd need additional configure options to enable/disable
> certain components, e.g. testdisplay with cairo/libudev deps.
 
That point seems a bit moot to me, we already need to make some parts of
i-g-t optional and the assembler would be one instance of that. I'm also
willing to do the work to ensure that's done.

> While we are at re-organizing intel-gen4asm, I would also suggest to
> think about using Chris' libbrw as the actual backend to the GenX
> assembler. There is a need to generate code both at run-time and
> build-time, and it would be better to use the same assembler back-end,
> so that to simplify validation and enablement of future generations.
> IMHO, that'd a more sensible move than deciding on where to host
> intel-gen4asm. :)

Yes, I agree, and I think having that effort done in i-g-t makes sense
as I'd like to chip away at things like a tool to analyse the last batch
buffer after a hang for instance. Sharing that code better is definitely
what is motivating to look at that.

I've started a series to synchronize the brw_*[ch] files with Mesa's but
I have to admit I did not look at Chris' libbrw. The changes I have so
far are definitely a step towards the right direction though as Chris
seems to have started with more recent versions of Mesa's files than the
ones we have in the assembler.

Hopefully I'll have a first pass done this week. I also had a looks at
Chris' libbrw. It does look good but diverges a bit from Mesa now, what
I'd love to have is the same thing everywhere, but I guess that's a long
shot.

-- 
Damien



More information about the Intel-gfx mailing list