On Wednesday, May 13, 2015 07:33:22 PM Francisco Jerez wrote:
>  - Nothing prevents you from "evaluating" the builder framework
>    independent from the tiling and format conversion code.
[snip]

I think you misunderstand - we all largely agree that the fs_builder
infrastructure is a good idea.  I think we've mostly evaluated it and
concluded that it's worth doing.

But the array_utils stuff - a new subclass of backend_reg -
emit_collect, emit_extract, emit_zip, emit_send, all of that is a bunch
of new infrastructure that's only used in your new code, and is hard to
evaluate.

> That's not nearly what I meant.  Of course whether and how we carry out
> the transition will still be open for discussion, that statement you
> quote was intended to show that I'm willing to do the hard work and not
> going to abandon the new infrastructure while the transition is half-way
> done (as Ken rudely insinuated on IRC).

That's good news.  In the past, many instances of "we'll fix it later"
have turned into "we never got around to it" - not from you, but from
other developers.  We've also been fixing bugs with subreg_offset for
about a year now...which was added for ARB_image_load_store (IIRC)...
and not accounted for in a lot of places (i.e. dead code elimination
thought writing g7.2 kills a previous write to any of g7).

So it's a real concern, no matter who's writing the code.

Attachment: signature.asc
Description: This is a digitally signed message part.

_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev

Reply via email to