[Mesa-dev] i965: Broadwell using brw_eu_emit.c, part 2

[Chris Forbes]

I obviously can't test on Broadwell, but via eyeballs only:

For the series:

Reviewed-by: Chris Forbes <chrisf at ijw.co.nz>

On Tue, Aug 12, 2014 at 5:18 PM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> Hello,
>
> This series finishes getting Broadwell to use brw_eu_emit.c, and finally
> kills off gen8_*_generator.* and gen8_instruction.*.  The diffstat is
> quite encouraging:
>
>    16 files changed, 147 insertions(+), 4152 deletions(-)
>
> While doing this work, I've been diffing the assembly generated via both
> methods across an entire Piglit run.  Because of this, I was able to find
> a bunch of bugs in both generators, which was really useful.  I've fixed
> those before unifying the two generators, and Cc'd stable.
>
> As is, this series doesn't get the two generating identical assembly.
> There are a bunch of small differences that shouldn't matter, and some
> compaction-related optimizations Matt did in the brw_eu_emit.c code that
> I never did in the Gen8 code.  But, we want those - if I wanted to have
> zero differences, I'd disable them on gen >= 8, throw the switch, and then
> immediately revert those changes.  But it's pretty close.
>
> This series is available as the 'kill-generators' branch of
> git://people.freedesktop.org/~kwg/mesa/
>
> The 'kill-generators-zerodiff' branch is also available, and has
> zero differences in the generated assembly.  I'm not proposing to upstream
> those, but people can at least see the remaining differences.
>
> Patch 11 also improves performance of Trine 2 on Haswell by 45%, so...
> that's always nice, right? :)
>
> --Ken
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev