[Mesa-dev] [PATCH] radv: detect command buffers that do no work and drop them

[Dave Airlie]

On 10 February 2017 at 15:29, Jason Ekstrand <jason at jlekstrand.net> wrote:
> On Feb 9, 2017 8:25 PM, "Dave Airlie" <airlied at gmail.com> wrote:
>
> From: Dave Airlie <airlied at redhat.com>
>
> If a buffer is just full of flushes we flush things on command
> buffer submission, so don't bother submitting these.
>
> This will reduce some CPU overhead on dota2, which submits a fair
> few command streams that don't end up drawing anything.
>
>
> I wrote basically the same patch for our driver earlier this year when I was
> preparing for our GDC Dota 2 demo.  I noticed an improvement at the time but
> I'm pretty sure it was just because of the stalls we had due to relocations.
> Now that those stalls are gone, I'm not convinced it would do much.  Did you
> actually measure a performance improvement or was this just a little CPU
> usage reduction?

No it looks like just a CPU usage reduction, since radeon had a bit of kernel
overhead on each command buffer, reducing the total number can reduce the
overhead. Though mostly the overhead is when submit a lot of buffers with the
command buffer, which none of these do.

But I'm not against reducing CPU overhead in any form, the less we use
the better!

Dave.