[Mesa-dev] i965: Turn UBOs as push constants

[Ben Widawsky]

On Fri, Oct 09, 2015 at 06:06:30PM +0300, Abdiel Janulgue wrote:
> Ping! No thoughts/consensus if this approach is worth considering or
> not? The patches are getting stale. Should I re-base or just discard
> these optimisations?
> 

I am not the right person to ask, but I think it's certainly worth considering.
Like I said previously I do suspect there are benchmarks that would have a very
large benefit and that the shader-db results are underselling this.

Having an option to turn things off easily so we can one day perhaps use
heuristics to determine when to use this and when not seem like an important to
have if it isn't already there (maybe not an ENV var, but an easy knob in the
code).

> On 09/15/2015 08:59 PM, Ben Widawsky wrote:
> > On Tue, Sep 15, 2015 at 12:24:00PM +0300, Abdiel Janulgue wrote:
> >> Here's a more comprehensive shader-db run:
> >>
> >> total instructions in shared programs: 6394485 -> 6374865 (-0.31%)
> >> instructions in affected programs:     261322 -> 241702 (-7.51%)
> >> helped:                                3210
> >> HURT:                                  0
> >> GAINED:                                5
> >> LOST:                                  2
> >>
> >>
> >> - Abdiel
> > 
> > Thanks. The interesting thing to me isn't so much that you reduced instruction
> > count but hopefully that you reduced a whole bunch of pull loads. I suspect
> > there should be benchmarks out there which have a greater benefit than this
> > (which was why I asked), but I don't know what off the top of my head.
> > 
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > http://lists.freedesktop.org/mailman/listinfo/mesa-dev
> > 

-- 
Ben Widawsky, Intel Open Source Technology Center