[Mesa-dev] [PATCH 3/6] i965: always run the post-RA scheduler

[Jason Ekstrand]

On Sat, Oct 3, 2015 at 11:32 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
> On Sat, Oct 3, 2015 at 2:28 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>> On Sat, Oct 3, 2015 at 11:26 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>>> On Sat, Oct 3, 2015 at 2:13 PM, Jason Ekstrand <jason at jlekstrand.net> wrote:
>>>> On Fri, Oct 2, 2015 at 2:37 PM, Connor Abbott <cwabbott0 at gmail.com> wrote:
>>>>> Before, we would only do scheduling after register allocation if we
>>>>> spilled, despite the fact that the pre-RA scheduler was only supposed to
>>>>> be for register pressure and set the latencies of every instruction to
>>>>> 1. This meant that unless we spilled, which we rarely do, then we never
>>>>> considered instruction latencies at all, and we usually never bothered
>>>>> to try and hide texture fetch latency. Although a later commit removes
>>>>> the setting the latency to 1 part, we still want to always run the
>>>>> post-RA scheduler since it's able to take the false dependencies that
>>>>> the register allocator creates into account, and it can be more
>>>>> aggressive than the pre-RA scheduler since it doesn't have to worry
>>>>> about register pressure at all.
>>>>>
>>>>> XXX perf data
>>>>
>>>> Test                   master      post-ra-sched     diff       %diff
>>>> bench_OglPSBump2       396.730     402.386           5.656      +1.400%
>>>> bench_OglPSBump8       244.370     247.591           3.221      +1.300%
>>>> bench_OglPSPhong       241.117     242.002           0.885      +0.300%
>>>> bench_OglPSPom         59.555      59.725            0.170      +0.200%
>>>> bench_OglShMapPcf      86.149      102.346           16.197     +18.800%
>>>> bench_OglVSTangent     388.849     395.489           6.640      +1.700%
>>>> bench_trex             65.471      65.862            0.390      +0.500%
>>>> bench_trexoff          69.562      70.150            0.588      +0.800%
>>>>
>>>> Unfortunately, neither of the unigin benchmarks (heaven or vally)
>>>> seemed to render correctly.  I just got white on both master and your
>>>> branch.  Not sure if we have a bug or if they just weren't running
>>>> right.  In any case, ministat didn't notice any difference in them.
>>>
>>> I believe they're called "features" :) Try with disable_blend_func_extended=true
>>
>> I pulled in a more recent drirc and am re-running those two.
>
> They're not in the latest drirc... probably should be added back in.

I used the drirc corresponding to the mesa commits I was testing.
Given that it's actually rendering stuff, I'm going to say it's
probably ok.

--Jason