Add tileSKP option to nanobench

Add tileSKP option to nanobench

Adds an option to draw SKPs untiled, directly into the nanobench
canvas.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1541983002

[Chris Dalton, 2015-12-21 22:08:46]

Description was changed from

==========
Add tileSKP option to nanobench

Adds an option to draw SKPs untiled, directly into the nanobench
canvas.

BUG=skia:
==========

to

==========
Add tileSKP option to nanobench

Adds an option to draw SKPs untiled, directly into the nanobench
canvas.

BUG=skia:
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

[Chris Dalton, 2015-12-21 22:15:29]

cdalton@nvidia.com changed reviewers:
+ bsalomon@google.com, joshualitt@google.com

[bsalomon, 2015-12-22 17:04:25]

https://codereview.chromium.org/1541983002/diff/20001/bench/SKPBench.cpp
File bench/SKPBench.cpp (right):

https://codereview.chromium.org/1541983002/diff/20001/bench/SKPBench.cpp#newc...
bench/SKPBench.cpp:142: // Ensure the GrContext doesn't batch across draw loops.
It seems like now we will not flush after the last loop.

Also, the previous code only flushes after a set of loops. The comment is kind
of ambiguous so I can't tell if that was intentional or not. I'd kind of like
Josh to weigh in but he won't be back in the office until 12/28.

[Chris Dalton, 2015-12-22 18:47:05]

https://codereview.chromium.org/1541983002/diff/20001/bench/SKPBench.cpp
File bench/SKPBench.cpp (right):

https://codereview.chromium.org/1541983002/diff/20001/bench/SKPBench.cpp#newc...
bench/SKPBench.cpp:142: // Ensure the GrContext doesn't batch across draw loops.
On 2015/12/22 17:04:25, bsalomon wrote:
> Also, the previous code only flushes after a set of loops.

Oops :-\ the previous code was wrong.

> It seems like now we will not flush after the last loop.

I added that flushing code a couple months back:
https://codereview.chromium.org/1427533002/ . Before, it didn't flush at all.
The idea was that without the flushing, Gr could cheat and reorder/batch
together draws that originated from different loops.

I purposely removed the flush after the final loop because nanobench itself
appears to take care of it:


    canvas = target->beginTiming(canvas);
    bench->draw(loops, canvas);
    if (canvas) {
        canvas->flush();
    }
    target->endTiming();


(Is it possible for "canvas->flush()" to not flush the GrContext?)


Do you think I should split this into two different commits? One to fix the
flushing thing and one to add the --tileSKP flag?

[bsalomon, 2016-01-04 14:51:01]

On 2015/12/22 18:47:05, Chris Dalton wrote:
> https://codereview.chromium.org/1541983002/diff/20001/bench/SKPBench.cpp
> File bench/SKPBench.cpp (right):
> 
>
https://codereview.chromium.org/1541983002/diff/20001/bench/SKPBench.cpp#newc...
> bench/SKPBench.cpp:142: // Ensure the GrContext doesn't batch across draw
loops.
> On 2015/12/22 17:04:25, bsalomon wrote:
> > Also, the previous code only flushes after a set of loops.
> 
> Oops :-\ the previous code was wrong.
> 
> > It seems like now we will not flush after the last loop.
> 
> I added that flushing code a couple months back:
> https://codereview.chromium.org/1427533002/ . Before, it didn't flush at all.
> The idea was that without the flushing, Gr could cheat and reorder/batch
> together draws that originated from different loops.
> 
> I purposely removed the flush after the final loop because nanobench itself
> appears to take care of it:
> 
> 
>     canvas = target->beginTiming(canvas);
>     bench->draw(loops, canvas);
>     if (canvas) {
>         canvas->flush();
>     }
>     target->endTiming();
> 
> 
> (Is it possible for "canvas->flush()" to not flush the GrContext?)
> 
> 
> Do you think I should split this into two different commits? One to fix the
> flushing thing and one to add the --tileSKP flag?

Yeah, I think that is a good idea. I think the flush per loop makes sense for
many SKPs because they represent a full frames worth of work. Some SKPs are
really small and more like microbenchs and there it probably makes sense to
allow batching across loops. Or maybe those SKPs should be deleted... or maybe
SKPs should have an inner repeat count that is run loops number of times.
Anyway, I think it's a separate issue from adding tiling.

[Chris Dalton, 2016-02-22 22:37:13]

Turns out we can do basically the same thing by tweaking GPUbenchTile{W,H}.
Abandoning.

[Chris Dalton, 2016-02-22 22:37:18]

Description was changed from

==========
Add tileSKP option to nanobench

Adds an option to draw SKPs untiled, directly into the nanobench
canvas.

BUG=skia:
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========

to

==========
Add tileSKP option to nanobench

Adds an option to draw SKPs untiled, directly into the nanobench
canvas.

BUG=skia:
GOLD_TRYBOT_URL=
https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&is...
==========