Refactor ShapesBench to make command-line version compile-time optional.

Refactor ShapesBench to make command-line version compile-time optional.

Create two instances using the new fully-specified constructor.

BUG=skia:
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1677253002

patch from issue 1677253002 at patchset 20001 (http://crrev.com/1677253002#ps20001)

[Brian Osman, 2016-02-09 20:24:57]

Description was changed from

==========
Refactor ShapesBench to make command-line version compile-time optional.

Create two instances using the new fully-specified constructor.

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

patch from issue 1677253002 at patchset 20001
(http://crrev.com/1677253002#ps20001)
==========

to

==========
Refactor ShapesBench to make command-line version compile-time optional.

Create two instances using the new fully-specified constructor.

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

patch from issue 1677253002 at patchset 20001
(http://crrev.com/1677253002#ps20001)
==========

[Brian Osman, 2016-02-09 20:24:57]

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

[Brian Osman, 2016-02-09 20:28:41]

I don't think we want to be using command-line parameters normally (given that
bots are usually the ones running the benchmarks, etc...). However, it's
potentially quite useful for experimenting to find new (interesting)
combinations of parameters.

So, I propose we do something like this... Make the command line settings
optional (at compile time), off by default. Then just have whatever combinations
we decide are interesting.

I just put in two (with small/large shapes to theoretically stress CPU vs. GPU),
but there's clearly more work to refine the set of benchmarks. I think this
gives us a good starting point, though?

[bsalomon, 2016-02-10 14:55:46]

It seems like a good idea to have variations that aren't mixed, perhaps a large
and small batch for each outer type. It would be useful to see how mixing
affects perf on different platforms. Once the shape rendering changes land we
expect that the systems that support it will show parity for mixed and non-mixed
while others wont. For the donut style I think interesting variations are inner
and outer are both rects and both are rrects.

https://codereview.chromium.org/1686673002/diff/1/bench/ShapesBench.cpp
File bench/ShapesBench.cpp (right):

https://codereview.chromium.org/1686673002/diff/1/bench/ShapesBench.cpp#newcode8
bench/ShapesBench.cpp:8: #include "Benchmark.h"
tiny nit, there's usually a \n after the copyright (and in other file)

https://codereview.chromium.org/1686673002/diff/1/gm/shapes.cpp
File gm/shapes.cpp (right):

https://codereview.chromium.org/1686673002/diff/1/gm/shapes.cpp#newcode177
gm/shapes.cpp:177: static GM* ShapesFactorySimple(void*) { return new
SimpleShapesGM(true); }
Let's use DEF_GM here

[Brian Osman, 2016-02-10 16:49:21]

Not sure if this is working.I tried to update the review with my latest commit
(addresses review feedback, adds a bunch of hard-coded parameter combinations).

https://codereview.chromium.org/1686673002/diff/1/bench/ShapesBench.cpp
File bench/ShapesBench.cpp (right):

https://codereview.chromium.org/1686673002/diff/1/bench/ShapesBench.cpp#newcode8
bench/ShapesBench.cpp:8: #include "Benchmark.h"
On 2016/02/10 14:55:46, bsalomon wrote:
> tiny nit, there's usually a \n after the copyright (and in other file)

Done.

https://codereview.chromium.org/1686673002/diff/1/gm/shapes.cpp
File gm/shapes.cpp (right):

https://codereview.chromium.org/1686673002/diff/1/gm/shapes.cpp#newcode177
gm/shapes.cpp:177: static GM* ShapesFactorySimple(void*) { return new
SimpleShapesGM(true); }
On 2016/02/10 14:55:46, bsalomon wrote:
> Let's use DEF_GM here

Done.

[Brian Osman, 2016-02-10 17:01:41]

Okay. Now the new patchset is uploaded...

[Chris Dalton, 2016-02-10 21:04:56]

On 2016/02/10 17:01:41, Brian Osman wrote:
> Okay. Now the new patchset is uploaded...

Thanks for working out the interesting set of benches for nanobench. I like your
set (didn't realize we could make so many new tests!)

I've added a few more modifications back over at
https://codereview.chromium.org/1677253002/ . I dropped your changes in directly
then uploaded my extra modifications as their own patchset, so if you look at
the delta from patchset 3 you should see just my changes.