Stroked line path bench

Add a wide stroked line path benchmark to bench.  It uses the same stroke width and endcap style as the stroked curved path.

[bsalomon, 2012-01-11 19:15:25]

On 2012/01/11 19:01:54, Stephen White wrote:

You're one bench up on me. :) It looks fine to me but I'll defer to Mike since
I'm not familiar with it.

[reed1, 2012-01-11 21:23:25]

lgtm with simplification-request and rand-seed-question.

http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp
File bench/PathBench.cpp (right):

http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp#newcode195
bench/PathBench.cpp:195: SkRandom rand(12);
Why seed rand with a specific value? It is, by default, seeded with 0. Just
curious.

Can simplify your rand coordinate calculation:

path->moveTo(rand.nextUScalar1() * 640, rand.nextUScalar1() * 480);
etc.

[Stephen White, 2012-01-12 17:33:47]

On 2012/01/11 21:23:25, reed1 wrote:
> lgtm with simplification-request and rand-seed-question.
> 
> http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp
> File bench/PathBench.cpp (right):
> 
> http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp#newcode195
> bench/PathBench.cpp:195: SkRandom rand(12);
> Why seed rand with a specific value? It is, by default, seeded with 0. Just
> curious.
> 
> Can simplify your rand coordinate calculation:
> 
> path->moveTo(rand.nextUScalar1() * 640, rand.nextUScalar1() * 480);
> etc.

Is it still compatible with fixed-point this way?  I actually changed it to
match the LongCurved test.

[bsalomon, 2012-01-12 18:26:52]

On 2012/01/12 17:33:47, Stephen White wrote:
> On 2012/01/11 21:23:25, reed1 wrote:
> > lgtm with simplification-request and rand-seed-question.
> > 
> > http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp
> > File bench/PathBench.cpp (right):
> > 
> > http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp#newcode195
> > bench/PathBench.cpp:195: SkRandom rand(12);
> > Why seed rand with a specific value? It is, by default, seeded with 0. Just
> > curious.
> > 
> > Can simplify your rand coordinate calculation:
> > 
> > path->moveTo(rand.nextUScalar1() * 640, rand.nextUScalar1() * 480);
> > etc.
> 
> Is it still compatible with fixed-point this way?  I actually changed it to
> match the LongCurved test.

Stephen, I tested this benchmark with the sw-rasterize-paths patch and got a
nice bump:

Offscreen AA paths:
running bench [640 480]  path_stroke_small_long_line   GPU: cmsecs =  37.44
gmsecs = 255.96
running bench [640 480]    path_fill_small_long_line   GPU: cmsecs =  30.81
gmsecs = 200.41

SW-to-texture paths:
running bench [640 480]  path_stroke_small_long_line   GPU: cmsecs =  69.81
gmsecs =  69.39
running bench [640 480]    path_fill_small_long_line   GPU: cmsecs =  31.20
gmsecs =  31.10

[reed1, 2012-01-12 18:27:29]

These invariants are true with either fixed or floats:

SkScalar * int == SkScalar
SkScalar / int == SkScalar

[Stephen White, 2012-01-12 18:39:31]

undo cargo cult programming

[Stephen White, 2012-01-12 18:41:41]

On 2012/01/12 18:26:52, bsalomon wrote:
> On 2012/01/12 17:33:47, Stephen White wrote:
> > On 2012/01/11 21:23:25, reed1 wrote:
> > > lgtm with simplification-request and rand-seed-question.
> > > 
> > > http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp
> > > File bench/PathBench.cpp (right):
> > > 
> > >
http://codereview.appspot.com/5529069/diff/1/bench/PathBench.cpp#newcode195
> > > bench/PathBench.cpp:195: SkRandom rand(12);
> > > Why seed rand with a specific value? It is, by default, seeded with 0.
Just
> > > curious.
> > > 
> > > Can simplify your rand coordinate calculation:
> > > 
> > > path->moveTo(rand.nextUScalar1() * 640, rand.nextUScalar1() * 480);
> > > etc.
> > 
> > Is it still compatible with fixed-point this way?  I actually changed it to
> > match the LongCurved test.
> 
> Stephen, I tested this benchmark with the sw-rasterize-paths patch and got a
> nice bump:
> 
> Offscreen AA paths:
> running bench [640 480]  path_stroke_small_long_line   GPU: cmsecs =  37.44
> gmsecs = 255.96
> running bench [640 480]    path_fill_small_long_line   GPU: cmsecs =  30.81
> gmsecs = 200.41
> 
> SW-to-texture paths:
> running bench [640 480]  path_stroke_small_long_line   GPU: cmsecs =  69.81
> gmsecs =  69.39
> running bench [640 480]    path_fill_small_long_line   GPU: cmsecs =  31.20
> gmsecs =  31.10

Fantastic!  That should take care of the Maps API case, I think.

[reed1, 2012-01-12 19:08:15]

lgtm