Add bench for hairline paths

Add bench for hairline paths.

R=bsalomon@google.com

Committed: https://code.google.com/p/skia/source/detail?r=10128

[egdaniel, 2013-07-15 17:26:44]

Here is the bench I created for hairline paths. Can I get some feed back and
what I should change or add for this bench?

[bsalomon, 2013-07-15 17:42:56]

In the GPU we do things a little differently depending on the actual controls
points (e.g. splitting quads at the vertex). Maybe we should generate a path
with at least a few different segments?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp
File bench/HairlinePathBench.cpp (right):

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:2: * Copyright 2011 Google Inc.
2013

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:21: SkPaint     fPaint;
Move these down to private below?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:30: virtual void appendName(SkString*) = 0;
do these need to be public?

[tfarina, 2013-07-16 02:03:16]

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp
File bench/HairlinePathBench.cpp (right):

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:13: #define HYPERBOLA 2.0f
can these be const floats instead?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:17: 10, 10, 15, 5, 20, 20
indent 4 spaces only?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:18: };
mv to first column?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:29: 
virtual destructor?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:32: virtual int complexity() { return 0; }
const?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:92: ConicPathBench(void* param, const SkScalar
weight) : INHERITED(param), fWeight(weight) {}
can we try to keep these in 80 columns (I know we don't stay adhere to this much
though)?

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:96: if (fWeight == HYPERBOLA) {
I think in skia people do comparisons in the reverse other, i.e., HYPEROLA ==
fWeight

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:100: };
micronit: no need of ; here

[bsalomon, 2013-07-16 12:57:17]

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp
File bench/HairlinePathBench.cpp (right):

https://codereview.chromium.org/19238003/diff/1/bench/HairlinePathBench.cpp#n...
bench/HairlinePathBench.cpp:92: ConicPathBench(void* param, const SkScalar
weight) : INHERITED(param), fWeight(weight) {}
On 2013/07/16 02:03:16, tfarina wrote:
> can we try to keep these in 80 columns (I know we don't stay adhere to this
much
> though)?

Skia's rough limit is 100 cols.

[egdaniel, 2013-07-16 19:17:20]

Added more varying paths for each type, and added cubics. Also there are now
flags to test big/small versions and AA and no AA.

[bsalomon, 2013-07-16 19:25:36]

For my nits I don't need to see another patch, but I am curious about the non-AA
question.

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.cpp
File bench/HairlinePathBench.cpp (right):

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:31: static const int maxPathSize = 10;
kMaxPathSize

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:45: fName.printf("path_hairline_%s_%s_",
trailing _ in name?

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:162: int base1 = 2*rand.nextRangeU(0, hSize-1);
There is a nextULessThan(hSize) that might make this a bit easier to read.

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:230: // DEF_BENCH( return new ConicPathBench(p,
FLAGS00); )
What happens if we run this? Do we get conics chopped into quads?

[egdaniel, 2013-07-16 19:38:38]

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.cpp
File bench/HairlinePathBench.cpp (right):

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:31: static const int maxPathSize = 10;
changed
On 2013/07/16 19:25:36, bsalomon wrote:
> kMaxPathSize

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:45: fName.printf("path_hairline_%s_%s_",
It will get filled in by the "appendName" function in each child class.

On 2013/07/16 19:25:36, bsalomon wrote:
> trailing _ in name?

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:162: int base1 = 2*rand.nextRangeU(0, hSize-1);
Okay changed it to that.

On 2013/07/16 19:25:36, bsalomon wrote:
> There is a nextULessThan(hSize) that might make this a bit easier to read.

https://codereview.chromium.org/19238003/diff/10001/bench/HairlinePathBench.c...
bench/HairlinePathBench.cpp:230: // DEF_BENCH( return new ConicPathBench(p,
FLAGS00); )
We get 5 lines drawn on screen. The lines connects the last control point of
conic[i] to the first control point of conic[i+1] for i = 0,2,4,6,8 

On 2013/07/16 19:25:36, bsalomon wrote:
> What happens if we run this? Do we get conics chopped into quads?

[bsalomon, 2013-07-16 20:26:14]

lgtm with nit fixes/explanations

[egdaniel, 2013-07-17 19:30:44]

Committed patchset #4 manually as r10128 (presubmit successful).