Add benchmark for PremultiplyAlpha and UnpremultiplyAlpha in Skia

Add benchmark for PremultiplyAlpha and UnpremultiplyAlpha in Skia

BUG=248549
Committed: http://code.google.com/p/skia/source/detail?r=9569

[Jun Jiang, 2013-06-11 15:36:20]

This benchmark is separated form the CL at
https://codereview.chromium.org/15402003/ with some changes that move some setup
code out of the loop.

[bsalomon, 2013-06-11 17:15:00]

On 2013/06/11 15:36:20, Jun Jiang wrote:
> This benchmark is separated form the CL at
> https://codereview.chromium.org/15402003/ with some changes that move some
setup
> code out of the loop.

lgtm

[tomhudson, 2013-06-11 17:31:35]

Nit: line lengths wasn't fixed.

Otherwise LGTM

[junj, 2013-06-12 03:34:49]

On 2013/06/11 17:31:35, tomhudson wrote:
> Nit: line lengths wasn't fixed.
> 
> Otherwise LGTM

Hi, Tom. Thanks for your comments. Yes, there are still some lines didn't meet
the line length. Refined the code and verified using cpplint.py.
PTAL.

[Noel Gordon, 2013-06-12 04:04:52]

LGTM

[tomhudson, 2013-06-12 09:23:34]

I'm interested to see the effects on timing, but I think the change in the inner
loop makes this a much better *micro*benchmark. Thanks for doing that.

https://codereview.chromium.org/16654004/diff/8001/bench/PremulAndUnpremulAlp...
File bench/PremulAndUnpremulAlphaOpsBench.cpp (right):

https://codereview.chromium.org/16654004/diff/8001/bench/PremulAndUnpremulAlp...
bench/PremulAndUnpremulAlphaOpsBench.cpp:58: const char* benchName;
I wasn't reading carefully enough earlier - another nit from the Skia style
guide is that member variables should all start with 'f', so these should be
fUnPremulConfig and fBenchName.

It's more in keeping with the SkBenchmark idiom to
 - have a SkString fName;
 - dynamically assemble the name in the constructor.

You shouldn't need to be passing the name in to the constructor; your
constructor knows everything it needs to create the name. SkAAClipBench is a
good a simple example. We prefer this style for change-proofing (along the lines
of Don't Repeat Yourself).

onGetName() would then return fName.c_str()

[Jun Jiang, 2013-06-12 16:59:26]

Hi, Tom. Thanks for your comments. I have made changes accordingly. PTAL.

https://codereview.chromium.org/16654004/diff/8001/bench/PremulAndUnpremulAlp...
File bench/PremulAndUnpremulAlphaOpsBench.cpp (right):

https://codereview.chromium.org/16654004/diff/8001/bench/PremulAndUnpremulAlp...
bench/PremulAndUnpremulAlphaOpsBench.cpp:58: const char* benchName;
On 2013/06/12 09:23:34, tomhudson wrote:
> I wasn't reading carefully enough earlier - another nit from the Skia style
> guide is that member variables should all start with 'f', so these should be
> fUnPremulConfig and fBenchName.
> 
> It's more in keeping with the SkBenchmark idiom to
>  - have a SkString fName;
>  - dynamically assemble the name in the constructor.
> 
> You shouldn't need to be passing the name in to the constructor; your
> constructor knows everything it needs to create the name. SkAAClipBench is a
> good a simple example. We prefer this style for change-proofing (along the
lines
> of Don't Repeat Yourself).
> 
> onGetName() would then return fName.c_str()

Done.

[tomhudson, 2013-06-13 09:56:26]

lgtm

[commit-bot: I haz the power, 2013-06-13 09:56:29]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/jun.a.jiang@intel.com/16654004/13001

[commit-bot: I haz the power, 2013-06-13 10:18:05]

Change committed as 9569