make OpenCL optional for skpdiff

make OpenCL optional for skpdiff

- add CPU based different pixels metric

R=djsollen@google.com

Committed: https://code.google.com/p/skia/source/detail?r=10127

[Zach Reizner, 2013-07-16 19:38:36]

This one looks like a lot of lines, but most of them were just refactoring
things into new files.

[djsollen, 2013-07-16 19:45:10]

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/SkDif...
File experimental/skpdiff/SkDifferentPixelsMetric.h (right):

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/SkDif...
experimental/skpdiff/SkDifferentPixelsMetric.h:27: public SkCLImageDiffer {
this seems backward.  Shouldn't CL and the CPU differ be subclasses of the
metric?

[Zach Reizner, 2013-07-16 19:58:32]

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/SkDif...
File experimental/skpdiff/SkDifferentPixelsMetric.h (right):

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/SkDif...
experimental/skpdiff/SkDifferentPixelsMetric.h:27: public SkCLImageDiffer {
On 2013/07/16 19:45:10, djsollen wrote:
> this seems backward.  Shouldn't CL and the CPU differ be subclasses of the
> metric?

I made it so that both versions are the metric, and the one that is compiled is
chosen at compile time. This simplified the main.cpp logic.

[djsollen, 2013-07-16 20:11:46]

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/SkDif...
File experimental/skpdiff/SkDifferentPixelsMetric.h (right):

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/SkDif...
experimental/skpdiff/SkDifferentPixelsMetric.h:13: #include "SkCLImageDiffer.h"
remove this

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/main.cpp
File experimental/skpdiff/main.cpp (right):

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/main....
experimental/skpdiff/main.cpp:13: #include "SkCLImageDiffer.h"
this is not needed.

https://codereview.chromium.org/19374006/diff/2001/experimental/skpdiff/main....
experimental/skpdiff/main.cpp:115: init_cl_diff,
why not have skImageDiffer have a static function for supportsOpenCL() and use
that to determine whether or not to call init_cl_diff?

[djsollen, 2013-07-17 12:51:02]

https://codereview.chromium.org/19374006/diff/10001/experimental/skpdiff/main...
File experimental/skpdiff/main.cpp (right):

https://codereview.chromium.org/19374006/diff/10001/experimental/skpdiff/main...
experimental/skpdiff/main.cpp:132: chosenDiffers.push(differ);
consolidate to 

if (differ->requiresOpenCL()) {
#if...
  init_cl...
#endif
}
chosenDiffers.push(differ

https://codereview.chromium.org/19374006/diff/10001/experimental/skpdiff/main...
experimental/skpdiff/main.cpp:144: chosenDiffers.push(differ);
same as above

[Zach Reizner, 2013-07-17 13:02:11]

https://codereview.chromium.org/19374006/diff/10001/experimental/skpdiff/main...
File experimental/skpdiff/main.cpp (right):

https://codereview.chromium.org/19374006/diff/10001/experimental/skpdiff/main...
experimental/skpdiff/main.cpp:132: chosenDiffers.push(differ);
On 2013/07/17 12:51:02, djsollen wrote:
> consolidate to 
> 
> if (differ->requiresOpenCL()) {
> #if...
>   init_cl...
> #endif
> }
> chosenDiffers.push(differ
If it requires OpenCL, but we don't support it, I want to avoid adding it to the
list of differs because it won't be initialized properly.

[djsollen, 2013-07-17 14:15:18]

lgtm

[Zach Reizner, 2013-07-17 15:50:10]

I caught some unguarded references to OpenCL and made it so kernel source is
embedded inside the metric.

[Zach Reizner, 2013-07-17 19:29:23]

Committed patchset #5 manually as r10127 (presubmit successful).