gm: report max-pixel-error if comparing against PNG files (not checksums)

gm: report max-pixel-error if comparing against PNG files (not checksums)
Committed: https://code.google.com/p/skia/source/detail?r=7526

[epoger, 2013-02-01 17:46:06]

Eric- please review the code

Mike and Brian- please let me know if the output will be sufficient.

https://codereview.appspot.com/7241064/diff/1/gm/gm_expectations.h
File gm/gm_expectations.h (right):

https://codereview.appspot.com/7241064/diff/1/gm/gm_expectations.h#newcode1
gm/gm_expectations.h:1: /*
This CL restores the pixel-diff reporting capability (but NOT the --diffPath
option) that was removed from gm in
https://code.google.com/p/skia/source/detail?r=7093 .

(This CL is based on Mike's https://codereview.appspot.com/7238043/ )

The output looks slightly different now than it used to, with more detail. 
Here's a sample:

$ out/Debug/gm --hierarchy --match dashing2 --config 8888 --config 565 -r
gm/tests/inputs/different-pixels
Non-default runtime configuration options:
reading from gm/tests/inputs/different-pixels
drawing... dashing2 [640 480]
---- 8888/dashing2: max per-channel pixel mismatch is R=255 G=255 B=255 A=0
---- 565/dashing2: not computing max per-channel pixel mismatch because non-8888
Ran 1 tests: 0 passed, 1 failed, 0 missing reference images
		8888/dashing2 pixel_error
		565/dashing2 pixel_error

https://codereview.appspot.com/7241064/diff/1/gm/gmmain.cpp
File gm/gmmain.cpp (right):

https://codereview.appspot.com/7241064/diff/1/gm/gmmain.cpp#newcode549
gm/gmmain.cpp:549: void report_bitmap_diffs(const SkBitmap& expectedBitmap,
const SkBitmap& actualBitmap,
As discussed in https://codereview.appspot.com/7238043/ , I looked into sharing
skdiff's compute_diff() function instead of adding this one.  But it ended up
being more trouble than it was worth... compute_diff() requires a lot of glue
code that gm doesn't currently import, and it performs operations (like
generating delta images) that we don't need.

[EricB, 2013-02-01 18:02:20]

LGTM assuming the output format is acceptable.

[bsalomon, 2013-02-01 18:27:21]

A count (or actually prolly a %) of the pixels that differed would be nice.

Otherwise lgtm.

[epoger, 2013-02-01 20:40:38]

On 2013/02/01 18:27:21, bsalomon wrote:
> A count (or actually prolly a %) of the pixels that differed would be nice.
> 
> Otherwise lgtm.

Thanks for the idea.  Here's what it looks like now... if anyone wants to adjust
the output, please feel free!

$ out/Debug/gm --hierarchy --match dashing2 --config 8888 --config 565 -r
gm/tests/inputs/different-pixels
Non-default runtime configuration options:
reading from gm/tests/inputs/different-pixels
drawing... dashing2 [640 480]
---- 8888/dashing2: 41051 (of 307200) differing pixels, max per-channel mismatch
R=255 G=255 B=255 A=0
---- 565/dashing2: not computing max per-channel pixel mismatch because non-8888
Ran 1 tests: 0 passed, 1 failed, 0 missing reference images
		8888/dashing2 pixel_error
		565/dashing2 pixel_error