Unable to enable box filtering for downscaling

[GoogleCodeExporter]

What steps will reproduce the problem?
1. Call libyuv::ScalePlane(in, instride, inwidth, inheight, out, outstride, 
outwidth, outheight, libyuv::kFilterBox), where the sizes don't match any 
special-cased scaling
2. Look at the output, notice that it does not look like it should (having the 
same look as with bilinear scaling), or check the code whether actually calls 
ScalePlaneBox

What is the expected output? What do you see instead?

If scaling down and using kFilterBox, one would expect to have the better 
visual quality than with bilinear. 

What version of the product are you using? On what operating system?

SVN rev 561, on Linux.

Please provide any additional information below.

The attached patch fixes the issue.

Original issue reported on code.google.com by mar...@martin.st on 6 Feb 2013 at 6:08

Attachments:

• 0001-Fix-the-condition-for-using-bilinear-scaling-for-sca.patch

[GoogleCodeExporter]

doh!  The *2 comparison is a little more efficient than /2, but is more 
confusing to read, which lead to getting it backwards.
Good catch.  Thanks for the patch!
Fixed in r562.

Original comment by fbarch...@chromium.org on 6 Feb 2013 at 7:22

• Changed state: Fixed

[GoogleCodeExporter]

Re-opening, as one of the scaling tests crash.  Seems unrelated

[ RUN      ] libyuvTest.ScaleDownBy8_Box
unknown file: error: SEH exception with code 0xc0000005 thrown in the test body.
[  FAILED  ] libyuvTest.ScaleDownBy8_Box (1 ms)
[ RUN      ] libyuvTest.ScaleDownBy16_None
filter 0 -        5 us C -        4 us OPT
[       OK ] libyuvTest.ScaleDownBy16_None (1 ms)
[ RUN      ] libyuvTest.ScaleDownBy16_Bilinear
filter 1 -        6 us C -        5 us OPT
[       OK ] libyuvTest.ScaleDownBy16_Bilinear (1 ms)
[ RUN      ] libyuvTest.ScaleDownBy16_Box
unknown file: error: SEH exception with code 0xc0000005 thrown in the test body.
[  FAILED  ] libyuvTest.ScaleDownBy16_Box (1 ms)

Its something to do with box filter scaling down with odd scale factors.

Original comment by fbarch...@chromium.org on 6 Feb 2013 at 11:34

• Changed state: Started

[GoogleCodeExporter]

Yes, the box filter downscaling probably hasn't been tested much at all, since 
the generic box filter downscaling function was impossible to invoke prior to 
this patch.

Original comment by mar...@martin.st on 7 Feb 2013 at 9:08

[GoogleCodeExporter]

Yikes, I think you're right.  We should reintroduce the bug you found :-)
r565 disables tests affected.  Which allows other changes to be tested.
Seems to be the box filter horizontally reading too much.
If the sizes work out to integers, it should be okay.  Or you can use 
Filter_None.
Should have a fix shortly though.

Original comment by fbarch...@chromium.org on 8 Feb 2013 at 2:33

• Added labels: Priority-High
• Removed labels: Priority-Medium

[GoogleCodeExporter]

Fixed in r571.  Box down sampling was starting the initial sampling at middle 
of box.  This is correct for point or bilinear but not box filter.

Original comment by fbarch...@google.com on 12 Feb 2013 at 7:07

• Changed state: Fixed