Avoid recomputing two consecutive identical 1D filters.

Avoid recomputing two consecutive identical 1D filters.

If the arguments to the X and Y filter computation are identical,
the results will be identical; copying is much faster than recomputing.

With a change like https://codereview.chromium.org/183763047/ applied
this speeds up BitmapScaleBench on Linux by around 10%.

BUG=skia:2236
R=humper@google.com

Committed: http://code.google.com/p/skia/source/detail?r=13687

[tomhudson, 2014-03-06 14:59:48]

PTAL?

[humper, 2014-03-06 15:18:30]

On 2014/03/06 14:59:48, tomhudson wrote:
> PTAL?

LGTM -- there's still some optimization that can probably be done here, since
your change will really only help square images.  The output of ComputeFilters
is an array of discretized filters; most of the filters will be identical except
as they start to near the edge of the image.

In addition to avoiding duplicating effort for square images as you've done
here, I wonder if we can detect (inside computeFilters) that we're away from the
edges of the image by more than one pixel (so we've already computed the filter
profile) and just copy the filter profile from the previous loop iteration.

[tomhudson, 2014-03-06 17:51:13]

The CQ bit was checked by tomhudson@google.com

[commit-bot: I haz the power, 2014-03-06 17:51:22]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/tomhudson@chromium.org/188743002/1

[tomhudson, 2014-03-06 17:51:43]

On 2014/03/06 15:18:30, humper wrote:
> LGTM -- there's still some optimization that can probably be done here, since
> your change will really only help square images.  The output of ComputeFilters
> is an array of discretized filters; most of the filters will be identical
except
> as they start to near the edge of the image.
> 
> In addition to avoiding duplicating effort for square images as you've done
> here, I wonder if we can detect (inside computeFilters) that we're away from
the
> edges of the image by more than one pixel (so we've already computed the
filter
> profile) and just copy the filter profile from the previous loop iteration.

Filed http://skbug.com/2261 to capture this idea.

[commit-bot: I haz the power, 2014-03-06 19:10:46]

Change committed as 13687