Fix fixed-pos jitter related to physical pixel snapping.

Fix fixed-pos jitter related to physical pixel snapping.

The effective scroll delta is corrected for rounding to nearest screen
pixel, but this was done in the wrong coordinate space (scroll delta is
in CSS pixels, not physical pixels).

NOTRY=true
BUG=357889
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=261057

[aelias_OOO_until_Jul13, 2014-03-29 05:48:37]

PTAL.  This is the mystery jitter I wrongly blamed on wjmaclean's changes a
month back.  The bug dates back to october, but it only started triggering more
consistently recently -- I assume it has been varying based on other changes
that make physical pixel snapping more or less likely.

[enne (OOO), 2014-03-31 22:45:22]

Could you also add a unit test similar to the one you already added where the
scroll layer itself has a non-integer scale and not just a parent in the tree?

https://codereview.chromium.org/216973006/diff/10001/cc/trees/layer_tree_host...
File cc/trees/layer_tree_host_common.cc (right):

https://codereview.chromium.org/216973006/diff/10001/cc/trees/layer_tree_host...
cc/trees/layer_tree_host_common.cc:1556: // in the scroll compensation matrix.
Can you also mention why you're scaling the offset?  If I understand this bug
correctly, it's because there's a scale earlier in the tree, so the scroll delta
gets multiplied by that scale when applied to children, and so you have to
correct for that here.

https://codereview.chromium.org/216973006/diff/10001/cc/trees/layer_tree_host...
cc/trees/layer_tree_host_common.cc:1559: 1 / combined_transform_scales.x(),
style nit: 1.f

[aelias_OOO_until_Jul13, 2014-04-01 06:24:20]

Added another unit test for the scale-on-layer-itself case.  I needed to change
the implementation to make that work -- I switched to getting the scale out of
the parent matrix, since the scroll is applied as part of the position
transformation which precedes the layer transform.

https://codereview.chromium.org/216973006/diff/10001/cc/trees/layer_tree_host...
File cc/trees/layer_tree_host_common.cc (right):

https://codereview.chromium.org/216973006/diff/10001/cc/trees/layer_tree_host...
cc/trees/layer_tree_host_common.cc:1556: // in the scroll compensation matrix.
On 2014/03/31 22:45:22, enne wrote:
> Can you also mention why you're scaling the offset?  If I understand this bug
> correctly, it's because there's a scale earlier in the tree, so the scroll
delta
> gets multiplied by that scale when applied to children, and so you have to
> correct for that here.

Added a longer comment.

https://codereview.chromium.org/216973006/diff/10001/cc/trees/layer_tree_host...
cc/trees/layer_tree_host_common.cc:1559: 1 / combined_transform_scales.x(),
On 2014/03/31 22:45:22, enne wrote:
> style nit: 1.f

Done.

[enne (OOO), 2014-04-01 17:13:23]

lgtm.  Thanks for the new test; I was worried about that edge case.

[aelias_OOO_until_Jul13, 2014-04-01 22:04:53]

The CQ bit was checked by aelias@chromium.org

[commit-bot: I haz the power, 2014-04-01 22:55:51]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/aelias@chromium.org/216973006/80001

[commit-bot: I haz the power, 2014-04-02 02:08:18]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/aelias@chromium.org/216973006/80001

[commit-bot: I haz the power, 2014-04-02 03:59:30]

Change committed as 261057