More robust matrix decomposition

More robust matrix decomposition

We decompose matrix, containing scale.
So we lose precision and get rotation due to
multiplication on 1.0/x instead of divide on x.

R=vollick@chromium.org

TEST=XFormTest.DecomposeTransform

Committed: https://crrev.com/d6e4728f3c0785c3e0e6ce8e5e33f579cee71c7b
Cr-Commit-Position: refs/heads/master@{#302084}

[Ian Vollick, 2014-10-30 14:12:55]

On 2014/10/29 at 21:51:08, lof84 wrote:
> 

The hope was that multiplying by 1/x would be faster, but if that causes your
test to fail due to floating point error, I agree that we should just do
division.

lgtm, and thank you for the test.

[Ian Vollick, 2014-10-30 14:13:06]

The CQ bit was checked by vollick@chromium.org

[commit-bot: I haz the power, 2014-10-30 14:13:57]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/692593003/1

[commit-bot: I haz the power, 2014-10-30 15:07:38]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2014-10-30 15:08:17]

Patchset 1 (id:??) landed as
https://crrev.com/d6e4728f3c0785c3e0e6ce8e5e33f579cee71c7b
Cr-Commit-Position: refs/heads/master@{#302084}