Handle near-colinear case in WSOLAS QuadraticInterpolation better.

Handle near-colinear cases in WSOLAS QuadraticInterpolation better.

BUG=357870
Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=261033

[DaleCurtis, 2014-03-29 01:10:01]

You'll want to add a BUG for this since I think we'll need to merge to M35.

lgtm, nice work!

cc:turajs in case he has any comments.

https://codereview.chromium.org/217553004/diff/1/media/filters/audio_renderer...
File media/filters/audio_renderer_algorithm_unittest.cc (right):

https://codereview.chromium.org/217553004/diff/1/media/filters/audio_renderer...
media/filters/audio_renderer_algorithm_unittest.cc:627: EXPECT_NEAR(extremum,
0.0, 1.0);
within 1 seems pretty large?

[sandersd (OOO until July 31), 2014-03-29 01:19:37]

https://codereview.chromium.org/217553004/diff/1/media/filters/audio_renderer...
File media/filters/audio_renderer_algorithm_unittest.cc (right):

https://codereview.chromium.org/217553004/diff/1/media/filters/audio_renderer...
media/filters/audio_renderer_algorithm_unittest.cc:627: EXPECT_NEAR(extremum,
0.0, 1.0);
On 2014/03/29 01:10:01, DaleCurtis wrote:
> within 1 seems pretty large?

Since the choice of x = 0 was arbitrary, I'm only testing that the returned
value satisfies -1 <= x <= 1.

[turajs1, 2014-03-31 23:18:41]

Please see my comment.

https://codereview.chromium.org/217553004/diff/1/media/filters/wsola_internal...
File media/filters/wsola_internals.cc (right):

https://codereview.chromium.org/217553004/diff/1/media/filters/wsola_internal...
media/filters/wsola_internals.cc:97: if (a == 0.f || fabsf(2.f * a) < fabsf(b))
{
abs(2*a) is always larger or equal to abs(b), because of the if-statement on
line 159. So if you are concerned with numerical precision, would it make sense
to actually compare abs(b / (2*a)) with 1?

[sandersd (OOO until July 31), 2014-04-01 00:27:48]

https://codereview.chromium.org/217553004/diff/1/media/filters/wsola_internal...
File media/filters/wsola_internals.cc (right):

https://codereview.chromium.org/217553004/diff/1/media/filters/wsola_internal...
media/filters/wsola_internals.cc:97: if (a == 0.f || fabsf(2.f * a) < fabsf(b))
{
On 2014/03/31 23:18:42, turajs1 wrote:
> abs(2*a) is always larger or equal to abs(b), because of the if-statement on
> line 159. So if you are concerned with numerical precision, would it make
sense
> to actually compare abs(b / (2*a)) with 1? 

After some experimentation, I have concluded that you are quite correct, it is
not possible to go outside those bounds while still having y[0] <= y[1] >= y[2],
even when accounting for floating point rounding.

I've simplified the code and improved the comments.

[sandersd (OOO until July 31), 2014-04-01 20:24:35]

The CQ bit was checked by sandersd@chromium.org

[commit-bot: I haz the power, 2014-04-01 20:28:36]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/sandersd@chromium.org/217553004/20001

[commit-bot: I haz the power, 2014-04-01 22:49:42]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/sandersd@chromium.org/217553004/20001

[commit-bot: I haz the power, 2014-04-02 01:52:41]

Change committed as 261033