rounding for arm filter

rounding for arm filter

R=wangcheng@google.com, harryjin@google.com
BUG=libyuv:607
Committed: https://chromium.googlesource.com/libyuv/libyuv/+/b8ddb5a2a789a3e062283fb509938b4eac5d41cc

[fbarchard1, 2016-06-24 21:37:56]

fbarchard@google.com changed reviewers:
+ wangcheng@google.com

[fbarchard1, 2016-06-24 21:37:57]

use rounding shift on arm.

[Diony Rosa, 2016-06-24 21:41:36]

On 2016/06/24 21:37:57, fbarchard1 wrote:
> use rounding shift on arm.

How do we know that this change improves anything? Does this change allow you to
tighten the error bounds on any of the unit tests?

[fbarchard1, 2016-06-24 21:49:31]

On 2016/06/24 21:41:36, Diony Rosa wrote:
> On 2016/06/24 21:37:57, fbarchard1 wrote:
> > use rounding shift on arm.
> 
> How do we know that this change improves anything? Does this change allow you
to
> tighten the error bounds on any of the unit tests?

I actually tried to test that by making the C mismatch the arm on purpose, and
my default test cant tell the difference.  So I think with a reasonable scale
factor, the low bits of the step are insignificant and dont actually produce a
difference.

I'm inclined to use rounding except when theres a good reason it cant be done.

[wangcheng, 2016-06-24 22:54:51]

lgtm

[wangcheng, 2016-06-24 22:54:53]

lgtm

[fbarchard1, 2016-06-24 23:07:51]

Description was changed from

==========
rounding for arm filter

R=harryjin@google.com
BUG=libyuv:607
==========

to

==========
rounding for arm filter

R=wangcheng@google.com, harryjin@google.com
BUG=libyuv:607

Committed:
https://chromium.googlesource.com/libyuv/libyuv/+/b8ddb5a2a789a3e062283fb5099...
==========

[fbarchard1, 2016-06-24 23:07:53]

Committed patchset #4 (id:60001) manually as
b8ddb5a2a789a3e062283fb509938b4eac5d41cc (presubmit successful).