Port ARGBToRGB565 from aarch64 neon to 32 bit

Port ARGBToRGB565 from aarch64 neon to 32 bit

The 64 bit version of ARGBToRGB565 to 32 bit. 64 bit is using sri which shifts and inserts, saving some masking.  The instruction is available for neon 32 bit as well.

R=magjed@chromium.org, harryjin@google.com
BUG=libyuv:571
Committed: https://chromium.googlesource.com/libyuv/libyuv/+/ee99b85126aeafe64ba3da8f28aafcac80a595ac

[fbarchard1, 2016-02-24 04:07:48]

Description was changed from

==========
port 565 sri code to neon 32 bit

R=harryjin@google.com
BUG=libyuv:571
==========

to

==========
Port ARGBToRGB565 from aarch64 neon to 32 bit

The 64 bit version of ARGBToRGB565 to 32 bit. 64 bit is using sri which shifts
and inserts, saving some masking.  The instruction is available for neon 32 bit
as well.

R=harryjin@google.com
BUG=libyuv:571
==========

[fbarchard1, 2016-02-24 05:35:53]

fbarchard@google.com changed reviewers:
+ magjed@chromium.org, yang.zhang@arm.com

[fbarchard1, 2016-02-24 05:35:53]

Make use for sri instruction to improve 565 for Neon

[fbarchard1, 2016-02-24 18:49:18]

comparing 64 and 32 bit I spotted an off by 1 shift.

https://codereview.chromium.org/1724393002/diff/20001/source/row_neon.cc
File source/row_neon.cc (right):

https://codereview.chromium.org/1724393002/diff/20001/source/row_neon.cc#newc...
source/row_neon.cc:362: "vsri.16     q0, q9, #5                    \n"  /* ARG  
               */ \
This shift should be 6.  1 for alpha + 5 for R.

[fbarchard1, 2016-02-24 18:57:01]

fixed shift.

https://codereview.chromium.org/1724393002/diff/20001/source/row_neon.cc
File source/row_neon.cc (right):

https://codereview.chromium.org/1724393002/diff/20001/source/row_neon.cc#newc...
source/row_neon.cc:362: "vsri.16     q0, q9, #5                    \n"  /* ARG  
               */ \
On 2016/02/24 18:49:18, fbarchard1 wrote:
> This shift should be 6.  1 for alpha + 5 for R.

Done.

[magjed_chromium, 2016-02-24 20:06:37]

magjed@chromium.org changed reviewers:
+ torbjorng@chromium.org

[magjed_chromium, 2016-02-24 20:08:21]

On 2016/02/24 18:57:01, fbarchard1 wrote:
> fixed shift.
> 
> https://codereview.chromium.org/1724393002/diff/20001/source/row_neon.cc
> File source/row_neon.cc (right):
> 
>
https://codereview.chromium.org/1724393002/diff/20001/source/row_neon.cc#newc...
> source/row_neon.cc:362: "vsri.16     q0, q9, #5                    \n"  /* ARG
 
>                */ \
> On 2016/02/24 18:49:18, fbarchard1 wrote:
> > This shift should be 6.  1 for alpha + 5 for R.
> 
> Done.

I need to study neon assembly more before reviewing this. I added torbjorng@ as
a reviewer.

[magjed_chromium, 2016-02-26 14:14:00]

I followed the tutorial and explanation in the bug. This lgtm.

[fbarchard1, 2016-02-29 20:22:28]

Description was changed from

==========
Port ARGBToRGB565 from aarch64 neon to 32 bit

The 64 bit version of ARGBToRGB565 to 32 bit. 64 bit is using sri which shifts
and inserts, saving some masking.  The instruction is available for neon 32 bit
as well.

R=harryjin@google.com
BUG=libyuv:571
==========

to

==========
Port ARGBToRGB565 from aarch64 neon to 32 bit

The 64 bit version of ARGBToRGB565 to 32 bit. 64 bit is using sri which shifts
and inserts, saving some masking.  The instruction is available for neon 32 bit
as well.

R=magjed@chromium.org, harryjin@google.com
BUG=libyuv:571

Committed:
https://chromium.googlesource.com/libyuv/libyuv/+/ee99b85126aeafe64ba3da8f28a...
==========

[fbarchard1, 2016-02-29 20:22:30]

Committed patchset #4 (id:60001) manually as
ee99b85126aeafe64ba3da8f28aafcac80a595ac (presubmit successful).