ARMv7 Neon version of I420AlphaToARGB

Neon versions of I420AlphaToARGB

Add alpha version of YUV to RGB to neon code for ARMv7 and aarch64.
For other YUV to RGB conversions, hoist alpha set to 255 out of loop.

TBR=harryjin@google.com
BUG=libyuv:516
Committed: https://chromium.googlesource.com/libyuv/libyuv/+/860cc0357a4b0d224be673c0ccfad192745a4192

[fbarchard, 2015-11-03 20:38:22]

fbarchard@chromium.org changed reviewers:
+ fbarchard@chromium.org

[fbarchard, 2015-11-03 20:38:23]

[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Any
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Any (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Unaligned
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Unaligned (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Invert
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Invert (2 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Opt
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Opt (2 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Premult
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Premult (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToABGR_Any
[       OK ] LibYUVConvertTest.I420AlphaToABGR_Any (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToABGR_Unaligned
[       OK ] LibYUVConvertTest.I420AlphaToABGR_Unaligned (2 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToABGR_Invert
[       OK ] LibYUVConvertTest.I420AlphaToABGR_Invert (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToABGR_Opt
[       OK ] LibYUVConvertTest.I420AlphaToABGR_Opt (2 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToABGR_Premult
[       OK ] LibYUVConvertTest.I420AlphaToABGR_Premult (5 ms)
[----------] 10 tests from LibYUVConvertTest (30 ms total)

[fbarchard, 2015-11-04 02:30:55]

64 bit version ported 

[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Any
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Any (2 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Unaligned
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Unaligned (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Invert
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Invert (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Opt
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Opt (3 ms)
[ RUN      ] LibYUVConvertTest.I420AlphaToARGB_Premult
[       OK ] LibYUVConvertTest.I420AlphaToARGB_Premult (2 ms)

[fbarchard, 2015-11-04 02:41:50]

hoisted alpha set out of inner loops on all YUV to RGB functions for aarch64.

aarch64 bit before (C version)
LibYUVConvertTest.I420AlphaToARGB_Any (38862 ms)
LibYUVConvertTest.I420AlphaToARGB_Unaligned (38987 ms)
LibYUVConvertTest.I420AlphaToARGB_Invert (38683 ms)
LibYUVConvertTest.I420AlphaToARGB_Opt (38709 ms)
LibYUVConvertTest.I420AlphaToARGB_Premult (46954 ms)

aarch64 bit after
LibYUVConvertTest.I420AlphaToARGB_Premult (4491 ms)
LibYUVConvertTest.I420AlphaToARGB_Any (3695 ms)
LibYUVConvertTest.I420AlphaToARGB_Unaligned (3473 ms)
LibYUVConvertTest.I420AlphaToARGB_Invert (3155 ms)
LibYUVConvertTest.I420AlphaToARGB_Opt (3119 ms)

[fbarchard, 2015-11-04 03:16:38]

armv7

Now with Neon
util/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release
--gtest_filter=*I420*ToARGB_Opt* -a "--libyuv_width=1280 --libyuv_height=720
--libyuv_repeat=999 --libyuv_flags=0 --libyuv_cpu_info=0" | grep ms
I420ToARGB_Opt (2926 ms)
I420AlphaToARGB_Opt (2896 ms)

Was with C
util/android/test_runner.py gtest -s libyuv_unittest -t 7200 --verbose --release
--gtest_filter=*I420*ToARGB_Opt* -a "--libyuv_width=1280 --libyuv_height=720
--libyuv_repeat=999 --libyuv_flags=1 --libyuv_cpu_info=1" | grep ms
I420ToARGB_Opt (39051 ms)
I420AlphaToARGB_Opt (38905 ms)

[fbarchard, 2015-11-04 03:21:03]

Description was changed from

==========
ARMv7 Neon version of I420AlphaToARGB

For other conversions, hoist alpha set to 255 out of loop.

R=xhwang@chromium.org
BUG=libyuv:516
==========

to

==========
Neon versions of I420AlphaToARGB

Add alpha version of YUV to RGB to neon code for ARMv7 and aarch64.
For other YUV to RGB conversions, hoist alpha set to 255 out of loop.

TBR=harryjin@google.com
BUG=libyuv:516
==========

[fbarchard1, 2015-11-04 03:21:41]

Committed patchset #6 (id:100001) manually as
860cc0357a4b0d224be673c0ccfad192745a4192 (presubmit successful).