Add rounding to InterpolateRow for improved quality and consistency.

Add rounding to InterpolateRow for improved quality and consistency.

Remove inaccurate specializations for 1/4 and 3/4, since they round
incorrectly.  Specialize for 100% and 50% are kept due to performance.
Make C and ARM code match SSSE3.
Make unittests expect zero difference.

BUG=libyuv:535
R=harryjin@google.com

Committed: https://chromium.googlesource.com/libyuv/libyuv/+/f4447745ae646749562cff00e9d8fbc3b9043580

[fbarchard, 2015-12-17 23:09:30]

fbarchard@chromium.org changed reviewers:
+ dhrosa@google.com, harryjin@google.com

[fbarchard, 2015-12-17 23:09:31]

arm code matches ssse3/avx2

[harryjin, 2015-12-17 23:13:40]

lgtm

[fbarchard, 2015-12-17 23:16:31]

93.00%  libyuv_unittest  libyuv_unittest      [.] InterpolateRow_AVX2           
                                                                             ◆
  0.90%  libyuv_unittest  [kernel.kallsyms]    [k] 0xffffffff810a7d10           
                                                                              ▒
  0.39%  libyuv_unittest  libc-2.19.so         [.] __memmove_ssse3_back         
                                                                              ▒
  0.23%  libyuv_unittest  libyuv_unittest      [.] InterpolatePlane       


 1.92 │49:   vmovdq (%rsi),%ymm0
 19.25 │      vmovdq (%rsi,%rdx,1),%ymm2
 16.05 │      vpunpc %ymm2,%ymm0,%ymm1
  3.89 │      vpunpc %ymm2,%ymm0,%ymm0
  1.97 │      vpmadd %ymm5,%ymm0,%ymm0
  6.76 │      vpmadd %ymm5,%ymm1,%ymm1
  2.89 │      vpaddw %ymm4,%ymm0,%ymm0
  1.10 │      vpaddw %ymm4,%ymm1,%ymm1
  1.78 │      vpsrlw $0x7,%ymm0,%ymm0
  2.85 │      vpsrlw $0x7,%ymm1,%ymm1
  3.32 │      vpacku %ymm1,%ymm0,%ymm0
  2.09 │      vmovdq %ymm0,(%rsi,%rdi,1)
  3.51 │      lea    0x20(%rsi),%rsi
  0.11 │      sub    $0x20,%ecx
  0.01 │    ↑ jg     49

[fbarchard, 2015-12-17 23:24:10]

Description was changed from

==========
Add rounding to InterpolateRow for improved quality and consistency.

Remove inaccurate specializations for 1/4 and 3/4, since they round
incorrectly.  Specialize for 100% and 50% are kept due to performance.
Make C and ARM code match SSSE3.
Make unittests expect zero difference.

BUG=libyuv:535
==========

to

==========
Add rounding to InterpolateRow for improved quality and consistency.

Remove inaccurate specializations for 1/4 and 3/4, since they round
incorrectly.  Specialize for 100% and 50% are kept due to performance.
Make C and ARM code match SSSE3.
Make unittests expect zero difference.

BUG=libyuv:535
R=harryjin@google.com

Committed:
https://chromium.googlesource.com/libyuv/libyuv/+/f4447745ae646749562cff00e9d...
==========

[fbarchard, 2015-12-17 23:24:12]

Committed patchset #3 (id:40001) manually as
f4447745ae646749562cff00e9d8fbc3b9043580 (presubmit successful).