Add SSE and NEON intrinsics for WSOLA's MultiChannelDotProduct().

Add SSE and NEON intrinsics for WSOLA's MultiChannelDotProduct().

When adapting 10 seconds of 192kHz stereo audio at a 2.0 playback
rate, this provides a ~9x speedup on ARM and a ~3x speedup on X86.

NEON: 24986ms -> 2695ms
SSE: 3014ms -> 894ms

No tests added since the existing tests verify this functionality
correctly and it continues to pass as before.

BUG=666813
TEST=n5x for ARM, z620 for X86.

Committed: https://crrev.com/77fd2aecaa7e1db995e158fb095a72be385e333c
Cr-Commit-Position: refs/heads/master@{#434036}

[DaleCurtis, 2016-11-22 21:26:57]

Description was changed from

==========
Add SSE and NEON intrinsics for WSOLA's MultiChannelDotProduct().

When adapting 10 seconds of 192kHz stereo audio at a 2.0 playback
rate, this provides a ~9x speedup on ARM and a ~3x speedup on X86.

No tests added since the existing tests verify this functionality
correctly and it continues to pass as before.

BUG=666813
TEST=n5x for ARM, z620 for X86.
==========

to

==========
Add SSE and NEON intrinsics for WSOLA's MultiChannelDotProduct().

When adapting 10 seconds of 192kHz stereo audio at a 2.0 playback
rate, this provides a ~9x speedup on ARM and a ~3x speedup on X86.

NEON: 24986ms -> 2695ms
SSE: 3014ms -> 894ms

No tests added since the existing tests verify this functionality
correctly and it continues to pass as before.

BUG=666813
TEST=n5x for ARM, z620 for X86.
==========

[DaleCurtis, 2016-11-22 21:27:02]

dalecurtis@chromium.org changed reviewers:
+ sandersd@chromium.org

[DaleCurtis, 2016-11-22 21:27:32]

See sinc_resampler.cc for reviewing assistance since these operations are very
similar.

[sandersd (OOO until July 31), 2016-11-22 22:13:55]

lgtm

https://codereview.chromium.org/2527533002/diff/1/media/filters/wsola_interna...
File media/filters/wsola_internals.cc (right):

https://codereview.chromium.org/2527533002/diff/1/media/filters/wsola_interna...
media/filters/wsola_internals.cc:77: // horizontal sum function, so we have to
condense manually.
1 -> _MM_SHUFFLE(0, 0, 0, 1)

Probably also worth a comment that the only purpose of the shuffle is to copy
the second-least-significant float to the last position.

[DaleCurtis, 2016-11-22 22:46:50]

The CQ bit was checked by dalecurtis@chromium.org

[commit-bot: I haz the power, 2016-11-22 22:47:36]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-11-23 00:08:17]

CQ is committing da patch.

Bot data: {"patchset_id": 20001, "attempt_start_ts": 1479854810745900,
"parent_rev": "61de69e9af98ecae27e451a9813dcd97b3676ca5", "commit_rev":
"a91182448cf4f6fb4089a7ea3afa1e65a4fbe4d1"}

[commit-bot: I haz the power, 2016-11-23 00:08:41]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2016-11-23 00:11:26]

Description was changed from

==========
Add SSE and NEON intrinsics for WSOLA's MultiChannelDotProduct().

When adapting 10 seconds of 192kHz stereo audio at a 2.0 playback
rate, this provides a ~9x speedup on ARM and a ~3x speedup on X86.

NEON: 24986ms -> 2695ms
SSE: 3014ms -> 894ms

No tests added since the existing tests verify this functionality
correctly and it continues to pass as before.

BUG=666813
TEST=n5x for ARM, z620 for X86.
==========

to

==========
Add SSE and NEON intrinsics for WSOLA's MultiChannelDotProduct().

When adapting 10 seconds of 192kHz stereo audio at a 2.0 playback
rate, this provides a ~9x speedup on ARM and a ~3x speedup on X86.

NEON: 24986ms -> 2695ms
SSE: 3014ms -> 894ms

No tests added since the existing tests verify this functionality
correctly and it continues to pass as before.

BUG=666813
TEST=n5x for ARM, z620 for X86.

Committed: https://crrev.com/77fd2aecaa7e1db995e158fb095a72be385e333c
Cr-Commit-Position: refs/heads/master@{#434036}
==========

[commit-bot: I haz the power, 2016-11-23 00:11:27]

Patchset 2 (id:??) landed as
https://crrev.com/77fd2aecaa7e1db995e158fb095a72be385e333c
Cr-Commit-Position: refs/heads/master@{#434036}