(This is a re-land---without the real_fourier.h changes---of 11716, which was reverted in 11726.)
BUG=webrtc:5520
Review URL: https://codereview.webrtc.org/1731153002
Cr-Commit-Position: refs/heads/master@{#11742}
Reason for revert:
Breaks downstream compilation using webrtc/common_audio/real_fourier.h. Let's chat tomorrow on how to coordinate a re-land.
Original issue's description:
> Replace scoped_ptr with unique_ptr in webrtc/common_audio/
>
> BUG=webrtc:5520
>
> Committed: https://crrev.com/79d7a499c0c3e1de8f5ad1138236f0386701053f
> Cr-Commit-Position: refs/heads/master@{#11716}
TBR=henrik.lundin@webrtc.org,kwiberg@webrtc.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=webrtc:5520
Review URL: https://codereview.webrtc.org/1726043002
Cr-Commit-Position: refs/heads/master@{#11726}
As part of the review, refactored AudioConverter into internal derived
classes, each focused on one type of conversion. A factory method
returns the correct converter (or chain of converters, via
CompositionConverter).
BUG=b/18938079
R=rojer@google.com
Review URL: https://webrtc-codereview.appspot.com/35699004
Cr-Commit-Position: refs/heads/master@{#8322}
git-svn-id: http://webrtc.googlecode.com/svn/trunk@8322 4adac7df-926f-26a2-2b94-8c16560cd09d
Ideally, PushSincResampler would have very little overhead on
SincResampler. This gets closer to that ideal.
Replace std::min/max and floor with inline functions. Add a benchmark
test to verify the improvement.
On a MacBook Retina, this results in PushSincResampler::Resample()
accounting for ~1% of CPU usage on voe_cmd_test vs the earlier ~2%
(with ISAC16 and 48 kHz audio devices).
Using the new benchmark, this results in a performance improvement of:
16 -> 44.1 : 1.7x
16 -> 48 : 1.9x
32 -> 44.1 : 1.6x
32 -> 48 : 1.7x
44.1 -> 16 : 1.5x
44.1 -> 32 : 1.7x
44.1 -> 48 : 1.7x
48 -> 16 : 1.5x
48 -> 32 : 1.5x
48 -> 44.1 : 1.8x
R=turaj@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/2157005
git-svn-id: http://webrtc.googlecode.com/svn/trunk@4695 4adac7df-926f-26a2-2b94-8c16560cd09d
Replace the BlockSize() workaround we were using previously to support
the push wrapper with the upstream request_frames interface. This
requires a bit of a trick to ensure we don't add more delay than
necessary. On the first pass we use a dummy Resample() call in order to
prime the buffer such that all later calls only require a single input
request through Run().
Notably, this brings in an optimized loop condition, improving
performance by ~2% - 3% on tested platforms and avoids a 20% performance
hit with clang. This addresses issue2041.
Only negligible changes to the PushSincResamplerTest SNR thresholds, due
to a fractional sample adjustment in output delay.
This still retains the per-instance CPU detection, as webrtc lacks a
LazyInstance helper for static initialization.
Ideally, we would adopt SetRatio() in PushSincResampler's
InitializeIfNeeded() for on-the-fly changes, but this will require a way
to update request_frames.
The diff against Chromium upstream is available here:
https://codereview.chromium.org/19470003
BUG=2041
TESTED=unit tests, voe_cmd_test in loopback running through all codecs
with 44.1 kHz and 48 kHz device formats using a stereo mic.
R=dalecurtis@chromium.org
Review URL: https://webrtc-codereview.appspot.com/1838004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@4406 4adac7df-926f-26a2-2b94-8c16560cd09d
