John Koleszar [Wed, 16 May 2012 23:25:51 +0000 (16:25 -0700)]
Build unit tests monolithically
Rework unit tests to have a single executable rather than many, which
should avoid pollution of the visual studio project namespace, improve
build times, and make it easier to use the gtest test sharding system
when we get these going on the continuous build cluster.
John Koleszar [Tue, 22 May 2012 17:37:20 +0000 (10:37 -0700)]
configure: initial support for CXX, CXXFLAGS variables
Use CXX rather than assuming g++ to invoke the compiler. Also introduce
separate CXXFLAGS, as certain CFLAGS we enable by default cause warnings
with g++.
Alexis Ballier [Sat, 12 May 2012 19:45:13 +0000 (15:45 -0400)]
Allow target autodetection to work when cross-compiling.
Allow CHOST to override the gcc -dumpmachine output. This allows to
use the target autodetection code when cross compiling by setting the
CHOST variable.
On Gentoo, we would like to support easy cross-compilation, and for
libvpx this would basically mean copying the code in
build/make/configure.sh to setup the right --target option. It seems a
lot easier to let it guess by itself.
Another option I considered was using CROSS-gcc instead but this would
not work for our multilib setups: They use gcc -m32 to build 32bits
binaries and gcc -m32 -dumpmachine will output the 64bits version,
which would then make libvpx wrongly believe it is building for a
64bits architecture.
John Koleszar [Tue, 8 May 2012 18:23:42 +0000 (11:23 -0700)]
Use consistent range for VP8E_SET_NOISE_SENSITIVITY
Accept the same range of inputs for the VP8E_SET_NOISE_SENSITIVITY
control, regardless of whether temporal denoising is enabled or not.
This is important for maintaining compatibility with existing
applications.
John Koleszar [Fri, 4 May 2012 16:46:57 +0000 (09:46 -0700)]
Formalize encodeframe.c forward delclarations
Change If4321cc5 fixed a bug caused by forward declarations not being
kept in sync across C files, resulting in a function call with the
wrong arguments. The commit moves the affected function declarations
into a header file, along with the other symbols from encodeframe.c
that were being sloppily shared.
Attila Nagy [Mon, 16 Apr 2012 07:23:57 +0000 (10:23 +0300)]
Make global data const
Removes all runtime initialization of global data. This commit is a
squashed version of the following series cherry-picked from master.
This is necessary because of a change that was merged to the tester
that depends on the scaler being moved to the RTCD framework, which
is a worthwhile thing to include in Eider anyway.
- a91b42f02 Makes all global data in entropy.c const
- b35a0db0e Makes all global data in tokenize.c const
- 441cac8ea Makes all mode token tables const
- 5948a0210 Ports vpx_xcaler to new RTCD method
- 317d4244c Makes all mode token tables const part 2
John Koleszar [Thu, 19 Apr 2012 17:00:33 +0000 (10:00 -0700)]
multi-res: restore v1.0.0 API
Move the notion of 0 bitrate implying skip deeper into the codec,
rather than doing it at the multi-encoder API level. This preserves
v1.0.0 ABI compatibility, rather than forcing a bump to v2.0.0 over a
minor change. Also, this allows the case where the application can
selectively enable and disable the larger resolution(s) without having
to reinitialize the codec instace (for instance, if no target is
receiving the full resolution stream).
It's not clear how deep to push this check. It may be valuable to
allow the framerate adaptation code to run, for example. Currently put
the check as early as possible for simplicity, should reevaluate this
as this feature gains real use.
Original patch by Ginn Chen <ginn.chen@oracle.com> against libvpx
v0.9.0.
I've forward-ported it to the current version (which mostly
involved removing hunks that were no longer relevant), since I've
given up on getting Ginn to submit this upstream himself.
Besides imposing a performance penalty at startup in most
configurations, these relocations break the dynamic linker for
native Fennec, since it does not support them at all.
Adrian Grange [Fri, 27 Apr 2012 16:01:17 +0000 (09:01 -0700)]
Modified ARNR MC-filter to ignore ARF frame
The ARNR filter uses MC to find the best match between the
ARF and other nearby frames in the filter-set. Since the
ARF is a member of the filter-set, MC in that case is
unnecesssary. This patch modifies the filter so it does
not apply MC in this case.
Adrian Grange [Thu, 26 Apr 2012 22:41:38 +0000 (15:41 -0700)]
Removed MV costing from ARNR filtering
The ARNR filter uses a motion compensated temporal filter,
but the motion estimation implementation accounts for the
cost of the mv in its decision making process. The ARNR
filter uses a dummy cost table initialized to 0 as a way
to ignore the mv costs (which are irrelevant to the filter).
This CL modifies the ARNR filter implementation so that
the mv costing is ignored without the requirement for
dummy tables.
Attila Nagy [Tue, 24 Apr 2012 12:33:44 +0000 (15:33 +0300)]
Moves error concealment allocations from common parts to decoder
The backup MODE_INFO buffer used in the error concealment was
allocated in the codec common parts allocation even though this is a
decoder only resource. Moved the allocation to the decoder side.
No need to update_mode_info_border as mode_info buffers are zero
allocated.
This fixes also a potential memory leak as the EC overlaps buffer was not
properly released before reallocation after a frame size change.
Ralph Giles [Thu, 5 Jan 2012 21:05:05 +0000 (15:05 -0600)]
Have vpxenc use a default kf_max_rate of 5 seconds.
Rather than using the static default maximum keyframe spacing
provided by vpx_codec_enc_config_default() set the default
value to 5 times the frame rate. Five seconds is too long for
live streaming applications, but is a compromise between seek
efficiency and giving the encoder freedom to choose keyframe
locations.
The five second value is from James Zern's suggestion in
http://article.gmane.org/gmane.comp.multimedia.webm.user/2945
Attila Nagy [Wed, 25 Apr 2012 11:04:55 +0000 (14:04 +0300)]
Fix loopfilter race condition in multithreaded encoder
Race was introduced by https://gerrit.chromium.org/gerrit/15563.
If loopfilter related config params were changed between frames, or
after a KEY frame, there could be a mismatch between encoder's and
decoder's recontructed image. In worst case, when frame buffers are
reallocated because of a size change, the loopfilter could
do an invalid data access (segmentation fault).
Fixes:
Sync with the loopfilter before applying any encoder changes in
vp8_change_config().
Moved the loopfilter synching to the top of
encode_frame_to_data_rate() so that it's done before any alteration of
the encoder.
Attila Nagy [Tue, 17 Apr 2012 07:40:56 +0000 (10:40 +0300)]
Shares one set of RD costs tables between all encoding threads
RD costs were local to MACROBLOCK data and had to be copied all the
time to each thread's MACROBLOCK data. Tables moved to a common place
and only pointers are setup for each encoding thread.
vp8_cost_tokens() generates 'int' costs so changed all types to be
int (i.e. removed unsigned).
NOTE: Could do some more cleaning in vp8cx_init_mbrthread_data().
Attila Nagy [Mon, 23 Apr 2012 12:20:07 +0000 (15:20 +0300)]
Optimizes precalculated decoder block ptrs&offs
The block pointers and offset do not need to be calculated for every
frame. Block internal predictors can be update once when decoder is
allocated. Destination and previous buffer offsets have to be updated
just when frame size is changing.
Attila Nagy [Mon, 23 Apr 2012 09:53:30 +0000 (12:53 +0300)]
Removes duplication of key frame mode probabilities
Key frame macrobock and block mode probabilities are constant.
Remove the allocation of tables for each codec instance and use
instead the default const prob tables.
John Koleszar [Thu, 19 Apr 2012 17:00:33 +0000 (10:00 -0700)]
multi-res: restore v1.0.0 API
Move the notion of 0 bitrate implying skip deeper into the codec,
rather than doing it at the multi-encoder API level. This preserves
v1.0.0 ABI compatibility, rather than forcing a bump to v2.0.0 over a
minor change. Also, this allows the case where the application can
selectively enable and disable the larger resolution(s) without having
to reinitialize the codec instace (for instance, if no target is
receiving the full resolution stream).
It's not clear how deep to push this check. It may be valuable to
allow the framerate adaptation code to run, for example. Currently put
the check as early as possible for simplicity, should reevaluate this
as this feature gains real use.
John Koleszar [Fri, 20 Apr 2012 17:17:57 +0000 (10:17 -0700)]
vp8_change_config: don't force kf with spatial resampling
Look for changes in the codec's configured w/h instead of its active
w/h when forcing keyframes. Otherwise calls to vp8_change_config()
will force a keyframe when spatial resampling is active.
Scott LaVarnway [Thu, 19 Apr 2012 21:35:20 +0000 (17:35 -0400)]
Makes all mode token tables const part 2
(see Change I9b2ccc88: Makes all mode token tables const)
Further remove runtime table initialization and use
precalculated const data. Data footprint reduced
by 4112 bytes.
John Koleszar [Wed, 18 Apr 2012 23:03:40 +0000 (16:03 -0700)]
rtcd: serialize function pointer initialization
Ensure that RTCD function pointers are set at most once, to silence
some data race warnings. Implementation provided for POSIX threads and
Win32, with the prior unsynchronized behavior left in place for other
platforms.
John Koleszar [Mon, 16 Apr 2012 19:22:59 +0000 (12:22 -0700)]
correct alt-ref contribution to frame rate
When producing an invisible ARF, the time stamp counters aren't
updated since the last time stamp is seen by the codec twice. The
prior code was trapping this case with refresh_alt_ref, but this isn't
correct for other uses of the ARF. Instead, use the show_frame flag.
Attila Nagy [Wed, 11 Apr 2012 12:20:13 +0000 (15:20 +0300)]
Fix vpx_rtcd.h dependency in Android.mk
Failed to build on Linux (as described in Android.mk) with NDK r7b.
Set vpx_rtcd.h dependency after libvpx sources are added to
LOCAL_SRC_FILES so that vpx_rtcd.h is generated before any libvpx file
is touched.
Deb Mukherjee [Tue, 3 Apr 2012 21:02:58 +0000 (14:02 -0700)]
Fixes to disable MFQE when there is motion.
This patch includes:
1. fixes to disable block based termporal mixing when motion
is detected (because this version of mfqe only handles zero motion).
2. The criterion used for determining whether to mix or
not are changed to use squared differences rather than
absolute differences.
3. Additional checks on color mismatch and excessive block
flatness added. If the block as decoded has very low activity
it is unlikely to yield benefits for mixing.
John Koleszar [Fri, 30 Mar 2012 18:34:36 +0000 (11:34 -0700)]
MFQE: apply threshold to subblocks and chroma.
In cases where you have a flat background occluded by a moving object
of similar luminosity in the foreground, it was likely that the
foreground blocks would persist for a few frames after the background
is uncovered. This is particularly noticable when the object has a
different color than the background, so add the chroma planes in as an
additional check.
In addition, for block sizes of 8 and 16, the luma threshold is
applied on four subblocks independently, which helps when only part of
the background in the block has been uncovered.
This fixes issue #392, which includes a test clip to reproduce the
issue.
Johann [Mon, 2 Apr 2012 22:08:18 +0000 (15:08 -0700)]
Allow disabling disabled codecs
When using 'make dist' after --disable-vp8[encoder|decoder] it would
fail to recognize the option. This would only occur when also specifying
--enable-install-docs and --enable-install-srcs but not
--enable-codec-srcs
Including vpx/ fixes builds with --enable-codec-srcs
James Zern [Fri, 30 Mar 2012 01:13:27 +0000 (18:13 -0700)]
tools/wrap-commit-msg.py: fix file truncation
truncate() operates from the current file pointer position. On at least
Linux specifying 0 without resetting the pointer will pad the file with
zeros to the current offset.
John Koleszar [Wed, 28 Mar 2012 23:41:16 +0000 (16:41 -0700)]
FTFY: support wordwrapping commit messages
It's common for commit messages to be wrapped at odd places. git-gui
is often to blame. Adds support for automatically fixing up these
messages if running ftfy --amend, and adds a new option --msg-only for
fixing only the commit message.
Scott LaVarnway [Fri, 23 Mar 2012 20:55:22 +0000 (16:55 -0400)]
MB_MODE_INFO size reduction
Reduced the size of the struct by 8 bytes, which would be
a memory savings of 64800 bytes for 1080 resolutions. Had
an extra byte, so created an is_4x4 for B_PRED or SPLITMV
modes. This simplified the mode checks in
vp8_reset_mb_tokens_context and vp8_decode_mb_tokens.
John Koleszar [Tue, 27 Mar 2012 23:28:41 +0000 (16:28 -0700)]
FTFY: an automated style corrector
This is a utility for applying a limited amount of style correction on
a change-by-change basis. Rather than a big-bang reformatting, this
tool attempts to only correct the style in diff hunks that you touch.
This should make the cosmetic changes small enough that we can mix them
with functional changes without destroying the diffs, and there's an
escape hatch for separating the reformatting to a second commit for
purists and cases where it hurts readability.
At this time, the script requires a clean working tree, so run it after
you've commited your changes. Run without arguments, the style
corrections will be applied and left unstaged in your working copy. It
also supports the --amend option, which will automatically amend your
HEAD with the corrected style, and --commit, which will create a new
change dependent on your HEAD that contains only the whitespace changes.
There are a number of ways this could be applied in an automated manner
if this proves to be useful, either on a project-wide or per-user
basis. This doesn't buy anything in terms of real code quality, the
intent here would be to keep formatting nits out of review comments in
favor of more meaningful ones and help people whose habitual style
doesn't match the baseline.
Requires astyle[1] 1.24 or newer.
[1]: http://astyle.sourceforge.net/
Scott LaVarnway [Fri, 23 Mar 2012 19:50:08 +0000 (15:50 -0400)]
New vp8_decode_mb_tokens()
This new vp8_decode_mb_tokens() uses a modified version of
WebP's GetCoeffs function. For now, the dequant does not
occur in GetCoeffs.
Tests showed performance improvements up to 2.5% depending
on material.
projects / libvpx / log