12013-11-15 v1.3.0 "Forest"
2  This release introduces the VP9 codec in a backward-compatible way.
3  All existing users of VP8 can continue to use the library without
4  modification. However, some VP8 options do not map to VP9 in the same manner.
5
6  The VP9 encoder in this release is not feature complete. Users interested in
7  the encoder are advised to use the git master branch and discuss issues on
8  libvpx mailing lists.
9
10  - Upgrading:
11    This release is ABI and API compatible with Duclair (v1.0.0). Users
12    of older releases should refer to the Upgrading notes in this document
13    for that release.
14
15  - Enhancements:
16      Get rid of bashisms in the main build scripts
17      Added usage info on command line options
18      Add lossless compression mode
19      Dll build of libvpx
20      Add additional Mac OS X targets: 10.7, 10.8 and 10.9 (darwin11-13)
21      Add option to disable documentation
22      configure: add --enable-external-build support
23      make: support V=1 as short form of verbose=yes
24      configure: support mingw-w64
25      configure: support hardfloat armv7 CHOSTS
26      configure: add support for android x86
27      Add estimated completion time to vpxenc
28      Don't exit on decode errors in vpxenc
29      vpxenc: support scaling prior to encoding
30      vpxdec: support scaling output
31      vpxenc: improve progress indicators with --skip
32      msvs: Don't link to winmm.lib
33      Add a new script for producing vcxproj files
34      Produce Visual Studio 10 and 11 project files
35      Produce Windows Phone project files
36      msvs-build: use msbuild for vs >= 2005
37      configure: default configure log to config.log
38      Add encoding option --static-thresh
39
40  - Speed:
41      Miscellaneous speed optimizations for VP8 and VP9.
42
43  - Quality:
44      In general, quality is consistent with the Eider release.
45
46  - Bug Fixes:
47      This release represents approximately a year of engineering effort,
48      and contains multiple bug fixes. Please refer to git history for details.
49
50
512012-12-21 v1.2.0
52  This release acts as a checkpoint for a large amount of internal refactoring
53  and testing. It also contains a number of small bugfixes, so all users are
54  encouraged to upgrade.
55
56  - Upgrading:
57    This release is ABI and API compatible with Duclair (v1.0.0). Users
58    of older releases should refer to the Upgrading notes in this
59    document for that release.
60
61  - Enhancements:
62      VP8 optimizations for MIPS dspr2
63      vpxenc: add -quiet option
64
65  - Speed:
66      Encoder and decoder speed is consistent with the Eider release.
67
68  - Quality:
69      In general, quality is consistent with the Eider release.
70
71      Minor tweaks to ARNR filtering
72      Minor improvements to real time encoding with multiple temporal layers
73
74  - Bug Fixes:
75      Fixes multithreaded encoder race condition in loopfilter
76      Fixes multi-resolution threaded encoding
77      Fix potential encoder dead-lock after picture resize
78
79
802012-05-09 v1.1.0 "Eider"
81  This introduces a number of enhancements, mostly focused on real-time
82  encoding. In addition, it fixes a decoder bug (first introduced in
83  Duclair) so all users of that release are encouraged to upgrade.
84
85  - Upgrading:
86    This release is ABI and API compatible with Duclair (v1.0.0). Users
87    of older releases should refer to the Upgrading notes in this
88    document for that release.
89
90    This release introduces a new temporal denoiser, controlled by the
91    VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not
92    currently take a strength parameter, so the control is effectively
93    a boolean - zero (off) or non-zero (on). For compatibility with
94    existing applications, the values accepted are the same as those
95    for the spatial denoiser (0-6). The temporal denoiser is enabled
96    by default, and the older spatial denoiser may be restored by
97    configuring with --disable-temporal-denoising. The temporal denoiser
98    is more computationally intensive than the spatial one.
99
100    This release removes support for a legacy, decode only API that was
101    supported, but deprecated, at the initial release of libvpx
102    (v0.9.0). This is not expected to have any impact. If you are
103    impacted, you can apply a reversion to commit 2bf8fb58 locally.
104    Please update to the latest libvpx API if you are affected.
105
106  - Enhancements:
107      Adds a motion compensated temporal denoiser to the encoder, which
108      gives higher quality than the older spatial denoiser. (See above
109      for notes on upgrading).
110
111      In addition, support for new compilers and platforms were added,
112      including:
113        improved support for XCode
114        Android x86 NDK build
115        OS/2 support
116        SunCC support
117
118      Changing resolution with vpx_codec_enc_config_set() is now
119      supported. Previously, reinitializing the codec was required to
120      change the input resolution.
121
122      The vpxenc application has initial support for producing multiple
123      encodes from the same input in one call. Resizing is not yet
124      supported, but varying other codec parameters is. Use -- to
125      delineate output streams. Options persist from one stream to the
126      next.
127
128      Also, the vpxenc application will now use a keyframe interval of
129      5 seconds by default. Use the --kf-max-dist option to override.
130
131  - Speed:
132      Decoder performance improved 2.5% versus Duclair. Encoder speed is
133      consistent with Duclair for most material. Two pass encoding of
134      slideshow-like material will see significant improvements.
135
136      Large realtime encoding speed gains at a small quality expense are
137      possible by configuring the on-the-fly bitpacking experiment with
138      --enable-onthefly-bitpacking. Realtime encoder can be up to 13%
139      faster (ARM) depending on the number of threads and bitrate
140      settings. This technique sees constant gain over the 5-16 speed
141      range. For VC style input the loss seen is up to 0.2dB. See commit
142      52cf4dca for further details.
143
144  - Quality:
145      On the whole, quality is consistent with the Duclair release. Some
146      tweaks:
147
148        Reduced blockiness in easy sections by applying a penalty to
149        intra modes.
150
151        Improved quality of static sections (like slideshows) with
152        two pass encoding.
153
154        Improved keyframe sizing with multiple temporal layers
155
156  - Bug Fixes:
157      Corrected alt-ref contribution to frame rate for visible updates
158      to the alt-ref buffer. This affected applications making manual
159      usage of the frame reference flags, or temporal layers.
160
161      Additional constraints were added to disable multi-frame quality
162      enhancement (MFQE) in sections of the frame where there is motion.
163      (#392)
164
165      Fixed corruption issues when vpx_codec_enc_config_set() was called
166      with spatial resampling enabled.
167
168      Fixed a decoder error introduced in Duclair where the segmentation
169      map was not being reinitialized on keyframes (#378)
170
171
1722012-01-27 v1.0.0 "Duclair"
173  Our fourth named release, focused on performance and features related to
174  real-time encoding. It also fixes a decoder crash bug introduced in
175  v0.9.7, so all users of that release are encouraged to upgrade.
176
177  - Upgrading:
178      This release is ABI incompatible with prior releases of libvpx, so the
179      "major" version number has been bumped to 1. You must recompile your
180      applications against the latest version of the libvpx headers. The
181      API remains compatible, and this should not require code changes in most
182      applications.
183
184  - Enhancements:
185      This release introduces several substantial new features to the encoder,
186      of particular interest to real time streaming applications.
187
188      Temporal scalability allows the encoder to produce a stream that can
189      be decimated to different frame rates, with independent rate targetting
190      for each substream.
191
192      Multiframe quality enhancement postprocessing can make visual quality
193      more consistent in the presence of frames that are substantially
194      different quality than the surrounding frames, as in the temporal
195      scalability case and in some forced keyframe scenarios.
196
197      Multiple-resolution encoding support allows the encoding of the
198      same content at different resolutions faster than encoding them
199      separately.
200
201  - Speed:
202      Optimization targets for this release included the decoder and the real-
203      time modes of the encoder. Decoder speed on x86 has improved 10.5% with
204      this release. Encoder improvements followed a curve where speeds 1-3
205      improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
206      1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
207      Cayuga release.
208
209  - Quality:
210      Encoder quality in the single stream case is consistent with the Cayuga
211      release.
212
213  - Bug Fixes:
214      This release fixes an OOB read decoder crash bug present in v0.9.7
215      related to the clamping of motion vectors in SPLITMV blocks. This
216      behavior could be triggered by corrupt input or by starting
217      decoding from a P-frame.
218
219
2202011-08-15 v0.9.7-p1 "Cayuga" patch 1
221  This is an incremental bugfix release against Cayuga. All users of that
222  release are strongly encouraged to upgrade.
223
224    - Fix potential OOB reads (cdae03a)
225
226          An unbounded out of bounds read was discovered when the
227          decoder was requested to perform error concealment (new in
228          Cayuga) given a frame with corrupt partition sizes.
229
230          A bounded out of bounds read was discovered affecting all
231          versions of libvpx. Given an multipartition input frame that
232          is truncated between the mode/mv partition and the first
233          residiual paritition (in the block of partition offsets), up
234          to 3 extra bytes could have been read from the source buffer.
235          The code will not take any action regardless of the contents
236          of these undefined bytes, as the truncated buffer is detected
237          immediately following the read based on the calculated
238          starting position of the coefficient partition.
239
240    - Fix potential error concealment crash when the very first frame
241      is missing or corrupt (a609be5)
242
243    - Fix significant artifacts in error concealment (a4c2211, 99d870a)
244
245    - Revert 1-pass CBR rate control changes (e961317)
246      Further testing showed this change produced undesirable visual
247      artifacts, rolling back for now.
248
249
2502011-08-02 v0.9.7 "Cayuga"
251  Our third named release, focused on a faster, higher quality, encoder.
252
253  - Upgrading:
254    This release is backwards compatible with Aylesbury (v0.9.5) and
255    Bali (v0.9.6). Users of older releases should refer to the Upgrading
256    notes in this document for that release.
257
258  - Enhancements:
259          Stereo 3D format support for vpxenc
260          Runtime detection of available processor cores.
261          Allow specifying --end-usage by enum name
262          vpxdec: test for frame corruption
263          vpxenc: add quantizer histogram display
264          vpxenc: add rate histogram display
265          Set VPX_FRAME_IS_DROPPABLE
266          update configure for ios sdk 4.3
267          Avoid text relocations in ARM vp8 decoder
268          Generate a vpx.pc file for pkg-config.
269          New ways of passing encoded data between encoder and decoder.
270
271  - Speed:
272      This release includes across-the-board speed improvements to the
273      encoder. On x86, these measure at approximately 11.5% in Best mode,
274      21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
275      On ARM Cortex A9 with Neon extensions, real-time encoding of video
276      telephony content is 35% faster than Bali on single core and 48%
277      faster on multi-core. On the NVidia Tegra2 platform, real time
278      encoding is 40% faster than Bali.
279
280      Decoder speed was not a priority for this release, but improved
281      approximately 8.4% on x86.
282
283          Reduce motion vector search on alt-ref frame.
284          Encoder loopfilter running in its own thread
285          Reworked loopfilter to precalculate more parameters
286          SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
287          Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
288          Removed redundant checks
289          Reduced structure sizes
290          utilize preload in ARMv6 MC/LPF/Copy routines
291          ARM optimized quantization, dfct, variance, subtract
292          Increase chrow row alignment to 16 bytes.
293          disable trellis optimization for first pass
294          Write SSSE3 sub-pixel filter function
295          Improve SSE2 half-pixel filter funtions
296          Add vp8_sub_pixel_variance16x8_ssse3 function
297          Reduce unnecessary distortion computation
298          Use diamond search to replace full search
299          Preload reference area in sub-pixel motion search (real-time mode)
300
301  - Quality:
302      This release focused primarily on one-pass use cases, including
303      video conferencing. Low latency data rate control was significantly
304      improved, improving streamability over bandwidth constrained links.
305      Added support for error concealment, allowing frames to maintain
306      visual quality in the presence of substantial packet loss.
307
308          Add rc_max_intra_bitrate_pct control
309          Limit size of initial keyframe in one-pass.
310          Improve framerate adaptation
311          Improved 1-pass CBR rate control
312          Improved KF insertion after fades to still.
313          Improved key frame detection.
314          Improved activity masking (lower PSNR impact for same SSIM boost)
315          Improved interaction between GF and ARFs
316          Adding error-concealment to the decoder.
317          Adding support for independent partitions
318          Adjusted rate-distortion constants
319
320
321  - Bug Fixes:
322          Removed firstpass motion map
323          Fix parallel make install
324          Fix multithreaded encoding for 1 MB wide frame
325          Fixed iwalsh_neon build problems with RVDS4.1
326          Fix semaphore emulation, spin-wait intrinsics on Windows
327          Fix build with xcode4 and simplify GLOBAL.
328          Mark ARM asm objects as allowing a non-executable stack.
329          Fix vpxenc encoding incorrect webm file header on big endian
330
331
3322011-03-07 v0.9.6 "Bali"
333  Our second named release, focused on a faster, higher quality, encoder.
334
335  - Upgrading:
336    This release is backwards compatible with Aylesbury (v0.9.5). Users
337    of older releases should refer to the Upgrading notes in this
338    document for that release.
339
340  - Enhancements:
341      vpxenc --psnr shows a summary when encode completes
342      --tune=ssim option to enable activity masking
343      improved postproc visualizations for development
344      updated support for Apple iOS to SDK 4.2
345      query decoder to determine which reference frames were updated
346      implemented error tracking in the decoder
347      fix pipe support on windows
348
349  - Speed:
350      Primary focus was on good quality mode, speed 0. Average improvement
351      on x86 about 40%, up to 100% on user-generated content at that speed.
352      Best quality mode speed improved 35%, and realtime speed 10-20%. This
353      release also saw significant improvement in realtime encoding speed
354      on ARM platforms.
355
356        Improved encoder threading
357        Dont pick encoder filter level when loopfilter is disabled.
358        Avoid double copying of key frames into alt and golden buffer
359        FDCT optimizations.
360        x86 sse2 temporal filter
361        SSSE3 version of fast quantizer
362        vp8_rd_pick_best_mbsegmentation code restructure
363        Adjusted breakout RD for SPLITMV
364        Changed segmentation check order
365        Improved rd_pick_intra4x4block
366        Adds armv6 optimized variance calculation
367        ARMv6 optimized sad16x16
368        ARMv6 optimized half pixel variance calculations
369        Full search SAD function optimization in SSE4.1
370        Improve MV prediction accuracy to achieve performance gain
371        Improve MV prediction in vp8_pick_inter_mode() for speed>3
372
373  - Quality:
374      Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
375      also includes support for "activity masking," which greatly improves
376      SSIM at the expense of PSNR. For now, this feature is available with
377      the --tune=ssim option. Further experimentation in this area
378      is ongoing. This release also introduces a new rate control mode
379      called "CQ," which changes the allocation of bits within a clip to
380      the sections where they will have the most visual impact.
381
382        Tuning for the more exact quantizer.
383        Relax rate control for last few frames
384        CQ Mode
385        Limit key frame quantizer for forced key frames.
386        KF/GF Pulsing
387        Add simple version of activity masking.
388        make rdmult adaptive for intra in quantizer RDO
389        cap the best quantizer for 2nd order DC
390        change the threshold of DC check for encode breakout
391
392  - Bug Fixes:
393      Fix crash on Sparc Solaris.
394      Fix counter of fixed keyframe distance
395      ARNR filter pointer update bug fix
396      Fixed use of motion percentage in KF/GF group calc
397      Changed condition for using RD in Intra Mode
398      Fix encoder real-time only configuration.
399      Fix ARM encoder crash with multiple token partitions
400      Fixed bug first cluster timecode of webm file is wrong.
401      Fixed various encoder bugs with odd-sized images
402      vp8e_get_preview fixed when spatial resampling enabled
403      quantizer: fix assertion in fast quantizer path
404      Allocate source buffers to be multiples of 16
405      Fix for manual Golden frame frequency
406      Fix drastic undershoot in long form content
407
408
4092010-10-28 v0.9.5 "Aylesbury"
410  Our first named release, focused on a faster decoder, and a better encoder.
411
412  - Upgrading:
413    This release incorporates backwards-incompatible changes to the
414    ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
415
416    vpxdec
417      * the -q (quiet) option has been removed, and replaced with
418        -v (verbose). the output is quiet by default. Use -v to see
419        the version number of the binary.
420
421      * The default behavior is now to write output to a single file
422        instead of individual frames. The -y option has been removed.
423        Y4M output is the default.
424
425      * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
426        options must be specified.
427
428          $ ivfdec -o OUTPUT INPUT
429          $ vpxdec --i420 -o OUTPUT INPUT
430
431      * If an output file is not specified, the default is to write
432        Y4M to stdout. This makes piping more natural.
433
434          $ ivfdec -y -o - INPUT | ...
435          $ vpxdec INPUT | ...
436
437      * The output file has additional flexibility for formatting the
438        filename. It supports escape characters for constructing a
439        filename from the width, height, and sequence number. This
440        replaces the -p option. To get the equivalent:
441
442          $ ivfdec -p frame INPUT
443          $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
444
445    vpxenc
446      * The output file must be specified with -o, rather than as the
447        last argument.
448
449          $ ivfenc <options> INPUT OUTPUT
450          $ vpxenc <options> -o OUTPUT INPUT
451
452      * The output defaults to webm. To get IVF output, use the --ivf
453        option.
454
455          $ ivfenc <options> INPUT OUTPUT.ivf
456          $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
457
458
459  - Enhancements:
460      ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
461      vpxdec supports .webm input
462      vpxdec writes .y4m by default
463      vpxenc writes .webm output by default
464      vpxenc --psnr now shows the average/overall PSNR at the end
465      ARM platforms now support runtime cpu detection
466      vpxdec visualizations added for motion vectors, block modes, references
467      vpxdec now silent by default
468      vpxdec --progress shows frame-by-frame timing information
469      vpxenc supports the distinction between --fps and --timebase
470      NASM is now a supported assembler
471      configure: enable PIC for shared libs by default
472      configure: add --enable-small
473      configure: support for ppc32-linux-gcc
474      configure: support for sparc-solaris-gcc
475
476  - Bugs:
477      Improve handling of invalid frames
478      Fix valgrind errors in the NEON loop filters.
479      Fix loopfilter delta zero transitions
480      Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
481      Build fixes for darwin-icc
482
483  - Speed:
484      20-40% (average 28%) improvement in libvpx decoder speed,
485      including:
486        Rewrite vp8_short_walsh4x4_sse2()
487        Optimizations on the loopfilters.
488        Miscellaneous improvements for Atom
489        Add 4-tap version of 2nd-pass ARMv6 MC filter.
490        Improved multithread utilization
491        Better instruction choices on x86
492        reorder data to use wider instructions
493        Update NEON wide idcts
494        Make block access to frame buffer sequential
495        Improved subset block search
496        Bilinear subpixel optimizations for ssse3.
497        Decrease memory footprint
498
499      Encoder speed improvements (percentage gain not measured):
500        Skip unnecessary search of identical frames
501        Add SSE2 subtract functions
502        Improve bounds checking in vp8_diamond_search_sadx4()
503        Added vp8_fast_quantize_b_sse2
504
505  - Quality:
506      Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
507      encoding mode, and up to 60% improvement on very noisy, still
508      or slow moving source video
509
510        Motion compensated temporal filter for Alt-Ref Noise Reduction
511        Improved use of trellis quantization on 2nd order Y blocks
512        Tune effect of motion on KF/GF boost in two pass
513        Allow coefficient optimization for good quality speed 0.
514        Improved control of active min quantizer for two pass.
515        Enable ARFs for non-lagged compress
516
5172010-09-02 v0.9.2
518  - Enhancements:
519      Disable frame dropping by default
520      Improved multithreaded performance
521      Improved Force Key Frame Behaviour
522      Increased rate control buffer level precision
523      Fix bug in 1st pass motion compensation
524      ivfenc: correct fixed kf interval, --disable-kf
525  - Speed:
526      Changed above and left context data layout
527      Rework idct calling structure.
528      Removed unnecessary MB_MODE_INFO copies
529      x86: SSSE3 sixtap prediction
530      Reworked IDCT to include reconstruction (add) step
531      Swap alt/gold/new/last frame buffer ptrs instead of copying.
532      Improve SSE2 loopfilter functions
533      Change bitreader to use a larger window.
534      Avoid loopfilter reinitialization when possible
535  - Quality:
536      Normalize quantizer's zero bin and rounding factors
537      Add trellis quantization.
538      Make the quantizer exact.
539      Updates to ARNR filtering algorithm
540      Fix breakout thresh computation for golden & AltRef frames
541      Redo the forward 4x4 dct
542      Improve the accuracy of forward walsh-hadamard transform
543      Further adjustment of RD behaviour with Q and Zbin.
544  - Build System:
545      Allow linking of libs built with MinGW to MSVC
546      Fix target auto-detection on mingw32
547      Allow --cpu= to work for x86.
548      configure: pass original arguments through to make dist
549      Fix builds without runtime CPU detection
550      msvs: fix install of codec sources
551      msvs: Change devenv.com command line for better msys support
552      msvs: Add vs9 targets.
553      Add x86_64-linux-icc target
554  - Bugs:
555      Potential crashes on older MinGW builds
556      Fix two-pass framrate for Y4M input.
557      Fixed simple loop filter, other crashes on ARM v6
558      arm: fix missing dependency with --enable-shared
559      configure: support directories containing .o
560      Replace pinsrw (SSE) with MMX instructions
561      apple: include proper mach primatives
562      Fixed rate control bug with long key frame interval.
563      Fix DSO link errors on x86-64 when not using a version script
564      Fixed buffer selection for UV in AltRef filtering
565
566
5672010-06-17 v0.9.1
568  - Enhancements:
569      * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
570      * Speed optimizations
571  - Bugfixes:
572      * Rate control
573      * Prevent out-of-bounds accesses on invalid data
574  - Build system updates:
575      * Detect toolchain to be used automatically for native builds
576      * Support building shared libraries
577      * Better autotools emulation (--prefix, --libdir, DESTDIR)
578  - Updated LICENSE
579      * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
580
581
5822010-05-18 v0.9.0
583  - Initial open source release. Welcome to WebM and VP8!
584
585