1 2017-01-04 v1.7.0 "Mandarin Duck"
2 This release focused on high bit depth performance (10/12 bit) and vp9
6 This release is ABI incompatible due to new vp9 encoder features.
8 Frame parallel decoding for vp9 has been removed.
11 vp9 encoding supports additional threads with --row-mt. This can be greater
12 than the number of tiles.
14 Two new vp9 encoder options have been added:
18 Additional tooling for respecting the vp9 "level" profiles has been added.
21 A variety of fuzzing issues.
22 vp8 threading fix for ARM.
23 Codec control VP9_SET_SKIP_LOOP_FILTER fixed.
24 Reject invalid multi resolution configurations.
26 2017-01-09 v1.6.1 "Long Tailed Duck"
27 This release improves upon the VP9 encoder and speeds up the encoding and
31 This release is ABI compatible with 1.6.0.
34 Faster VP9 encoding and decoding.
35 High bit depth builds now provide similar speed for 8 bit encode and decode
36 for x86 targets. Other platforms and higher bit depth improvements are in
40 A variety of fuzzing issues.
42 2016-07-20 v1.6.0 "Khaki Campbell Duck"
43 This release improves upon the VP9 encoder and speeds up the encoding and
47 This release is ABI incompatible with 1.5.0 due to a new 'color_range' enum
48 in vpx_image and some minor changes to the VP8_COMP structure.
50 The default key frame interval for VP9 has changed from 128 to 9999.
53 A core focus has been performance for low end Intel processors. SSSE3
54 instructions such as 'pshufb' have been avoided and instructions have been
55 reordered to better accommodate the more constrained pipelines.
57 As a result, devices based on Celeron processors have seen substantial
58 decoding improvements. From Indian Runner Duck to Javan Whistling Duck,
59 decoding speed improved between 10 and 30%. Between Javan Whistling Duck
60 and Khaki Campbell Duck, it improved another 10 to 15%.
62 While Celeron benefited most, Core-i5 also improved 5% and 10% between the
65 Realtime performance for WebRTC for both speed and quality has received a
69 A number of fuzzing issues, found variously by Mozilla, Chromium and others,
70 have been fixed and we strongly recommend updating.
72 2015-11-09 v1.5.0 "Javan Whistling Duck"
73 This release improves upon the VP9 encoder and speeds up the encoding and
77 This release is ABI incompatible with 1.4.0. It drops deprecated VP8
78 controls and adds a variety of VP9 controls for testing.
80 The vpxenc utility now prefers VP9 by default.
83 Faster VP9 encoding and decoding
84 Smaller library size by combining functions used by VP8 and VP9
87 A variety of fuzzing issues
89 2015-04-03 v1.4.0 "Indian Runner Duck"
90 This release includes significant improvements to the VP9 codec.
93 This release is ABI incompatible with 1.3.0. It drops the compatibility
94 layer, requiring VPX_IMG_FMT_* instead of IMG_FMT_*, and adds several codec
98 Faster VP9 encoding and decoding
99 Multithreaded VP9 decoding (tile and frame-based)
100 Multithreaded VP9 encoding - on by default
101 YUV 4:2:2 and 4:4:4 support in VP9
102 10 and 12bit support in VP9
103 64bit ARM support by replacing ARM assembly with intrinsics
106 Fixes a VP9 bitstream issue in Profile 1. This only affected non-YUV 4:2:0
110 Frame Parallel decoding fails for segmented and non-420 files.
112 2013-11-15 v1.3.0 "Forest"
113 This release introduces the VP9 codec in a backward-compatible way.
114 All existing users of VP8 can continue to use the library without
115 modification. However, some VP8 options do not map to VP9 in the same manner.
117 The VP9 encoder in this release is not feature complete. Users interested in
118 the encoder are advised to use the git master branch and discuss issues on
119 libvpx mailing lists.
122 This release is ABI and API compatible with Duclair (v1.0.0). Users
123 of older releases should refer to the Upgrading notes in this document
127 Get rid of bashisms in the main build scripts
128 Added usage info on command line options
129 Add lossless compression mode
131 Add additional Mac OS X targets: 10.7, 10.8 and 10.9 (darwin11-13)
132 Add option to disable documentation
133 configure: add --enable-external-build support
134 make: support V=1 as short form of verbose=yes
135 configure: support mingw-w64
136 configure: support hardfloat armv7 CHOSTS
137 configure: add support for android x86
138 Add estimated completion time to vpxenc
139 Don't exit on decode errors in vpxenc
140 vpxenc: support scaling prior to encoding
141 vpxdec: support scaling output
142 vpxenc: improve progress indicators with --skip
143 msvs: Don't link to winmm.lib
144 Add a new script for producing vcxproj files
145 Produce Visual Studio 10 and 11 project files
146 Produce Windows Phone project files
147 msvs-build: use msbuild for vs >= 2005
148 configure: default configure log to config.log
149 Add encoding option --static-thresh
152 Miscellaneous speed optimizations for VP8 and VP9.
155 In general, quality is consistent with the Eider release.
158 This release represents approximately a year of engineering effort,
159 and contains multiple bug fixes. Please refer to git history for details.
163 This release acts as a checkpoint for a large amount of internal refactoring
164 and testing. It also contains a number of small bugfixes, so all users are
165 encouraged to upgrade.
168 This release is ABI and API compatible with Duclair (v1.0.0). Users
169 of older releases should refer to the Upgrading notes in this
170 document for that release.
173 VP8 optimizations for MIPS dspr2
174 vpxenc: add -quiet option
177 Encoder and decoder speed is consistent with the Eider release.
180 In general, quality is consistent with the Eider release.
182 Minor tweaks to ARNR filtering
183 Minor improvements to real time encoding with multiple temporal layers
186 Fixes multithreaded encoder race condition in loopfilter
187 Fixes multi-resolution threaded encoding
188 Fix potential encoder dead-lock after picture resize
191 2012-05-09 v1.1.0 "Eider"
192 This introduces a number of enhancements, mostly focused on real-time
193 encoding. In addition, it fixes a decoder bug (first introduced in
194 Duclair) so all users of that release are encouraged to upgrade.
197 This release is ABI and API compatible with Duclair (v1.0.0). Users
198 of older releases should refer to the Upgrading notes in this
199 document for that release.
201 This release introduces a new temporal denoiser, controlled by the
202 VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not
203 currently take a strength parameter, so the control is effectively
204 a boolean - zero (off) or non-zero (on). For compatibility with
205 existing applications, the values accepted are the same as those
206 for the spatial denoiser (0-6). The temporal denoiser is enabled
207 by default, and the older spatial denoiser may be restored by
208 configuring with --disable-temporal-denoising. The temporal denoiser
209 is more computationally intensive than the spatial one.
211 This release removes support for a legacy, decode only API that was
212 supported, but deprecated, at the initial release of libvpx
213 (v0.9.0). This is not expected to have any impact. If you are
214 impacted, you can apply a reversion to commit 2bf8fb58 locally.
215 Please update to the latest libvpx API if you are affected.
218 Adds a motion compensated temporal denoiser to the encoder, which
219 gives higher quality than the older spatial denoiser. (See above
220 for notes on upgrading).
222 In addition, support for new compilers and platforms were added,
224 improved support for XCode
225 Android x86 NDK build
229 Changing resolution with vpx_codec_enc_config_set() is now
230 supported. Previously, reinitializing the codec was required to
231 change the input resolution.
233 The vpxenc application has initial support for producing multiple
234 encodes from the same input in one call. Resizing is not yet
235 supported, but varying other codec parameters is. Use -- to
236 delineate output streams. Options persist from one stream to the
239 Also, the vpxenc application will now use a keyframe interval of
240 5 seconds by default. Use the --kf-max-dist option to override.
243 Decoder performance improved 2.5% versus Duclair. Encoder speed is
244 consistent with Duclair for most material. Two pass encoding of
245 slideshow-like material will see significant improvements.
247 Large realtime encoding speed gains at a small quality expense are
248 possible by configuring the on-the-fly bitpacking experiment with
249 --enable-onthefly-bitpacking. Realtime encoder can be up to 13%
250 faster (ARM) depending on the number of threads and bitrate
251 settings. This technique sees constant gain over the 5-16 speed
252 range. For VC style input the loss seen is up to 0.2dB. See commit
253 52cf4dca for further details.
256 On the whole, quality is consistent with the Duclair release. Some
259 Reduced blockiness in easy sections by applying a penalty to
262 Improved quality of static sections (like slideshows) with
265 Improved keyframe sizing with multiple temporal layers
268 Corrected alt-ref contribution to frame rate for visible updates
269 to the alt-ref buffer. This affected applications making manual
270 usage of the frame reference flags, or temporal layers.
272 Additional constraints were added to disable multi-frame quality
273 enhancement (MFQE) in sections of the frame where there is motion.
276 Fixed corruption issues when vpx_codec_enc_config_set() was called
277 with spatial resampling enabled.
279 Fixed a decoder error introduced in Duclair where the segmentation
280 map was not being reinitialized on keyframes (#378)
283 2012-01-27 v1.0.0 "Duclair"
284 Our fourth named release, focused on performance and features related to
285 real-time encoding. It also fixes a decoder crash bug introduced in
286 v0.9.7, so all users of that release are encouraged to upgrade.
289 This release is ABI incompatible with prior releases of libvpx, so the
290 "major" version number has been bumped to 1. You must recompile your
291 applications against the latest version of the libvpx headers. The
292 API remains compatible, and this should not require code changes in most
296 This release introduces several substantial new features to the encoder,
297 of particular interest to real time streaming applications.
299 Temporal scalability allows the encoder to produce a stream that can
300 be decimated to different frame rates, with independent rate targetting
303 Multiframe quality enhancement postprocessing can make visual quality
304 more consistent in the presence of frames that are substantially
305 different quality than the surrounding frames, as in the temporal
306 scalability case and in some forced keyframe scenarios.
308 Multiple-resolution encoding support allows the encoding of the
309 same content at different resolutions faster than encoding them
313 Optimization targets for this release included the decoder and the real-
314 time modes of the encoder. Decoder speed on x86 has improved 10.5% with
315 this release. Encoder improvements followed a curve where speeds 1-3
316 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
317 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
321 Encoder quality in the single stream case is consistent with the Cayuga
325 This release fixes an OOB read decoder crash bug present in v0.9.7
326 related to the clamping of motion vectors in SPLITMV blocks. This
327 behavior could be triggered by corrupt input or by starting
328 decoding from a P-frame.
331 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
332 This is an incremental bugfix release against Cayuga. All users of that
333 release are strongly encouraged to upgrade.
335 - Fix potential OOB reads (cdae03a)
337 An unbounded out of bounds read was discovered when the
338 decoder was requested to perform error concealment (new in
339 Cayuga) given a frame with corrupt partition sizes.
341 A bounded out of bounds read was discovered affecting all
342 versions of libvpx. Given an multipartition input frame that
343 is truncated between the mode/mv partition and the first
344 residiual paritition (in the block of partition offsets), up
345 to 3 extra bytes could have been read from the source buffer.
346 The code will not take any action regardless of the contents
347 of these undefined bytes, as the truncated buffer is detected
348 immediately following the read based on the calculated
349 starting position of the coefficient partition.
351 - Fix potential error concealment crash when the very first frame
352 is missing or corrupt (a609be5)
354 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
356 - Revert 1-pass CBR rate control changes (e961317)
357 Further testing showed this change produced undesirable visual
358 artifacts, rolling back for now.
361 2011-08-02 v0.9.7 "Cayuga"
362 Our third named release, focused on a faster, higher quality, encoder.
365 This release is backwards compatible with Aylesbury (v0.9.5) and
366 Bali (v0.9.6). Users of older releases should refer to the Upgrading
367 notes in this document for that release.
370 Stereo 3D format support for vpxenc
371 Runtime detection of available processor cores.
372 Allow specifying --end-usage by enum name
373 vpxdec: test for frame corruption
374 vpxenc: add quantizer histogram display
375 vpxenc: add rate histogram display
376 Set VPX_FRAME_IS_DROPPABLE
377 update configure for ios sdk 4.3
378 Avoid text relocations in ARM vp8 decoder
379 Generate a vpx.pc file for pkg-config.
380 New ways of passing encoded data between encoder and decoder.
383 This release includes across-the-board speed improvements to the
384 encoder. On x86, these measure at approximately 11.5% in Best mode,
385 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
386 On ARM Cortex A9 with Neon extensions, real-time encoding of video
387 telephony content is 35% faster than Bali on single core and 48%
388 faster on multi-core. On the NVidia Tegra2 platform, real time
389 encoding is 40% faster than Bali.
391 Decoder speed was not a priority for this release, but improved
392 approximately 8.4% on x86.
394 Reduce motion vector search on alt-ref frame.
395 Encoder loopfilter running in its own thread
396 Reworked loopfilter to precalculate more parameters
397 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
398 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
399 Removed redundant checks
400 Reduced structure sizes
401 utilize preload in ARMv6 MC/LPF/Copy routines
402 ARM optimized quantization, dfct, variance, subtract
403 Increase chrow row alignment to 16 bytes.
404 disable trellis optimization for first pass
405 Write SSSE3 sub-pixel filter function
406 Improve SSE2 half-pixel filter funtions
407 Add vp8_sub_pixel_variance16x8_ssse3 function
408 Reduce unnecessary distortion computation
409 Use diamond search to replace full search
410 Preload reference area in sub-pixel motion search (real-time mode)
413 This release focused primarily on one-pass use cases, including
414 video conferencing. Low latency data rate control was significantly
415 improved, improving streamability over bandwidth constrained links.
416 Added support for error concealment, allowing frames to maintain
417 visual quality in the presence of substantial packet loss.
419 Add rc_max_intra_bitrate_pct control
420 Limit size of initial keyframe in one-pass.
421 Improve framerate adaptation
422 Improved 1-pass CBR rate control
423 Improved KF insertion after fades to still.
424 Improved key frame detection.
425 Improved activity masking (lower PSNR impact for same SSIM boost)
426 Improved interaction between GF and ARFs
427 Adding error-concealment to the decoder.
428 Adding support for independent partitions
429 Adjusted rate-distortion constants
433 Removed firstpass motion map
434 Fix parallel make install
435 Fix multithreaded encoding for 1 MB wide frame
436 Fixed iwalsh_neon build problems with RVDS4.1
437 Fix semaphore emulation, spin-wait intrinsics on Windows
438 Fix build with xcode4 and simplify GLOBAL.
439 Mark ARM asm objects as allowing a non-executable stack.
440 Fix vpxenc encoding incorrect webm file header on big endian
443 2011-03-07 v0.9.6 "Bali"
444 Our second named release, focused on a faster, higher quality, encoder.
447 This release is backwards compatible with Aylesbury (v0.9.5). Users
448 of older releases should refer to the Upgrading notes in this
449 document for that release.
452 vpxenc --psnr shows a summary when encode completes
453 --tune=ssim option to enable activity masking
454 improved postproc visualizations for development
455 updated support for Apple iOS to SDK 4.2
456 query decoder to determine which reference frames were updated
457 implemented error tracking in the decoder
458 fix pipe support on windows
461 Primary focus was on good quality mode, speed 0. Average improvement
462 on x86 about 40%, up to 100% on user-generated content at that speed.
463 Best quality mode speed improved 35%, and realtime speed 10-20%. This
464 release also saw significant improvement in realtime encoding speed
467 Improved encoder threading
468 Dont pick encoder filter level when loopfilter is disabled.
469 Avoid double copying of key frames into alt and golden buffer
471 x86 sse2 temporal filter
472 SSSE3 version of fast quantizer
473 vp8_rd_pick_best_mbsegmentation code restructure
474 Adjusted breakout RD for SPLITMV
475 Changed segmentation check order
476 Improved rd_pick_intra4x4block
477 Adds armv6 optimized variance calculation
478 ARMv6 optimized sad16x16
479 ARMv6 optimized half pixel variance calculations
480 Full search SAD function optimization in SSE4.1
481 Improve MV prediction accuracy to achieve performance gain
482 Improve MV prediction in vp8_pick_inter_mode() for speed>3
485 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
486 also includes support for "activity masking," which greatly improves
487 SSIM at the expense of PSNR. For now, this feature is available with
488 the --tune=ssim option. Further experimentation in this area
489 is ongoing. This release also introduces a new rate control mode
490 called "CQ," which changes the allocation of bits within a clip to
491 the sections where they will have the most visual impact.
493 Tuning for the more exact quantizer.
494 Relax rate control for last few frames
496 Limit key frame quantizer for forced key frames.
498 Add simple version of activity masking.
499 make rdmult adaptive for intra in quantizer RDO
500 cap the best quantizer for 2nd order DC
501 change the threshold of DC check for encode breakout
504 Fix crash on Sparc Solaris.
505 Fix counter of fixed keyframe distance
506 ARNR filter pointer update bug fix
507 Fixed use of motion percentage in KF/GF group calc
508 Changed condition for using RD in Intra Mode
509 Fix encoder real-time only configuration.
510 Fix ARM encoder crash with multiple token partitions
511 Fixed bug first cluster timecode of webm file is wrong.
512 Fixed various encoder bugs with odd-sized images
513 vp8e_get_preview fixed when spatial resampling enabled
514 quantizer: fix assertion in fast quantizer path
515 Allocate source buffers to be multiples of 16
516 Fix for manual Golden frame frequency
517 Fix drastic undershoot in long form content
520 2010-10-28 v0.9.5 "Aylesbury"
521 Our first named release, focused on a faster decoder, and a better encoder.
524 This release incorporates backwards-incompatible changes to the
525 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
528 * the -q (quiet) option has been removed, and replaced with
529 -v (verbose). the output is quiet by default. Use -v to see
530 the version number of the binary.
532 * The default behavior is now to write output to a single file
533 instead of individual frames. The -y option has been removed.
534 Y4M output is the default.
536 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
537 options must be specified.
539 $ ivfdec -o OUTPUT INPUT
540 $ vpxdec --i420 -o OUTPUT INPUT
542 * If an output file is not specified, the default is to write
543 Y4M to stdout. This makes piping more natural.
545 $ ivfdec -y -o - INPUT | ...
548 * The output file has additional flexibility for formatting the
549 filename. It supports escape characters for constructing a
550 filename from the width, height, and sequence number. This
551 replaces the -p option. To get the equivalent:
553 $ ivfdec -p frame INPUT
554 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
557 * The output file must be specified with -o, rather than as the
560 $ ivfenc <options> INPUT OUTPUT
561 $ vpxenc <options> -o OUTPUT INPUT
563 * The output defaults to webm. To get IVF output, use the --ivf
566 $ ivfenc <options> INPUT OUTPUT.ivf
567 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
571 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
572 vpxdec supports .webm input
573 vpxdec writes .y4m by default
574 vpxenc writes .webm output by default
575 vpxenc --psnr now shows the average/overall PSNR at the end
576 ARM platforms now support runtime cpu detection
577 vpxdec visualizations added for motion vectors, block modes, references
578 vpxdec now silent by default
579 vpxdec --progress shows frame-by-frame timing information
580 vpxenc supports the distinction between --fps and --timebase
581 NASM is now a supported assembler
582 configure: enable PIC for shared libs by default
583 configure: add --enable-small
584 configure: support for ppc32-linux-gcc
585 configure: support for sparc-solaris-gcc
588 Improve handling of invalid frames
589 Fix valgrind errors in the NEON loop filters.
590 Fix loopfilter delta zero transitions
591 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
592 Build fixes for darwin-icc
595 20-40% (average 28%) improvement in libvpx decoder speed,
597 Rewrite vp8_short_walsh4x4_sse2()
598 Optimizations on the loopfilters.
599 Miscellaneous improvements for Atom
600 Add 4-tap version of 2nd-pass ARMv6 MC filter.
601 Improved multithread utilization
602 Better instruction choices on x86
603 reorder data to use wider instructions
604 Update NEON wide idcts
605 Make block access to frame buffer sequential
606 Improved subset block search
607 Bilinear subpixel optimizations for ssse3.
608 Decrease memory footprint
610 Encoder speed improvements (percentage gain not measured):
611 Skip unnecessary search of identical frames
612 Add SSE2 subtract functions
613 Improve bounds checking in vp8_diamond_search_sadx4()
614 Added vp8_fast_quantize_b_sse2
617 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
618 encoding mode, and up to 60% improvement on very noisy, still
619 or slow moving source video
621 Motion compensated temporal filter for Alt-Ref Noise Reduction
622 Improved use of trellis quantization on 2nd order Y blocks
623 Tune effect of motion on KF/GF boost in two pass
624 Allow coefficient optimization for good quality speed 0.
625 Improved control of active min quantizer for two pass.
626 Enable ARFs for non-lagged compress
630 Disable frame dropping by default
631 Improved multithreaded performance
632 Improved Force Key Frame Behaviour
633 Increased rate control buffer level precision
634 Fix bug in 1st pass motion compensation
635 ivfenc: correct fixed kf interval, --disable-kf
637 Changed above and left context data layout
638 Rework idct calling structure.
639 Removed unnecessary MB_MODE_INFO copies
640 x86: SSSE3 sixtap prediction
641 Reworked IDCT to include reconstruction (add) step
642 Swap alt/gold/new/last frame buffer ptrs instead of copying.
643 Improve SSE2 loopfilter functions
644 Change bitreader to use a larger window.
645 Avoid loopfilter reinitialization when possible
647 Normalize quantizer's zero bin and rounding factors
648 Add trellis quantization.
649 Make the quantizer exact.
650 Updates to ARNR filtering algorithm
651 Fix breakout thresh computation for golden & AltRef frames
652 Redo the forward 4x4 dct
653 Improve the accuracy of forward walsh-hadamard transform
654 Further adjustment of RD behaviour with Q and Zbin.
656 Allow linking of libs built with MinGW to MSVC
657 Fix target auto-detection on mingw32
658 Allow --cpu= to work for x86.
659 configure: pass original arguments through to make dist
660 Fix builds without runtime CPU detection
661 msvs: fix install of codec sources
662 msvs: Change devenv.com command line for better msys support
663 msvs: Add vs9 targets.
664 Add x86_64-linux-icc target
666 Potential crashes on older MinGW builds
667 Fix two-pass framrate for Y4M input.
668 Fixed simple loop filter, other crashes on ARM v6
669 arm: fix missing dependency with --enable-shared
670 configure: support directories containing .o
671 Replace pinsrw (SSE) with MMX instructions
672 apple: include proper mach primatives
673 Fixed rate control bug with long key frame interval.
674 Fix DSO link errors on x86-64 when not using a version script
675 Fixed buffer selection for UV in AltRef filtering
680 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
681 * Speed optimizations
684 * Prevent out-of-bounds accesses on invalid data
685 - Build system updates:
686 * Detect toolchain to be used automatically for native builds
687 * Support building shared libraries
688 * Better autotools emulation (--prefix, --libdir, DESTDIR)
690 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
694 - Initial open source release. Welcome to WebM and VP8!