1 2016-07-20 v1.6.0 "Khaki Campbell Duck"
2 This release improves upon the VP9 encoder and speeds up the encoding and
6 This release is ABI incompatible with 1.5.0 due to a new 'color_range' enum
7 in vpx_image and some minor changes to the VP8_COMP structure.
9 The default key frame interval for VP9 has changed from 128 to 9999.
12 A core focus has been performance for low end Intel processors. SSSE3
13 instructions such as 'pshufb' have been avoided and instructions have been
14 reordered to better accommodate the more constrained pipelines.
16 As a result, devices based on Celeron processors have seen substantial
17 decoding improvements. From Indian Runner Duck to Javan Whistling Duck,
18 decoding speed improved between 10 and 30%. Between Javan Whistling Duck
19 and Khaki Campbell Duck, it improved another 10 to 15%.
21 While Celeron benefited most, Core-i5 also improved 5% and 10% between the
24 Realtime performance for WebRTC for both speed and quality has received a
28 A number of fuzzing issues, found variously by Mozilla, Chromium and others,
29 have been fixed and we strongly recommend updating.
31 2015-11-09 v1.5.0 "Javan Whistling Duck"
32 This release improves upon the VP9 encoder and speeds up the encoding and
36 This release is ABI incompatible with 1.4.0. It drops deprecated VP8
37 controls and adds a variety of VP9 controls for testing.
39 The vpxenc utility now prefers VP9 by default.
42 Faster VP9 encoding and decoding
43 Smaller library size by combining functions used by VP8 and VP9
46 A variety of fuzzing issues
48 2015-04-03 v1.4.0 "Indian Runner Duck"
49 This release includes significant improvements to the VP9 codec.
52 This release is ABI incompatible with 1.3.0. It drops the compatibility
53 layer, requiring VPX_IMG_FMT_* instead of IMG_FMT_*, and adds several codec
57 Faster VP9 encoding and decoding
58 Multithreaded VP9 decoding (tile and frame-based)
59 Multithreaded VP9 encoding - on by default
60 YUV 4:2:2 and 4:4:4 support in VP9
61 10 and 12bit support in VP9
62 64bit ARM support by replacing ARM assembly with intrinsics
65 Fixes a VP9 bitstream issue in Profile 1. This only affected non-YUV 4:2:0
69 Frame Parallel decoding fails for segmented and non-420 files.
71 2013-11-15 v1.3.0 "Forest"
72 This release introduces the VP9 codec in a backward-compatible way.
73 All existing users of VP8 can continue to use the library without
74 modification. However, some VP8 options do not map to VP9 in the same manner.
76 The VP9 encoder in this release is not feature complete. Users interested in
77 the encoder are advised to use the git master branch and discuss issues on
81 This release is ABI and API compatible with Duclair (v1.0.0). Users
82 of older releases should refer to the Upgrading notes in this document
86 Get rid of bashisms in the main build scripts
87 Added usage info on command line options
88 Add lossless compression mode
90 Add additional Mac OS X targets: 10.7, 10.8 and 10.9 (darwin11-13)
91 Add option to disable documentation
92 configure: add --enable-external-build support
93 make: support V=1 as short form of verbose=yes
94 configure: support mingw-w64
95 configure: support hardfloat armv7 CHOSTS
96 configure: add support for android x86
97 Add estimated completion time to vpxenc
98 Don't exit on decode errors in vpxenc
99 vpxenc: support scaling prior to encoding
100 vpxdec: support scaling output
101 vpxenc: improve progress indicators with --skip
102 msvs: Don't link to winmm.lib
103 Add a new script for producing vcxproj files
104 Produce Visual Studio 10 and 11 project files
105 Produce Windows Phone project files
106 msvs-build: use msbuild for vs >= 2005
107 configure: default configure log to config.log
108 Add encoding option --static-thresh
111 Miscellaneous speed optimizations for VP8 and VP9.
114 In general, quality is consistent with the Eider release.
117 This release represents approximately a year of engineering effort,
118 and contains multiple bug fixes. Please refer to git history for details.
122 This release acts as a checkpoint for a large amount of internal refactoring
123 and testing. It also contains a number of small bugfixes, so all users are
124 encouraged to upgrade.
127 This release is ABI and API compatible with Duclair (v1.0.0). Users
128 of older releases should refer to the Upgrading notes in this
129 document for that release.
132 VP8 optimizations for MIPS dspr2
133 vpxenc: add -quiet option
136 Encoder and decoder speed is consistent with the Eider release.
139 In general, quality is consistent with the Eider release.
141 Minor tweaks to ARNR filtering
142 Minor improvements to real time encoding with multiple temporal layers
145 Fixes multithreaded encoder race condition in loopfilter
146 Fixes multi-resolution threaded encoding
147 Fix potential encoder dead-lock after picture resize
150 2012-05-09 v1.1.0 "Eider"
151 This introduces a number of enhancements, mostly focused on real-time
152 encoding. In addition, it fixes a decoder bug (first introduced in
153 Duclair) so all users of that release are encouraged to upgrade.
156 This release is ABI and API compatible with Duclair (v1.0.0). Users
157 of older releases should refer to the Upgrading notes in this
158 document for that release.
160 This release introduces a new temporal denoiser, controlled by the
161 VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not
162 currently take a strength parameter, so the control is effectively
163 a boolean - zero (off) or non-zero (on). For compatibility with
164 existing applications, the values accepted are the same as those
165 for the spatial denoiser (0-6). The temporal denoiser is enabled
166 by default, and the older spatial denoiser may be restored by
167 configuring with --disable-temporal-denoising. The temporal denoiser
168 is more computationally intensive than the spatial one.
170 This release removes support for a legacy, decode only API that was
171 supported, but deprecated, at the initial release of libvpx
172 (v0.9.0). This is not expected to have any impact. If you are
173 impacted, you can apply a reversion to commit 2bf8fb58 locally.
174 Please update to the latest libvpx API if you are affected.
177 Adds a motion compensated temporal denoiser to the encoder, which
178 gives higher quality than the older spatial denoiser. (See above
179 for notes on upgrading).
181 In addition, support for new compilers and platforms were added,
183 improved support for XCode
184 Android x86 NDK build
188 Changing resolution with vpx_codec_enc_config_set() is now
189 supported. Previously, reinitializing the codec was required to
190 change the input resolution.
192 The vpxenc application has initial support for producing multiple
193 encodes from the same input in one call. Resizing is not yet
194 supported, but varying other codec parameters is. Use -- to
195 delineate output streams. Options persist from one stream to the
198 Also, the vpxenc application will now use a keyframe interval of
199 5 seconds by default. Use the --kf-max-dist option to override.
202 Decoder performance improved 2.5% versus Duclair. Encoder speed is
203 consistent with Duclair for most material. Two pass encoding of
204 slideshow-like material will see significant improvements.
206 Large realtime encoding speed gains at a small quality expense are
207 possible by configuring the on-the-fly bitpacking experiment with
208 --enable-onthefly-bitpacking. Realtime encoder can be up to 13%
209 faster (ARM) depending on the number of threads and bitrate
210 settings. This technique sees constant gain over the 5-16 speed
211 range. For VC style input the loss seen is up to 0.2dB. See commit
212 52cf4dca for further details.
215 On the whole, quality is consistent with the Duclair release. Some
218 Reduced blockiness in easy sections by applying a penalty to
221 Improved quality of static sections (like slideshows) with
224 Improved keyframe sizing with multiple temporal layers
227 Corrected alt-ref contribution to frame rate for visible updates
228 to the alt-ref buffer. This affected applications making manual
229 usage of the frame reference flags, or temporal layers.
231 Additional constraints were added to disable multi-frame quality
232 enhancement (MFQE) in sections of the frame where there is motion.
235 Fixed corruption issues when vpx_codec_enc_config_set() was called
236 with spatial resampling enabled.
238 Fixed a decoder error introduced in Duclair where the segmentation
239 map was not being reinitialized on keyframes (#378)
242 2012-01-27 v1.0.0 "Duclair"
243 Our fourth named release, focused on performance and features related to
244 real-time encoding. It also fixes a decoder crash bug introduced in
245 v0.9.7, so all users of that release are encouraged to upgrade.
248 This release is ABI incompatible with prior releases of libvpx, so the
249 "major" version number has been bumped to 1. You must recompile your
250 applications against the latest version of the libvpx headers. The
251 API remains compatible, and this should not require code changes in most
255 This release introduces several substantial new features to the encoder,
256 of particular interest to real time streaming applications.
258 Temporal scalability allows the encoder to produce a stream that can
259 be decimated to different frame rates, with independent rate targetting
262 Multiframe quality enhancement postprocessing can make visual quality
263 more consistent in the presence of frames that are substantially
264 different quality than the surrounding frames, as in the temporal
265 scalability case and in some forced keyframe scenarios.
267 Multiple-resolution encoding support allows the encoding of the
268 same content at different resolutions faster than encoding them
272 Optimization targets for this release included the decoder and the real-
273 time modes of the encoder. Decoder speed on x86 has improved 10.5% with
274 this release. Encoder improvements followed a curve where speeds 1-3
275 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
276 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
280 Encoder quality in the single stream case is consistent with the Cayuga
284 This release fixes an OOB read decoder crash bug present in v0.9.7
285 related to the clamping of motion vectors in SPLITMV blocks. This
286 behavior could be triggered by corrupt input or by starting
287 decoding from a P-frame.
290 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
291 This is an incremental bugfix release against Cayuga. All users of that
292 release are strongly encouraged to upgrade.
294 - Fix potential OOB reads (cdae03a)
296 An unbounded out of bounds read was discovered when the
297 decoder was requested to perform error concealment (new in
298 Cayuga) given a frame with corrupt partition sizes.
300 A bounded out of bounds read was discovered affecting all
301 versions of libvpx. Given an multipartition input frame that
302 is truncated between the mode/mv partition and the first
303 residiual paritition (in the block of partition offsets), up
304 to 3 extra bytes could have been read from the source buffer.
305 The code will not take any action regardless of the contents
306 of these undefined bytes, as the truncated buffer is detected
307 immediately following the read based on the calculated
308 starting position of the coefficient partition.
310 - Fix potential error concealment crash when the very first frame
311 is missing or corrupt (a609be5)
313 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
315 - Revert 1-pass CBR rate control changes (e961317)
316 Further testing showed this change produced undesirable visual
317 artifacts, rolling back for now.
320 2011-08-02 v0.9.7 "Cayuga"
321 Our third named release, focused on a faster, higher quality, encoder.
324 This release is backwards compatible with Aylesbury (v0.9.5) and
325 Bali (v0.9.6). Users of older releases should refer to the Upgrading
326 notes in this document for that release.
329 Stereo 3D format support for vpxenc
330 Runtime detection of available processor cores.
331 Allow specifying --end-usage by enum name
332 vpxdec: test for frame corruption
333 vpxenc: add quantizer histogram display
334 vpxenc: add rate histogram display
335 Set VPX_FRAME_IS_DROPPABLE
336 update configure for ios sdk 4.3
337 Avoid text relocations in ARM vp8 decoder
338 Generate a vpx.pc file for pkg-config.
339 New ways of passing encoded data between encoder and decoder.
342 This release includes across-the-board speed improvements to the
343 encoder. On x86, these measure at approximately 11.5% in Best mode,
344 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
345 On ARM Cortex A9 with Neon extensions, real-time encoding of video
346 telephony content is 35% faster than Bali on single core and 48%
347 faster on multi-core. On the NVidia Tegra2 platform, real time
348 encoding is 40% faster than Bali.
350 Decoder speed was not a priority for this release, but improved
351 approximately 8.4% on x86.
353 Reduce motion vector search on alt-ref frame.
354 Encoder loopfilter running in its own thread
355 Reworked loopfilter to precalculate more parameters
356 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
357 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
358 Removed redundant checks
359 Reduced structure sizes
360 utilize preload in ARMv6 MC/LPF/Copy routines
361 ARM optimized quantization, dfct, variance, subtract
362 Increase chrow row alignment to 16 bytes.
363 disable trellis optimization for first pass
364 Write SSSE3 sub-pixel filter function
365 Improve SSE2 half-pixel filter funtions
366 Add vp8_sub_pixel_variance16x8_ssse3 function
367 Reduce unnecessary distortion computation
368 Use diamond search to replace full search
369 Preload reference area in sub-pixel motion search (real-time mode)
372 This release focused primarily on one-pass use cases, including
373 video conferencing. Low latency data rate control was significantly
374 improved, improving streamability over bandwidth constrained links.
375 Added support for error concealment, allowing frames to maintain
376 visual quality in the presence of substantial packet loss.
378 Add rc_max_intra_bitrate_pct control
379 Limit size of initial keyframe in one-pass.
380 Improve framerate adaptation
381 Improved 1-pass CBR rate control
382 Improved KF insertion after fades to still.
383 Improved key frame detection.
384 Improved activity masking (lower PSNR impact for same SSIM boost)
385 Improved interaction between GF and ARFs
386 Adding error-concealment to the decoder.
387 Adding support for independent partitions
388 Adjusted rate-distortion constants
392 Removed firstpass motion map
393 Fix parallel make install
394 Fix multithreaded encoding for 1 MB wide frame
395 Fixed iwalsh_neon build problems with RVDS4.1
396 Fix semaphore emulation, spin-wait intrinsics on Windows
397 Fix build with xcode4 and simplify GLOBAL.
398 Mark ARM asm objects as allowing a non-executable stack.
399 Fix vpxenc encoding incorrect webm file header on big endian
402 2011-03-07 v0.9.6 "Bali"
403 Our second named release, focused on a faster, higher quality, encoder.
406 This release is backwards compatible with Aylesbury (v0.9.5). Users
407 of older releases should refer to the Upgrading notes in this
408 document for that release.
411 vpxenc --psnr shows a summary when encode completes
412 --tune=ssim option to enable activity masking
413 improved postproc visualizations for development
414 updated support for Apple iOS to SDK 4.2
415 query decoder to determine which reference frames were updated
416 implemented error tracking in the decoder
417 fix pipe support on windows
420 Primary focus was on good quality mode, speed 0. Average improvement
421 on x86 about 40%, up to 100% on user-generated content at that speed.
422 Best quality mode speed improved 35%, and realtime speed 10-20%. This
423 release also saw significant improvement in realtime encoding speed
426 Improved encoder threading
427 Dont pick encoder filter level when loopfilter is disabled.
428 Avoid double copying of key frames into alt and golden buffer
430 x86 sse2 temporal filter
431 SSSE3 version of fast quantizer
432 vp8_rd_pick_best_mbsegmentation code restructure
433 Adjusted breakout RD for SPLITMV
434 Changed segmentation check order
435 Improved rd_pick_intra4x4block
436 Adds armv6 optimized variance calculation
437 ARMv6 optimized sad16x16
438 ARMv6 optimized half pixel variance calculations
439 Full search SAD function optimization in SSE4.1
440 Improve MV prediction accuracy to achieve performance gain
441 Improve MV prediction in vp8_pick_inter_mode() for speed>3
444 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
445 also includes support for "activity masking," which greatly improves
446 SSIM at the expense of PSNR. For now, this feature is available with
447 the --tune=ssim option. Further experimentation in this area
448 is ongoing. This release also introduces a new rate control mode
449 called "CQ," which changes the allocation of bits within a clip to
450 the sections where they will have the most visual impact.
452 Tuning for the more exact quantizer.
453 Relax rate control for last few frames
455 Limit key frame quantizer for forced key frames.
457 Add simple version of activity masking.
458 make rdmult adaptive for intra in quantizer RDO
459 cap the best quantizer for 2nd order DC
460 change the threshold of DC check for encode breakout
463 Fix crash on Sparc Solaris.
464 Fix counter of fixed keyframe distance
465 ARNR filter pointer update bug fix
466 Fixed use of motion percentage in KF/GF group calc
467 Changed condition for using RD in Intra Mode
468 Fix encoder real-time only configuration.
469 Fix ARM encoder crash with multiple token partitions
470 Fixed bug first cluster timecode of webm file is wrong.
471 Fixed various encoder bugs with odd-sized images
472 vp8e_get_preview fixed when spatial resampling enabled
473 quantizer: fix assertion in fast quantizer path
474 Allocate source buffers to be multiples of 16
475 Fix for manual Golden frame frequency
476 Fix drastic undershoot in long form content
479 2010-10-28 v0.9.5 "Aylesbury"
480 Our first named release, focused on a faster decoder, and a better encoder.
483 This release incorporates backwards-incompatible changes to the
484 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
487 * the -q (quiet) option has been removed, and replaced with
488 -v (verbose). the output is quiet by default. Use -v to see
489 the version number of the binary.
491 * The default behavior is now to write output to a single file
492 instead of individual frames. The -y option has been removed.
493 Y4M output is the default.
495 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
496 options must be specified.
498 $ ivfdec -o OUTPUT INPUT
499 $ vpxdec --i420 -o OUTPUT INPUT
501 * If an output file is not specified, the default is to write
502 Y4M to stdout. This makes piping more natural.
504 $ ivfdec -y -o - INPUT | ...
507 * The output file has additional flexibility for formatting the
508 filename. It supports escape characters for constructing a
509 filename from the width, height, and sequence number. This
510 replaces the -p option. To get the equivalent:
512 $ ivfdec -p frame INPUT
513 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
516 * The output file must be specified with -o, rather than as the
519 $ ivfenc <options> INPUT OUTPUT
520 $ vpxenc <options> -o OUTPUT INPUT
522 * The output defaults to webm. To get IVF output, use the --ivf
525 $ ivfenc <options> INPUT OUTPUT.ivf
526 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
530 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
531 vpxdec supports .webm input
532 vpxdec writes .y4m by default
533 vpxenc writes .webm output by default
534 vpxenc --psnr now shows the average/overall PSNR at the end
535 ARM platforms now support runtime cpu detection
536 vpxdec visualizations added for motion vectors, block modes, references
537 vpxdec now silent by default
538 vpxdec --progress shows frame-by-frame timing information
539 vpxenc supports the distinction between --fps and --timebase
540 NASM is now a supported assembler
541 configure: enable PIC for shared libs by default
542 configure: add --enable-small
543 configure: support for ppc32-linux-gcc
544 configure: support for sparc-solaris-gcc
547 Improve handling of invalid frames
548 Fix valgrind errors in the NEON loop filters.
549 Fix loopfilter delta zero transitions
550 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
551 Build fixes for darwin-icc
554 20-40% (average 28%) improvement in libvpx decoder speed,
556 Rewrite vp8_short_walsh4x4_sse2()
557 Optimizations on the loopfilters.
558 Miscellaneous improvements for Atom
559 Add 4-tap version of 2nd-pass ARMv6 MC filter.
560 Improved multithread utilization
561 Better instruction choices on x86
562 reorder data to use wider instructions
563 Update NEON wide idcts
564 Make block access to frame buffer sequential
565 Improved subset block search
566 Bilinear subpixel optimizations for ssse3.
567 Decrease memory footprint
569 Encoder speed improvements (percentage gain not measured):
570 Skip unnecessary search of identical frames
571 Add SSE2 subtract functions
572 Improve bounds checking in vp8_diamond_search_sadx4()
573 Added vp8_fast_quantize_b_sse2
576 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
577 encoding mode, and up to 60% improvement on very noisy, still
578 or slow moving source video
580 Motion compensated temporal filter for Alt-Ref Noise Reduction
581 Improved use of trellis quantization on 2nd order Y blocks
582 Tune effect of motion on KF/GF boost in two pass
583 Allow coefficient optimization for good quality speed 0.
584 Improved control of active min quantizer for two pass.
585 Enable ARFs for non-lagged compress
589 Disable frame dropping by default
590 Improved multithreaded performance
591 Improved Force Key Frame Behaviour
592 Increased rate control buffer level precision
593 Fix bug in 1st pass motion compensation
594 ivfenc: correct fixed kf interval, --disable-kf
596 Changed above and left context data layout
597 Rework idct calling structure.
598 Removed unnecessary MB_MODE_INFO copies
599 x86: SSSE3 sixtap prediction
600 Reworked IDCT to include reconstruction (add) step
601 Swap alt/gold/new/last frame buffer ptrs instead of copying.
602 Improve SSE2 loopfilter functions
603 Change bitreader to use a larger window.
604 Avoid loopfilter reinitialization when possible
606 Normalize quantizer's zero bin and rounding factors
607 Add trellis quantization.
608 Make the quantizer exact.
609 Updates to ARNR filtering algorithm
610 Fix breakout thresh computation for golden & AltRef frames
611 Redo the forward 4x4 dct
612 Improve the accuracy of forward walsh-hadamard transform
613 Further adjustment of RD behaviour with Q and Zbin.
615 Allow linking of libs built with MinGW to MSVC
616 Fix target auto-detection on mingw32
617 Allow --cpu= to work for x86.
618 configure: pass original arguments through to make dist
619 Fix builds without runtime CPU detection
620 msvs: fix install of codec sources
621 msvs: Change devenv.com command line for better msys support
622 msvs: Add vs9 targets.
623 Add x86_64-linux-icc target
625 Potential crashes on older MinGW builds
626 Fix two-pass framrate for Y4M input.
627 Fixed simple loop filter, other crashes on ARM v6
628 arm: fix missing dependency with --enable-shared
629 configure: support directories containing .o
630 Replace pinsrw (SSE) with MMX instructions
631 apple: include proper mach primatives
632 Fixed rate control bug with long key frame interval.
633 Fix DSO link errors on x86-64 when not using a version script
634 Fixed buffer selection for UV in AltRef filtering
639 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
640 * Speed optimizations
643 * Prevent out-of-bounds accesses on invalid data
644 - Build system updates:
645 * Detect toolchain to be used automatically for native builds
646 * Support building shared libraries
647 * Better autotools emulation (--prefix, --libdir, DESTDIR)
649 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
653 - Initial open source release. Welcome to WebM and VP8!