1 2015-04-03 v1.4.0 "Indian Runner Duck"
2 This release includes significant improvements to the VP9 codec.
5 This release is ABI incompatible with 1.3.0. It drops the compatibility
6 layer, requiring VPX_IMG_FMT_* instead of IMG_FMT_*, and adds several codec
10 Faster VP9 encoding and decoding
11 Multithreaded VP9 decoding (tile and frame-based)
12 Multithreaded VP9 encoding - on by default
13 YUV 4:2:2 and 4:4:4 support in VP9
14 10 and 12bit support in VP9
15 64bit ARM support by replacing ARM assembly with intrinsics
18 Fixes a VP9 bitstream issue in Profile 1. This only affected non-YUV 4:2:0
22 Frame Parallel decoding fails for segmented and non-420 files.
24 2013-11-15 v1.3.0 "Forest"
25 This release introduces the VP9 codec in a backward-compatible way.
26 All existing users of VP8 can continue to use the library without
27 modification. However, some VP8 options do not map to VP9 in the same manner.
29 The VP9 encoder in this release is not feature complete. Users interested in
30 the encoder are advised to use the git master branch and discuss issues on
34 This release is ABI and API compatible with Duclair (v1.0.0). Users
35 of older releases should refer to the Upgrading notes in this document
39 Get rid of bashisms in the main build scripts
40 Added usage info on command line options
41 Add lossless compression mode
43 Add additional Mac OS X targets: 10.7, 10.8 and 10.9 (darwin11-13)
44 Add option to disable documentation
45 configure: add --enable-external-build support
46 make: support V=1 as short form of verbose=yes
47 configure: support mingw-w64
48 configure: support hardfloat armv7 CHOSTS
49 configure: add support for android x86
50 Add estimated completion time to vpxenc
51 Don't exit on decode errors in vpxenc
52 vpxenc: support scaling prior to encoding
53 vpxdec: support scaling output
54 vpxenc: improve progress indicators with --skip
55 msvs: Don't link to winmm.lib
56 Add a new script for producing vcxproj files
57 Produce Visual Studio 10 and 11 project files
58 Produce Windows Phone project files
59 msvs-build: use msbuild for vs >= 2005
60 configure: default configure log to config.log
61 Add encoding option --static-thresh
64 Miscellaneous speed optimizations for VP8 and VP9.
67 In general, quality is consistent with the Eider release.
70 This release represents approximately a year of engineering effort,
71 and contains multiple bug fixes. Please refer to git history for details.
75 This release acts as a checkpoint for a large amount of internal refactoring
76 and testing. It also contains a number of small bugfixes, so all users are
77 encouraged to upgrade.
80 This release is ABI and API compatible with Duclair (v1.0.0). Users
81 of older releases should refer to the Upgrading notes in this
82 document for that release.
85 VP8 optimizations for MIPS dspr2
86 vpxenc: add -quiet option
89 Encoder and decoder speed is consistent with the Eider release.
92 In general, quality is consistent with the Eider release.
94 Minor tweaks to ARNR filtering
95 Minor improvements to real time encoding with multiple temporal layers
98 Fixes multithreaded encoder race condition in loopfilter
99 Fixes multi-resolution threaded encoding
100 Fix potential encoder dead-lock after picture resize
103 2012-05-09 v1.1.0 "Eider"
104 This introduces a number of enhancements, mostly focused on real-time
105 encoding. In addition, it fixes a decoder bug (first introduced in
106 Duclair) so all users of that release are encouraged to upgrade.
109 This release is ABI and API compatible with Duclair (v1.0.0). Users
110 of older releases should refer to the Upgrading notes in this
111 document for that release.
113 This release introduces a new temporal denoiser, controlled by the
114 VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not
115 currently take a strength parameter, so the control is effectively
116 a boolean - zero (off) or non-zero (on). For compatibility with
117 existing applications, the values accepted are the same as those
118 for the spatial denoiser (0-6). The temporal denoiser is enabled
119 by default, and the older spatial denoiser may be restored by
120 configuring with --disable-temporal-denoising. The temporal denoiser
121 is more computationally intensive than the spatial one.
123 This release removes support for a legacy, decode only API that was
124 supported, but deprecated, at the initial release of libvpx
125 (v0.9.0). This is not expected to have any impact. If you are
126 impacted, you can apply a reversion to commit 2bf8fb58 locally.
127 Please update to the latest libvpx API if you are affected.
130 Adds a motion compensated temporal denoiser to the encoder, which
131 gives higher quality than the older spatial denoiser. (See above
132 for notes on upgrading).
134 In addition, support for new compilers and platforms were added,
136 improved support for XCode
137 Android x86 NDK build
141 Changing resolution with vpx_codec_enc_config_set() is now
142 supported. Previously, reinitializing the codec was required to
143 change the input resolution.
145 The vpxenc application has initial support for producing multiple
146 encodes from the same input in one call. Resizing is not yet
147 supported, but varying other codec parameters is. Use -- to
148 delineate output streams. Options persist from one stream to the
151 Also, the vpxenc application will now use a keyframe interval of
152 5 seconds by default. Use the --kf-max-dist option to override.
155 Decoder performance improved 2.5% versus Duclair. Encoder speed is
156 consistent with Duclair for most material. Two pass encoding of
157 slideshow-like material will see significant improvements.
159 Large realtime encoding speed gains at a small quality expense are
160 possible by configuring the on-the-fly bitpacking experiment with
161 --enable-onthefly-bitpacking. Realtime encoder can be up to 13%
162 faster (ARM) depending on the number of threads and bitrate
163 settings. This technique sees constant gain over the 5-16 speed
164 range. For VC style input the loss seen is up to 0.2dB. See commit
165 52cf4dca for further details.
168 On the whole, quality is consistent with the Duclair release. Some
171 Reduced blockiness in easy sections by applying a penalty to
174 Improved quality of static sections (like slideshows) with
177 Improved keyframe sizing with multiple temporal layers
180 Corrected alt-ref contribution to frame rate for visible updates
181 to the alt-ref buffer. This affected applications making manual
182 usage of the frame reference flags, or temporal layers.
184 Additional constraints were added to disable multi-frame quality
185 enhancement (MFQE) in sections of the frame where there is motion.
188 Fixed corruption issues when vpx_codec_enc_config_set() was called
189 with spatial resampling enabled.
191 Fixed a decoder error introduced in Duclair where the segmentation
192 map was not being reinitialized on keyframes (#378)
195 2012-01-27 v1.0.0 "Duclair"
196 Our fourth named release, focused on performance and features related to
197 real-time encoding. It also fixes a decoder crash bug introduced in
198 v0.9.7, so all users of that release are encouraged to upgrade.
201 This release is ABI incompatible with prior releases of libvpx, so the
202 "major" version number has been bumped to 1. You must recompile your
203 applications against the latest version of the libvpx headers. The
204 API remains compatible, and this should not require code changes in most
208 This release introduces several substantial new features to the encoder,
209 of particular interest to real time streaming applications.
211 Temporal scalability allows the encoder to produce a stream that can
212 be decimated to different frame rates, with independent rate targetting
215 Multiframe quality enhancement postprocessing can make visual quality
216 more consistent in the presence of frames that are substantially
217 different quality than the surrounding frames, as in the temporal
218 scalability case and in some forced keyframe scenarios.
220 Multiple-resolution encoding support allows the encoding of the
221 same content at different resolutions faster than encoding them
225 Optimization targets for this release included the decoder and the real-
226 time modes of the encoder. Decoder speed on x86 has improved 10.5% with
227 this release. Encoder improvements followed a curve where speeds 1-3
228 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
229 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
233 Encoder quality in the single stream case is consistent with the Cayuga
237 This release fixes an OOB read decoder crash bug present in v0.9.7
238 related to the clamping of motion vectors in SPLITMV blocks. This
239 behavior could be triggered by corrupt input or by starting
240 decoding from a P-frame.
243 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
244 This is an incremental bugfix release against Cayuga. All users of that
245 release are strongly encouraged to upgrade.
247 - Fix potential OOB reads (cdae03a)
249 An unbounded out of bounds read was discovered when the
250 decoder was requested to perform error concealment (new in
251 Cayuga) given a frame with corrupt partition sizes.
253 A bounded out of bounds read was discovered affecting all
254 versions of libvpx. Given an multipartition input frame that
255 is truncated between the mode/mv partition and the first
256 residiual paritition (in the block of partition offsets), up
257 to 3 extra bytes could have been read from the source buffer.
258 The code will not take any action regardless of the contents
259 of these undefined bytes, as the truncated buffer is detected
260 immediately following the read based on the calculated
261 starting position of the coefficient partition.
263 - Fix potential error concealment crash when the very first frame
264 is missing or corrupt (a609be5)
266 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
268 - Revert 1-pass CBR rate control changes (e961317)
269 Further testing showed this change produced undesirable visual
270 artifacts, rolling back for now.
273 2011-08-02 v0.9.7 "Cayuga"
274 Our third named release, focused on a faster, higher quality, encoder.
277 This release is backwards compatible with Aylesbury (v0.9.5) and
278 Bali (v0.9.6). Users of older releases should refer to the Upgrading
279 notes in this document for that release.
282 Stereo 3D format support for vpxenc
283 Runtime detection of available processor cores.
284 Allow specifying --end-usage by enum name
285 vpxdec: test for frame corruption
286 vpxenc: add quantizer histogram display
287 vpxenc: add rate histogram display
288 Set VPX_FRAME_IS_DROPPABLE
289 update configure for ios sdk 4.3
290 Avoid text relocations in ARM vp8 decoder
291 Generate a vpx.pc file for pkg-config.
292 New ways of passing encoded data between encoder and decoder.
295 This release includes across-the-board speed improvements to the
296 encoder. On x86, these measure at approximately 11.5% in Best mode,
297 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
298 On ARM Cortex A9 with Neon extensions, real-time encoding of video
299 telephony content is 35% faster than Bali on single core and 48%
300 faster on multi-core. On the NVidia Tegra2 platform, real time
301 encoding is 40% faster than Bali.
303 Decoder speed was not a priority for this release, but improved
304 approximately 8.4% on x86.
306 Reduce motion vector search on alt-ref frame.
307 Encoder loopfilter running in its own thread
308 Reworked loopfilter to precalculate more parameters
309 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
310 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
311 Removed redundant checks
312 Reduced structure sizes
313 utilize preload in ARMv6 MC/LPF/Copy routines
314 ARM optimized quantization, dfct, variance, subtract
315 Increase chrow row alignment to 16 bytes.
316 disable trellis optimization for first pass
317 Write SSSE3 sub-pixel filter function
318 Improve SSE2 half-pixel filter funtions
319 Add vp8_sub_pixel_variance16x8_ssse3 function
320 Reduce unnecessary distortion computation
321 Use diamond search to replace full search
322 Preload reference area in sub-pixel motion search (real-time mode)
325 This release focused primarily on one-pass use cases, including
326 video conferencing. Low latency data rate control was significantly
327 improved, improving streamability over bandwidth constrained links.
328 Added support for error concealment, allowing frames to maintain
329 visual quality in the presence of substantial packet loss.
331 Add rc_max_intra_bitrate_pct control
332 Limit size of initial keyframe in one-pass.
333 Improve framerate adaptation
334 Improved 1-pass CBR rate control
335 Improved KF insertion after fades to still.
336 Improved key frame detection.
337 Improved activity masking (lower PSNR impact for same SSIM boost)
338 Improved interaction between GF and ARFs
339 Adding error-concealment to the decoder.
340 Adding support for independent partitions
341 Adjusted rate-distortion constants
345 Removed firstpass motion map
346 Fix parallel make install
347 Fix multithreaded encoding for 1 MB wide frame
348 Fixed iwalsh_neon build problems with RVDS4.1
349 Fix semaphore emulation, spin-wait intrinsics on Windows
350 Fix build with xcode4 and simplify GLOBAL.
351 Mark ARM asm objects as allowing a non-executable stack.
352 Fix vpxenc encoding incorrect webm file header on big endian
355 2011-03-07 v0.9.6 "Bali"
356 Our second named release, focused on a faster, higher quality, encoder.
359 This release is backwards compatible with Aylesbury (v0.9.5). Users
360 of older releases should refer to the Upgrading notes in this
361 document for that release.
364 vpxenc --psnr shows a summary when encode completes
365 --tune=ssim option to enable activity masking
366 improved postproc visualizations for development
367 updated support for Apple iOS to SDK 4.2
368 query decoder to determine which reference frames were updated
369 implemented error tracking in the decoder
370 fix pipe support on windows
373 Primary focus was on good quality mode, speed 0. Average improvement
374 on x86 about 40%, up to 100% on user-generated content at that speed.
375 Best quality mode speed improved 35%, and realtime speed 10-20%. This
376 release also saw significant improvement in realtime encoding speed
379 Improved encoder threading
380 Dont pick encoder filter level when loopfilter is disabled.
381 Avoid double copying of key frames into alt and golden buffer
383 x86 sse2 temporal filter
384 SSSE3 version of fast quantizer
385 vp8_rd_pick_best_mbsegmentation code restructure
386 Adjusted breakout RD for SPLITMV
387 Changed segmentation check order
388 Improved rd_pick_intra4x4block
389 Adds armv6 optimized variance calculation
390 ARMv6 optimized sad16x16
391 ARMv6 optimized half pixel variance calculations
392 Full search SAD function optimization in SSE4.1
393 Improve MV prediction accuracy to achieve performance gain
394 Improve MV prediction in vp8_pick_inter_mode() for speed>3
397 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
398 also includes support for "activity masking," which greatly improves
399 SSIM at the expense of PSNR. For now, this feature is available with
400 the --tune=ssim option. Further experimentation in this area
401 is ongoing. This release also introduces a new rate control mode
402 called "CQ," which changes the allocation of bits within a clip to
403 the sections where they will have the most visual impact.
405 Tuning for the more exact quantizer.
406 Relax rate control for last few frames
408 Limit key frame quantizer for forced key frames.
410 Add simple version of activity masking.
411 make rdmult adaptive for intra in quantizer RDO
412 cap the best quantizer for 2nd order DC
413 change the threshold of DC check for encode breakout
416 Fix crash on Sparc Solaris.
417 Fix counter of fixed keyframe distance
418 ARNR filter pointer update bug fix
419 Fixed use of motion percentage in KF/GF group calc
420 Changed condition for using RD in Intra Mode
421 Fix encoder real-time only configuration.
422 Fix ARM encoder crash with multiple token partitions
423 Fixed bug first cluster timecode of webm file is wrong.
424 Fixed various encoder bugs with odd-sized images
425 vp8e_get_preview fixed when spatial resampling enabled
426 quantizer: fix assertion in fast quantizer path
427 Allocate source buffers to be multiples of 16
428 Fix for manual Golden frame frequency
429 Fix drastic undershoot in long form content
432 2010-10-28 v0.9.5 "Aylesbury"
433 Our first named release, focused on a faster decoder, and a better encoder.
436 This release incorporates backwards-incompatible changes to the
437 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
440 * the -q (quiet) option has been removed, and replaced with
441 -v (verbose). the output is quiet by default. Use -v to see
442 the version number of the binary.
444 * The default behavior is now to write output to a single file
445 instead of individual frames. The -y option has been removed.
446 Y4M output is the default.
448 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
449 options must be specified.
451 $ ivfdec -o OUTPUT INPUT
452 $ vpxdec --i420 -o OUTPUT INPUT
454 * If an output file is not specified, the default is to write
455 Y4M to stdout. This makes piping more natural.
457 $ ivfdec -y -o - INPUT | ...
460 * The output file has additional flexibility for formatting the
461 filename. It supports escape characters for constructing a
462 filename from the width, height, and sequence number. This
463 replaces the -p option. To get the equivalent:
465 $ ivfdec -p frame INPUT
466 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
469 * The output file must be specified with -o, rather than as the
472 $ ivfenc <options> INPUT OUTPUT
473 $ vpxenc <options> -o OUTPUT INPUT
475 * The output defaults to webm. To get IVF output, use the --ivf
478 $ ivfenc <options> INPUT OUTPUT.ivf
479 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
483 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
484 vpxdec supports .webm input
485 vpxdec writes .y4m by default
486 vpxenc writes .webm output by default
487 vpxenc --psnr now shows the average/overall PSNR at the end
488 ARM platforms now support runtime cpu detection
489 vpxdec visualizations added for motion vectors, block modes, references
490 vpxdec now silent by default
491 vpxdec --progress shows frame-by-frame timing information
492 vpxenc supports the distinction between --fps and --timebase
493 NASM is now a supported assembler
494 configure: enable PIC for shared libs by default
495 configure: add --enable-small
496 configure: support for ppc32-linux-gcc
497 configure: support for sparc-solaris-gcc
500 Improve handling of invalid frames
501 Fix valgrind errors in the NEON loop filters.
502 Fix loopfilter delta zero transitions
503 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
504 Build fixes for darwin-icc
507 20-40% (average 28%) improvement in libvpx decoder speed,
509 Rewrite vp8_short_walsh4x4_sse2()
510 Optimizations on the loopfilters.
511 Miscellaneous improvements for Atom
512 Add 4-tap version of 2nd-pass ARMv6 MC filter.
513 Improved multithread utilization
514 Better instruction choices on x86
515 reorder data to use wider instructions
516 Update NEON wide idcts
517 Make block access to frame buffer sequential
518 Improved subset block search
519 Bilinear subpixel optimizations for ssse3.
520 Decrease memory footprint
522 Encoder speed improvements (percentage gain not measured):
523 Skip unnecessary search of identical frames
524 Add SSE2 subtract functions
525 Improve bounds checking in vp8_diamond_search_sadx4()
526 Added vp8_fast_quantize_b_sse2
529 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
530 encoding mode, and up to 60% improvement on very noisy, still
531 or slow moving source video
533 Motion compensated temporal filter for Alt-Ref Noise Reduction
534 Improved use of trellis quantization on 2nd order Y blocks
535 Tune effect of motion on KF/GF boost in two pass
536 Allow coefficient optimization for good quality speed 0.
537 Improved control of active min quantizer for two pass.
538 Enable ARFs for non-lagged compress
542 Disable frame dropping by default
543 Improved multithreaded performance
544 Improved Force Key Frame Behaviour
545 Increased rate control buffer level precision
546 Fix bug in 1st pass motion compensation
547 ivfenc: correct fixed kf interval, --disable-kf
549 Changed above and left context data layout
550 Rework idct calling structure.
551 Removed unnecessary MB_MODE_INFO copies
552 x86: SSSE3 sixtap prediction
553 Reworked IDCT to include reconstruction (add) step
554 Swap alt/gold/new/last frame buffer ptrs instead of copying.
555 Improve SSE2 loopfilter functions
556 Change bitreader to use a larger window.
557 Avoid loopfilter reinitialization when possible
559 Normalize quantizer's zero bin and rounding factors
560 Add trellis quantization.
561 Make the quantizer exact.
562 Updates to ARNR filtering algorithm
563 Fix breakout thresh computation for golden & AltRef frames
564 Redo the forward 4x4 dct
565 Improve the accuracy of forward walsh-hadamard transform
566 Further adjustment of RD behaviour with Q and Zbin.
568 Allow linking of libs built with MinGW to MSVC
569 Fix target auto-detection on mingw32
570 Allow --cpu= to work for x86.
571 configure: pass original arguments through to make dist
572 Fix builds without runtime CPU detection
573 msvs: fix install of codec sources
574 msvs: Change devenv.com command line for better msys support
575 msvs: Add vs9 targets.
576 Add x86_64-linux-icc target
578 Potential crashes on older MinGW builds
579 Fix two-pass framrate for Y4M input.
580 Fixed simple loop filter, other crashes on ARM v6
581 arm: fix missing dependency with --enable-shared
582 configure: support directories containing .o
583 Replace pinsrw (SSE) with MMX instructions
584 apple: include proper mach primatives
585 Fixed rate control bug with long key frame interval.
586 Fix DSO link errors on x86-64 when not using a version script
587 Fixed buffer selection for UV in AltRef filtering
592 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
593 * Speed optimizations
596 * Prevent out-of-bounds accesses on invalid data
597 - Build system updates:
598 * Detect toolchain to be used automatically for native builds
599 * Support building shared libraries
600 * Better autotools emulation (--prefix, --libdir, DESTDIR)
602 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
606 - Initial open source release. Welcome to WebM and VP8!