1 2012-01-27 v1.0.0 "Duclair"
2 Our fourth named release, focused on performance and features related to
3 real-time encoding. It also fixes a decoder crash bug introduced in
4 v0.9.7, so all users of that release are encouraged to upgrade.
7 This release is ABI incompatible with prior releases of libvpx, so the
8 "major" version number has been bumped to 1. You must recompile your
9 applications against the latest version of the libvpx headers. The
10 API remains compatible, and this should not require code changes in most
14 This release introduces several substantial new features to the encoder,
15 of particular interest to real time streaming applications.
17 Temporal scalability allows the encoder to produce a stream that can
18 be decimated to different frame rates, with independent rate targetting
21 Multiframe quality enhancement postprocessing can make visual quality
22 more consistent in the presence of frames that are substantially
23 different quality than the surrounding frames, as in the temporal
24 scalability case and in some forced keyframe scenarios.
26 Multiple-resolution encoding support allows the encoding of the
27 same content at different resolutions faster than encoding them
31 Optimization targets for this release included the decoder and the real-
32 time modes of the encoder. Decoder speed on x86 has improved 10.5% with
33 this release. Encoder improvements followed a curve where speeds 1-3
34 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
35 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
39 Encoder quality in the single stream case is consistent with the Cayuga
43 This release fixes an OOB read decoder crash bug present in v0.9.7
44 related to the clamping of motion vectors in SPLITMV blocks. This
45 behavior could be triggered by corrupt input or by starting
46 decoding from a P-frame.
49 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
50 This is an incremental bugfix release against Cayuga. All users of that
51 release are strongly encouraged to upgrade.
53 - Fix potential OOB reads (cdae03a)
55 An unbounded out of bounds read was discovered when the
56 decoder was requested to perform error concealment (new in
57 Cayuga) given a frame with corrupt partition sizes.
59 A bounded out of bounds read was discovered affecting all
60 versions of libvpx. Given an multipartition input frame that
61 is truncated between the mode/mv partition and the first
62 residiual paritition (in the block of partition offsets), up
63 to 3 extra bytes could have been read from the source buffer.
64 The code will not take any action regardless of the contents
65 of these undefined bytes, as the truncated buffer is detected
66 immediately following the read based on the calculated
67 starting position of the coefficient partition.
69 - Fix potential error concealment crash when the very first frame
70 is missing or corrupt (a609be5)
72 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
74 - Revert 1-pass CBR rate control changes (e961317)
75 Further testing showed this change produced undesirable visual
76 artifacts, rolling back for now.
79 2011-08-02 v0.9.7 "Cayuga"
80 Our third named release, focused on a faster, higher quality, encoder.
83 This release is backwards compatible with Aylesbury (v0.9.5) and
84 Bali (v0.9.6). Users of older releases should refer to the Upgrading
85 notes in this document for that release.
88 Stereo 3D format support for vpxenc
89 Runtime detection of available processor cores.
90 Allow specifying --end-usage by enum name
91 vpxdec: test for frame corruption
92 vpxenc: add quantizer histogram display
93 vpxenc: add rate histogram display
94 Set VPX_FRAME_IS_DROPPABLE
95 update configure for ios sdk 4.3
96 Avoid text relocations in ARM vp8 decoder
97 Generate a vpx.pc file for pkg-config.
98 New ways of passing encoded data between encoder and decoder.
101 This release includes across-the-board speed improvements to the
102 encoder. On x86, these measure at approximately 11.5% in Best mode,
103 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
104 On ARM Cortex A9 with Neon extensions, real-time encoding of video
105 telephony content is 35% faster than Bali on single core and 48%
106 faster on multi-core. On the NVidia Tegra2 platform, real time
107 encoding is 40% faster than Bali.
109 Decoder speed was not a priority for this release, but improved
110 approximately 8.4% on x86.
112 Reduce motion vector search on alt-ref frame.
113 Encoder loopfilter running in its own thread
114 Reworked loopfilter to precalculate more parameters
115 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
116 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
117 Removed redundant checks
118 Reduced structure sizes
119 utilize preload in ARMv6 MC/LPF/Copy routines
120 ARM optimized quantization, dfct, variance, subtract
121 Increase chrow row alignment to 16 bytes.
122 disable trellis optimization for first pass
123 Write SSSE3 sub-pixel filter function
124 Improve SSE2 half-pixel filter funtions
125 Add vp8_sub_pixel_variance16x8_ssse3 function
126 Reduce unnecessary distortion computation
127 Use diamond search to replace full search
128 Preload reference area in sub-pixel motion search (real-time mode)
131 This release focused primarily on one-pass use cases, including
132 video conferencing. Low latency data rate control was significantly
133 improved, improving streamability over bandwidth constrained links.
134 Added support for error concealment, allowing frames to maintain
135 visual quality in the presence of substantial packet loss.
137 Add rc_max_intra_bitrate_pct control
138 Limit size of initial keyframe in one-pass.
139 Improve framerate adaptation
140 Improved 1-pass CBR rate control
141 Improved KF insertion after fades to still.
142 Improved key frame detection.
143 Improved activity masking (lower PSNR impact for same SSIM boost)
144 Improved interaction between GF and ARFs
145 Adding error-concealment to the decoder.
146 Adding support for independent partitions
147 Adjusted rate-distortion constants
151 Removed firstpass motion map
152 Fix parallel make install
153 Fix multithreaded encoding for 1 MB wide frame
154 Fixed iwalsh_neon build problems with RVDS4.1
155 Fix semaphore emulation, spin-wait intrinsics on Windows
156 Fix build with xcode4 and simplify GLOBAL.
157 Mark ARM asm objects as allowing a non-executable stack.
158 Fix vpxenc encoding incorrect webm file header on big endian
161 2011-03-07 v0.9.6 "Bali"
162 Our second named release, focused on a faster, higher quality, encoder.
165 This release is backwards compatible with Aylesbury (v0.9.5). Users
166 of older releases should refer to the Upgrading notes in this
167 document for that release.
170 vpxenc --psnr shows a summary when encode completes
171 --tune=ssim option to enable activity masking
172 improved postproc visualizations for development
173 updated support for Apple iOS to SDK 4.2
174 query decoder to determine which reference frames were updated
175 implemented error tracking in the decoder
176 fix pipe support on windows
179 Primary focus was on good quality mode, speed 0. Average improvement
180 on x86 about 40%, up to 100% on user-generated content at that speed.
181 Best quality mode speed improved 35%, and realtime speed 10-20%. This
182 release also saw significant improvement in realtime encoding speed
185 Improved encoder threading
186 Dont pick encoder filter level when loopfilter is disabled.
187 Avoid double copying of key frames into alt and golden buffer
189 x86 sse2 temporal filter
190 SSSE3 version of fast quantizer
191 vp8_rd_pick_best_mbsegmentation code restructure
192 Adjusted breakout RD for SPLITMV
193 Changed segmentation check order
194 Improved rd_pick_intra4x4block
195 Adds armv6 optimized variance calculation
196 ARMv6 optimized sad16x16
197 ARMv6 optimized half pixel variance calculations
198 Full search SAD function optimization in SSE4.1
199 Improve MV prediction accuracy to achieve performance gain
200 Improve MV prediction in vp8_pick_inter_mode() for speed>3
203 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
204 also includes support for "activity masking," which greatly improves
205 SSIM at the expense of PSNR. For now, this feature is available with
206 the --tune=ssim option. Further experimentation in this area
207 is ongoing. This release also introduces a new rate control mode
208 called "CQ," which changes the allocation of bits within a clip to
209 the sections where they will have the most visual impact.
211 Tuning for the more exact quantizer.
212 Relax rate control for last few frames
214 Limit key frame quantizer for forced key frames.
216 Add simple version of activity masking.
217 make rdmult adaptive for intra in quantizer RDO
218 cap the best quantizer for 2nd order DC
219 change the threshold of DC check for encode breakout
222 Fix crash on Sparc Solaris.
223 Fix counter of fixed keyframe distance
224 ARNR filter pointer update bug fix
225 Fixed use of motion percentage in KF/GF group calc
226 Changed condition for using RD in Intra Mode
227 Fix encoder real-time only configuration.
228 Fix ARM encoder crash with multiple token partitions
229 Fixed bug first cluster timecode of webm file is wrong.
230 Fixed various encoder bugs with odd-sized images
231 vp8e_get_preview fixed when spatial resampling enabled
232 quantizer: fix assertion in fast quantizer path
233 Allocate source buffers to be multiples of 16
234 Fix for manual Golden frame frequency
235 Fix drastic undershoot in long form content
238 2010-10-28 v0.9.5 "Aylesbury"
239 Our first named release, focused on a faster decoder, and a better encoder.
242 This release incorporates backwards-incompatible changes to the
243 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
246 * the -q (quiet) option has been removed, and replaced with
247 -v (verbose). the output is quiet by default. Use -v to see
248 the version number of the binary.
250 * The default behavior is now to write output to a single file
251 instead of individual frames. The -y option has been removed.
252 Y4M output is the default.
254 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
255 options must be specified.
257 $ ivfdec -o OUTPUT INPUT
258 $ vpxdec --i420 -o OUTPUT INPUT
260 * If an output file is not specified, the default is to write
261 Y4M to stdout. This makes piping more natural.
263 $ ivfdec -y -o - INPUT | ...
266 * The output file has additional flexibility for formatting the
267 filename. It supports escape characters for constructing a
268 filename from the width, height, and sequence number. This
269 replaces the -p option. To get the equivalent:
271 $ ivfdec -p frame INPUT
272 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
275 * The output file must be specified with -o, rather than as the
278 $ ivfenc <options> INPUT OUTPUT
279 $ vpxenc <options> -o OUTPUT INPUT
281 * The output defaults to webm. To get IVF output, use the --ivf
284 $ ivfenc <options> INPUT OUTPUT.ivf
285 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
289 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
290 vpxdec supports .webm input
291 vpxdec writes .y4m by default
292 vpxenc writes .webm output by default
293 vpxenc --psnr now shows the average/overall PSNR at the end
294 ARM platforms now support runtime cpu detection
295 vpxdec visualizations added for motion vectors, block modes, references
296 vpxdec now silent by default
297 vpxdec --progress shows frame-by-frame timing information
298 vpxenc supports the distinction between --fps and --timebase
299 NASM is now a supported assembler
300 configure: enable PIC for shared libs by default
301 configure: add --enable-small
302 configure: support for ppc32-linux-gcc
303 configure: support for sparc-solaris-gcc
306 Improve handling of invalid frames
307 Fix valgrind errors in the NEON loop filters.
308 Fix loopfilter delta zero transitions
309 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
310 Build fixes for darwin-icc
313 20-40% (average 28%) improvement in libvpx decoder speed,
315 Rewrite vp8_short_walsh4x4_sse2()
316 Optimizations on the loopfilters.
317 Miscellaneous improvements for Atom
318 Add 4-tap version of 2nd-pass ARMv6 MC filter.
319 Improved multithread utilization
320 Better instruction choices on x86
321 reorder data to use wider instructions
322 Update NEON wide idcts
323 Make block access to frame buffer sequential
324 Improved subset block search
325 Bilinear subpixel optimizations for ssse3.
326 Decrease memory footprint
328 Encoder speed improvements (percentage gain not measured):
329 Skip unnecessary search of identical frames
330 Add SSE2 subtract functions
331 Improve bounds checking in vp8_diamond_search_sadx4()
332 Added vp8_fast_quantize_b_sse2
335 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
336 encoding mode, and up to 60% improvement on very noisy, still
337 or slow moving source video
339 Motion compensated temporal filter for Alt-Ref Noise Reduction
340 Improved use of trellis quantization on 2nd order Y blocks
341 Tune effect of motion on KF/GF boost in two pass
342 Allow coefficient optimization for good quality speed 0.
343 Improved control of active min quantizer for two pass.
344 Enable ARFs for non-lagged compress
348 Disable frame dropping by default
349 Improved multithreaded performance
350 Improved Force Key Frame Behaviour
351 Increased rate control buffer level precision
352 Fix bug in 1st pass motion compensation
353 ivfenc: correct fixed kf interval, --disable-kf
355 Changed above and left context data layout
356 Rework idct calling structure.
357 Removed unnecessary MB_MODE_INFO copies
358 x86: SSSE3 sixtap prediction
359 Reworked IDCT to include reconstruction (add) step
360 Swap alt/gold/new/last frame buffer ptrs instead of copying.
361 Improve SSE2 loopfilter functions
362 Change bitreader to use a larger window.
363 Avoid loopfilter reinitialization when possible
365 Normalize quantizer's zero bin and rounding factors
366 Add trellis quantization.
367 Make the quantizer exact.
368 Updates to ARNR filtering algorithm
369 Fix breakout thresh computation for golden & AltRef frames
370 Redo the forward 4x4 dct
371 Improve the accuracy of forward walsh-hadamard transform
372 Further adjustment of RD behaviour with Q and Zbin.
374 Allow linking of libs built with MinGW to MSVC
375 Fix target auto-detection on mingw32
376 Allow --cpu= to work for x86.
377 configure: pass original arguments through to make dist
378 Fix builds without runtime CPU detection
379 msvs: fix install of codec sources
380 msvs: Change devenv.com command line for better msys support
381 msvs: Add vs9 targets.
382 Add x86_64-linux-icc target
384 Potential crashes on older MinGW builds
385 Fix two-pass framrate for Y4M input.
386 Fixed simple loop filter, other crashes on ARM v6
387 arm: fix missing dependency with --enable-shared
388 configure: support directories containing .o
389 Replace pinsrw (SSE) with MMX instructions
390 apple: include proper mach primatives
391 Fixed rate control bug with long key frame interval.
392 Fix DSO link errors on x86-64 when not using a version script
393 Fixed buffer selection for UV in AltRef filtering
398 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
399 * Speed optimizations
402 * Prevent out-of-bounds accesses on invalid data
403 - Build system updates:
404 * Detect toolchain to be used automatically for native builds
405 * Support building shared libraries
406 * Better autotools emulation (--prefix, --libdir, DESTDIR)
408 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
412 - Initial open source release. Welcome to WebM and VP8!