1 NOTE FOR NEXT RELEASE: PLEASE DOCUMENT API CHANGE TO NOISE SENSITIVITY FEATURE
2 AND WARN PEOPLE THAT FOR NOW IT'S EITHER ON OR OFF
4 2012-01-27 v1.0.0 "Duclair"
5 Our fourth named release, focused on performance and features related to
6 real-time encoding. It also fixes a decoder crash bug introduced in
7 v0.9.7, so all users of that release are encouraged to upgrade.
10 This release is ABI incompatible with prior releases of libvpx, so the
11 "major" version number has been bumped to 1. You must recompile your
12 applications against the latest version of the libvpx headers. The
13 API remains compatible, and this should not require code changes in most
17 This release introduces several substantial new features to the encoder,
18 of particular interest to real time streaming applications.
20 Temporal scalability allows the encoder to produce a stream that can
21 be decimated to different frame rates, with independent rate targetting
24 Multiframe quality enhancement postprocessing can make visual quality
25 more consistent in the presence of frames that are substantially
26 different quality than the surrounding frames, as in the temporal
27 scalability case and in some forced keyframe scenarios.
29 Multiple-resolution encoding support allows the encoding of the
30 same content at different resolutions faster than encoding them
34 Optimization targets for this release included the decoder and the real-
35 time modes of the encoder. Decoder speed on x86 has improved 10.5% with
36 this release. Encoder improvements followed a curve where speeds 1-3
37 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
38 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
42 Encoder quality in the single stream case is consistent with the Cayuga
46 This release fixes an OOB read decoder crash bug present in v0.9.7
47 related to the clamping of motion vectors in SPLITMV blocks. This
48 behavior could be triggered by corrupt input or by starting
49 decoding from a P-frame.
52 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
53 This is an incremental bugfix release against Cayuga. All users of that
54 release are strongly encouraged to upgrade.
56 - Fix potential OOB reads (cdae03a)
58 An unbounded out of bounds read was discovered when the
59 decoder was requested to perform error concealment (new in
60 Cayuga) given a frame with corrupt partition sizes.
62 A bounded out of bounds read was discovered affecting all
63 versions of libvpx. Given an multipartition input frame that
64 is truncated between the mode/mv partition and the first
65 residiual paritition (in the block of partition offsets), up
66 to 3 extra bytes could have been read from the source buffer.
67 The code will not take any action regardless of the contents
68 of these undefined bytes, as the truncated buffer is detected
69 immediately following the read based on the calculated
70 starting position of the coefficient partition.
72 - Fix potential error concealment crash when the very first frame
73 is missing or corrupt (a609be5)
75 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
77 - Revert 1-pass CBR rate control changes (e961317)
78 Further testing showed this change produced undesirable visual
79 artifacts, rolling back for now.
82 2011-08-02 v0.9.7 "Cayuga"
83 Our third named release, focused on a faster, higher quality, encoder.
86 This release is backwards compatible with Aylesbury (v0.9.5) and
87 Bali (v0.9.6). Users of older releases should refer to the Upgrading
88 notes in this document for that release.
91 Stereo 3D format support for vpxenc
92 Runtime detection of available processor cores.
93 Allow specifying --end-usage by enum name
94 vpxdec: test for frame corruption
95 vpxenc: add quantizer histogram display
96 vpxenc: add rate histogram display
97 Set VPX_FRAME_IS_DROPPABLE
98 update configure for ios sdk 4.3
99 Avoid text relocations in ARM vp8 decoder
100 Generate a vpx.pc file for pkg-config.
101 New ways of passing encoded data between encoder and decoder.
104 This release includes across-the-board speed improvements to the
105 encoder. On x86, these measure at approximately 11.5% in Best mode,
106 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
107 On ARM Cortex A9 with Neon extensions, real-time encoding of video
108 telephony content is 35% faster than Bali on single core and 48%
109 faster on multi-core. On the NVidia Tegra2 platform, real time
110 encoding is 40% faster than Bali.
112 Decoder speed was not a priority for this release, but improved
113 approximately 8.4% on x86.
115 Reduce motion vector search on alt-ref frame.
116 Encoder loopfilter running in its own thread
117 Reworked loopfilter to precalculate more parameters
118 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
119 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
120 Removed redundant checks
121 Reduced structure sizes
122 utilize preload in ARMv6 MC/LPF/Copy routines
123 ARM optimized quantization, dfct, variance, subtract
124 Increase chrow row alignment to 16 bytes.
125 disable trellis optimization for first pass
126 Write SSSE3 sub-pixel filter function
127 Improve SSE2 half-pixel filter funtions
128 Add vp8_sub_pixel_variance16x8_ssse3 function
129 Reduce unnecessary distortion computation
130 Use diamond search to replace full search
131 Preload reference area in sub-pixel motion search (real-time mode)
134 This release focused primarily on one-pass use cases, including
135 video conferencing. Low latency data rate control was significantly
136 improved, improving streamability over bandwidth constrained links.
137 Added support for error concealment, allowing frames to maintain
138 visual quality in the presence of substantial packet loss.
140 Add rc_max_intra_bitrate_pct control
141 Limit size of initial keyframe in one-pass.
142 Improve framerate adaptation
143 Improved 1-pass CBR rate control
144 Improved KF insertion after fades to still.
145 Improved key frame detection.
146 Improved activity masking (lower PSNR impact for same SSIM boost)
147 Improved interaction between GF and ARFs
148 Adding error-concealment to the decoder.
149 Adding support for independent partitions
150 Adjusted rate-distortion constants
154 Removed firstpass motion map
155 Fix parallel make install
156 Fix multithreaded encoding for 1 MB wide frame
157 Fixed iwalsh_neon build problems with RVDS4.1
158 Fix semaphore emulation, spin-wait intrinsics on Windows
159 Fix build with xcode4 and simplify GLOBAL.
160 Mark ARM asm objects as allowing a non-executable stack.
161 Fix vpxenc encoding incorrect webm file header on big endian
164 2011-03-07 v0.9.6 "Bali"
165 Our second named release, focused on a faster, higher quality, encoder.
168 This release is backwards compatible with Aylesbury (v0.9.5). Users
169 of older releases should refer to the Upgrading notes in this
170 document for that release.
173 vpxenc --psnr shows a summary when encode completes
174 --tune=ssim option to enable activity masking
175 improved postproc visualizations for development
176 updated support for Apple iOS to SDK 4.2
177 query decoder to determine which reference frames were updated
178 implemented error tracking in the decoder
179 fix pipe support on windows
182 Primary focus was on good quality mode, speed 0. Average improvement
183 on x86 about 40%, up to 100% on user-generated content at that speed.
184 Best quality mode speed improved 35%, and realtime speed 10-20%. This
185 release also saw significant improvement in realtime encoding speed
188 Improved encoder threading
189 Dont pick encoder filter level when loopfilter is disabled.
190 Avoid double copying of key frames into alt and golden buffer
192 x86 sse2 temporal filter
193 SSSE3 version of fast quantizer
194 vp8_rd_pick_best_mbsegmentation code restructure
195 Adjusted breakout RD for SPLITMV
196 Changed segmentation check order
197 Improved rd_pick_intra4x4block
198 Adds armv6 optimized variance calculation
199 ARMv6 optimized sad16x16
200 ARMv6 optimized half pixel variance calculations
201 Full search SAD function optimization in SSE4.1
202 Improve MV prediction accuracy to achieve performance gain
203 Improve MV prediction in vp8_pick_inter_mode() for speed>3
206 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
207 also includes support for "activity masking," which greatly improves
208 SSIM at the expense of PSNR. For now, this feature is available with
209 the --tune=ssim option. Further experimentation in this area
210 is ongoing. This release also introduces a new rate control mode
211 called "CQ," which changes the allocation of bits within a clip to
212 the sections where they will have the most visual impact.
214 Tuning for the more exact quantizer.
215 Relax rate control for last few frames
217 Limit key frame quantizer for forced key frames.
219 Add simple version of activity masking.
220 make rdmult adaptive for intra in quantizer RDO
221 cap the best quantizer for 2nd order DC
222 change the threshold of DC check for encode breakout
225 Fix crash on Sparc Solaris.
226 Fix counter of fixed keyframe distance
227 ARNR filter pointer update bug fix
228 Fixed use of motion percentage in KF/GF group calc
229 Changed condition for using RD in Intra Mode
230 Fix encoder real-time only configuration.
231 Fix ARM encoder crash with multiple token partitions
232 Fixed bug first cluster timecode of webm file is wrong.
233 Fixed various encoder bugs with odd-sized images
234 vp8e_get_preview fixed when spatial resampling enabled
235 quantizer: fix assertion in fast quantizer path
236 Allocate source buffers to be multiples of 16
237 Fix for manual Golden frame frequency
238 Fix drastic undershoot in long form content
241 2010-10-28 v0.9.5 "Aylesbury"
242 Our first named release, focused on a faster decoder, and a better encoder.
245 This release incorporates backwards-incompatible changes to the
246 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
249 * the -q (quiet) option has been removed, and replaced with
250 -v (verbose). the output is quiet by default. Use -v to see
251 the version number of the binary.
253 * The default behavior is now to write output to a single file
254 instead of individual frames. The -y option has been removed.
255 Y4M output is the default.
257 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
258 options must be specified.
260 $ ivfdec -o OUTPUT INPUT
261 $ vpxdec --i420 -o OUTPUT INPUT
263 * If an output file is not specified, the default is to write
264 Y4M to stdout. This makes piping more natural.
266 $ ivfdec -y -o - INPUT | ...
269 * The output file has additional flexibility for formatting the
270 filename. It supports escape characters for constructing a
271 filename from the width, height, and sequence number. This
272 replaces the -p option. To get the equivalent:
274 $ ivfdec -p frame INPUT
275 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
278 * The output file must be specified with -o, rather than as the
281 $ ivfenc <options> INPUT OUTPUT
282 $ vpxenc <options> -o OUTPUT INPUT
284 * The output defaults to webm. To get IVF output, use the --ivf
287 $ ivfenc <options> INPUT OUTPUT.ivf
288 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
292 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
293 vpxdec supports .webm input
294 vpxdec writes .y4m by default
295 vpxenc writes .webm output by default
296 vpxenc --psnr now shows the average/overall PSNR at the end
297 ARM platforms now support runtime cpu detection
298 vpxdec visualizations added for motion vectors, block modes, references
299 vpxdec now silent by default
300 vpxdec --progress shows frame-by-frame timing information
301 vpxenc supports the distinction between --fps and --timebase
302 NASM is now a supported assembler
303 configure: enable PIC for shared libs by default
304 configure: add --enable-small
305 configure: support for ppc32-linux-gcc
306 configure: support for sparc-solaris-gcc
309 Improve handling of invalid frames
310 Fix valgrind errors in the NEON loop filters.
311 Fix loopfilter delta zero transitions
312 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
313 Build fixes for darwin-icc
316 20-40% (average 28%) improvement in libvpx decoder speed,
318 Rewrite vp8_short_walsh4x4_sse2()
319 Optimizations on the loopfilters.
320 Miscellaneous improvements for Atom
321 Add 4-tap version of 2nd-pass ARMv6 MC filter.
322 Improved multithread utilization
323 Better instruction choices on x86
324 reorder data to use wider instructions
325 Update NEON wide idcts
326 Make block access to frame buffer sequential
327 Improved subset block search
328 Bilinear subpixel optimizations for ssse3.
329 Decrease memory footprint
331 Encoder speed improvements (percentage gain not measured):
332 Skip unnecessary search of identical frames
333 Add SSE2 subtract functions
334 Improve bounds checking in vp8_diamond_search_sadx4()
335 Added vp8_fast_quantize_b_sse2
338 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
339 encoding mode, and up to 60% improvement on very noisy, still
340 or slow moving source video
342 Motion compensated temporal filter for Alt-Ref Noise Reduction
343 Improved use of trellis quantization on 2nd order Y blocks
344 Tune effect of motion on KF/GF boost in two pass
345 Allow coefficient optimization for good quality speed 0.
346 Improved control of active min quantizer for two pass.
347 Enable ARFs for non-lagged compress
351 Disable frame dropping by default
352 Improved multithreaded performance
353 Improved Force Key Frame Behaviour
354 Increased rate control buffer level precision
355 Fix bug in 1st pass motion compensation
356 ivfenc: correct fixed kf interval, --disable-kf
358 Changed above and left context data layout
359 Rework idct calling structure.
360 Removed unnecessary MB_MODE_INFO copies
361 x86: SSSE3 sixtap prediction
362 Reworked IDCT to include reconstruction (add) step
363 Swap alt/gold/new/last frame buffer ptrs instead of copying.
364 Improve SSE2 loopfilter functions
365 Change bitreader to use a larger window.
366 Avoid loopfilter reinitialization when possible
368 Normalize quantizer's zero bin and rounding factors
369 Add trellis quantization.
370 Make the quantizer exact.
371 Updates to ARNR filtering algorithm
372 Fix breakout thresh computation for golden & AltRef frames
373 Redo the forward 4x4 dct
374 Improve the accuracy of forward walsh-hadamard transform
375 Further adjustment of RD behaviour with Q and Zbin.
377 Allow linking of libs built with MinGW to MSVC
378 Fix target auto-detection on mingw32
379 Allow --cpu= to work for x86.
380 configure: pass original arguments through to make dist
381 Fix builds without runtime CPU detection
382 msvs: fix install of codec sources
383 msvs: Change devenv.com command line for better msys support
384 msvs: Add vs9 targets.
385 Add x86_64-linux-icc target
387 Potential crashes on older MinGW builds
388 Fix two-pass framrate for Y4M input.
389 Fixed simple loop filter, other crashes on ARM v6
390 arm: fix missing dependency with --enable-shared
391 configure: support directories containing .o
392 Replace pinsrw (SSE) with MMX instructions
393 apple: include proper mach primatives
394 Fixed rate control bug with long key frame interval.
395 Fix DSO link errors on x86-64 when not using a version script
396 Fixed buffer selection for UV in AltRef filtering
401 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
402 * Speed optimizations
405 * Prevent out-of-bounds accesses on invalid data
406 - Build system updates:
407 * Detect toolchain to be used automatically for native builds
408 * Support building shared libraries
409 * Better autotools emulation (--prefix, --libdir, DESTDIR)
411 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
415 - Initial open source release. Welcome to WebM and VP8!