| blob | dcb9f738a680fc7ef83aaa9b1305cdf33989c071 |
1 2012-05-09 v1.1.0 "Eider"
2 This introduces a number of enhancements, mostly focused on real-time
3 encoding. In addition, it fixes a decoder bug (first introduced in
4 Duclair) so all users of that release are encouraged to upgrade.
6 - Upgrading:
7 This release is ABI and API compatible with Duclair (v1.0.0). Users
8 of older releases should refer to the Upgrading notes in this
9 document for that release.
11 This release introduces a new temporal denoiser, controlled by the
12 VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not
13 currently take a strength parameter, so the control is effectively
14 a boolean - zero (off) or non-zero (on). For compatibility with
15 existing applications, the values accepted are the same as those
16 for the spatial denoiser (0-6). The temporal denoiser is enabled
17 by default, and the older spatial denoiser may be restored by
18 configuring with --disable-temporal-denoising. The temporal denoiser
19 is more computationally intensive than the spatial one.
21 This release removes support for a legacy, decode only API that was
22 supported, but deprecated, at the initial release of libvpx
23 (v0.9.0). This is not expected to have any impact. If you are
24 impacted, you can apply a reversion to commit 2bf8fb58 locally.
25 Please update to the latest libvpx API if you are affected.
27 - Enhancements:
28 Adds a motion compensated temporal denoiser to the encoder, which
29 gives higher quality than the older spatial denoiser. (See above
30 for notes on upgrading).
32 In addition, support for new compilers and platforms were added,
33 including:
34 improved support for XCode
35 Android x86 NDK build
36 OS/2 support
37 SunCC support
39 Changing resolution with vpx_codec_enc_config_set() is now
40 supported. Previously, reinitializing the codec was required to
41 change the input resolution.
43 The vpxenc application has initial support for producing multiple
44 encodes from the same input in one call. Resizing is not yet
45 supported, but varying other codec parameters is. Use -- to
46 delineate output streams. Options persist from one stream to the
47 next.
49 Also, the vpxenc application will now use a keyframe interval of
50 5 seconds by default. Use the --kf-max-dist option to override.
52 - Speed:
53 Decoder performance improved 2.5% versus Duclair. Encoder speed is
54 consistent with Duclair for most material. Two pass encoding of
55 slideshow-like material will see significant improvements.
57 Large realtime encoding speed gains at a small quality expense are
58 possible by configuring the on-the-fly bitpacking experiment with
59 --enable-onthefly-bitpacking. Realtime encoder can be up to 13%
60 faster (ARM) depending on the number of threads and bitrate
61 settings. This technique sees constant gain over the 5-16 speed
62 range. For VC style input the loss seen is up to 0.2dB. See commit
63 52cf4dca for further details.
65 - Quality:
66 On the whole, quality is consistent with the Duclair release. Some
67 tweaks:
69 Reduced blockiness in easy sections by applying a penalty to
70 intra modes.
72 Improved quality of static sections (like slideshows) with
73 two pass encoding.
75 Improved keyframe sizing with multiple temporal layers
77 - Bug Fixes:
78 Corrected alt-ref contribution to frame rate for visible updates
79 to the alt-ref buffer. This affected applications making manual
80 usage of the frame reference flags, or temporal layers.
82 Additional constraints were added to disable multi-frame quality
83 enhancement (MFQE) in sections of the frame where there is motion.
84 (#392)
86 Fixed corruption issues when vpx_codec_enc_config_set() was called
87 with spatial resampling enabled.
89 Fixed a decoder error introduced in Duclair where the segmentation
90 map was not being reinitialized on keyframes (#378)
93 2012-01-27 v1.0.0 "Duclair"
94 Our fourth named release, focused on performance and features related to
95 real-time encoding. It also fixes a decoder crash bug introduced in
96 v0.9.7, so all users of that release are encouraged to upgrade.
98 - Upgrading:
99 This release is ABI incompatible with prior releases of libvpx, so the
100 "major" version number has been bumped to 1. You must recompile your
101 applications against the latest version of the libvpx headers. The
102 API remains compatible, and this should not require code changes in most
103 applications.
105 - Enhancements:
106 This release introduces several substantial new features to the encoder,
107 of particular interest to real time streaming applications.
109 Temporal scalability allows the encoder to produce a stream that can
110 be decimated to different frame rates, with independent rate targetting
111 for each substream.
113 Multiframe quality enhancement postprocessing can make visual quality
114 more consistent in the presence of frames that are substantially
115 different quality than the surrounding frames, as in the temporal
116 scalability case and in some forced keyframe scenarios.
118 Multiple-resolution encoding support allows the encoding of the
119 same content at different resolutions faster than encoding them
120 separately.
122 - Speed:
123 Optimization targets for this release included the decoder and the real-
124 time modes of the encoder. Decoder speed on x86 has improved 10.5% with
125 this release. Encoder improvements followed a curve where speeds 1-3
126 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved
127 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the
128 Cayuga release.
130 - Quality:
131 Encoder quality in the single stream case is consistent with the Cayuga
132 release.
134 - Bug Fixes:
135 This release fixes an OOB read decoder crash bug present in v0.9.7
136 related to the clamping of motion vectors in SPLITMV blocks. This
137 behavior could be triggered by corrupt input or by starting
138 decoding from a P-frame.
141 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
142 This is an incremental bugfix release against Cayuga. All users of that
143 release are strongly encouraged to upgrade.
145 - Fix potential OOB reads (cdae03a)
147 An unbounded out of bounds read was discovered when the
148 decoder was requested to perform error concealment (new in
149 Cayuga) given a frame with corrupt partition sizes.
151 A bounded out of bounds read was discovered affecting all
152 versions of libvpx. Given an multipartition input frame that
153 is truncated between the mode/mv partition and the first
154 residiual paritition (in the block of partition offsets), up
155 to 3 extra bytes could have been read from the source buffer.
156 The code will not take any action regardless of the contents
157 of these undefined bytes, as the truncated buffer is detected
158 immediately following the read based on the calculated
159 starting position of the coefficient partition.
161 - Fix potential error concealment crash when the very first frame
162 is missing or corrupt (a609be5)
164 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
166 - Revert 1-pass CBR rate control changes (e961317)
167 Further testing showed this change produced undesirable visual
168 artifacts, rolling back for now.
171 2011-08-02 v0.9.7 "Cayuga"
172 Our third named release, focused on a faster, higher quality, encoder.
174 - Upgrading:
175 This release is backwards compatible with Aylesbury (v0.9.5) and
176 Bali (v0.9.6). Users of older releases should refer to the Upgrading
177 notes in this document for that release.
179 - Enhancements:
180 Stereo 3D format support for vpxenc
181 Runtime detection of available processor cores.
182 Allow specifying --end-usage by enum name
183 vpxdec: test for frame corruption
184 vpxenc: add quantizer histogram display
185 vpxenc: add rate histogram display
186 Set VPX_FRAME_IS_DROPPABLE
187 update configure for ios sdk 4.3
188 Avoid text relocations in ARM vp8 decoder
189 Generate a vpx.pc file for pkg-config.
190 New ways of passing encoded data between encoder and decoder.
192 - Speed:
193 This release includes across-the-board speed improvements to the
194 encoder. On x86, these measure at approximately 11.5% in Best mode,
195 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
196 On ARM Cortex A9 with Neon extensions, real-time encoding of video
197 telephony content is 35% faster than Bali on single core and 48%
198 faster on multi-core. On the NVidia Tegra2 platform, real time
199 encoding is 40% faster than Bali.
201 Decoder speed was not a priority for this release, but improved
202 approximately 8.4% on x86.
204 Reduce motion vector search on alt-ref frame.
205 Encoder loopfilter running in its own thread
206 Reworked loopfilter to precalculate more parameters
207 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
208 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
209 Removed redundant checks
210 Reduced structure sizes
211 utilize preload in ARMv6 MC/LPF/Copy routines
212 ARM optimized quantization, dfct, variance, subtract
213 Increase chrow row alignment to 16 bytes.
214 disable trellis optimization for first pass
215 Write SSSE3 sub-pixel filter function
216 Improve SSE2 half-pixel filter funtions
217 Add vp8_sub_pixel_variance16x8_ssse3 function
218 Reduce unnecessary distortion computation
219 Use diamond search to replace full search
220 Preload reference area in sub-pixel motion search (real-time mode)
222 - Quality:
223 This release focused primarily on one-pass use cases, including
224 video conferencing. Low latency data rate control was significantly
225 improved, improving streamability over bandwidth constrained links.
226 Added support for error concealment, allowing frames to maintain
227 visual quality in the presence of substantial packet loss.
229 Add rc_max_intra_bitrate_pct control
230 Limit size of initial keyframe in one-pass.
231 Improve framerate adaptation
232 Improved 1-pass CBR rate control
233 Improved KF insertion after fades to still.
234 Improved key frame detection.
235 Improved activity masking (lower PSNR impact for same SSIM boost)
236 Improved interaction between GF and ARFs
237 Adding error-concealment to the decoder.
238 Adding support for independent partitions
239 Adjusted rate-distortion constants
242 - Bug Fixes:
243 Removed firstpass motion map
244 Fix parallel make install
245 Fix multithreaded encoding for 1 MB wide frame
246 Fixed iwalsh_neon build problems with RVDS4.1
247 Fix semaphore emulation, spin-wait intrinsics on Windows
248 Fix build with xcode4 and simplify GLOBAL.
249 Mark ARM asm objects as allowing a non-executable stack.
250 Fix vpxenc encoding incorrect webm file header on big endian
253 2011-03-07 v0.9.6 "Bali"
254 Our second named release, focused on a faster, higher quality, encoder.
256 - Upgrading:
257 This release is backwards compatible with Aylesbury (v0.9.5). Users
258 of older releases should refer to the Upgrading notes in this
259 document for that release.
261 - Enhancements:
262 vpxenc --psnr shows a summary when encode completes
263 --tune=ssim option to enable activity masking
264 improved postproc visualizations for development
265 updated support for Apple iOS to SDK 4.2
266 query decoder to determine which reference frames were updated
267 implemented error tracking in the decoder
268 fix pipe support on windows
270 - Speed:
271 Primary focus was on good quality mode, speed 0. Average improvement
272 on x86 about 40%, up to 100% on user-generated content at that speed.
273 Best quality mode speed improved 35%, and realtime speed 10-20%. This
274 release also saw significant improvement in realtime encoding speed
275 on ARM platforms.
277 Improved encoder threading
278 Dont pick encoder filter level when loopfilter is disabled.
279 Avoid double copying of key frames into alt and golden buffer
280 FDCT optimizations.
281 x86 sse2 temporal filter
282 SSSE3 version of fast quantizer
283 vp8_rd_pick_best_mbsegmentation code restructure
284 Adjusted breakout RD for SPLITMV
285 Changed segmentation check order
286 Improved rd_pick_intra4x4block
287 Adds armv6 optimized variance calculation
288 ARMv6 optimized sad16x16
289 ARMv6 optimized half pixel variance calculations
290 Full search SAD function optimization in SSE4.1
291 Improve MV prediction accuracy to achieve performance gain
292 Improve MV prediction in vp8_pick_inter_mode() for speed>3
294 - Quality:
295 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
296 also includes support for "activity masking," which greatly improves
297 SSIM at the expense of PSNR. For now, this feature is available with
298 the --tune=ssim option. Further experimentation in this area
299 is ongoing. This release also introduces a new rate control mode
300 called "CQ," which changes the allocation of bits within a clip to
301 the sections where they will have the most visual impact.
303 Tuning for the more exact quantizer.
304 Relax rate control for last few frames
305 CQ Mode
306 Limit key frame quantizer for forced key frames.
307 KF/GF Pulsing
308 Add simple version of activity masking.
309 make rdmult adaptive for intra in quantizer RDO
310 cap the best quantizer for 2nd order DC
311 change the threshold of DC check for encode breakout
313 - Bug Fixes:
314 Fix crash on Sparc Solaris.
315 Fix counter of fixed keyframe distance
316 ARNR filter pointer update bug fix
317 Fixed use of motion percentage in KF/GF group calc
318 Changed condition for using RD in Intra Mode
319 Fix encoder real-time only configuration.
320 Fix ARM encoder crash with multiple token partitions
321 Fixed bug first cluster timecode of webm file is wrong.
322 Fixed various encoder bugs with odd-sized images
323 vp8e_get_preview fixed when spatial resampling enabled
324 quantizer: fix assertion in fast quantizer path
325 Allocate source buffers to be multiples of 16
326 Fix for manual Golden frame frequency
327 Fix drastic undershoot in long form content
330 2010-10-28 v0.9.5 "Aylesbury"
331 Our first named release, focused on a faster decoder, and a better encoder.
333 - Upgrading:
334 This release incorporates backwards-incompatible changes to the
335 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
337 vpxdec
338 * the -q (quiet) option has been removed, and replaced with
339 -v (verbose). the output is quiet by default. Use -v to see
340 the version number of the binary.
342 * The default behavior is now to write output to a single file
343 instead of individual frames. The -y option has been removed.
344 Y4M output is the default.
346 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
347 options must be specified.
349 $ ivfdec -o OUTPUT INPUT
350 $ vpxdec --i420 -o OUTPUT INPUT
352 * If an output file is not specified, the default is to write
353 Y4M to stdout. This makes piping more natural.
355 $ ivfdec -y -o - INPUT | ...
356 $ vpxdec INPUT | ...
358 * The output file has additional flexibility for formatting the
359 filename. It supports escape characters for constructing a
360 filename from the width, height, and sequence number. This
361 replaces the -p option. To get the equivalent:
363 $ ivfdec -p frame INPUT
364 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
366 vpxenc
367 * The output file must be specified with -o, rather than as the
368 last argument.
370 $ ivfenc <options> INPUT OUTPUT
371 $ vpxenc <options> -o OUTPUT INPUT
373 * The output defaults to webm. To get IVF output, use the --ivf
374 option.
376 $ ivfenc <options> INPUT OUTPUT.ivf
377 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
380 - Enhancements:
381 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
382 vpxdec supports .webm input
383 vpxdec writes .y4m by default
384 vpxenc writes .webm output by default
385 vpxenc --psnr now shows the average/overall PSNR at the end
386 ARM platforms now support runtime cpu detection
387 vpxdec visualizations added for motion vectors, block modes, references
388 vpxdec now silent by default
389 vpxdec --progress shows frame-by-frame timing information
390 vpxenc supports the distinction between --fps and --timebase
391 NASM is now a supported assembler
392 configure: enable PIC for shared libs by default
393 configure: add --enable-small
394 configure: support for ppc32-linux-gcc
395 configure: support for sparc-solaris-gcc
397 - Bugs:
398 Improve handling of invalid frames
399 Fix valgrind errors in the NEON loop filters.
400 Fix loopfilter delta zero transitions
401 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
402 Build fixes for darwin-icc
404 - Speed:
405 20-40% (average 28%) improvement in libvpx decoder speed,
406 including:
407 Rewrite vp8_short_walsh4x4_sse2()
408 Optimizations on the loopfilters.
409 Miscellaneous improvements for Atom
410 Add 4-tap version of 2nd-pass ARMv6 MC filter.
411 Improved multithread utilization
412 Better instruction choices on x86
413 reorder data to use wider instructions
414 Update NEON wide idcts
415 Make block access to frame buffer sequential
416 Improved subset block search
417 Bilinear subpixel optimizations for ssse3.
418 Decrease memory footprint
420 Encoder speed improvements (percentage gain not measured):
421 Skip unnecessary search of identical frames
422 Add SSE2 subtract functions
423 Improve bounds checking in vp8_diamond_search_sadx4()
424 Added vp8_fast_quantize_b_sse2
426 - Quality:
427 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
428 encoding mode, and up to 60% improvement on very noisy, still
429 or slow moving source video
431 Motion compensated temporal filter for Alt-Ref Noise Reduction
432 Improved use of trellis quantization on 2nd order Y blocks
433 Tune effect of motion on KF/GF boost in two pass
434 Allow coefficient optimization for good quality speed 0.
435 Improved control of active min quantizer for two pass.
436 Enable ARFs for non-lagged compress
438 2010-09-02 v0.9.2
439 - Enhancements:
440 Disable frame dropping by default
441 Improved multithreaded performance
442 Improved Force Key Frame Behaviour
443 Increased rate control buffer level precision
444 Fix bug in 1st pass motion compensation
445 ivfenc: correct fixed kf interval, --disable-kf
446 - Speed:
447 Changed above and left context data layout
448 Rework idct calling structure.
449 Removed unnecessary MB_MODE_INFO copies
450 x86: SSSE3 sixtap prediction
451 Reworked IDCT to include reconstruction (add) step
452 Swap alt/gold/new/last frame buffer ptrs instead of copying.
453 Improve SSE2 loopfilter functions
454 Change bitreader to use a larger window.
455 Avoid loopfilter reinitialization when possible
456 - Quality:
457 Normalize quantizer's zero bin and rounding factors
458 Add trellis quantization.
459 Make the quantizer exact.
460 Updates to ARNR filtering algorithm
461 Fix breakout thresh computation for golden & AltRef frames
462 Redo the forward 4x4 dct
463 Improve the accuracy of forward walsh-hadamard transform
464 Further adjustment of RD behaviour with Q and Zbin.
465 - Build System:
466 Allow linking of libs built with MinGW to MSVC
467 Fix target auto-detection on mingw32
468 Allow --cpu= to work for x86.
469 configure: pass original arguments through to make dist
470 Fix builds without runtime CPU detection
471 msvs: fix install of codec sources
472 msvs: Change devenv.com command line for better msys support
473 msvs: Add vs9 targets.
474 Add x86_64-linux-icc target
475 - Bugs:
476 Potential crashes on older MinGW builds
477 Fix two-pass framrate for Y4M input.
478 Fixed simple loop filter, other crashes on ARM v6
479 arm: fix missing dependency with --enable-shared
480 configure: support directories containing .o
481 Replace pinsrw (SSE) with MMX instructions
482 apple: include proper mach primatives
483 Fixed rate control bug with long key frame interval.
484 Fix DSO link errors on x86-64 when not using a version script
485 Fixed buffer selection for UV in AltRef filtering
488 2010-06-17 v0.9.1
489 - Enhancements:
490 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
491 * Speed optimizations
492 - Bugfixes:
493 * Rate control
494 * Prevent out-of-bounds accesses on invalid data
495 - Build system updates:
496 * Detect toolchain to be used automatically for native builds
497 * Support building shared libraries
498 * Better autotools emulation (--prefix, --libdir, DESTDIR)
499 - Updated LICENSE
500 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
503 2010-05-18 v0.9.0
504 - Initial open source release. Welcome to WebM and VP8!