1 2011-08-15 v0.9.7-p1 "Cayuga" patch 1
2 This is an incremental bugfix release against Cayuga. All users of that
3 release are strongly encouraged to upgrade.
5 - Fix potential OOB reads (cdae03a)
7 An unbounded out of bounds read was discovered when the
8 decoder was requested to perform error concealment (new in
9 Cayuga) given a frame with corrupt partition sizes.
11 A bounded out of bounds read was discovered affecting all
12 versions of libvpx. Given an multipartition input frame that
13 is truncated between the mode/mv partition and the first
14 residiual paritition (in the block of partition offsets), up
15 to 3 extra bytes could have been read from the source buffer.
16 The code will not take any action regardless of the contents
17 of these undefined bytes, as the truncated buffer is detected
18 immediately following the read based on the calculated
19 starting position of the coefficient partition.
21 - Fix potential error concealment crash when the very first frame
22 is missing or corrupt (a609be5)
24 - Fix significant artifacts in error concealment (a4c2211, 99d870a)
26 - Revert 1-pass CBR rate control changes (e961317)
27 Further testing showed this change produced undesirable visual
28 artifacts, rolling back for now.
31 2011-08-02 v0.9.7 "Cayuga"
32 Our third named release, focused on a faster, higher quality, encoder.
35 This release is backwards compatible with Aylesbury (v0.9.5) and
36 Bali (v0.9.6). Users of older releases should refer to the Upgrading
37 notes in this document for that release.
40 Stereo 3D format support for vpxenc
41 Runtime detection of available processor cores.
42 Allow specifying --end-usage by enum name
43 vpxdec: test for frame corruption
44 vpxenc: add quantizer histogram display
45 vpxenc: add rate histogram display
46 Set VPX_FRAME_IS_DROPPABLE
47 update configure for ios sdk 4.3
48 Avoid text relocations in ARM vp8 decoder
49 Generate a vpx.pc file for pkg-config.
50 New ways of passing encoded data between encoder and decoder.
53 This release includes across-the-board speed improvements to the
54 encoder. On x86, these measure at approximately 11.5% in Best mode,
55 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
56 On ARM Cortex A9 with Neon extensions, real-time encoding of video
57 telephony content is 35% faster than Bali on single core and 48%
58 faster on multi-core. On the NVidia Tegra2 platform, real time
59 encoding is 40% faster than Bali.
61 Decoder speed was not a priority for this release, but improved
62 approximately 8.4% on x86.
64 Reduce motion vector search on alt-ref frame.
65 Encoder loopfilter running in its own thread
66 Reworked loopfilter to precalculate more parameters
67 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
68 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
69 Removed redundant checks
70 Reduced structure sizes
71 utilize preload in ARMv6 MC/LPF/Copy routines
72 ARM optimized quantization, dfct, variance, subtract
73 Increase chrow row alignment to 16 bytes.
74 disable trellis optimization for first pass
75 Write SSSE3 sub-pixel filter function
76 Improve SSE2 half-pixel filter funtions
77 Add vp8_sub_pixel_variance16x8_ssse3 function
78 Reduce unnecessary distortion computation
79 Use diamond search to replace full search
80 Preload reference area in sub-pixel motion search (real-time mode)
83 This release focused primarily on one-pass use cases, including
84 video conferencing. Low latency data rate control was significantly
85 improved, improving streamability over bandwidth constrained links.
86 Added support for error concealment, allowing frames to maintain
87 visual quality in the presence of substantial packet loss.
89 Add rc_max_intra_bitrate_pct control
90 Limit size of initial keyframe in one-pass.
91 Improve framerate adaptation
92 Improved 1-pass CBR rate control
93 Improved KF insertion after fades to still.
94 Improved key frame detection.
95 Improved activity masking (lower PSNR impact for same SSIM boost)
96 Improved interaction between GF and ARFs
97 Adding error-concealment to the decoder.
98 Adding support for independent partitions
99 Adjusted rate-distortion constants
103 Removed firstpass motion map
104 Fix parallel make install
105 Fix multithreaded encoding for 1 MB wide frame
106 Fixed iwalsh_neon build problems with RVDS4.1
107 Fix semaphore emulation, spin-wait intrinsics on Windows
108 Fix build with xcode4 and simplify GLOBAL.
109 Mark ARM asm objects as allowing a non-executable stack.
110 Fix vpxenc encoding incorrect webm file header on big endian
113 2011-03-07 v0.9.6 "Bali"
114 Our second named release, focused on a faster, higher quality, encoder.
117 This release is backwards compatible with Aylesbury (v0.9.5). Users
118 of older releases should refer to the Upgrading notes in this
119 document for that release.
122 vpxenc --psnr shows a summary when encode completes
123 --tune=ssim option to enable activity masking
124 improved postproc visualizations for development
125 updated support for Apple iOS to SDK 4.2
126 query decoder to determine which reference frames were updated
127 implemented error tracking in the decoder
128 fix pipe support on windows
131 Primary focus was on good quality mode, speed 0. Average improvement
132 on x86 about 40%, up to 100% on user-generated content at that speed.
133 Best quality mode speed improved 35%, and realtime speed 10-20%. This
134 release also saw significant improvement in realtime encoding speed
137 Improved encoder threading
138 Dont pick encoder filter level when loopfilter is disabled.
139 Avoid double copying of key frames into alt and golden buffer
141 x86 sse2 temporal filter
142 SSSE3 version of fast quantizer
143 vp8_rd_pick_best_mbsegmentation code restructure
144 Adjusted breakout RD for SPLITMV
145 Changed segmentation check order
146 Improved rd_pick_intra4x4block
147 Adds armv6 optimized variance calculation
148 ARMv6 optimized sad16x16
149 ARMv6 optimized half pixel variance calculations
150 Full search SAD function optimization in SSE4.1
151 Improve MV prediction accuracy to achieve performance gain
152 Improve MV prediction in vp8_pick_inter_mode() for speed>3
155 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
156 also includes support for "activity masking," which greatly improves
157 SSIM at the expense of PSNR. For now, this feature is available with
158 the --tune=ssim option. Further experimentation in this area
159 is ongoing. This release also introduces a new rate control mode
160 called "CQ," which changes the allocation of bits within a clip to
161 the sections where they will have the most visual impact.
163 Tuning for the more exact quantizer.
164 Relax rate control for last few frames
166 Limit key frame quantizer for forced key frames.
168 Add simple version of activity masking.
169 make rdmult adaptive for intra in quantizer RDO
170 cap the best quantizer for 2nd order DC
171 change the threshold of DC check for encode breakout
174 Fix crash on Sparc Solaris.
175 Fix counter of fixed keyframe distance
176 ARNR filter pointer update bug fix
177 Fixed use of motion percentage in KF/GF group calc
178 Changed condition for using RD in Intra Mode
179 Fix encoder real-time only configuration.
180 Fix ARM encoder crash with multiple token partitions
181 Fixed bug first cluster timecode of webm file is wrong.
182 Fixed various encoder bugs with odd-sized images
183 vp8e_get_preview fixed when spatial resampling enabled
184 quantizer: fix assertion in fast quantizer path
185 Allocate source buffers to be multiples of 16
186 Fix for manual Golden frame frequency
187 Fix drastic undershoot in long form content
190 2010-10-28 v0.9.5 "Aylesbury"
191 Our first named release, focused on a faster decoder, and a better encoder.
194 This release incorporates backwards-incompatible changes to the
195 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
198 * the -q (quiet) option has been removed, and replaced with
199 -v (verbose). the output is quiet by default. Use -v to see
200 the version number of the binary.
202 * The default behavior is now to write output to a single file
203 instead of individual frames. The -y option has been removed.
204 Y4M output is the default.
206 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
207 options must be specified.
209 $ ivfdec -o OUTPUT INPUT
210 $ vpxdec --i420 -o OUTPUT INPUT
212 * If an output file is not specified, the default is to write
213 Y4M to stdout. This makes piping more natural.
215 $ ivfdec -y -o - INPUT | ...
218 * The output file has additional flexibility for formatting the
219 filename. It supports escape characters for constructing a
220 filename from the width, height, and sequence number. This
221 replaces the -p option. To get the equivalent:
223 $ ivfdec -p frame INPUT
224 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
227 * The output file must be specified with -o, rather than as the
230 $ ivfenc <options> INPUT OUTPUT
231 $ vpxenc <options> -o OUTPUT INPUT
233 * The output defaults to webm. To get IVF output, use the --ivf
236 $ ivfenc <options> INPUT OUTPUT.ivf
237 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
241 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
242 vpxdec supports .webm input
243 vpxdec writes .y4m by default
244 vpxenc writes .webm output by default
245 vpxenc --psnr now shows the average/overall PSNR at the end
246 ARM platforms now support runtime cpu detection
247 vpxdec visualizations added for motion vectors, block modes, references
248 vpxdec now silent by default
249 vpxdec --progress shows frame-by-frame timing information
250 vpxenc supports the distinction between --fps and --timebase
251 NASM is now a supported assembler
252 configure: enable PIC for shared libs by default
253 configure: add --enable-small
254 configure: support for ppc32-linux-gcc
255 configure: support for sparc-solaris-gcc
258 Improve handling of invalid frames
259 Fix valgrind errors in the NEON loop filters.
260 Fix loopfilter delta zero transitions
261 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
262 Build fixes for darwin-icc
265 20-40% (average 28%) improvement in libvpx decoder speed,
267 Rewrite vp8_short_walsh4x4_sse2()
268 Optimizations on the loopfilters.
269 Miscellaneous improvements for Atom
270 Add 4-tap version of 2nd-pass ARMv6 MC filter.
271 Improved multithread utilization
272 Better instruction choices on x86
273 reorder data to use wider instructions
274 Update NEON wide idcts
275 Make block access to frame buffer sequential
276 Improved subset block search
277 Bilinear subpixel optimizations for ssse3.
278 Decrease memory footprint
280 Encoder speed improvements (percentage gain not measured):
281 Skip unnecessary search of identical frames
282 Add SSE2 subtract functions
283 Improve bounds checking in vp8_diamond_search_sadx4()
284 Added vp8_fast_quantize_b_sse2
287 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
288 encoding mode, and up to 60% improvement on very noisy, still
289 or slow moving source video
291 Motion compensated temporal filter for Alt-Ref Noise Reduction
292 Improved use of trellis quantization on 2nd order Y blocks
293 Tune effect of motion on KF/GF boost in two pass
294 Allow coefficient optimization for good quality speed 0.
295 Improved control of active min quantizer for two pass.
296 Enable ARFs for non-lagged compress
300 Disable frame dropping by default
301 Improved multithreaded performance
302 Improved Force Key Frame Behaviour
303 Increased rate control buffer level precision
304 Fix bug in 1st pass motion compensation
305 ivfenc: correct fixed kf interval, --disable-kf
307 Changed above and left context data layout
308 Rework idct calling structure.
309 Removed unnecessary MB_MODE_INFO copies
310 x86: SSSE3 sixtap prediction
311 Reworked IDCT to include reconstruction (add) step
312 Swap alt/gold/new/last frame buffer ptrs instead of copying.
313 Improve SSE2 loopfilter functions
314 Change bitreader to use a larger window.
315 Avoid loopfilter reinitialization when possible
317 Normalize quantizer's zero bin and rounding factors
318 Add trellis quantization.
319 Make the quantizer exact.
320 Updates to ARNR filtering algorithm
321 Fix breakout thresh computation for golden & AltRef frames
322 Redo the forward 4x4 dct
323 Improve the accuracy of forward walsh-hadamard transform
324 Further adjustment of RD behaviour with Q and Zbin.
326 Allow linking of libs built with MinGW to MSVC
327 Fix target auto-detection on mingw32
328 Allow --cpu= to work for x86.
329 configure: pass original arguments through to make dist
330 Fix builds without runtime CPU detection
331 msvs: fix install of codec sources
332 msvs: Change devenv.com command line for better msys support
333 msvs: Add vs9 targets.
334 Add x86_64-linux-icc target
336 Potential crashes on older MinGW builds
337 Fix two-pass framrate for Y4M input.
338 Fixed simple loop filter, other crashes on ARM v6
339 arm: fix missing dependency with --enable-shared
340 configure: support directories containing .o
341 Replace pinsrw (SSE) with MMX instructions
342 apple: include proper mach primatives
343 Fixed rate control bug with long key frame interval.
344 Fix DSO link errors on x86-64 when not using a version script
345 Fixed buffer selection for UV in AltRef filtering
350 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
351 * Speed optimizations
354 * Prevent out-of-bounds accesses on invalid data
355 - Build system updates:
356 * Detect toolchain to be used automatically for native builds
357 * Support building shared libraries
358 * Better autotools emulation (--prefix, --libdir, DESTDIR)
360 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
364 - Initial open source release. Welcome to WebM and VP8!