1 2011-08-02 v0.9.7 "Cayuga"
2 Our third named release, focused on a faster, higher quality, encoder.
5 This release is backwards compatible with Aylesbury (v0.9.5) and
6 Bali (v0.9.6). Users of older releases should refer to the Upgrading
7 notes in this document for that release.
10 Stereo 3D format support for vpxenc
11 Runtime detection of available processor cores.
12 Allow specifying --end-usage by enum name
13 vpxdec: test for frame corruption
14 vpxenc: add quantizer histogram display
15 vpxenc: add rate histogram display
16 Set VPX_FRAME_IS_DROPPABLE
17 update configure for ios sdk 4.3
18 Avoid text relocations in ARM vp8 decoder
19 Generate a vpx.pc file for pkg-config.
20 New ways of passing encoded data between encoder and decoder.
23 This release includes across-the-board speed improvements to the
24 encoder. On x86, these measure at approximately 11.5% in Best mode,
25 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6).
26 On ARM Cortex A9 with Neon extensions, real-time encoding of video
27 telephony content is 35% faster than Bali on single core and 48%
28 faster on multi-core. On the NVidia Tegra2 platform, real time
29 encoding is 40% faster than Bali.
31 Decoder speed was not a priority for this release, but improved
32 approximately 8.4% on x86.
34 Reduce motion vector search on alt-ref frame.
35 Encoder loopfilter running in its own thread
36 Reworked loopfilter to precalculate more parameters
37 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}().
38 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3.
39 Removed redundant checks
40 Reduced structure sizes
41 utilize preload in ARMv6 MC/LPF/Copy routines
42 ARM optimized quantization, dfct, variance, subtract
43 Increase chrow row alignment to 16 bytes.
44 disable trellis optimization for first pass
45 Write SSSE3 sub-pixel filter function
46 Improve SSE2 half-pixel filter funtions
47 Add vp8_sub_pixel_variance16x8_ssse3 function
48 Reduce unnecessary distortion computation
49 Use diamond search to replace full search
50 Preload reference area in sub-pixel motion search (real-time mode)
53 This release focused primarily on one-pass use cases, including
54 video conferencing. Low latency data rate control was significantly
55 improved, improving streamability over bandwidth constrained links.
56 Added support for error concealment, allowing frames to maintain
57 visual quality in the presence of substantial packet loss.
59 Add rc_max_intra_bitrate_pct control
60 Limit size of initial keyframe in one-pass.
61 Improve framerate adaptation
62 Improved 1-pass CBR rate control
63 Improved KF insertion after fades to still.
64 Improved key frame detection.
65 Improved activity masking (lower PSNR impact for same SSIM boost)
66 Improved interaction between GF and ARFs
67 Adding error-concealment to the decoder.
68 Adding support for independent partitions
69 Adjusted rate-distortion constants
73 Removed firstpass motion map
74 Fix parallel make install
75 Fix multithreaded encoding for 1 MB wide frame
76 Fixed iwalsh_neon build problems with RVDS4.1
77 Fix semaphore emulation, spin-wait intrinsics on Windows
78 Fix build with xcode4 and simplify GLOBAL.
79 Mark ARM asm objects as allowing a non-executable stack.
80 Fix vpxenc encoding incorrect webm file header on big endian
83 2011-03-07 v0.9.6 "Bali"
84 Our second named release, focused on a faster, higher quality, encoder.
87 This release is backwards compatible with Aylesbury (v0.9.5). Users
88 of older releases should refer to the Upgrading notes in this
89 document for that release.
92 vpxenc --psnr shows a summary when encode completes
93 --tune=ssim option to enable activity masking
94 improved postproc visualizations for development
95 updated support for Apple iOS to SDK 4.2
96 query decoder to determine which reference frames were updated
97 implemented error tracking in the decoder
98 fix pipe support on windows
101 Primary focus was on good quality mode, speed 0. Average improvement
102 on x86 about 40%, up to 100% on user-generated content at that speed.
103 Best quality mode speed improved 35%, and realtime speed 10-20%. This
104 release also saw significant improvement in realtime encoding speed
107 Improved encoder threading
108 Dont pick encoder filter level when loopfilter is disabled.
109 Avoid double copying of key frames into alt and golden buffer
111 x86 sse2 temporal filter
112 SSSE3 version of fast quantizer
113 vp8_rd_pick_best_mbsegmentation code restructure
114 Adjusted breakout RD for SPLITMV
115 Changed segmentation check order
116 Improved rd_pick_intra4x4block
117 Adds armv6 optimized variance calculation
118 ARMv6 optimized sad16x16
119 ARMv6 optimized half pixel variance calculations
120 Full search SAD function optimization in SSE4.1
121 Improve MV prediction accuracy to achieve performance gain
122 Improve MV prediction in vp8_pick_inter_mode() for speed>3
125 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release
126 also includes support for "activity masking," which greatly improves
127 SSIM at the expense of PSNR. For now, this feature is available with
128 the --tune=ssim option. Further experimentation in this area
129 is ongoing. This release also introduces a new rate control mode
130 called "CQ," which changes the allocation of bits within a clip to
131 the sections where they will have the most visual impact.
133 Tuning for the more exact quantizer.
134 Relax rate control for last few frames
136 Limit key frame quantizer for forced key frames.
138 Add simple version of activity masking.
139 make rdmult adaptive for intra in quantizer RDO
140 cap the best quantizer for 2nd order DC
141 change the threshold of DC check for encode breakout
144 Fix crash on Sparc Solaris.
145 Fix counter of fixed keyframe distance
146 ARNR filter pointer update bug fix
147 Fixed use of motion percentage in KF/GF group calc
148 Changed condition for using RD in Intra Mode
149 Fix encoder real-time only configuration.
150 Fix ARM encoder crash with multiple token partitions
151 Fixed bug first cluster timecode of webm file is wrong.
152 Fixed various encoder bugs with odd-sized images
153 vp8e_get_preview fixed when spatial resampling enabled
154 quantizer: fix assertion in fast quantizer path
155 Allocate source buffers to be multiples of 16
156 Fix for manual Golden frame frequency
157 Fix drastic undershoot in long form content
160 2010-10-28 v0.9.5 "Aylesbury"
161 Our first named release, focused on a faster decoder, and a better encoder.
164 This release incorporates backwards-incompatible changes to the
165 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec.
168 * the -q (quiet) option has been removed, and replaced with
169 -v (verbose). the output is quiet by default. Use -v to see
170 the version number of the binary.
172 * The default behavior is now to write output to a single file
173 instead of individual frames. The -y option has been removed.
174 Y4M output is the default.
176 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12
177 options must be specified.
179 $ ivfdec -o OUTPUT INPUT
180 $ vpxdec --i420 -o OUTPUT INPUT
182 * If an output file is not specified, the default is to write
183 Y4M to stdout. This makes piping more natural.
185 $ ivfdec -y -o - INPUT | ...
188 * The output file has additional flexibility for formatting the
189 filename. It supports escape characters for constructing a
190 filename from the width, height, and sequence number. This
191 replaces the -p option. To get the equivalent:
193 $ ivfdec -p frame INPUT
194 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT
197 * The output file must be specified with -o, rather than as the
200 $ ivfenc <options> INPUT OUTPUT
201 $ vpxenc <options> -o OUTPUT INPUT
203 * The output defaults to webm. To get IVF output, use the --ivf
206 $ ivfenc <options> INPUT OUTPUT.ivf
207 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT
211 ivfenc and ivfdec have been renamed to vpxenc, vpxdec.
212 vpxdec supports .webm input
213 vpxdec writes .y4m by default
214 vpxenc writes .webm output by default
215 vpxenc --psnr now shows the average/overall PSNR at the end
216 ARM platforms now support runtime cpu detection
217 vpxdec visualizations added for motion vectors, block modes, references
218 vpxdec now silent by default
219 vpxdec --progress shows frame-by-frame timing information
220 vpxenc supports the distinction between --fps and --timebase
221 NASM is now a supported assembler
222 configure: enable PIC for shared libs by default
223 configure: add --enable-small
224 configure: support for ppc32-linux-gcc
225 configure: support for sparc-solaris-gcc
228 Improve handling of invalid frames
229 Fix valgrind errors in the NEON loop filters.
230 Fix loopfilter delta zero transitions
231 Fix valgrind errors in vp8_sixtap_predict8x4_armv6().
232 Build fixes for darwin-icc
235 20-40% (average 28%) improvement in libvpx decoder speed,
237 Rewrite vp8_short_walsh4x4_sse2()
238 Optimizations on the loopfilters.
239 Miscellaneous improvements for Atom
240 Add 4-tap version of 2nd-pass ARMv6 MC filter.
241 Improved multithread utilization
242 Better instruction choices on x86
243 reorder data to use wider instructions
244 Update NEON wide idcts
245 Make block access to frame buffer sequential
246 Improved subset block search
247 Bilinear subpixel optimizations for ssse3.
248 Decrease memory footprint
250 Encoder speed improvements (percentage gain not measured):
251 Skip unnecessary search of identical frames
252 Add SSE2 subtract functions
253 Improve bounds checking in vp8_diamond_search_sadx4()
254 Added vp8_fast_quantize_b_sse2
257 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality
258 encoding mode, and up to 60% improvement on very noisy, still
259 or slow moving source video
261 Motion compensated temporal filter for Alt-Ref Noise Reduction
262 Improved use of trellis quantization on 2nd order Y blocks
263 Tune effect of motion on KF/GF boost in two pass
264 Allow coefficient optimization for good quality speed 0.
265 Improved control of active min quantizer for two pass.
266 Enable ARFs for non-lagged compress
270 Disable frame dropping by default
271 Improved multithreaded performance
272 Improved Force Key Frame Behaviour
273 Increased rate control buffer level precision
274 Fix bug in 1st pass motion compensation
275 ivfenc: correct fixed kf interval, --disable-kf
277 Changed above and left context data layout
278 Rework idct calling structure.
279 Removed unnecessary MB_MODE_INFO copies
280 x86: SSSE3 sixtap prediction
281 Reworked IDCT to include reconstruction (add) step
282 Swap alt/gold/new/last frame buffer ptrs instead of copying.
283 Improve SSE2 loopfilter functions
284 Change bitreader to use a larger window.
285 Avoid loopfilter reinitialization when possible
287 Normalize quantizer's zero bin and rounding factors
288 Add trellis quantization.
289 Make the quantizer exact.
290 Updates to ARNR filtering algorithm
291 Fix breakout thresh computation for golden & AltRef frames
292 Redo the forward 4x4 dct
293 Improve the accuracy of forward walsh-hadamard transform
294 Further adjustment of RD behaviour with Q and Zbin.
296 Allow linking of libs built with MinGW to MSVC
297 Fix target auto-detection on mingw32
298 Allow --cpu= to work for x86.
299 configure: pass original arguments through to make dist
300 Fix builds without runtime CPU detection
301 msvs: fix install of codec sources
302 msvs: Change devenv.com command line for better msys support
303 msvs: Add vs9 targets.
304 Add x86_64-linux-icc target
306 Potential crashes on older MinGW builds
307 Fix two-pass framrate for Y4M input.
308 Fixed simple loop filter, other crashes on ARM v6
309 arm: fix missing dependency with --enable-shared
310 configure: support directories containing .o
311 Replace pinsrw (SSE) with MMX instructions
312 apple: include proper mach primatives
313 Fixed rate control bug with long key frame interval.
314 Fix DSO link errors on x86-64 when not using a version script
315 Fixed buffer selection for UV in AltRef filtering
320 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O
321 * Speed optimizations
324 * Prevent out-of-bounds accesses on invalid data
325 - Build system updates:
326 * Detect toolchain to be used automatically for native builds
327 * Support building shared libraries
328 * Better autotools emulation (--prefix, --libdir, DESTDIR)
330 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html
334 - Initial open source release. Welcome to WebM and VP8!