12011-03-07 v0.9.6 "Bali" 2 Our second named release, focused on a faster, higher quality, encoder. 3 4 - Upgrading: 5 This release is backwards compatible with Aylesbury (v0.9.5). Users 6 of older releases should refer to the Upgrading notes in this 7 document for that release. 8 9 - Enhancements: 10 vpxenc --psnr shows a summary when encode completes 11 --tune=ssim option to enable activity masking 12 improved postproc visualizations for development 13 updated support for Apple iOS to SDK 4.2 14 query decoder to determine which reference frames were updated 15 implemented error tracking in the decoder 16 fix pipe support on windows 17 18 - Speed: 19 Primary focus was on good quality mode, speed 0. Average improvement 20 on x86 about 40%, up to 100% on user-generated content at that speed. 21 Best quality mode speed improved 35%, and realtime speed 10-20%. This 22 release also saw significant improvement in realtime encoding speed 23 on ARM platforms. 24 25 Improved encoder threading 26 Dont pick encoder filter level when loopfilter is disabled. 27 Avoid double copying of key frames into alt and golden buffer 28 FDCT optimizations. 29 x86 sse2 temporal filter 30 SSSE3 version of fast quantizer 31 vp8_rd_pick_best_mbsegmentation code restructure 32 Adjusted breakout RD for SPLITMV 33 Changed segmentation check order 34 Improved rd_pick_intra4x4block 35 Adds armv6 optimized variance calculation 36 ARMv6 optimized sad16x16 37 ARMv6 optimized half pixel variance calculations 38 Full search SAD function optimization in SSE4.1 39 Improve MV prediction accuracy to achieve performance gain 40 Improve MV prediction in vp8_pick_inter_mode() for speed>3 41 42 - Quality: 43 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release 44 also includes support for "activity masking," which greatly improves 45 SSIM at the expense of PSNR. For now, this feature is available with 46 the --tune=ssim option. Further experimentation in this area 47 is ongoing. This release also introduces a new rate control mode 48 called "CQ," which changes the allocation of bits within a clip to 49 the sections where they will have the most visual impact. 50 51 Tuning for the more exact quantizer. 52 Relax rate control for last few frames 53 CQ Mode 54 Limit key frame quantizer for forced key frames. 55 KF/GF Pulsing 56 Add simple version of activity masking. 57 make rdmult adaptive for intra in quantizer RDO 58 cap the best quantizer for 2nd order DC 59 change the threshold of DC check for encode breakout 60 61 - Bug Fixes: 62 Fix crash on Sparc Solaris. 63 Fix counter of fixed keyframe distance 64 ARNR filter pointer update bug fix 65 Fixed use of motion percentage in KF/GF group calc 66 Changed condition for using RD in Intra Mode 67 Fix encoder real-time only configuration. 68 Fix ARM encoder crash with multiple token partitions 69 Fixed bug first cluster timecode of webm file is wrong. 70 Fixed various encoder bugs with odd-sized images 71 vp8e_get_preview fixed when spatial resampling enabled 72 quantizer: fix assertion in fast quantizer path 73 Allocate source buffers to be multiples of 16 74 Fix for manual Golden frame frequency 75 Fix drastic undershoot in long form content 76 77 782010-10-28 v0.9.5 "Aylesbury" 79 Our first named release, focused on a faster decoder, and a better encoder. 80 81 - Upgrading: 82 This release incorporates backwards-incompatible changes to the 83 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec. 84 85 vpxdec 86 * the -q (quiet) option has been removed, and replaced with 87 -v (verbose). the output is quiet by default. Use -v to see 88 the version number of the binary. 89 90 * The default behavior is now to write output to a single file 91 instead of individual frames. The -y option has been removed. 92 Y4M output is the default. 93 94 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12 95 options must be specified. 96 97 $ ivfdec -o OUTPUT INPUT 98 $ vpxdec --i420 -o OUTPUT INPUT 99 100 * If an output file is not specified, the default is to write 101 Y4M to stdout. This makes piping more natural. 102 103 $ ivfdec -y -o - INPUT | ... 104 $ vpxdec INPUT | ... 105 106 * The output file has additional flexibility for formatting the 107 filename. It supports escape characters for constructing a 108 filename from the width, height, and sequence number. This 109 replaces the -p option. To get the equivalent: 110 111 $ ivfdec -p frame INPUT 112 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT 113 114 vpxenc 115 * The output file must be specified with -o, rather than as the 116 last argument. 117 118 $ ivfenc <options> INPUT OUTPUT 119 $ vpxenc <options> -o OUTPUT INPUT 120 121 * The output defaults to webm. To get IVF output, use the --ivf 122 option. 123 124 $ ivfenc <options> INPUT OUTPUT.ivf 125 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT 126 127 128 - Enhancements: 129 ivfenc and ivfdec have been renamed to vpxenc, vpxdec. 130 vpxdec supports .webm input 131 vpxdec writes .y4m by default 132 vpxenc writes .webm output by default 133 vpxenc --psnr now shows the average/overall PSNR at the end 134 ARM platforms now support runtime cpu detection 135 vpxdec visualizations added for motion vectors, block modes, references 136 vpxdec now silent by default 137 vpxdec --progress shows frame-by-frame timing information 138 vpxenc supports the distinction between --fps and --timebase 139 NASM is now a supported assembler 140 configure: enable PIC for shared libs by default 141 configure: add --enable-small 142 configure: support for ppc32-linux-gcc 143 configure: support for sparc-solaris-gcc 144 145 - Bugs: 146 Improve handling of invalid frames 147 Fix valgrind errors in the NEON loop filters. 148 Fix loopfilter delta zero transitions 149 Fix valgrind errors in vp8_sixtap_predict8x4_armv6(). 150 Build fixes for darwin-icc 151 152 - Speed: 153 20-40% (average 28%) improvement in libvpx decoder speed, 154 including: 155 Rewrite vp8_short_walsh4x4_sse2() 156 Optimizations on the loopfilters. 157 Miscellaneous improvements for Atom 158 Add 4-tap version of 2nd-pass ARMv6 MC filter. 159 Improved multithread utilization 160 Better instruction choices on x86 161 reorder data to use wider instructions 162 Update NEON wide idcts 163 Make block access to frame buffer sequential 164 Improved subset block search 165 Bilinear subpixel optimizations for ssse3. 166 Decrease memory footprint 167 168 Encoder speed improvements (percentage gain not measured): 169 Skip unnecessary search of identical frames 170 Add SSE2 subtract functions 171 Improve bounds checking in vp8_diamond_search_sadx4() 172 Added vp8_fast_quantize_b_sse2 173 174 - Quality: 175 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality 176 encoding mode, and up to 60% improvement on very noisy, still 177 or slow moving source video 178 179 Motion compensated temporal filter for Alt-Ref Noise Reduction 180 Improved use of trellis quantization on 2nd order Y blocks 181 Tune effect of motion on KF/GF boost in two pass 182 Allow coefficient optimization for good quality speed 0. 183 Improved control of active min quantizer for two pass. 184 Enable ARFs for non-lagged compress 185 1862010-09-02 v0.9.2 187 - Enhancements: 188 Disable frame dropping by default 189 Improved multithreaded performance 190 Improved Force Key Frame Behaviour 191 Increased rate control buffer level precision 192 Fix bug in 1st pass motion compensation 193 ivfenc: correct fixed kf interval, --disable-kf 194 - Speed: 195 Changed above and left context data layout 196 Rework idct calling structure. 197 Removed unnecessary MB_MODE_INFO copies 198 x86: SSSE3 sixtap prediction 199 Reworked IDCT to include reconstruction (add) step 200 Swap alt/gold/new/last frame buffer ptrs instead of copying. 201 Improve SSE2 loopfilter functions 202 Change bitreader to use a larger window. 203 Avoid loopfilter reinitialization when possible 204 - Quality: 205 Normalize quantizer's zero bin and rounding factors 206 Add trellis quantization. 207 Make the quantizer exact. 208 Updates to ARNR filtering algorithm 209 Fix breakout thresh computation for golden & AltRef frames 210 Redo the forward 4x4 dct 211 Improve the accuracy of forward walsh-hadamard transform 212 Further adjustment of RD behaviour with Q and Zbin. 213 - Build System: 214 Allow linking of libs built with MinGW to MSVC 215 Fix target auto-detection on mingw32 216 Allow --cpu= to work for x86. 217 configure: pass original arguments through to make dist 218 Fix builds without runtime CPU detection 219 msvs: fix install of codec sources 220 msvs: Change devenv.com command line for better msys support 221 msvs: Add vs9 targets. 222 Add x86_64-linux-icc target 223 - Bugs: 224 Potential crashes on older MinGW builds 225 Fix two-pass framrate for Y4M input. 226 Fixed simple loop filter, other crashes on ARM v6 227 arm: fix missing dependency with --enable-shared 228 configure: support directories containing .o 229 Replace pinsrw (SSE) with MMX instructions 230 apple: include proper mach primatives 231 Fixed rate control bug with long key frame interval. 232 Fix DSO link errors on x86-64 when not using a version script 233 Fixed buffer selection for UV in AltRef filtering 234 235 2362010-06-17 v0.9.1 237 - Enhancements: 238 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O 239 * Speed optimizations 240 - Bugfixes: 241 * Rate control 242 * Prevent out-of-bounds accesses on invalid data 243 - Build system updates: 244 * Detect toolchain to be used automatically for native builds 245 * Support building shared libraries 246 * Better autotools emulation (--prefix, --libdir, DESTDIR) 247 - Updated LICENSE 248 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html 249 250 2512010-05-18 v0.9.0 252 - Initial open source release. Welcome to WebM and VP8! 253 254