CHANGELOG revision ba164dffc5a6795bce97fae02b51ccf3330e15e4
12012-12-21 v1.2.0 2 This release acts as a checkpoint for a large amount of internal refactoring 3 and testing. It also contains a number of small bugfixes, so all users are 4 encouraged to upgrade. 5 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. 10 11 - Enhancements: 12 VP8 optimizations for MIPS dspr2 13 vpxenc: add -quiet option 14 15 - Speed: 16 Encoder and decoder speed is consistent with the Eider release. 17 18 - Quality: 19 In general, quality is consistent with the Eider release. 20 21 Minor tweaks to ARNR filtering 22 Minor improvements to real time encoding with multiple temporal layers 23 24 - Bug Fixes: 25 Fixes multithreaded encoder race condition in loopfilter 26 Fixes multi-resolution threaded encoding 27 Fix potential encoder dead-lock after picture resize 28 29 302012-05-09 v1.1.0 "Eider" 31 This introduces a number of enhancements, mostly focused on real-time 32 encoding. In addition, it fixes a decoder bug (first introduced in 33 Duclair) so all users of that release are encouraged to upgrade. 34 35 - Upgrading: 36 This release is ABI and API compatible with Duclair (v1.0.0). Users 37 of older releases should refer to the Upgrading notes in this 38 document for that release. 39 40 This release introduces a new temporal denoiser, controlled by the 41 VP8E_SET_NOISE_SENSITIVITY control. The temporal denoiser does not 42 currently take a strength parameter, so the control is effectively 43 a boolean - zero (off) or non-zero (on). For compatibility with 44 existing applications, the values accepted are the same as those 45 for the spatial denoiser (0-6). The temporal denoiser is enabled 46 by default, and the older spatial denoiser may be restored by 47 configuring with --disable-temporal-denoising. The temporal denoiser 48 is more computationally intensive than the spatial one. 49 50 This release removes support for a legacy, decode only API that was 51 supported, but deprecated, at the initial release of libvpx 52 (v0.9.0). This is not expected to have any impact. If you are 53 impacted, you can apply a reversion to commit 2bf8fb58 locally. 54 Please update to the latest libvpx API if you are affected. 55 56 - Enhancements: 57 Adds a motion compensated temporal denoiser to the encoder, which 58 gives higher quality than the older spatial denoiser. (See above 59 for notes on upgrading). 60 61 In addition, support for new compilers and platforms were added, 62 including: 63 improved support for XCode 64 Android x86 NDK build 65 OS/2 support 66 SunCC support 67 68 Changing resolution with vpx_codec_enc_config_set() is now 69 supported. Previously, reinitializing the codec was required to 70 change the input resolution. 71 72 The vpxenc application has initial support for producing multiple 73 encodes from the same input in one call. Resizing is not yet 74 supported, but varying other codec parameters is. Use -- to 75 delineate output streams. Options persist from one stream to the 76 next. 77 78 Also, the vpxenc application will now use a keyframe interval of 79 5 seconds by default. Use the --kf-max-dist option to override. 80 81 - Speed: 82 Decoder performance improved 2.5% versus Duclair. Encoder speed is 83 consistent with Duclair for most material. Two pass encoding of 84 slideshow-like material will see significant improvements. 85 86 Large realtime encoding speed gains at a small quality expense are 87 possible by configuring the on-the-fly bitpacking experiment with 88 --enable-onthefly-bitpacking. Realtime encoder can be up to 13% 89 faster (ARM) depending on the number of threads and bitrate 90 settings. This technique sees constant gain over the 5-16 speed 91 range. For VC style input the loss seen is up to 0.2dB. See commit 92 52cf4dca for further details. 93 94 - Quality: 95 On the whole, quality is consistent with the Duclair release. Some 96 tweaks: 97 98 Reduced blockiness in easy sections by applying a penalty to 99 intra modes. 100 101 Improved quality of static sections (like slideshows) with 102 two pass encoding. 103 104 Improved keyframe sizing with multiple temporal layers 105 106 - Bug Fixes: 107 Corrected alt-ref contribution to frame rate for visible updates 108 to the alt-ref buffer. This affected applications making manual 109 usage of the frame reference flags, or temporal layers. 110 111 Additional constraints were added to disable multi-frame quality 112 enhancement (MFQE) in sections of the frame where there is motion. 113 (#392) 114 115 Fixed corruption issues when vpx_codec_enc_config_set() was called 116 with spatial resampling enabled. 117 118 Fixed a decoder error introduced in Duclair where the segmentation 119 map was not being reinitialized on keyframes (#378) 120 121 1222012-01-27 v1.0.0 "Duclair" 123 Our fourth named release, focused on performance and features related to 124 real-time encoding. It also fixes a decoder crash bug introduced in 125 v0.9.7, so all users of that release are encouraged to upgrade. 126 127 - Upgrading: 128 This release is ABI incompatible with prior releases of libvpx, so the 129 "major" version number has been bumped to 1. You must recompile your 130 applications against the latest version of the libvpx headers. The 131 API remains compatible, and this should not require code changes in most 132 applications. 133 134 - Enhancements: 135 This release introduces several substantial new features to the encoder, 136 of particular interest to real time streaming applications. 137 138 Temporal scalability allows the encoder to produce a stream that can 139 be decimated to different frame rates, with independent rate targetting 140 for each substream. 141 142 Multiframe quality enhancement postprocessing can make visual quality 143 more consistent in the presence of frames that are substantially 144 different quality than the surrounding frames, as in the temporal 145 scalability case and in some forced keyframe scenarios. 146 147 Multiple-resolution encoding support allows the encoding of the 148 same content at different resolutions faster than encoding them 149 separately. 150 151 - Speed: 152 Optimization targets for this release included the decoder and the real- 153 time modes of the encoder. Decoder speed on x86 has improved 10.5% with 154 this release. Encoder improvements followed a curve where speeds 1-3 155 improved 4.0%-1.5%, speeds 4-8 improved <1%, and speeds 9-16 improved 156 1.5% to 10.5%, respectively. "Best" mode speed is consistent with the 157 Cayuga release. 158 159 - Quality: 160 Encoder quality in the single stream case is consistent with the Cayuga 161 release. 162 163 - Bug Fixes: 164 This release fixes an OOB read decoder crash bug present in v0.9.7 165 related to the clamping of motion vectors in SPLITMV blocks. This 166 behavior could be triggered by corrupt input or by starting 167 decoding from a P-frame. 168 169 1702011-08-15 v0.9.7-p1 "Cayuga" patch 1 171 This is an incremental bugfix release against Cayuga. All users of that 172 release are strongly encouraged to upgrade. 173 174 - Fix potential OOB reads (cdae03a) 175 176 An unbounded out of bounds read was discovered when the 177 decoder was requested to perform error concealment (new in 178 Cayuga) given a frame with corrupt partition sizes. 179 180 A bounded out of bounds read was discovered affecting all 181 versions of libvpx. Given an multipartition input frame that 182 is truncated between the mode/mv partition and the first 183 residiual paritition (in the block of partition offsets), up 184 to 3 extra bytes could have been read from the source buffer. 185 The code will not take any action regardless of the contents 186 of these undefined bytes, as the truncated buffer is detected 187 immediately following the read based on the calculated 188 starting position of the coefficient partition. 189 190 - Fix potential error concealment crash when the very first frame 191 is missing or corrupt (a609be5) 192 193 - Fix significant artifacts in error concealment (a4c2211, 99d870a) 194 195 - Revert 1-pass CBR rate control changes (e961317) 196 Further testing showed this change produced undesirable visual 197 artifacts, rolling back for now. 198 199 2002011-08-02 v0.9.7 "Cayuga" 201 Our third named release, focused on a faster, higher quality, encoder. 202 203 - Upgrading: 204 This release is backwards compatible with Aylesbury (v0.9.5) and 205 Bali (v0.9.6). Users of older releases should refer to the Upgrading 206 notes in this document for that release. 207 208 - Enhancements: 209 Stereo 3D format support for vpxenc 210 Runtime detection of available processor cores. 211 Allow specifying --end-usage by enum name 212 vpxdec: test for frame corruption 213 vpxenc: add quantizer histogram display 214 vpxenc: add rate histogram display 215 Set VPX_FRAME_IS_DROPPABLE 216 update configure for ios sdk 4.3 217 Avoid text relocations in ARM vp8 decoder 218 Generate a vpx.pc file for pkg-config. 219 New ways of passing encoded data between encoder and decoder. 220 221 - Speed: 222 This release includes across-the-board speed improvements to the 223 encoder. On x86, these measure at approximately 11.5% in Best mode, 224 21.5% in Good mode (speed 0), and 22.5% in Realtime mode (speed 6). 225 On ARM Cortex A9 with Neon extensions, real-time encoding of video 226 telephony content is 35% faster than Bali on single core and 48% 227 faster on multi-core. On the NVidia Tegra2 platform, real time 228 encoding is 40% faster than Bali. 229 230 Decoder speed was not a priority for this release, but improved 231 approximately 8.4% on x86. 232 233 Reduce motion vector search on alt-ref frame. 234 Encoder loopfilter running in its own thread 235 Reworked loopfilter to precalculate more parameters 236 SSE2/SSSE3 optimizations for build_predictors_mbuv{,_s}(). 237 Make hor UV predict ~2x faster (73 vs 132 cycles) using SSSE3. 238 Removed redundant checks 239 Reduced structure sizes 240 utilize preload in ARMv6 MC/LPF/Copy routines 241 ARM optimized quantization, dfct, variance, subtract 242 Increase chrow row alignment to 16 bytes. 243 disable trellis optimization for first pass 244 Write SSSE3 sub-pixel filter function 245 Improve SSE2 half-pixel filter funtions 246 Add vp8_sub_pixel_variance16x8_ssse3 function 247 Reduce unnecessary distortion computation 248 Use diamond search to replace full search 249 Preload reference area in sub-pixel motion search (real-time mode) 250 251 - Quality: 252 This release focused primarily on one-pass use cases, including 253 video conferencing. Low latency data rate control was significantly 254 improved, improving streamability over bandwidth constrained links. 255 Added support for error concealment, allowing frames to maintain 256 visual quality in the presence of substantial packet loss. 257 258 Add rc_max_intra_bitrate_pct control 259 Limit size of initial keyframe in one-pass. 260 Improve framerate adaptation 261 Improved 1-pass CBR rate control 262 Improved KF insertion after fades to still. 263 Improved key frame detection. 264 Improved activity masking (lower PSNR impact for same SSIM boost) 265 Improved interaction between GF and ARFs 266 Adding error-concealment to the decoder. 267 Adding support for independent partitions 268 Adjusted rate-distortion constants 269 270 271 - Bug Fixes: 272 Removed firstpass motion map 273 Fix parallel make install 274 Fix multithreaded encoding for 1 MB wide frame 275 Fixed iwalsh_neon build problems with RVDS4.1 276 Fix semaphore emulation, spin-wait intrinsics on Windows 277 Fix build with xcode4 and simplify GLOBAL. 278 Mark ARM asm objects as allowing a non-executable stack. 279 Fix vpxenc encoding incorrect webm file header on big endian 280 281 2822011-03-07 v0.9.6 "Bali" 283 Our second named release, focused on a faster, higher quality, encoder. 284 285 - Upgrading: 286 This release is backwards compatible with Aylesbury (v0.9.5). Users 287 of older releases should refer to the Upgrading notes in this 288 document for that release. 289 290 - Enhancements: 291 vpxenc --psnr shows a summary when encode completes 292 --tune=ssim option to enable activity masking 293 improved postproc visualizations for development 294 updated support for Apple iOS to SDK 4.2 295 query decoder to determine which reference frames were updated 296 implemented error tracking in the decoder 297 fix pipe support on windows 298 299 - Speed: 300 Primary focus was on good quality mode, speed 0. Average improvement 301 on x86 about 40%, up to 100% on user-generated content at that speed. 302 Best quality mode speed improved 35%, and realtime speed 10-20%. This 303 release also saw significant improvement in realtime encoding speed 304 on ARM platforms. 305 306 Improved encoder threading 307 Dont pick encoder filter level when loopfilter is disabled. 308 Avoid double copying of key frames into alt and golden buffer 309 FDCT optimizations. 310 x86 sse2 temporal filter 311 SSSE3 version of fast quantizer 312 vp8_rd_pick_best_mbsegmentation code restructure 313 Adjusted breakout RD for SPLITMV 314 Changed segmentation check order 315 Improved rd_pick_intra4x4block 316 Adds armv6 optimized variance calculation 317 ARMv6 optimized sad16x16 318 ARMv6 optimized half pixel variance calculations 319 Full search SAD function optimization in SSE4.1 320 Improve MV prediction accuracy to achieve performance gain 321 Improve MV prediction in vp8_pick_inter_mode() for speed>3 322 323 - Quality: 324 Best quality mode improved PSNR 6.3%, and SSIM 6.1%. This release 325 also includes support for "activity masking," which greatly improves 326 SSIM at the expense of PSNR. For now, this feature is available with 327 the --tune=ssim option. Further experimentation in this area 328 is ongoing. This release also introduces a new rate control mode 329 called "CQ," which changes the allocation of bits within a clip to 330 the sections where they will have the most visual impact. 331 332 Tuning for the more exact quantizer. 333 Relax rate control for last few frames 334 CQ Mode 335 Limit key frame quantizer for forced key frames. 336 KF/GF Pulsing 337 Add simple version of activity masking. 338 make rdmult adaptive for intra in quantizer RDO 339 cap the best quantizer for 2nd order DC 340 change the threshold of DC check for encode breakout 341 342 - Bug Fixes: 343 Fix crash on Sparc Solaris. 344 Fix counter of fixed keyframe distance 345 ARNR filter pointer update bug fix 346 Fixed use of motion percentage in KF/GF group calc 347 Changed condition for using RD in Intra Mode 348 Fix encoder real-time only configuration. 349 Fix ARM encoder crash with multiple token partitions 350 Fixed bug first cluster timecode of webm file is wrong. 351 Fixed various encoder bugs with odd-sized images 352 vp8e_get_preview fixed when spatial resampling enabled 353 quantizer: fix assertion in fast quantizer path 354 Allocate source buffers to be multiples of 16 355 Fix for manual Golden frame frequency 356 Fix drastic undershoot in long form content 357 358 3592010-10-28 v0.9.5 "Aylesbury" 360 Our first named release, focused on a faster decoder, and a better encoder. 361 362 - Upgrading: 363 This release incorporates backwards-incompatible changes to the 364 ivfenc and ivfdec tools. These tools are now called vpxenc and vpxdec. 365 366 vpxdec 367 * the -q (quiet) option has been removed, and replaced with 368 -v (verbose). the output is quiet by default. Use -v to see 369 the version number of the binary. 370 371 * The default behavior is now to write output to a single file 372 instead of individual frames. The -y option has been removed. 373 Y4M output is the default. 374 375 * For raw I420/YV12 output instead of Y4M, the --i420 or --yv12 376 options must be specified. 377 378 $ ivfdec -o OUTPUT INPUT 379 $ vpxdec --i420 -o OUTPUT INPUT 380 381 * If an output file is not specified, the default is to write 382 Y4M to stdout. This makes piping more natural. 383 384 $ ivfdec -y -o - INPUT | ... 385 $ vpxdec INPUT | ... 386 387 * The output file has additional flexibility for formatting the 388 filename. It supports escape characters for constructing a 389 filename from the width, height, and sequence number. This 390 replaces the -p option. To get the equivalent: 391 392 $ ivfdec -p frame INPUT 393 $ vpxdec --i420 -o frame-%wx%h-%4.i420 INPUT 394 395 vpxenc 396 * The output file must be specified with -o, rather than as the 397 last argument. 398 399 $ ivfenc <options> INPUT OUTPUT 400 $ vpxenc <options> -o OUTPUT INPUT 401 402 * The output defaults to webm. To get IVF output, use the --ivf 403 option. 404 405 $ ivfenc <options> INPUT OUTPUT.ivf 406 $ vpxenc <options> -o OUTPUT.ivf --ivf INPUT 407 408 409 - Enhancements: 410 ivfenc and ivfdec have been renamed to vpxenc, vpxdec. 411 vpxdec supports .webm input 412 vpxdec writes .y4m by default 413 vpxenc writes .webm output by default 414 vpxenc --psnr now shows the average/overall PSNR at the end 415 ARM platforms now support runtime cpu detection 416 vpxdec visualizations added for motion vectors, block modes, references 417 vpxdec now silent by default 418 vpxdec --progress shows frame-by-frame timing information 419 vpxenc supports the distinction between --fps and --timebase 420 NASM is now a supported assembler 421 configure: enable PIC for shared libs by default 422 configure: add --enable-small 423 configure: support for ppc32-linux-gcc 424 configure: support for sparc-solaris-gcc 425 426 - Bugs: 427 Improve handling of invalid frames 428 Fix valgrind errors in the NEON loop filters. 429 Fix loopfilter delta zero transitions 430 Fix valgrind errors in vp8_sixtap_predict8x4_armv6(). 431 Build fixes for darwin-icc 432 433 - Speed: 434 20-40% (average 28%) improvement in libvpx decoder speed, 435 including: 436 Rewrite vp8_short_walsh4x4_sse2() 437 Optimizations on the loopfilters. 438 Miscellaneous improvements for Atom 439 Add 4-tap version of 2nd-pass ARMv6 MC filter. 440 Improved multithread utilization 441 Better instruction choices on x86 442 reorder data to use wider instructions 443 Update NEON wide idcts 444 Make block access to frame buffer sequential 445 Improved subset block search 446 Bilinear subpixel optimizations for ssse3. 447 Decrease memory footprint 448 449 Encoder speed improvements (percentage gain not measured): 450 Skip unnecessary search of identical frames 451 Add SSE2 subtract functions 452 Improve bounds checking in vp8_diamond_search_sadx4() 453 Added vp8_fast_quantize_b_sse2 454 455 - Quality: 456 Over 7% overall PSNR improvement (6.3% SSIM) in "best" quality 457 encoding mode, and up to 60% improvement on very noisy, still 458 or slow moving source video 459 460 Motion compensated temporal filter for Alt-Ref Noise Reduction 461 Improved use of trellis quantization on 2nd order Y blocks 462 Tune effect of motion on KF/GF boost in two pass 463 Allow coefficient optimization for good quality speed 0. 464 Improved control of active min quantizer for two pass. 465 Enable ARFs for non-lagged compress 466 4672010-09-02 v0.9.2 468 - Enhancements: 469 Disable frame dropping by default 470 Improved multithreaded performance 471 Improved Force Key Frame Behaviour 472 Increased rate control buffer level precision 473 Fix bug in 1st pass motion compensation 474 ivfenc: correct fixed kf interval, --disable-kf 475 - Speed: 476 Changed above and left context data layout 477 Rework idct calling structure. 478 Removed unnecessary MB_MODE_INFO copies 479 x86: SSSE3 sixtap prediction 480 Reworked IDCT to include reconstruction (add) step 481 Swap alt/gold/new/last frame buffer ptrs instead of copying. 482 Improve SSE2 loopfilter functions 483 Change bitreader to use a larger window. 484 Avoid loopfilter reinitialization when possible 485 - Quality: 486 Normalize quantizer's zero bin and rounding factors 487 Add trellis quantization. 488 Make the quantizer exact. 489 Updates to ARNR filtering algorithm 490 Fix breakout thresh computation for golden & AltRef frames 491 Redo the forward 4x4 dct 492 Improve the accuracy of forward walsh-hadamard transform 493 Further adjustment of RD behaviour with Q and Zbin. 494 - Build System: 495 Allow linking of libs built with MinGW to MSVC 496 Fix target auto-detection on mingw32 497 Allow --cpu= to work for x86. 498 configure: pass original arguments through to make dist 499 Fix builds without runtime CPU detection 500 msvs: fix install of codec sources 501 msvs: Change devenv.com command line for better msys support 502 msvs: Add vs9 targets. 503 Add x86_64-linux-icc target 504 - Bugs: 505 Potential crashes on older MinGW builds 506 Fix two-pass framrate for Y4M input. 507 Fixed simple loop filter, other crashes on ARM v6 508 arm: fix missing dependency with --enable-shared 509 configure: support directories containing .o 510 Replace pinsrw (SSE) with MMX instructions 511 apple: include proper mach primatives 512 Fixed rate control bug with long key frame interval. 513 Fix DSO link errors on x86-64 when not using a version script 514 Fixed buffer selection for UV in AltRef filtering 515 516 5172010-06-17 v0.9.1 518 - Enhancements: 519 * ivfenc/ivfdec now support YUV4MPEG2 input and pipe I/O 520 * Speed optimizations 521 - Bugfixes: 522 * Rate control 523 * Prevent out-of-bounds accesses on invalid data 524 - Build system updates: 525 * Detect toolchain to be used automatically for native builds 526 * Support building shared libraries 527 * Better autotools emulation (--prefix, --libdir, DESTDIR) 528 - Updated LICENSE 529 * http://webmproject.blogspot.com/2010/06/changes-to-webm-open-source-license.html 530 531 5322010-05-18 v0.9.0 533 - Initial open source release. Welcome to WebM and VP8! 534 535