libmad - MPEG audio decoder library Copyright (C) 2000-2004 Underbit Technologies, Inc. $Id: CHANGES,v 1.14 2004/02/17 02:02:03 rob Exp $ =============================================================================== Version 0.15.1 (beta) * Updated to autoconf 2.59, automake 1.8.2, libtool 1.5.2. * Replaced Layer III IMDCT routine with one based on a faster algorithm, improving both speed and accuracy. * Improved portability of the Huffman table initialization. * Fixed a problem that could result in an assertion failure in layer3.c due to an invalid Layer III free format bitrate. * Improved the robustness of Layer II bitrate/mode combinations, and added a new MAD_ERROR_BADMODE error enum. The allowability of low-bitrate stereo streams is influenced by the --enable-strict-iso option to `configure'. Version 0.15.0 (beta) * Updated to autoconf 2.57, automake 1.7.5, libtool 1.4.3. * Added new mad_f_div() API routine. * Added a 64th entry to the Layer I/Layer II scalefactor table, for better compatibility with existing streams. The --enable-strict-iso option to `configure' can be used to disable use of this entry. * Modified the header decoding routine to allow the reserved emphasis value, for better compatibility with existing streams. The --enable-strict-iso option to `configure' can be used to restore the previous behavior of reporting this value as an error. * Added new MAD_EMPHASIS_RESERVED enumeration constant. * Fixed a bug in the ARM version of mad_f_scale64() discovered by Andre McCurdy. * Rewrote PowerPC assembly for minor gains. * Modified mad_timer_fraction() to avoid the possibility of division by zero when 0 is passed as the second argument. * Fixed a non-fatal problem caused by attempting to designate ancillary bits in Layer III after a decoding error. * Changed to build a shared library by default. * Changed to use native Cygwin build by default; give --host=mingw32 to `configure' to use MinGW (and avoid a dependency on the Cygwin DLL). Version 0.14.2 (beta) * Changed Cygwin builds to use MinGW; resulting Win32 executables no longer have a dependency on Cygwin DLLs. * Added a new mad_stream_errorstr() API function to libmad for retrieving a string description of the current error condition. Version 0.14.1 (beta) * Updated config.guess and config.sub to latest upstream versions. * Enabled libtool versioning rather than release numbering. * Improved the documentation in minimad.c. * Several other small fixes. Version 0.14.0 (beta) * Added a 64-bit FPM negation operation to improve performance of subband synthesis on some platforms. * Improved MSVC++ portability and added MSVC++ project files. * Added rounding to Layer III requantization for slightly better accuracy. Version 0.13.0 (beta) * Ancillary data is now properly extracted from Layer III streams. * Rewrote the Layer III joint stereo decoding routine to correct a major MPEG-2 problem and a minor MPEG-1 problem decoding intensity stereo. * Eliminated the dependency on sign-extending right shifts for Layer I and Layer II. * Renamed `private' field to `private_bits' for better C++ compatibility. * Gratuitously renamed `sfreq' field to `samplerate' and MAD_ERROR_BADSAMPLEFREQ constant to MAD_ERROR_BADSAMPLERATE. * Added `samplerate' and `channels' fields to synth.pcm struct to allow these to be different from the decoded frame, and for simpler access. * Added new mad_stream_options() and mad_decoder_options() API entries for special runtime decoding options. * Added new MAD_OPTION_IGNORECRC and MAD_OPTION_HALFSAMPLERATE options. * Added new MAD_FLAG_FREEFORMAT indicator flag. * Fixed some bugs in the async decoder. * Added a new mad_timer_multiply() API routine. * Eliminated `+' from asm constraints under Intel for better compatibility with some compilers. * Fixed a PIC-related problem in imdct_l_arm.S. * Eliminated a static variable to make libmad thread-safe. Version 0.12.5 (beta) * Modified Layer III requantization to occur during Huffman decoding for significant performance gains. * Optimized short block IMDCT by eliminating redundant calculations. * Made several other Layer III performance improvements; added ASO_INTERLEAVE1, ASO_INTERLEAVE2, and ASO_ZEROCHECK architecture-specific options for best performance on various architectures. * Optimized synthesis DCT to store result values as soon as they are calculated. Version 0.12.4 (beta) * New PowerPC fixed-point assembly courtesy of David Blythe. * Reorganized fixed-point assembly routines for easier maintenance and better performance. * Improved performance of subband synthesis through better indexing and fewer local variables. * Added alias reduction for the lower two subbands of mixed short blocks, per a report of ambiguity with ISO/IEC 11172-3 and for uniformity with most other implementations. Also improved alias reduction performance using multiply/accumulate. * Added --enable-strict-iso option to `configure' to override best accepted practices such as the alias reduction for mixed short blocks. * Improved performance of Layer III IMDCT by using longer multiply/accumulate runs where possible. Version 0.12.3 (beta) * Added MPEG 2.5 support. * Added preliminary support for parameterizing the binary point position in the fixed-point representation. * Added multiply/accumulate optimization to the Layer III IMDCT for long blocks. * Fixed a bug in the handling of Layer III mixed_block_flag. * Fixed a configure problem when multiple -O CFLAGS are present. Version 0.12.2 (beta) * Rearranged the synthesis polyphase filterbank memory vector for better locality of reference, and rewrote mad_synth_frame() to accommodate, resulting in improved performance. * Discovered a combination of compiler optimization flags that further improve performance. * Changed some array references in layer3.c to pointer derefs. Version 0.12.1 (beta) * Resolved the intensity + MS joint stereo issue (a simple bug). OPT_ISKLUGE is no longer considered to be a kluge. * Fixed another, hopefully last main_data memory bug. * Split part of struct mad_frame into struct mad_header for convenience and size. Version 0.12.0 (alpha) * Changed the build environment to use automake and libtool. A libmad shared library can now be built using the --enable-shared option to `configure'. * Added another callback to MAD's high-level decoder API after the frame header has been read but before the frame's audio data is decoded. * Streamlined header processing so that mad_frame_decode() can be called with or without having already called mad_frame_header(). * Fixed some other header reading miscellany, including CRC handling and free bitrate detection, and frame length verification with free bitrates. * Fixed a problem with Layer III free bitrates > 320 kbps. The main_data buffer size should now be large enough to handle any size frame, by virtue of the maximum possible part2_3_length. * Further developed the async API; arbitrary messages can now be passed to the subsidiary decoding process. * Streamlined timer.c and extended its interface. It now has support for video frame/field lengths, including output support for drop-frame encoding. * Replaced many constant integer preprocessor defines with enums. Version 0.11.4 (beta) * Fixed free format bitrate discovery. * Changed the timer implementation and extended its interface. * Integrated Nicolas Pitre's patch for pre-shifting at compile-time and for better multiply/accumulate code output. * Applied Simon Burge's patch to imdct_l_arm.S for a.out compatibility. * Added -mtune=strongarm for all ARM targets. Version 0.11.3 (beta) * Added new --enable-speed and --enable-accuracy options for `configure' to automatically select appropriate SSO/ASO options, et al. * Modified subband synthesis to use multiply/accumulate optimization (if available) for better speed and/or accuracy. * Incorporated Andre McCurdy's changes for further rounding optimizations in the rest of his code. Version 0.11.2 (beta) * Incorporated Nicolas Pitre's ARM assembly and parameterized scaling changes. * Incorporated Andre McCurdy's ARM assembly optimization (used only if --enable-aso is given to `configure' to enable architecture-specific optimizations.) * Reduced FPM_INTEL assembly to two instructions. * Fixed accuracy problems with certain FPM modes in synth.c. * Improved the accuracy of FPM_APPROX. * Improved the accuracy of SSO. * Improved sync discovery by checking for a sync word in the following frame. * Minor code clean-up. * Added experimental rules for generating a libmad.so shared library. Version 0.11.1 (beta) * Moved libmad code into a separate directory. * Changed SSO to be disabled by default, as output accuracy is deemed to be more important than speed in the general case. * Fixed a bug in Layer III sanity checking that could cause a crash on certain random data input. * Extended the Layer III requantization table from 8191 to 8206 as some encoders are known to use these values, even though ISO/IEC 11172-3 suggests the maximum should be 8191. Version 0.11.0 (beta) * Implemented MPEG-2 extension to Lower Sampling Frequencies. * Improved Layer III performance by avoiding IMDCT calculation when all input samples are zero. * Significantly reduced size of Layer II tables. Version 0.10.3 (beta) * Improved SSO output quality. * Made portable to cygwin. * Localized memory references in III_huffdecode() for better performance. Version 0.10.2 (beta) * Rewrote Layer III long block 36-point IMDCT routine for better performance. * Improved subband synthesis fixed-point games somewhat. Version 0.10.1 (beta) * Added a subband synthesis optimization (SSO) which involves modifying the fixed-point multiplication method during windowing. This produces subtle differences in the output but improves performance greatly. * Added I_STEREO and MS_STEREO flags to frame struct. * Eliminated privately-used CRCFAILED flag. * Fixed a bug where Layer III decoding could crash on some badly-formatted (e.g. non-MPEG) bitstreams. * Miscellaneous code clean-up. Version 0.10.0 (beta) * Added SPARC fixed-point math support. * Revamped libmad API for better high- and low-level support. * Documented more of the code. * Changed sync semantics such that new stream buffers are assumed to be sync-aligned. * Changed Layer III to dynamically allocate static memory so as not to waste it (about 6.4K) when only decoding Layer I or Layer II. ===============================================================================