Changes for Version 8.5 relative to Version 8.4
-----------------------------------------------

* reformat software to comply with coding style (KS)

* convert comments to documentation comments (KS)

* Bug fix: bit counter (DM,KS)

* Bug fix: qp=0 not accepted (GS)

* Bug fix: mismatch if different frame qp settings (JL,KS)

* memory management cleanups (KS)
  - fix some memory leaks
  - check returned pointer after _every_ malloc/calloc

* use size checking string functions (KS)
  - replace sprintf and strcpy with snprintf and strncpy

* clean up exit calls (KS)
  - use common error() function for all error exits


Changes for Version 8.4 relative to Version 8.0
-----------------------------------------------

* Datapartitioning according to VCEG-M52 supported (TO)

* Output in InterimFileFormat supported (TO)

* Bugfixes for SliceMode 2 (fixed Bytes/MB) (TO)
  - adaptions for RLC of SkipMBs 

* Changed initialisation values for CABAC contexts (TO)

* PictureHeader and SliceHeader at the beginning of every slice (TO)

* RunLength and constrained intra prediction (RS)

* Implementation of delta-QP (IL)

* loopfilter simplification including a macroblock-based approach (PL)

* Removal of RD Quant (IL)

* Chroma coefficient removal (IL)


Changes for Version 8.0 relative to Version 6.1
-----------------------------------------------

* Fast Integer Pixel Motion Estimation (HS)
  - precalculation of SAD for all 4x4 blocks and all vector positions
  - SAD calculation of nxm blocks is based on 4x4 SAD's
  - search range center is the predictor of 16x16 block for all block sizes

* Extended Search range (HS, SW)
  - search range can be 39x39 at maximum
  - search range reductions for older reference frames or older reference frames
    and smaller blocks can be enabled in the configuration file

* Unrestricted Motion Vectors (HS, SW)
  - Unrestricted motion vectors for P- and B-frames
  - no range checking in Direct mode

* Adaption of P-Interval and Single Quantizer Change (primarily for MPEG-test) (HS)
  - the interval P-P can be adapted at the very end of a sequence to hit the last
    frame which then have to be specified in the configuration file
  - the quantization parameters can be changed once per sequence (config file)

* Rate-Distortion Optimization (HS, TW)
  - a rd-optimized mode can be enabled in the configuration file
  - in rd-opt. mode:
    - slightly modified rate-constrained motion estimation
    - rate-constrained macroblock mode decision
    - rate-constrained mode decision for 4x4 intra modes
  - the rd-opt. mode decision need to code (DCT, Q, Q^-1, IDCT, entopy coding)
    each block/macroblock several times

* Increased Frame Buffer (HS, TW)
  - the frame buffer for long term memory prediction is no longer restricted to
    five reference frames
  - the buffer size in the encoder will be adapted according to the number of reference
    frames given (config file) and the config file parameter "AdditionalReferenceFrame"
  - the buffer size in the decoder has to be specified in the config file
  - the parameter "AdditionalReferenceFrame" specifies a reference frame index which is
    considered during motion estimation
  - E.g:  NumberReferenceFrames=5, AdditionalReferenceFrame=19
    -> 20 reference frames are stored at encoder side
    -> the frames with indizes 0,1,2,3,4, and 19 are searched during motion estimation

* NAL/Slices (TO)
  - support of slices in both symbol modes
  - at the decoder UVLC symbols are decoded by the time they are needed, not for the whole 
    slice at one time (to be in line with the encoder)
  

* Added Motion Compensated Prediction with 1/8-pel MV-resolution: (TWe)
  - New Parameter in CFG-file to switch between 1/4- and 1/8-pel
  - New interplation filter for 1/8-pel MV-resolution added.
    a) reference frames are sampled up and stpred by a factor of 4 (1/4-pel resolution)
    b) 1/8-pel postitions are calculated on the fly by bilinear interolation
    c) all chroma values are interpolated with a bilinear filter
    d) items b) and c) lead to a slightly reduced gain compared to implementation
     in M20, but still a gain up to 1.0 dB is obtained comared to TML591. 
  Thus, further improvements are possible.

* General (all / KS)
  - Some clean-ups of data structures
  - removal of superfluous source and include files in the decoder

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
People:
KS      Karsten Shring         suehring@hhi.de

DM      Detlev Marpe            marpe@hhi.de
GB      Gabi Blttermann        blaetter@hhi.de
GS      Gary Sullivan		garysull@microsoft.com
HS      Heiko Schwarz           hschwarz@hhi.de
IL      Inge Lille-Langoy       Inge.Lille-Langoy@telenor.com
JL      Jani Lainema            Jani.Lainema@nokia.com
PL      Peter List              Peter.List@t-systems.de
RS      Rickard Sjberg         Rickard.Sjoberg@era.ericsson.se
SW      Stephan Wenger          stewe@cs.tu-berlin.de
TO      Tobias Oelbaum          oelbaum@hhi.de
TW      Thomas Wiegand          wiegand@hhi.de
TWe     Thomas Wedi             wedi@tnt.uni-hannover.de
