Draft: [NonBE] FIX 2492 unifies decode_indexes in LSF

  • Related issues: #2492
  • Requested reviewers:

Reason why this change is needed

There exist 2 overlapping functions decode_indexes_ivas_fx() and decode_indexes_fx(). Within these functions there are also several table addresses computations that need to be made with BASOP ops.

Also at least the related variables table_no_cv and table_no_cv_fx are duplicated.

Description of the change

This change evolved from decode_indexes_fx and decode_indexes_ivas_fx are unified into decode_indexes_fx

to

EVS and IVAS versions of LSF quantization functions are unified, to use a single version as it is in the IVAS floating code.

The offset_scales input parameters have been removed (link to #2303 ): It saves 9,892 bytes of heap RAM per transport channel, both at the encoder and the decoder.

There were differences in rounding to integer between IVAS and EVS scales and sigma values that made the change nonBE for all case where ACELP is used.

Affected operating points

  • Describe here as well as possible what operating points are affected and how
  • In minimum, there should be a status for bitstream compatibility and output bit exactness
  • For bitstream compatibility, the following levels are helpful for describing encoder
    1. Produced bitstream is BE compared to previous state only for EVS modes
    2. Produced bitstream is nonBE with previous modes, but bitstream compatible.
Edited by vaclav

Merge request reports

Loading