Resolve "USAN: value out of range for int16 in FEC_HQ_phase_ecu.c"

• Requested reviewers: @kiene

Reason why this change is needed

• "USAN: value out of range for cast to int16_t from float(24bit mantissa) in FEC_HQ_phase_ecu.c"
• Slightly incorrect extraction of fractional phase in C-Float, somewhat unsafe cast
• Incorrect use of modulo "%" in extraction of fractional phase in C-Float L1095

Description of the change

ALT1 (active, proposed)

• added cast to signed 32 bit from float(24 bit mantissa) in the phase dithering section
• replaced WMC-costly "%32768" with L_and(0x000003ff ) in the regular peak evolution section
• new cost approx. ~4 cycles

ALT2 (not active, not proposed)

• range control(within a 16 bit signed integer) and rounding, not BE, but within 1 table index in the 1024 resolution sine/cosine table , i.e. a 1 corresponds to a phase change of (2*pi)/1024
• This C-float solution costs about 6 cycles(with rounding), compared to the EVS-BASOP 3-4 cycles extraction,
• not BE with previous IVAS code, not BE with EVS-MONO

misc:

• the USAN reported error L1139 was occurring in a region where the phase was scrambled anyway. The undefined casting was not an actual problem in MSVC simulations
• The inactive ALT2 the corrected extractions with rounding of the fraction for line 1095 improves precision and would guarantee that there is no overflow in the cast to int32_t (L1095), nor in the cast to int16_t(L1139)
• the BE ALT1 is proposed due to avoiding the extra cycles and the added branching cost of checking for EVS_MONO in-loop

Affected operating points

All Phase ECU PLC conditions.

1. (ALT1) BE

( 2. (ALT2) Produced bitstream is non-BE but it is fully backwards compatible for decoding. Decoded output may differ. )

Related to issue #1002 (closed)