Corrupted accumulation in reduce_metadata_further_fx

Basic info

• Float reference:
  • Encoder (float): 2a32cceaf0d7ce3482cb05e5c12951e95a8a552a
  • Decoder (float):
• Fixed point:
  • Encoder (fixed): a8ff39cd
  • Decoder (fixed):

Bug description

A temp variable is not reset properly in reduce_metadata_further_fx which causes wrong accumulation and corruption. See below:

Float

for ( band = 0; band < numCodingBands; band++ )
{
    xSum = 0.0f;
    ySum = 0.0f;
    zSum = 0.0f;
    bandSumEnergy = 0.0f;

    for ( sf = 0; sf < MAX_PARAM_SPATIAL_SUBFRAMES; sf++ )
    {
        bandSumEnergy += energy[sf][band];
    }
    spreadCoh[0] = (uint8_t) roundf( spreadCohSum / ( bandSumEnergy + EPSILON ) * 255.0f );
}

BASOP

W_tmp = 0; <== BUG as reset does not happen properly
move64();
FOR( band = 0; band < numCodingBands; band++ )
{
    xSum = 0; ySum = 0; zSum = 0;
    bandSumEnergy = 0;
    move32(); move32(); move32(); move32();

    FOR( sf = 0; sf < MAX_PARAM_SPATIAL_SUBFRAMES; sf++ )
    {
        ...
        W_tmp = W_add( W_tmp, energy[sf][band] );
    }

    tmp2 = W_norm( W_tmp );
    tmp2 = s_min( 32, tmp2 );
    bandSumEnergy = W_extract_h( W_shl( W_tmp, tmp2 ) );
    ...
    W_xSum_sq = W_mult0_32_32( xSum, xSum );
    W_ySum_sq = W_mult0_32_32( ySum, ySum );
    W_tmp = W_add( W_xSum_sq, W_ySum_sq );
    ...
    tmp = BASOP_Util_Divide3232_Scale( spreadCohSum, bandSumEnergy, &exp );
}