Commit 6e315e53 authored by Sandesh Venkatesh's avatar Sandesh Venkatesh
Browse files

Merge branch '1072-complexity-high-complexity-overhead-for-osba-decoding' into 'main' 

Resolve "Complexity: High complexity overhead for OSBA Decoding"

Closes #1072

See merge request !1091
parents 9ae2ae46 c9e2364c

lib_com/ivas_dirac_com.c

+48 −0
Original line number Diff line number Diff line
@@ -967,6 +967,28 @@ void computeDiffuseness_fixed( 


        q_tmp = add( q_factor_energy[i], min_q_shift1 );





#ifdef FIX_1072_SPEEDUP_COMPUTEDIFUSENESS

        Word16 shift_q = sub( q_tmp, q_ene );

        Word32 shiftEquiv;

        Word16 shift_qtotal;

        if ( shift_q < 0 )

        {

            shiftEquiv = L_lshl( 0x80000000, shift_q );

        }

        if ( shift_q >= 0 )

        {

            shiftEquiv = L_add( 0x7FFFFFFF, 0 );

        }

        shift_qtotal = sub( min_q_shift1, s_max( shift_q, 0 ) );



        FOR( k = 0; k < num_freq_bands; k++ )

        {

            tmp = L_shl( p_tmp_c[k], shift_qtotal );

            energy_slow[k] = Madd_32_32_r( tmp, energy_slow[k], shiftEquiv );

            move32();

        }

#else

        Word16 shift_q = sub( q_tmp, q_ene );

        IF( shift_q < 0 )

        {
@@ -986,6 +1008,8 @@ void computeDiffuseness_fixed( 
                move32();

            }

        }

#endif





        q_ene = s_min( q_ene, q_tmp );
@@ -993,6 +1017,28 @@ void computeDiffuseness_fixed( 
        q_tmp = add( q_factor_intensity[i], min_q_shift2 );



        shift_q = sub( q_tmp, q_intensity );

#ifdef FIX_1072_SPEEDUP_COMPUTEDIFUSENESS

        if ( shift_q < 0 )

        {

            shiftEquiv = L_lshl( 0x80000000, shift_q );

        }

        if ( shift_q >= 0 )

        {

            shiftEquiv = L_lshl( 0x7FFFFFFF, 0 );

        }

        shift_qtotal = sub( min_q_shift2, s_max( shift_q, 0 ) );



        FOR( j = 0; j < DIRAC_NUM_DIMS; ++j )

        {

            p_tmp = buffer_intensity[j][i];

            FOR( k = 0; k < num_freq_bands; k++ )

            {

                tmp = L_shl( p_tmp[k], shift_qtotal );

                intensity_slow[j * num_freq_bands + k] = Madd_32_32_r( tmp, intensity_slow[j * num_freq_bands + k], shiftEquiv );

                move32();

            }

        }

#else

        IF( shift_q > 0 )

        {

            FOR( j = 0; j < DIRAC_NUM_DIMS; ++j )
@@ -1019,6 +1065,8 @@ void computeDiffuseness_fixed( 
                }

            }

        }

#endif



        q_intensity = s_min( q_intensity, q_tmp );

    }

lib_com/options.h

100755 → 100644
+2 −1
Original line number Diff line number Diff line
@@ -165,6 +165,7 @@ 
#define FIX_ISSUE_1247

#define NONBE_FIX_1087_OOB_SBA_DTX_RS                   /* VA: issue 1087: Extend the length of the buffer for MCT decoding to avoid out-of-bound writing in SBA SID bitrate switching decoding */

#define FIX_1285_DECODER_CRASH



#define FIX_1072_SPEEDUP_gainpanning            /* FhG: Minor WMOPS tuning, nonbe */

#define FIX_1072_SPEEDUP_COMPUTEDIFUSENESS      /* FhG: Minor WMOPS tuning, nonbe */

#define FIX_1320_LOWRATE_ACELP

#endif

lib_rend/ivas_dirac_output_synthesis_dec.c

+53 −0
Original line number Diff line number Diff line
@@ -1081,6 +1081,16 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx( 
            Word16 temp_q = sub( add( h_dirac_output_synthesis_state->direct_power_factor_q, h_dirac_output_synthesis_state->direct_responses_q ), 31 );

            IF( LT_16( temp_q, h_dirac_output_synthesis_state->q_cy_cross_dir_smooth ) )

            {

#ifdef FIX_1072_SPEEDUP_gainpanning /*is there any difference in any bitstream?*/

                Word16 temp_q1 = sub( temp_q, h_dirac_output_synthesis_state->q_cy_cross_dir_smooth );

                FOR( Word16 kk = 0; kk < tmp16; kk++ )

                {

                    h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[kk] = L_shl( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[kk], temp_q1 ); /*h_dirac_output_synthesis_state->q_cy_cross_dir_smooth ->temp_q*/

                    move32();

                }

                h_dirac_output_synthesis_state->q_cy_cross_dir_smooth = temp_q;

                move16();

#else

                FOR( Word16 kk = 0; kk < tmp16; kk++ )

                {

                    h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[kk] = L_shl( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[kk], sub( temp_q, h_dirac_output_synthesis_state->q_cy_cross_dir_smooth ) ); /*h_dirac_output_synthesis_state->q_cy_cross_dir_smooth ->temp_q*/
@@ -1088,7 +1098,49 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx( 
                }

                h_dirac_output_synthesis_state->q_cy_cross_dir_smooth = temp_q;

                move16();

#endif

            }

#ifdef FIX_1072_SPEEDUP_gainpanning

            Word16 temp_q1 = sub( h_dirac_output_synthesis_state->q_cy_cross_dir_smooth, temp_q );

            FOR( ch_idx = s_min( 4, nchan_transport ); ch_idx < num_channels_dir; ch_idx++ )

            {

                IF( NE_16( temp_q, h_dirac_output_synthesis_state->q_cy_cross_dir_smooth ) )

                {

                    Word16 i;

                    Word32 aux;

                    IF( temp_q1 < 0 )

                    {

                        Word32 temp_q1_equiv = L_lshl( 0x80000000, temp_q1 );

                        FOR( i = 0; i < num_freq_bands; i++ )

                        {

                            aux = Mpy_32_32( h_dirac_output_synthesis_state->direct_power_factor_fx[i], h_dirac_output_synthesis_state->direct_responses_fx[ch_idx * num_freq_bands + i] );

                            h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + i] = Madd_32_32( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + i], aux, temp_q1_equiv );

                            move32();

                        }

                    }

                    ELSE

                    {

                        FOR( i = 0; i < num_freq_bands; i++ )

                        {

                            aux = Mpy_32_32( h_dirac_output_synthesis_state->direct_power_factor_fx[i], h_dirac_output_synthesis_state->direct_responses_fx[ch_idx * num_freq_bands + i] );

                            aux = L_shl( aux, temp_q1 );

                            h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + i] = L_add( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + i], aux );

                            move32();

                        }

                    }

                }

                ELSE

                {

                    Word16 i;

                    FOR( i = 0; i < num_freq_bands; i++ )

                    {

                        h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + i] = Madd_32_32( h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands + i], h_dirac_output_synthesis_state->direct_power_factor_fx[i], h_dirac_output_synthesis_state->direct_responses_fx[ch_idx * num_freq_bands + i] );

                        move32();

                    }

                }

            }



#else

            FOR( ch_idx = s_min( 4, nchan_transport ); ch_idx < num_channels_dir; ch_idx++ )

            {

                v_mult_fixed( h_dirac_output_synthesis_state->direct_power_factor_fx,
@@ -1106,6 +1158,7 @@ void ivas_dirac_dec_output_synthesis_process_slot_fx( 
                             &h_dirac_output_synthesis_state->cy_cross_dir_smooth_fx[ch_idx * num_freq_bands],

                             num_freq_bands, 0 ); /*Q(h_dirac_output_synthesis_state->q_cy_cross_dir_smooth)*/

            }

#endif



            /*Diffuse gain*/

            FOR( ch_idx = s_min( 4, nchan_transport ); ch_idx < num_channels_diff; ch_idx++ )