From 67f74ae7d4354cfb68eac5ca5b4746d60694c5ed Mon Sep 17 00:00:00 2001 From: Sandesh Venkatesh Date: Wed, 28 May 2025 11:36:09 +0530 Subject: [PATCH] Fix for 3GPP issue 1504: BASOP decoder to MONO output for MASA 2TC LTV input exhibits higher MLD Link #1504 --- lib_rend/ivas_dirac_rend_fx.c | 122 ++++++++++++---------------------- lib_rend/ivas_stat_rend.h | 4 +- 2 files changed, 44 insertions(+), 82 deletions(-) diff --git a/lib_rend/ivas_dirac_rend_fx.c b/lib_rend/ivas_dirac_rend_fx.c index 063c70f66..99a84bf52 100644 --- a/lib_rend/ivas_dirac_rend_fx.c +++ b/lib_rend/ivas_dirac_rend_fx.c @@ -1759,17 +1759,19 @@ void protoSignalComputation2_fx( Word32 lr_total_bb_ratio_fx, lr_total_hi_ratio_fx; Word32 min_sum_total_ratio_fx, min_sum_total_ratio_db_fx; Word32 sum_total_ratio_fx[MASA_SUM_FREQ_RANGE_BINS]; - Word16 q_sum_total_ratio; + Word16 exp_sum_total_ratio_fx[MASA_SUM_FREQ_RANGE_BINS]; + Word16 q_sum_total_ratio = 0; + move16(); Word32 a_fx, b_fx, a2_fx, b2_fx; Word16 interpolatorSpaced_fx, interpolatorDmx_fx; Word32 tempSpaced_fx, tempDmx_fx; #ifdef FIX_867_CLDFB_NRG_SCALE - Word16 q_shift, min_q_shift[2], exp, q_temp[2], temp_q_shift, q_temp2; + Word16 q_shift, min_q_shift[2], exp, q_temp[2]; #else Word16 q_shift, min_q_shift, exp, q_temp, temp_q_shift, q_temp2; #endif Word32 temp; - Word64 W_tmp1, W_tmp2; + Word64 W_tmp1, W_tmp2, W_tmp3; Word64 reference_power_64fx[CLDFB_NO_CHANNELS_MAX]; Word16 q_reference_power_64fx; Word16 head_room, q_Left_Right_power; @@ -1786,9 +1788,7 @@ void protoSignalComputation2_fx( min_q_shift = Q31; move16(); #endif - temp_q_shift = Q31; - move16(); - q_sum_total_ratio = Q31; + q_sum_total_ratio = 0; move16(); exp = 0; move16(); @@ -1813,7 +1813,6 @@ void protoSignalComputation2_fx( min_q_shift = s_min( min_q_shift, q_shift ); #endif q_shift = s_min( L_norm_arr( RealBuffer_fx[l][0], s_min( num_freq_bands, MASA_SUM_FREQ_RANGE_BINS ) ), L_norm_arr( ImagBuffer_fx[l][0], s_min( num_freq_bands, MASA_SUM_FREQ_RANGE_BINS ) ) ); - temp_q_shift = s_min( temp_q_shift, q_shift ); } #ifdef FIX_867_CLDFB_NRG_SCALE @@ -1822,7 +1821,6 @@ void protoSignalComputation2_fx( #else min_q_shift = sub( min_q_shift, 2 ); // guard bits #endif - temp_q_shift = sub( temp_q_shift, 2 ); // guard bits /* Upscaling of the buffer proto_power_smooth_fx */ #ifdef FIX_867_CLDFB_NRG_SCALE @@ -2072,7 +2070,7 @@ void protoSignalComputation2_fx( } #ifdef FIX_867_CLDFB_NRG_SCALE - Word16 total_shift[2], q_temp_total; + Word16 total_shift[2]; /* total_shift shift required to get common Q of sum power values */ total_shift[0] = shl( s_max( 0, sub( min_q_shift[0], min_q_shift[1] ) ), 1 ); total_shift[1] = shl( s_max( 0, sub( min_q_shift[1], min_q_shift[0] ) ), 1 ); @@ -2080,13 +2078,11 @@ void protoSignalComputation2_fx( min_q_shift[1] = sub( min_q_shift[1], idiv1616( find_guarded_bits_fx( num_freq_bands ), 2 ) ); q_temp[0] = sub( add( add( q_cldfb, min_q_shift[0] ), add( q_cldfb, min_q_shift[0] ) ), 31 ); q_temp[1] = sub( add( add( q_cldfb, min_q_shift[1] ), add( q_cldfb, min_q_shift[1] ) ), 31 ); - q_temp_total = s_min( q_temp[0], q_temp[1] ); #else min_q_shift = sub( min_q_shift, idiv1616( find_guarded_bits_fx( num_freq_bands ), 2 ) ); q_temp = sub( add( add( q_cldfb, min_q_shift ), add( q_cldfb, min_q_shift ) ), 31 ); #endif - q_temp2 = sub( add( add( q_cldfb, temp_q_shift ), add( q_cldfb, temp_q_shift ) ), 31 ); head_room = 63; move16(); @@ -2195,77 +2191,50 @@ void protoSignalComputation2_fx( IF( LT_16( l, s_min( num_freq_bands, MASA_SUM_FREQ_RANGE_BINS ) ) ) { -#ifdef FIX_867_CLDFB_NRG_SCALE - re_aux = L_shl( Real_aux_fx, sub( temp_q_shift, min_q_shift[0] ) ); // q_cldfb+temp_q_shift - im_aux = L_shl( Imag_aux_fx, sub( temp_q_shift, min_q_shift[0] ) ); // q_cldfb+temp_q_shift -#else - re_aux = L_shl( Real_aux_fx, sub( temp_q_shift, min_q_shift ) ); // q_cldfb+temp_q_shift - im_aux = L_shl( Imag_aux_fx, sub( temp_q_shift, min_q_shift ) ); // q_cldfb+temp_q_shift -#endif + re_aux = L_add( L_shr( RealBuffer_fx[0][0][l], 1 ), L_shr( RealBuffer_fx[1][0][l], 1 ) ); + im_aux = L_add( L_shr( ImagBuffer_fx[0][0][l], 1 ), L_shr( ImagBuffer_fx[1][0][l], 1 ) ); - sum_power_fx = Madd_32_32( Mpy_32_32( re_aux, re_aux ), im_aux, im_aux ); // 2*(q_cldfb+temp_q_shift)-31 - temp = Mpy_32_32( a_fx, sum_power_fx ); // 2*(q_cldfb+temp_q_shift)-31 + W_tmp3 = W_add( W_mult0_32_32( re_aux, re_aux ), W_mult0_32_32( im_aux, im_aux ) ); + q_shift = W_norm( W_tmp3 ); + sum_power_fx = W_extract_h( W_shl( W_tmp3, q_shift ) ); + exp_temppp = sub( 31, sub( add( shl( sub( q_cldfb, 1 ), 1 ), q_shift ), 32 ) ); - IF( LT_16( q_temp2, stereo_type_detect->q_sum_power ) ) - { - stereo_type_detect->sum_power_fx[l] = L_add( temp, L_shr( Mpy_32_32( b_fx, stereo_type_detect->sum_power_fx[l] ), sub( stereo_type_detect->q_sum_power, q_temp2 ) ) ); // q_temp2 - move32(); - } - ELSE - { - stereo_type_detect->sum_power_fx[l] = L_add( L_shr( temp, sub( q_temp2, stereo_type_detect->q_sum_power ) ), Mpy_32_32( b_fx, stereo_type_detect->sum_power_fx[l] ) ); // stereo_type_detect->q_sum_power - move32(); - } + stereo_type_detect->sum_power_fx[l] = BASOP_Util_Add_Mant32Exp( Mpy_32_32( a_fx, sum_power_fx ), exp_temppp, Mpy_32_32( b_fx, stereo_type_detect->sum_power_fx[l] ), stereo_type_detect->exp_sum_power[l], &stereo_type_detect->exp_sum_power[l] ); + move32(); + W_tmp1 = W_add( W_mult0_32_32( RealBuffer_fx[0][0][l], RealBuffer_fx[0][0][l] ), W_mult0_32_32( ImagBuffer_fx[0][0][l], ImagBuffer_fx[0][0][l] ) ); + W_tmp2 = W_add( W_mult0_32_32( RealBuffer_fx[1][0][l], RealBuffer_fx[1][0][l] ), W_mult0_32_32( ImagBuffer_fx[1][0][l], ImagBuffer_fx[1][0][l] ) ); -#ifdef FIX_867_CLDFB_NRG_SCALE - temp = Mpy_32_32( a_fx, W_extract_l( W_shr( reference_power_64fx[l], add( 31, total_shift[qidx] ) ) ) ); // 2*(q_cldfb+min_q_shift) -31 - IF( LT_16( q_temp_total, stereo_type_detect->q_total_power ) ) - { - stereo_type_detect->total_power_fx[l] = L_add( temp, L_shr( Mpy_32_32( b_fx, stereo_type_detect->total_power_fx[l] ), sub( stereo_type_detect->q_total_power, q_temp_total ) ) ); // q_temp - move32(); - } - ELSE - { - stereo_type_detect->total_power_fx[l] = L_add( L_shr( temp, sub( q_temp_total, stereo_type_detect->q_total_power ) ), Mpy_32_32( b_fx, stereo_type_detect->total_power_fx[l] ) ); // stereo_type_detect->q_total_power - move32(); - } -#else - temp = Mpy_32_32( a_fx, W_extract_l( W_shr( reference_power_64fx[l], 31 ) ) ); // 2*(q_cldfb+min_q_shift) -31 - IF( LT_16( q_temp, stereo_type_detect->q_total_power ) ) - { - stereo_type_detect->total_power_fx[l] = L_add( temp, L_shr( Mpy_32_32( b_fx, stereo_type_detect->total_power_fx[l] ), sub( stereo_type_detect->q_total_power, q_temp ) ) ); // q_temp - move32(); - } - ELSE - { - stereo_type_detect->total_power_fx[l] = L_add( L_shr( temp, sub( q_temp, stereo_type_detect->q_total_power ) ), Mpy_32_32( b_fx, stereo_type_detect->total_power_fx[l] ) ); // stereo_type_detect->q_total_power - move32(); - } -#endif + W_tmp2 = W_add( W_tmp1, W_tmp2 ); + q_shift = W_norm( W_tmp2 ); + Total_power_fx = W_extract_h( W_shl( W_tmp2, q_shift ) ); + exp_temppp = sub( 31, sub( add( shl( q_cldfb, 1 ), q_shift ), 32 ) ); + + stereo_type_detect->total_power_fx[l] = BASOP_Util_Add_Mant32Exp( Mpy_32_32( a_fx, Total_power_fx ), exp_temppp, Mpy_32_32( b_fx, stereo_type_detect->total_power_fx[l] ), stereo_type_detect->exp_total_power[l], &stereo_type_detect->exp_total_power[l] ); + move32(); test(); IF( ( stereo_type_detect->sum_power_fx[l] == 0 ) && ( stereo_type_detect->total_power_fx[l] == 0 ) ) { sum_total_ratio_fx[l] = MAX_32; // q15 move32(); + exp_sum_total_ratio_fx[l] = 16; + move16(); } ELSE IF( stereo_type_detect->total_power_fx[l] == 0 ) { sum_total_ratio_fx[l] = MAX_32; // q15 move32(); + exp_sum_total_ratio_fx[l] = 16; + move16(); } ELSE { - sum_total_ratio_fx[l] = BASOP_Util_Divide3232_Scale( stereo_type_detect->sum_power_fx[l], stereo_type_detect->total_power_fx[l], &exp ); // 15-(exp+s_min( stereo_type_detect->q_total_power, q_temp )-s_min( stereo_type_detect->q_sum_power, q_temp2 )) - move32(); -#ifdef FIX_867_CLDFB_NRG_SCALE - q_sum_total_ratio = add( sub( 15, exp ), sub( s_min( stereo_type_detect->q_sum_power, q_temp2 ), s_min( stereo_type_detect->q_total_power, q_temp_total ) ) ); -#else - q_sum_total_ratio = add( sub( 15, exp ), sub( s_min( stereo_type_detect->q_sum_power, q_temp2 ), s_min( stereo_type_detect->q_total_power, q_temp ) ) ); -#endif - sum_total_ratio_fx[l] = L_shl( sum_total_ratio_fx[l], sub( Q15, q_sum_total_ratio ) ); // q15 + sum_total_ratio_fx[l] = BASOP_Util_Divide3232_Scale_newton( stereo_type_detect->sum_power_fx[l], stereo_type_detect->total_power_fx[l], &exp ); // 15-(exp+s_min( stereo_type_detect->q_total_power, q_temp )-s_min( stereo_type_detect->q_sum_power, q_temp2 )) move32(); + exp_sum_total_ratio_fx[l] = add( exp, sub( stereo_type_detect->exp_sum_power[l], stereo_type_detect->exp_total_power[l] ) ); + move16(); } + q_sum_total_ratio = s_max( q_sum_total_ratio, exp_sum_total_ratio_fx[l] ); } IF( l == 0 ) @@ -2626,18 +2595,6 @@ void protoSignalComputation2_fx( move32(); } - stereo_type_detect->q_sum_power = s_min( stereo_type_detect->q_sum_power, q_temp2 ); - move16(); -#ifdef FIX_867_CLDFB_NRG_SCALE - stereo_type_detect->q_total_power = s_min( stereo_type_detect->q_total_power, q_temp_total ); - move16(); -#else - stereo_type_detect->q_total_power = s_min( stereo_type_detect->q_total_power, q_temp ); - move16(); -#endif - q_sum_total_ratio = Q15; - move16(); - IF( stereo_type_detect->interpolator > 0 ) { stereo_type_detect->interpolator++; @@ -2767,8 +2724,15 @@ void protoSignalComputation2_fx( // 20480 = 10 in Q11 lr_total_hi_ratio_fx = Mpy_32_16_1( temp, 20480 ); // Q21 + FOR( Word16 i = 0; i < s_min( num_freq_bands, MASA_SUM_FREQ_RANGE_BINS ); i++ ) + { + sum_total_ratio_fx[i] = L_shl( sum_total_ratio_fx[i], sub( exp_sum_total_ratio_fx[i], q_sum_total_ratio ) ); + move32(); + } + minimum_l( sum_total_ratio_fx, s_min( num_freq_bands, MASA_SUM_FREQ_RANGE_BINS ), &min_sum_total_ratio_fx ); // q_sum_total_ratio - exp = sub( 31, q_sum_total_ratio ); + exp = q_sum_total_ratio; + move16(); temp = BASOP_Util_Log2( min_sum_total_ratio_fx ); // q25 IF( NE_32( temp, MIN_32 ) ) { @@ -3537,10 +3501,8 @@ void ivas_masa_init_stereotype_detection_fx( set32_fx( stereo_type_detect->sum_power_fx, 0, MASA_SUM_FREQ_RANGE_BINS ); set32_fx( stereo_type_detect->total_power_fx, 0, MASA_SUM_FREQ_RANGE_BINS ); - stereo_type_detect->q_sum_power = Q31; - move16(); - stereo_type_detect->q_total_power = Q31; - move16(); + set16_fx( stereo_type_detect->exp_sum_power, 0, MASA_SUM_FREQ_RANGE_BINS ); + set16_fx( stereo_type_detect->exp_total_power, 0, MASA_SUM_FREQ_RANGE_BINS ); stereo_type_detect->subtract_power_y_fx = 0; move32(); diff --git a/lib_rend/ivas_stat_rend.h b/lib_rend/ivas_stat_rend.h index 90e4f5d69..a13ee27e7 100644 --- a/lib_rend/ivas_stat_rend.h +++ b/lib_rend/ivas_stat_rend.h @@ -395,9 +395,9 @@ typedef struct Word16 q_total_hi_power; Word32 sum_power_fx[MASA_SUM_FREQ_RANGE_BINS]; /* Q(q_sum_power) */ - Word16 q_sum_power; + Word16 exp_sum_power[MASA_SUM_FREQ_RANGE_BINS]; Word32 total_power_fx[MASA_SUM_FREQ_RANGE_BINS]; /* Q(q_total_power) */ - Word16 q_total_power; + Word16 exp_total_power[MASA_SUM_FREQ_RANGE_BINS]; Word32 subtract_power_y_fx; /* Q(q_subtract_power_y) */ Word16 q_subtract_power_y; -- GitLab