completed tuning of MC param: no more BE now (91f046f4) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+1 −1

Original line number	Diff line number	Diff line
		@@ -158,5 +158,5 @@
		#define FIX_MINOR_SVD_WMOPS_MR1010X /* FhG: Minor WMOPS tuning, bit-exact to previous version, saves about 8.2 WMOPS for MR1010 */
		#define SVD_WMOPS_OPT /* Ittiam : SVD related optimizations */
		#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 IMPROVE_HIGH_COMPLEXITY_PARAM_MC_PRM_EST /* FhG: reduces WMOPS of param_mc_prm_est, bit-exact to previous version */
		#define IMPROVE_HIGH_COMPLEXITY_PARAM_MC_PRM_EST /* FhG: reduces WMOPS of param_mc_prm_est, not bit-exact to previous version */
		#endif

lib_enc/ivas_mc_param_enc.c

100644 → 100755

+49 −10

Original line number	Diff line number	Diff line
		@@ -815,6 +815,24 @@ static void ivas_param_mc_param_est_enc_fx(
		/* Cy for input channels */
		FOR( ch_idx1 = 0; ch_idx1 < nchan_input; ++ch_idx1 )
		{
		#ifdef IMPROVE_HIGH_COMPLEXITY_PARAM_MC_PRM_EST
		a_e = norm_l( slot_frame_f_real_fx[ch_idx1][cur_cldfb_band] );
		a_fx = L_shl( slot_frame_f_real_fx[ch_idx1][cur_cldfb_band], a_e );
		a_e = sub( add20gb, a_e );
		if ( a_fx == 0 )
		{
		a_e = 0;
		move16();
		}
		b_e = norm_l( slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band] );
		b_fx = L_shl( slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band], b_e );
		b_e = sub( add20gb, b_e );
		if ( b_fx == 0 )
		{
		b_e = 0;
		move16();
		}
		#endif
		FOR( ch_idx2 = ch_idx1; ch_idx2 < nchan_input; ++ch_idx2 )
		{
		#ifndef IMPROVE_HIGH_COMPLEXITY_PARAM_MC_PRM_EST
		@@ -823,20 +841,23 @@ static void ivas_param_mc_param_est_enc_fx(
		c_fx = BASOP_Util_Add_Mant32Exp( slot_frame_f_real_fx[ch_idx2][cur_cldfb_band], add( 20, gb ), 0, 0, &c_e );
		d_fx = BASOP_Util_Add_Mant32Exp( slot_frame_f_imag_fx[ch_idx2][cur_cldfb_band], add( 20, gb ), 0, 0, &d_e );
		#else
		a_e = norm_l( slot_frame_f_real_fx[ch_idx1][cur_cldfb_band] );
		a_fx = L_shl( slot_frame_f_real_fx[ch_idx1][cur_cldfb_band], a_e );
		a_e = sub( add20gb, a_e );
		b_e = norm_l( slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band] );
		b_fx = L_shl( slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band], b_e );
		b_e = sub( add20gb, b_e );
		c_e = norm_l( slot_frame_f_real_fx[ch_idx2][cur_cldfb_band] );
		c_fx = L_shl( slot_frame_f_real_fx[ch_idx2][cur_cldfb_band], c_e );
		c_e = sub( add20gb, c_e );
		if ( c_fx == 0 )
		{
		c_e = 0;
		move16();
		}
		d_e = norm_l( slot_frame_f_imag_fx[ch_idx2][cur_cldfb_band] );
		d_fx = L_shl( slot_frame_f_imag_fx[ch_idx2][cur_cldfb_band], d_e );
		d_e = sub( add20gb, d_e );
		if ( d_fx == 0 )
		{
		d_e = 0;
		move16();
		}
		#endif

		/* (a-ib)(c+id) = ac + bd + i(ad-bc) */
		L_tmp = BASOP_Util_Add_Mant32Exp( Mpy_32_32( a_fx, c_fx ), add( a_e, c_e ), Mpy_32_32( b_fx, d_fx ), add( b_e, d_e ), &tmp_e );
		Cy_sum_fx[cur_param_band][ch_idx1][ch_idx2] = BASOP_Util_Add_Mant32Exp( Cy_sum_fx[cur_param_band][ch_idx1][ch_idx2], Cy_sum_e[cur_param_band][ch_idx1][ch_idx2],
		@@ -899,7 +920,7 @@ static void ivas_param_mc_param_est_enc_fx(
		L_tmp = Mpy_32_32( slot_frame_f_real_fx[inp_ch][cur_cldfb_band], ( *p_dmx_fac_fx ) );
		real_fx = BASOP_Util_Add_Mant32Exp( real_fx, real_e, L_tmp, add20gb, &real_e );
		L_tmp = Mpy_32_32( slot_frame_f_imag_fx[inp_ch][cur_cldfb_band], ( *p_dmx_fac_fx ) );
		imag_fx = BASOP_Util_Add_Mant32Exp( imag_fx, imag_e, L_tmp, add( 20, gb ), &imag_e );
		imag_fx = BASOP_Util_Add_Mant32Exp( imag_fx, imag_e, L_tmp, add20gb, &imag_e );
		p_dmx_fac_fx++;
		}
		dmx_real_fx[ch_idx1] = real_fx;
		@@ -967,9 +988,19 @@ static void ivas_param_mc_param_est_enc_fx(
		a_e = norm_l( slot_frame_f_real_fx[ch_idx1][cur_cldfb_band] );
		a_fx = L_shl( slot_frame_f_real_fx[ch_idx1][cur_cldfb_band], a_e );
		a_e = sub( add20gb, a_e );
		if ( a_fx == 0 )
		{
		a_e = 0;
		move16();
		}
		b_e = norm_l( slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band] );
		b_fx = L_shl( slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band], b_e );
		b_e = sub( add20gb, b_e );
		if ( b_fx == 0 )
		{
		b_e = 0;
		move16();
		}
		#endif
		FOR( ch_idx2 = ch_idx1; ch_idx2 < nchan_input; ++ch_idx2 )
		{
		@@ -979,15 +1010,23 @@ static void ivas_param_mc_param_est_enc_fx(
		c_fx = BASOP_Util_Add_Mant32Exp( slot_frame_f_real_fx[ch_idx2][cur_cldfb_band], add( 20, gb ), 0, 0, &c_e );
		d_fx = BASOP_Util_Add_Mant32Exp( slot_frame_f_imag_fx[ch_idx2][cur_cldfb_band], add( 20, gb ), 0, 0, &d_e );
		#else

		c_e = norm_l( slot_frame_f_real_fx[ch_idx2][cur_cldfb_band] );
		c_fx = L_shl( slot_frame_f_real_fx[ch_idx2][cur_cldfb_band], c_e );
		c_e = sub( add20gb, c_e );
		if ( c_fx == 0 )
		{
		c_e = 0;
		move16();
		}
		d_e = norm_l( slot_frame_f_imag_fx[ch_idx2][cur_cldfb_band] );
		d_fx = L_shl( slot_frame_f_imag_fx[ch_idx2][cur_cldfb_band], d_e );
		d_e = sub( add20gb, d_e );
		if ( d_fx == 0 )
		{
		d_e = 0;
		move16();
		}
		#endif

		/* (a-ib)(c+id) = ac + bd + i(ad-bc) */
		L_tmp = BASOP_Util_Add_Mant32Exp( Mpy_32_32( a_fx, c_fx ), add( a_e, c_e ), Mpy_32_32( b_fx, d_fx ), add( b_e, d_e ), &tmp_e );
		Cy_sum_fx[cur_param_band][ch_idx1][ch_idx2] = BASOP_Util_Add_Mant32Exp( Cy_sum_fx[cur_param_band][ch_idx1][ch_idx2], Cy_sum_e[cur_param_band][ch_idx1][ch_idx2], L_tmp, tmp_e,