[cleanup] accept NONBE_OPT_2193_EIG2X2 (32e2dc36) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+0 −1

Original line number	Diff line number	Diff line
		@@ -115,7 +115,6 @@

		/* #################### Start BASOP optimization switches ############################ */

		#define NONBE_OPT_2193_EIG2X2 /* Dolby: Issue 2193, optimize eig2x2_fx. */

		/* #################### End BASOP optimization switches ############################ */

lib_rend/ivas_dirac_dec_binaural_functions_fx.c

+0 −371

Original line number	Diff line number	Diff line
		@@ -3463,7 +3463,6 @@ static void ivas_dirac_dec_binaural_check_and_switch_transports_headtracked_fx(
		return;
		}

		#ifdef NONBE_OPT_2193_EIG2X2
		static Word32 eig2x2_div_fx( Word32 num, Word32 den );

		static Word32 eig2x2_div_fx( Word32 num, Word32 den )
		@@ -3474,7 +3473,6 @@ static Word32 eig2x2_div_fx( Word32 num, Word32 den )
		}
		return div_w_newton( num, den );
		}
		#endif

		static void eig2x2_fx(
		const Word32 E1_fx, /q_E/
		@@ -3489,7 +3487,6 @@ static void eig2x2_fx(
		Word32 D_fx[BINAURAL_CHANNELS], /q_D/
		Word16 *q_D )
		{
		#ifdef NONBE_OPT_2193_EIG2X2
		Word32 pm_fx, add_fx;
		Word32 tmp1, tmp2, e1, e2, c_re, c_im, c0_im, c1_im;
		Word32 s0_fx, s1_fx, nval0_fx, nval1_fx;
		@@ -3801,374 +3798,6 @@ static void eig2x2_fx(
		move32();
		Uim_fx[i11][1] = Mpy_32_32( c1_im, nval1_fx );
		move32();
		#else
		Word16 chA, chB, ch;
		Word32 s_fx, normVal_fx, crossSquare_fx, a_fx, pm_fx, add_fx;
		Word32 tmp1, tmp2, tmp3, e1, e2, c_re, c_im;
		Word16 q_crossSquare, q_tmp1, q_tmp2, exp_tmp3, exp, q_e, q_c, q_U_1, q_U_2;
		Word32 epsilon_mant = 1180591621;
		Word16 epsilon_exp = -39;
		move32();
		move16();

		pm_fx = 0;
		move32();
		add_fx = 0;
		move32();
		q_tmp2 = 0;
		move16();

		FOR( chA = 0; chA < BINAURAL_CHANNELS; chA++ )
		{
		FOR( chB = 0; chB < BINAURAL_CHANNELS; chB++ )
		{
		Ure_fx[chA][chB] = 0;
		move32();
		Uim_fx[chA][chB] = 0;
		move32();
		}
		}

		exp = sub( get_min_scalefactor( Cre_fx, Cim_fx ), 2 );
		c_re = L_shl( Cre_fx, exp );
		c_im = L_shl( Cim_fx, exp );
		q_c = add( q_C, exp );

		exp = sub( get_min_scalefactor( E1_fx, E2_fx ), 2 );
		e1 = L_shl( E1_fx, exp );
		e2 = L_shl( E2_fx, exp );
		q_e = add( q_E, exp );

		/crossSquare_fx = (c_re c_re) + (c_im * c_im)
		a_fx = (e1 + e2) * (e1 + e2) - 4.0f * ((e1 * e2) - crossSquare_fx) = (e1 - e2)^2 + 4 * crossSquare_fx
		pm_fx = 0.5f * sqrtf(max(0.0f, a_fx))
		add_fx = 0.5f * (e1 + e2)*/

		IF( L_and( c_re == 0, c_im == 0 ) )
		{
		/* if c_re = 0 and c_im = 0, then crossSquare_fx = (c_re * c_re) + (c_im * c_im) = 0
		a_fx = (E1 - E2)^2
		pm_fx = 0.5 * sqrt(max(0, a_fx)) = 0.5 * max(0, (e1 - e2)) */
		crossSquare_fx = 0;
		move32();
		q_crossSquare = Q31;
		move16();
		pm_fx = L_shr( L_max( 0, L_abs( L_sub( e1, e2 ) ) ), 1 );
		q_tmp2 = q_e;
		move16();
		}
		ELSE
		{
		crossSquare_fx = Madd_32_32( Mpy_32_32( c_re, c_re ), c_im, c_im );
		q_crossSquare = sub( add( q_c, q_c ), 31 );
		IF( EQ_32( e1, e2 ) )
		{
		/* if e1 - e2 = 0, then a_fx = 4 * crossSquare_fx
		pm_fx = 0.5 * sqrt(max(0, 4 * crossSquare_fx)) = sqrt(0, crossSquare_fx)*/
		test();
		test();
		IF( EQ_32( c_re, 0 ) \|\| LT_32( L_abs( c_re ), ONE_IN_Q15 ) )
		{
		/* if c_re = 0, then crossSquare_fx = c_im^2 => pm_fx = max(0, c_im) */
		pm_fx = L_max( 0, L_abs( c_im ) );
		q_tmp2 = q_c;
		move16();
		}
		ELSE IF( c_im == 0 \|\| LT_32( L_abs( c_im ), ONE_IN_Q15 ) )
		{
		/* if c_im = 0, then crossSquare_fx = c_re^2 => pm_fx = max(0, c_re) */
		pm_fx = L_max( 0, L_abs( c_re ) );
		q_tmp2 = q_c;
		move16();
		}
		ELSE
		{
		exp = sub( 31, q_crossSquare );
		pm_fx = Sqrt32( L_max( 0, crossSquare_fx ), &exp );
		q_tmp2 = sub( 31, exp );
		}
		}
		ELSE
		{
		/* if e1, e2 >> c_re, c_im then (e1 - e2)^2 ~ (e1 - e2)^2 + 4 * crossSquare_fx
		a_fx = (E1 - E2)^2
		pm_fx = 0.5 * sqrt(max(0, a_fx)) = 0.5 * max(0, (e1 - e2)) */

		IF( GT_16( sub( q_c, q_e ), Q15 ) )
		{
		pm_fx = L_shr( L_max( 0, L_abs( L_sub( e1, e2 ) ) ), 1 );
		q_tmp2 = q_e;
		move16();
		}
		ELSE
		{
		tmp2 = crossSquare_fx;
		move32();
		q_tmp2 = sub( q_crossSquare, 2 ); // crossSquare_fx = 4 * crossSquare_fx;

		tmp1 = Mpy_32_32( L_sub( e1, e2 ), L_sub( e1, e2 ) );
		q_tmp1 = sub( add( q_e, q_e ), 31 );

		a_fx = BASOP_Util_Add_Mant32Exp( tmp1, sub( 31, q_tmp1 ), tmp2, sub( 31, q_tmp2 ), &q_tmp2 ); // (E1 - E2)^2 + 4 * crossSquare_fx
		q_tmp2 = sub( 31, q_tmp2 );

		exp = sub( 31, q_tmp2 );
		pm_fx = Sqrt32( L_max( 0, L_sub( a_fx, 1 ) ), &exp );
		pm_fx = L_shr( pm_fx, 1 );
		q_tmp2 = sub( 31, exp );
		}
		}
		}
		// add_fx = 0.5 * (e1 + e2)
		add_fx = L_shr( L_add( e1, e2 ), 1 );
		q_tmp1 = q_e;
		move16();

		IF( LT_16( q_tmp1, q_tmp2 ) )
		{
		D_fx[0] = L_add( L_shr( add_fx, 1 ), L_shr( pm_fx, add( sub( q_tmp2, q_tmp1 ), 1 ) ) );
		move32();
		D_fx[1] = L_max( 0, L_sub( L_shr( add_fx, 1 ), L_shr( pm_fx, add( sub( q_tmp2, q_tmp1 ), 1 ) ) ) );
		move32();
		*q_D = sub( q_tmp1, 1 );
		move16();
		}
		ELSE
		{
		D_fx[0] = L_add( L_shr( add_fx, add( sub( q_tmp1, q_tmp2 ), 1 ) ), L_shr( pm_fx, 1 ) );
		move32();
		D_fx[1] = L_max( 0, L_sub( L_shr( add_fx, add( sub( q_tmp1, q_tmp2 ), 1 ) ), L_shr( pm_fx, 1 ) ) );
		move32();
		*q_D = sub( q_tmp2, 1 );
		move16();
		}

		/* Numeric case, when input is practically zeros */
		// IF( D_fx[0] < EPSILON_FX )

		IF( LT_32( L_shl_sat( D_fx[0], sub( sub( 31, *q_D ), EPSILON_EXP ) ), EPSILON_MANT ) )
		{
		Ure_fx[0][0] = ONE_IN_Q31;
		move32();
		Ure_fx[1][1] = ONE_IN_Q31;
		move32();
		*q_U = Q31;
		move16();

		return;
		}

		/* Numeric case, when input is near an identity matrix with a gain */
		tmp1 = Mpy_32_32( INV_1000_Q31, add_fx );

		IF( LT_16( q_tmp1, q_tmp2 ) )
		{
		IF( LT_32( L_shr( pm_fx, sub( q_tmp2, q_tmp1 ) ), tmp1 ) )
		{
		Ure_fx[0][0] = ONE_IN_Q30;
		move32();
		Ure_fx[1][1] = ONE_IN_Q30;
		move32();
		*q_U = Q30;
		move16();

		return;
		}
		}
		ELSE
		{
		IF( LT_32( pm_fx, L_shr( tmp1, sub( q_tmp1, q_tmp2 ) ) ) )
		{
		Ure_fx[0][0] = ONE_IN_Q30;
		move32();
		Ure_fx[1][1] = ONE_IN_Q30;
		move32();
		*q_U = Q30;
		move16();

		return;
		}
		}

		q_U_1 = 0;
		q_U_2 = 0;
		move16();
		move16();

		/* Eigenvectors */
		FOR( ch = 0; ch < BINAURAL_CHANNELS; ch++ )
		{
		Word16 q_diff = sub( q_e, *q_D );
		IF( q_diff > 0 )
		{
		tmp1 = L_sub( D_fx[ch], L_shr( e1, q_diff ) );
		tmp2 = L_sub( D_fx[ch], L_shr( e2, q_diff ) );
		q_tmp1 = *q_D;
		move16();
		}
		ELSE
		{
		tmp1 = L_sub( L_shl( D_fx[ch], q_diff ), e1 );
		tmp2 = L_sub( L_shl( D_fx[ch], q_diff ), e2 );
		q_tmp1 = q_e;
		move16();
		}

		IF( GT_32( L_abs( tmp2 ), L_abs( tmp1 ) ) )
		{
		s_fx = tmp2;
		move32();
		exp = sub( norm_l( s_fx ), 1 );
		tmp2 = Mpy_32_32( s_fx, s_fx );
		q_tmp2 = sub( add( q_tmp1, q_tmp1 ), 31 );

		tmp2 = BASOP_Util_Add_Mant32Exp( crossSquare_fx, sub( 31, q_crossSquare ), tmp2, sub( 31, q_tmp2 ), &q_tmp2 );
		q_tmp2 = sub( 31, q_tmp2 );

		tmp3 = BASOP_Util_Add_Mant32Exp( tmp2, sub( 31, q_tmp2 ), epsilon_mant, epsilon_exp, &exp_tmp3 );

		tmp2 = BASOP_Util_Divide3232_Scale_newton( ONE_IN_Q30, tmp3, &exp );
		exp = sub( exp, sub( Q30, sub( 31, exp_tmp3 ) ) );
		normVal_fx = Sqrt32( tmp2, &exp ); // q_tmp2
		q_tmp2 = sub( 31, exp );

		q_diff = sub( q_c, q_tmp1 );
		IF( q_diff > 0 )
		{
		c_re = L_shr( c_re, q_diff );
		c_im = L_shr( c_im, q_diff );
		q_c = q_tmp1;
		move16();
		}
		ELSE
		{
		s_fx = L_shl( s_fx, q_diff );
		q_tmp1 = q_c;
		move16();
		}

		Ure_fx[0][ch] = Mpy_32_32( s_fx, normVal_fx );
		move32();
		Ure_fx[1][ch] = Mpy_32_32( c_re, normVal_fx );
		move32();
		Uim_fx[1][ch] = Mpy_32_32( c_im, normVal_fx );
		move32();
		q_U_1 = sub( add( q_tmp1, q_tmp2 ), 31 );

		IF( q_U_2 != 0 )
		{
		q_diff = sub( q_U_2, q_U_1 );
		IF( q_diff > 0 )
		{
		Ure_fx[1][ch - 1] = L_shr( Ure_fx[1][ch - 1], q_diff );
		Ure_fx[0][ch - 1] = L_shr( Ure_fx[0][ch - 1], q_diff );
		Uim_fx[0][ch - 1] = L_shr( Uim_fx[0][ch - 1], q_diff );
		q_U_2 = q_U_1;
		move32();
		move32();
		move32();
		move16();
		}
		ELSE IF( GT_16( q_U_1, q_U_2 ) )
		{
		Ure_fx[1][ch] = L_shl( Ure_fx[1][ch], q_diff );
		Ure_fx[0][ch] = L_shl( Ure_fx[0][ch], q_diff );
		Uim_fx[1][ch] = L_shl( Uim_fx[1][ch], q_diff );
		q_U_1 = q_U_2;
		move32();
		move32();
		move32();
		move16();
		}
		}
		q_U_2 = q_U_1;
		move16();
		}
		ELSE
		{
		s_fx = tmp1;
		move32();

		exp = sub( norm_l( s_fx ), 1 );
		tmp2 = Mpy_32_32( s_fx, s_fx );
		q_tmp2 = sub( add( q_tmp1, q_tmp1 ), 31 );

		tmp2 = BASOP_Util_Add_Mant32Exp( crossSquare_fx, sub( 31, q_crossSquare ), tmp2, sub( 31, q_tmp2 ), &q_tmp2 );
		q_tmp2 = sub( 31, q_tmp2 );

		tmp3 = BASOP_Util_Add_Mant32Exp( tmp2, sub( 31, q_tmp2 ), epsilon_mant, epsilon_exp, &exp_tmp3 );

		tmp2 = BASOP_Util_Divide3232_Scale_newton( ONE_IN_Q30, tmp3, &exp );
		exp = sub( exp, sub( Q30, sub( 31, exp_tmp3 ) ) );
		normVal_fx = Sqrt32( tmp2, &exp ); // q_tmp2
		q_tmp2 = sub( 31, exp );

		q_diff = sub( q_c, q_tmp1 );
		IF( q_diff > 0 )
		{
		c_re = L_shr( c_re, q_diff );
		c_im = L_shr( c_im, q_diff );
		q_c = q_tmp1;
		move16();
		}
		ELSE
		{
		s_fx = L_shl( s_fx, q_diff );
		q_tmp1 = q_c;
		move16();
		}

		Ure_fx[1][ch] = Mpy_32_32( s_fx, normVal_fx );
		move32();
		Ure_fx[0][ch] = Mpy_32_32( c_re, normVal_fx );
		move32();
		Uim_fx[0][ch] = Mpy_32_32( L_negate( c_im ), normVal_fx );
		move32();
		q_U_2 = sub( add( q_tmp1, q_tmp2 ), 31 );

		IF( q_U_1 != 0 )
		{
		q_diff = sub( q_U_2, q_U_1 );
		IF( q_diff > 0 )
		{
		Ure_fx[1][ch] = L_shr( Ure_fx[1][ch], q_diff );
		Ure_fx[0][ch] = L_shr( Ure_fx[0][ch], q_diff );
		Uim_fx[0][ch] = L_shr( Uim_fx[0][ch], q_diff );
		q_U_2 = q_U_1;
		move32();
		move32();
		move32();
		move16();
		}
		ELSE IF( GT_16( q_U_1, q_U_2 ) )
		{
		Ure_fx[1][ch - 1] = L_shl( Ure_fx[1][ch - 1], q_diff );
		Ure_fx[0][ch - 1] = L_shl( Ure_fx[0][ch - 1], q_diff );
		Uim_fx[1][ch - 1] = L_shl( Uim_fx[1][ch - 1], q_diff );
		q_U_1 = q_U_2;
		move32();
		move32();
		move32();
		move16();
		}
		}
		q_U_1 = q_U_2;
		move16();
		}
		}

		if ( q_U_1 != 0 )
		{
		*q_U = q_U_1;
		move16();
		}

		if ( q_U_1 == 0 )
		{
		*q_U = q_U_2;
		move16();
		}
		#endif
		return;
		}