Complete optimisation. (2304d459) · Commits · SA4 / Audio / IVAS BASOP

lib_rend/ivas_dirac_dec_binaural_functions_fx.c

+123 −241

Original line number	Diff line number	Diff line
		@@ -4250,75 +4250,54 @@ static void matrixScale_fx(
		Word32 Aim_fx[BINAURAL_CHANNELS][BINAURAL_CHANNELS], /q_A/
		Word16 *q_A )
		{
		#if 0 // def OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS
		#ifdef OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS
		Word16 shift_tmp;
		Word16 shift = 31;
		move16();

		shift_tmp = norm_l( Are_fx[0][0] );
		if ( Are_fx[0][0] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		shift_tmp = norm_l( Are_fx[0][1] );
		if ( Are_fx[0][1] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		shift_tmp = norm_l( Are_fx[1][0] );
		if ( Are_fx[1][0] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		shift_tmp = norm_l( Are_fx[1][1] );
		if ( Are_fx[1][1] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		#define MATRIX_NORM_FX( mat, chA, chB ) \
		do \
		{ \
		shift_tmp = norm_l( mat[chA][chB] ); \
		if ( mat[chA][chB] != 0 ) \
		{ \
		shift = s_min( shift, shift_tmp ); \
		} \
		} while ( 0 )

		shift_tmp = norm_l( Aim_fx[0][0] );
		if ( Aim_fx[0][0] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		shift_tmp = norm_l( Aim_fx[0][1] );
		if ( Aim_fx[0][1] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		shift_tmp = norm_l( Aim_fx[1][0] );
		if ( Aim_fx[1][0] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		shift_tmp = norm_l( Aim_fx[1][1] );
		if ( Aim_fx[1][1] != 0 )
		{
		shift = s_min( shift, shift_tmp );
		}
		MATRIX_NORM_FX( Are_fx, 0, 0 );
		MATRIX_NORM_FX( Are_fx, 0, 1 );
		MATRIX_NORM_FX( Are_fx, 1, 0 );
		MATRIX_NORM_FX( Are_fx, 1, 1 );
		MATRIX_NORM_FX( Aim_fx, 0, 0 );
		MATRIX_NORM_FX( Aim_fx, 0, 1 );
		MATRIX_NORM_FX( Aim_fx, 1, 0 );
		MATRIX_NORM_FX( Aim_fx, 1, 1 );

		shift = sub( shift, 1 );
		#undef MATRIX_NORM_FX

		#define MATRIX_SHL_FX( mat, chA, chB ) \
		do \
		{ \
		mat[chA][chB] = L_shl( mat[chA][chB], shift ); \
		move32(); \
		} while ( 0 )

		shift = sub( shift, 1 );
		IF( shift != 0 )
		{
		Are_fx[0][0] = L_shl( Are_fx[0][0], shift );
		Are_fx[0][1] = L_shl( Are_fx[0][1], shift );
		Are_fx[1][0] = L_shl( Are_fx[1][0], shift );
		Are_fx[1][1] = L_shl( Are_fx[1][1], shift );
		Aim_fx[0][0] = L_shl( Aim_fx[0][0], shift );
		Aim_fx[0][1] = L_shl( Aim_fx[0][1], shift );
		Aim_fx[1][0] = L_shl( Aim_fx[1][0], shift );
		Aim_fx[1][1] = L_shl( Aim_fx[1][1], shift );
		move32();
		move32();
		move32();
		move32();
		move32();
		move32();
		move32();
		move32();
		MATRIX_SHL_FX( Are_fx, 0, 0 );
		MATRIX_SHL_FX( Are_fx, 0, 1 );
		MATRIX_SHL_FX( Are_fx, 1, 0 );
		MATRIX_SHL_FX( Are_fx, 1, 1 );
		MATRIX_SHL_FX( Aim_fx, 0, 0 );
		MATRIX_SHL_FX( Aim_fx, 0, 1 );
		MATRIX_SHL_FX( Aim_fx, 1, 0 );
		MATRIX_SHL_FX( Aim_fx, 1, 1 );
		}

		#undef MATRIX_SHL_FX

		q_A = add( q_A, shift );
		move16();
		#else
		@@ -4344,58 +4323,30 @@ static void matrixMul_fx(
		Word16 *q_out )
		{
		Word32 not_zero = 0;
		#if 0 // defined( OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS ) && !defined( IVAS_ENH64_CADENCE_CHANGES )
		outRe_fx[0][0] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Are_fx[0][0], Bre_fx[0][0] ),
		Are_fx[0][1], Bre_fx[1][0] ),
		Aim_fx[0][0], Bim_fx[0][0] ),
		Aim_fx[0][1], Bim_fx[1][0] );
		move32();
		outIm_fx[0][0] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Aim_fx[0][0], Bre_fx[0][0] ),
		Aim_fx[0][1], Bre_fx[1][0] ),
		Are_fx[0][0], Bim_fx[0][0] ),
		Are_fx[0][1], Bim_fx[1][0] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[0][0] );
		not_zero = L_or( not_zero, outIm_fx[0][0] );

		outRe_fx[0][1] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Are_fx[0][0], Bre_fx[0][1] ),
		Are_fx[0][1], Bre_fx[1][1] ),
		Aim_fx[0][0], Bim_fx[0][1] ),
		Aim_fx[0][1], Bim_fx[1][1] );
		move32();
		outIm_fx[0][1] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Aim_fx[0][0], Bre_fx[0][1] ),
		Aim_fx[0][1], Bre_fx[1][1] ),
		Are_fx[0][0], Bim_fx[0][1] ),
		Are_fx[0][1], Bim_fx[1][1] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[0][1] );
		not_zero = L_or( not_zero, outIm_fx[0][1] );

		outRe_fx[1][0] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Are_fx[1][0], Bre_fx[0][0] ),
		Are_fx[1][1], Bre_fx[1][0] ),
		Aim_fx[1][0], Bim_fx[0][0] ),
		Aim_fx[1][1], Bim_fx[1][0] );
		move32();
		outIm_fx[1][0] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Aim_fx[1][0], Bre_fx[0][0] ),
		Aim_fx[1][1], Bre_fx[1][0] ),
		Are_fx[1][0], Bim_fx[0][0] ),
		Are_fx[1][1], Bim_fx[1][0] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[1][0] );
		not_zero = L_or( not_zero, outIm_fx[1][0] );

		outRe_fx[1][1] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Are_fx[1][0], Bre_fx[0][1] ),
		Are_fx[1][1], Bre_fx[1][1] ),
		Aim_fx[1][0], Bim_fx[0][1] ),
		Aim_fx[1][1], Bim_fx[1][1] );
		move32();
		outIm_fx[1][1] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Aim_fx[1][0], Bre_fx[0][1] ),
		Aim_fx[1][1], Bre_fx[1][1] ),
		Are_fx[1][0], Bim_fx[0][1] ),
		Are_fx[1][1], Bim_fx[1][1] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[1][1] );
		not_zero = L_or( not_zero, outIm_fx[1][1] );
		#if defined( OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS ) && !defined( IVAS_ENH64_CADENCE_CHANGES )
		#define MATRIX_MUL_FX( chA, chB ) \
		do \
		{ \
		outRe_fx[chA][chB] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Are_fx[chA][0], Bre_fx[0][chB] ), \
		Are_fx[chA][1], Bre_fx[1][chB] ), \
		Aim_fx[chA][0], Bim_fx[0][chB] ), \
		Aim_fx[chA][1], Bim_fx[1][chB] ); \
		move32(); \
		outIm_fx[chA][chB] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Aim_fx[chA][0], Bre_fx[0][chB] ), \
		Aim_fx[chA][1], Bre_fx[1][chB] ), \
		Are_fx[chA][0], Bim_fx[0][chB] ), \
		Are_fx[chA][1], Bim_fx[1][chB] ); \
		move32(); \
		not_zero = L_or( not_zero, outRe_fx[chA][chB] ); \
		not_zero = L_or( not_zero, outIm_fx[chA][chB] ); \
		} while ( 0 )

		MATRIX_MUL_FX( 0, 0 );
		MATRIX_MUL_FX( 0, 1 );
		MATRIX_MUL_FX( 1, 0 );
		MATRIX_MUL_FX( 1, 1 );

		#undef MATRIX_MUL_FX
		#else
		Word16 chA, chB;

		@@ -4463,90 +4414,49 @@ static void matrixTransp1Mul_fx(
		Word32 not_zero = 0;
		move32();

		#if 0 // def OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS
		tmp64 = W_mult_32_32( Are_fx[0][0], Bre_fx[0][0] );
		tmp64 = W_mac_32_32( tmp64, Are_fx[1][0], Bre_fx[1][0] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[0][0], Bim_fx[0][0] );
		tmp_outRe_fx[0][0] = W_mac_32_32( tmp64, Aim_fx[1][0], Bim_fx[1][0] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outRe_fx[0][0] ) );

		tmp64 = W_mult_32_32( Aim_fx[0][0], Bre_fx[0][0] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[1][0], Bre_fx[1][0] );
		tmp64 = W_mac_32_32( W_neg( tmp64 ), Are_fx[0][0], Bim_fx[0][0] );
		tmp_outIm_fx[0][0] = W_mac_32_32( tmp64, Are_fx[1][0], Bim_fx[1][0] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outIm_fx[0][0] ) );

		tmp64 = W_mult_32_32( Are_fx[0][0], Bre_fx[0][1] );
		tmp64 = W_mac_32_32( tmp64, Are_fx[1][0], Bre_fx[1][1] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[0][0], Bim_fx[0][1] );
		tmp_outRe_fx[0][1] = W_mac_32_32( tmp64, Aim_fx[1][0], Bim_fx[1][1] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outRe_fx[0][1] ) );

		tmp64 = W_mult_32_32( Aim_fx[0][0], Bre_fx[0][1] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[1][0], Bre_fx[1][1] );
		tmp64 = W_mac_32_32( W_neg( tmp64 ), Are_fx[0][0], Bim_fx[0][1] );
		tmp_outIm_fx[0][1] = W_mac_32_32( tmp64, Are_fx[1][0], Bim_fx[1][1] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outIm_fx[0][1] ) );

		tmp64 = W_mult_32_32( Are_fx[0][1], Bre_fx[0][0] );
		tmp64 = W_mac_32_32( tmp64, Are_fx[1][1], Bre_fx[1][0] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[0][1], Bim_fx[0][0] );
		tmp_outRe_fx[1][0] = W_mac_32_32( tmp64, Aim_fx[1][1], Bim_fx[1][0] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outRe_fx[1][0] ) );

		tmp64 = W_mult_32_32( Aim_fx[0][1], Bre_fx[0][0] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[1][1], Bre_fx[1][0] );
		tmp64 = W_mac_32_32( W_neg( tmp64 ), Are_fx[0][1], Bim_fx[0][0] );
		tmp_outIm_fx[1][0] = W_mac_32_32( tmp64, Are_fx[1][1], Bim_fx[1][0] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outIm_fx[1][0] ) );
		#ifdef OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS
		#define MATRIX_TRANSP1_MUL_FX( chA, chB ) \
		do \
		{ \
		tmp64 = W_mult_32_32( Are_fx[0][chA], Bre_fx[0][chB] ); \
		tmp64 = W_mac_32_32( tmp64, Are_fx[1][chA], Bre_fx[1][chB] ); \
		tmp64 = W_mac_32_32( tmp64, Aim_fx[0][chA], Bim_fx[0][chB] ); \
		tmp_outRe_fx[chA][chB] = W_mac_32_32( tmp64, Aim_fx[1][chA], Bim_fx[1][chB] ); \
		move64(); \
		common_lsh = s_min( common_lsh, W_norm( tmp_outRe_fx[chA][chB] ) ); \
		\
		tmp64 = W_mult_32_32( Aim_fx[0][chA], Bre_fx[0][chB] ); \
		tmp64 = W_mac_32_32( tmp64, Aim_fx[1][chA], Bre_fx[1][chB] ); \
		tmp64 = W_mac_32_32( W_neg( tmp64 ), Are_fx[0][chA], Bim_fx[0][chB] ); \
		tmp_outIm_fx[chA][chB] = W_mac_32_32( tmp64, Are_fx[1][chA], Bim_fx[1][chB] ); \
		move64(); \
		common_lsh = s_min( common_lsh, W_norm( tmp_outIm_fx[chA][chB] ) ); \
		} while ( 0 )

		tmp64 = W_mult_32_32( Are_fx[0][1], Bre_fx[0][1] );
		tmp64 = W_mac_32_32( tmp64, Are_fx[1][1], Bre_fx[1][1] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[0][1], Bim_fx[0][1] );
		tmp_outRe_fx[1][1] = W_mac_32_32( tmp64, Aim_fx[1][1], Bim_fx[1][1] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outRe_fx[1][1] ) );
		MATRIX_TRANSP1_MUL_FX( 0, 0 );
		MATRIX_TRANSP1_MUL_FX( 0, 1 );
		MATRIX_TRANSP1_MUL_FX( 1, 0 );
		MATRIX_TRANSP1_MUL_FX( 1, 1 );

		tmp64 = W_mult_32_32( Aim_fx[0][1], Bre_fx[0][1] );
		tmp64 = W_mac_32_32( tmp64, Aim_fx[1][1], Bre_fx[1][1] );
		tmp64 = W_mac_32_32( W_neg( tmp64 ), Are_fx[0][1], Bim_fx[0][1] );
		tmp_outIm_fx[1][1] = W_mac_32_32( tmp64, Are_fx[1][1], Bim_fx[1][1] );
		move64();
		common_lsh = s_min( common_lsh, W_norm( tmp_outIm_fx[1][1] ) );
		#undef MATRIX_TRANSP1_MUL_FX

		outRe_fx[0][0] = W_extract_h( W_shl( tmp_outRe_fx[0][0], common_lsh ) );
		move32();
		outIm_fx[0][0] = W_extract_h( W_shl( tmp_outIm_fx[0][0], common_lsh ) );
		move32();
		not_zero = L_or( not_zero, outRe_fx[0][0] );
		not_zero = L_or( not_zero, outIm_fx[0][0] );

		outRe_fx[0][1] = W_extract_h( W_shl( tmp_outRe_fx[0][1], common_lsh ) );
		move32();
		outIm_fx[0][1] = W_extract_h( W_shl( tmp_outIm_fx[0][1], common_lsh ) );
		move32();
		not_zero = L_or( not_zero, outRe_fx[0][1] );
		not_zero = L_or( not_zero, outIm_fx[0][1] );
		#define MATRIX_TRANSP1_SCALE_FX( chA, chB ) \
		do \
		{ \
		outRe_fx[chA][chB] = W_extract_h( W_shl( tmp_outRe_fx[chA][chB], common_lsh ) ); \
		move32(); \
		outIm_fx[chA][chB] = W_extract_h( W_shl( tmp_outIm_fx[chA][chB], common_lsh ) ); \
		move32(); \
		not_zero = L_or( not_zero, outRe_fx[chA][chB] ); \
		not_zero = L_or( not_zero, outIm_fx[chA][chB] ); \
		} while ( 0 )

		outRe_fx[1][0] = W_extract_h( W_shl( tmp_outRe_fx[1][0], common_lsh ) );
		move32();
		outIm_fx[1][0] = W_extract_h( W_shl( tmp_outIm_fx[1][0], common_lsh ) );
		move32();
		not_zero = L_or( not_zero, outRe_fx[1][0] );
		not_zero = L_or( not_zero, outIm_fx[1][0] );
		MATRIX_TRANSP1_SCALE_FX( 0, 0 );
		MATRIX_TRANSP1_SCALE_FX( 0, 1 );
		MATRIX_TRANSP1_SCALE_FX( 1, 0 );
		MATRIX_TRANSP1_SCALE_FX( 1, 1 );

		outRe_fx[1][1] = W_extract_h( W_shl( tmp_outRe_fx[1][1], common_lsh ) );
		move32();
		outIm_fx[1][1] = W_extract_h( W_shl( tmp_outIm_fx[1][1], common_lsh ) );
		move32();
		not_zero = L_or( not_zero, outRe_fx[1][1] );
		not_zero = L_or( not_zero, outIm_fx[1][1] );
		#undef MATRIX_TRANSP1_SCALE_FX
		#else
		Word16 chA, chB;

		@@ -4607,58 +4517,30 @@ static void matrixTransp2Mul_fx(
		Word16 *q_out )
		{
		Word32 not_zero = 0;
		#if 0 // defined( OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS ) && !defined( IVAS_ENH64_CADENCE_CHANGES )
		outRe_fx[0][0] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Are_fx[0][0], Bre_fx[0][0] ),
		Are_fx[0][1], Bre_fx[0][1] ),
		Aim_fx[0][0], Bim_fx[0][0] ),
		Aim_fx[0][1], Bim_fx[0][1] );
		move32();
		outIm_fx[0][0] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Aim_fx[0][0], Bre_fx[0][0] ),
		Aim_fx[0][1], Bre_fx[0][1] ),
		Are_fx[0][0], Bim_fx[0][0] ),
		Are_fx[0][1], Bim_fx[0][1] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[0][0] );
		not_zero = L_or( not_zero, outIm_fx[0][0] );

		outRe_fx[0][1] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Are_fx[0][0], Bre_fx[1][0] ),
		Are_fx[0][1], Bre_fx[1][1] ),
		Aim_fx[0][0], Bim_fx[1][0] ),
		Aim_fx[0][1], Bim_fx[1][1] );
		move32();
		outIm_fx[0][1] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Aim_fx[0][0], Bre_fx[1][0] ),
		Aim_fx[0][1], Bre_fx[1][1] ),
		Are_fx[0][0], Bim_fx[1][0] ),
		Are_fx[0][1], Bim_fx[1][1] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[0][1] );
		not_zero = L_or( not_zero, outIm_fx[0][1] );

		outRe_fx[1][0] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Are_fx[1][0], Bre_fx[0][0] ),
		Are_fx[1][1], Bre_fx[0][1] ),
		Aim_fx[1][0], Bim_fx[0][0] ),
		Aim_fx[1][1], Bim_fx[0][1] );
		move32();
		outIm_fx[1][0] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Aim_fx[1][0], Bre_fx[0][0] ),
		Aim_fx[1][1], Bre_fx[0][1] ),
		Are_fx[1][0], Bim_fx[0][0] ),
		Are_fx[1][1], Bim_fx[0][1] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[1][0] );
		not_zero = L_or( not_zero, outIm_fx[1][0] );

		outRe_fx[1][1] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Are_fx[1][0], Bre_fx[1][0] ),
		Are_fx[1][1], Bre_fx[1][1] ),
		Aim_fx[1][0], Bim_fx[1][0] ),
		Aim_fx[1][1], Bim_fx[1][1] );
		move32();
		outIm_fx[1][1] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Aim_fx[1][0], Bre_fx[1][0] ),
		Aim_fx[1][1], Bre_fx[1][1] ),
		Are_fx[1][0], Bim_fx[1][0] ),
		Are_fx[1][1], Bim_fx[1][1] );
		move32();
		not_zero = L_or( not_zero, outRe_fx[1][1] );
		not_zero = L_or( not_zero, outIm_fx[1][1] );
		#if defined( OPT_2269_IVAS_DIRAC_DEC_BINAURAL_FUNCTIONS ) && !defined( IVAS_ENH64_CADENCE_CHANGES )
		#define MATRIX_TRANSP2_MUL_FX( chA, chB ) \
		do \
		{ \
		outRe_fx[chA][chB] = Madd_32_32( Madd_32_32( Madd_32_32( Mpy_32_32( Are_fx[chA][0], Bre_fx[chB][0] ), \
		Are_fx[chA][1], Bre_fx[chB][1] ), \
		Aim_fx[chA][0], Bim_fx[chB][0] ), \
		Aim_fx[chA][1], Bim_fx[chB][1] ); \
		move32(); \
		outIm_fx[chA][chB] = Msub_32_32( Msub_32_32( Madd_32_32( Mpy_32_32( Aim_fx[chA][0], Bre_fx[chB][0] ), \
		Aim_fx[chA][1], Bre_fx[chB][1] ), \
		Are_fx[chA][0], Bim_fx[chB][0] ), \
		Are_fx[chA][1], Bim_fx[chB][1] ); \
		move32(); \
		not_zero = L_or( not_zero, outRe_fx[chA][chB] ); \
		not_zero = L_or( not_zero, outIm_fx[chA][chB] ); \
		} while ( 0 )

		MATRIX_TRANSP2_MUL_FX( 0, 0 );
		MATRIX_TRANSP2_MUL_FX( 0, 1 );
		MATRIX_TRANSP2_MUL_FX( 1, 0 );
		MATRIX_TRANSP2_MUL_FX( 1, 1 );

		#undef MATRIX_TRANSP2_MUL_FX
		#else
		Word16 chA, chB;
		FOR( chA = 0; chA < BINAURAL_CHANNELS; chA++ )