[cleanup] accept MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE, remove... (ae433577) · Commits · SA4 / Audio / IVAS BASOP

lib_com/options.h

+0 −4

Original line number	Diff line number	Diff line
		@@ -77,10 +77,6 @@
		/* Note: each compile switch (FIX_1101_...) is independent from the other ones */
		#define OPT_MCT_ENC_V3_BE
		#define OPT_BIN_REND_V2_NBE
		/* Both following 2 macros (IMPROVE_HIGH_COMPLEXITY_PARAM_MC_PRM_EST) are independent from each other, they refer to different code blocks /
		//#define IMPROVE_HIGH_COMPLEXITY_PARAM_MC_PRM_EST_NONBE /* FhG: reduces WMOPS of param_mc_prm_est, not bit-exact to previous version. Obsoleted by MERGE_REQUEST_1378_SPEEDUP_ivas_mc_param_enc_fx_NONBE. */
		#define MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE /* FhG: reduce WMOPS of dmx calculation in ivas_param_mc_param_est_enc_fx() by using 64 Bit addition. Requires MERGE_REQUEST_1378_SPEEDUP_ivas_mc_param_enc_fx_NONBE. */


		#define FIX_1348_BIT_PRECISION_IMPROVEMENT
		#define FIX_1348_BIT_PRECISION_IMPROVEMENT_QWIN

lib_enc/ivas_mc_param_enc_fx.c

+0 −119

Original line number	Diff line number	Diff line
		@@ -664,15 +664,8 @@ static void ivas_param_mc_param_est_enc_fx(
		Word32 p_slot_frame_f_real_fx[MAX_CICP_CHANNELS]; / Output of the MDFT FB - real part */
		Word32 p_slot_frame_f_imag_fx[MAX_CICP_CHANNELS]; / Output of the MDFT FB - imag part */

		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		Word64 dmx_real_64[PARAM_MC_MAX_TRANSPORT_CHANS];
		Word64 dmx_imag_64[PARAM_MC_MAX_TRANSPORT_CHANS];
		#else
		Word32 dmx_real_fx[PARAM_MC_MAX_TRANSPORT_CHANS]; /* Downmix channel - Real Part */
		Word16 dmx_real_e[PARAM_MC_MAX_TRANSPORT_CHANS]; /* Downmix channel - Real Part */
		Word32 dmx_imag_fx[PARAM_MC_MAX_TRANSPORT_CHANS]; /* Downmix channel - Imag Part */
		Word16 dmx_imag_e[PARAM_MC_MAX_TRANSPORT_CHANS]; /* Downmix channel - Imag Part */
		#endif
		Word32 a_fx, b_fx, c_fx, d_fx; /* Tmp complex values */
		Word16 a_e, b_e, c_e, d_e; /* Tmp complex values */
		Word64 Cy_sum_real_64[PARAM_MC_MAX_PARAMETER_BANDS][MAX_CICP_CHANNELS][MAX_CICP_CHANNELS];
		@@ -749,9 +742,6 @@ static void ivas_param_mc_param_est_enc_fx(
		}

		Word16 gb = find_guarded_bits_fx( l_ts );
		#ifndef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		Word16 add20gb = add( 20, gb );
		#endif

		sub35gb = sub( 32, sub( 11, find_guarded_bits_fx( l_ts ) ) ); // 31 - (((11 - gb) + 31 + norm) - 32)
		sub62gb = sub( 63, shl( sub( 11, find_guarded_bits_fx( l_ts ) ), 1 ) ); // 31 - ((2*(11 - gb) + norm) - 32)
		@@ -784,7 +774,6 @@ static void ivas_param_mc_param_est_enc_fx(

		FOR( ch_idx1 = 0; ch_idx1 < nchan_transport; ++ch_idx1 )
		{
		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		Word64 real_64;
		Word64 imag_64;

		@@ -803,61 +792,17 @@ static void ivas_param_mc_param_est_enc_fx(
		move64();
		move64();

		#else
		Word32 real_fx = L_add( 0, 0 );
		Word16 real_e = add( 0, 0 );
		Word32 imag_fx = L_add( 0, 0 );
		Word16 imag_e = add( 0, 0 );
		FOR( inp_ch = 0; inp_ch < nchan_input; inp_ch++ )
		{
		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, add20gb, &imag_e );
		p_dmx_fac_fx++;
		}
		dmx_real_fx[ch_idx1] = real_fx;
		dmx_real_e[ch_idx1] = real_e;
		dmx_imag_fx[ch_idx1] = imag_fx;
		dmx_imag_e[ch_idx1] = imag_e;
		move32();
		move16();
		move32();
		move16();

		#endif
		}

		/* Cx for transport channels */
		FOR( ch_idx1 = 0; ch_idx1 < nchan_transport; ++ch_idx1 )
		{
		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		CONVERT_DMX( dmx_real_64[ch_idx1], a_fx, a_e );
		CONVERT_DMX( dmx_imag_64[ch_idx1], b_fx, b_e );
		#endif
		FOR( ch_idx2 = 0; ch_idx2 < nchan_transport; ++ch_idx2 )
		{
		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		CONVERT_DMX( dmx_real_64[ch_idx2], c_fx, c_e );
		CONVERT_DMX( dmx_imag_64[ch_idx2], d_fx, d_e );
		#else
		a_fx = dmx_real_fx[ch_idx1];
		move32();
		a_e = dmx_real_e[ch_idx1];
		move16();
		b_fx = dmx_imag_fx[ch_idx1];
		move32();
		b_e = dmx_imag_e[ch_idx1];
		move16();
		c_fx = dmx_real_fx[ch_idx2];
		move32();
		c_e = dmx_real_e[ch_idx2];
		move16();
		d_fx = dmx_imag_fx[ch_idx2];
		move32();
		d_e = dmx_imag_e[ch_idx2];
		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 );
		@@ -907,7 +852,6 @@ static void ivas_param_mc_param_est_enc_fx(

		FOR( ch_idx1 = 0; ch_idx1 < nchan_transport; ++ch_idx1 )
		{
		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		Word64 real_64;
		Word64 imag_64;

		@@ -926,54 +870,15 @@ static void ivas_param_mc_param_est_enc_fx(
		dmx_imag_64[ch_idx1] = imag_64;
		move64();
		move64();
		#else
		Word32 real_fx = L_add( 0, 0 );
		Word16 real_e = 0;
		move16();
		Word32 imag_fx = L_add( 0, 0 );
		Word16 imag_e = 0;
		move16();

		FOR( inp_ch = 0; inp_ch < nchan_input; inp_ch++ )
		{
		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, add20gb, &imag_e );
		p_dmx_fac_fx++;
		}
		dmx_real_fx[ch_idx1] = real_fx;
		move32();
		dmx_real_e[ch_idx1] = real_e;
		move16();
		dmx_imag_fx[ch_idx1] = imag_fx;
		move32();
		dmx_imag_e[ch_idx1] = imag_e;
		move16();

		#endif
		}

		/* Cx for transport channels */
		FOR( ch_idx1 = 0; ch_idx1 < nchan_transport; ++ch_idx1 )
		{
		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		CONVERT_DMX( dmx_real_64[ch_idx1], a_fx, a_e );
		CONVERT_DMX( dmx_imag_64[ch_idx1], b_fx, b_e );
		#else
		a_fx = dmx_real_fx[ch_idx1];
		move32();
		a_e = dmx_real_e[ch_idx1];
		move16();
		b_fx = dmx_imag_fx[ch_idx1];
		move32();
		b_e = dmx_imag_e[ch_idx1];
		move16();

		#endif
		FOR( ch_idx2 = 0; ch_idx2 < nchan_transport; ++ch_idx2 )
		{
		#ifdef MERGE_REQUEST_1472_SPEEDUP_ivas_mc_param_enc_fx_NONBE
		CONVERT_DMX( dmx_real_64[ch_idx2], c_fx, c_e );
		CONVERT_DMX( dmx_imag_64[ch_idx2], d_fx, d_e );

		@@ -981,12 +886,6 @@ static void ivas_param_mc_param_est_enc_fx(
		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 );
		Cx_sum_fx[cur_param_band][ch_idx1][ch_idx2] = BASOP_Util_Add_Mant32Exp( Cx_sum_fx[cur_param_band][ch_idx1][ch_idx2], Cx_sum_e[cur_param_band][ch_idx1][ch_idx2], L_tmp, tmp_e,
		&Cx_sum_e[cur_param_band][ch_idx1][ch_idx2] );
		#else
		/* (a-ib)(c+id) = ac + bd + i(ad-bc) */
		L_tmp = BASOP_Util_Add_Mant32Exp( Mpy_32_32( a_fx, dmx_real_fx[ch_idx2] ), add( a_e, dmx_real_e[ch_idx2] ), Mpy_32_32( b_fx, dmx_imag_fx[ch_idx2] ), add( b_e, dmx_imag_e[ch_idx2] ), &tmp_e );
		Cx_sum_fx[cur_param_band][ch_idx1][ch_idx2] = BASOP_Util_Add_Mant32Exp( Cx_sum_fx[cur_param_band][ch_idx1][ch_idx2], Cx_sum_e[cur_param_band][ch_idx1][ch_idx2], L_tmp, tmp_e,
		&Cx_sum_e[cur_param_band][ch_idx1][ch_idx2] );
		#endif
		move32();
		}
		}
		@@ -994,24 +893,6 @@ 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_NONBE
		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
		a_fx = slot_frame_f_real_fx[ch_idx1][cur_cldfb_band];
		b_fx = slot_frame_f_imag_fx[ch_idx1][cur_cldfb_band];
		move32();

Admin message