introduced FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2 and... (ff856655) · Commits · SA4 / Audio / IVAS BASOP

lib_com/prot_fx.h

+16 −0

Original line number	Diff line number	Diff line
		@@ -59,6 +59,10 @@
		#define TCX_IMDCT_SCALE 15
		#define TCX_IMDCT_HEADROOM 1

		#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE1
		#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2
		#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE3
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32

		/----------------------------------------------------------------------------------
		* Prototypes of global macros
		@@ -3258,6 +3262,17 @@ void interp_code_4over2_fx(

		void wb_tbe_extras_reset_synth_fx( Word16 state_lsyn_filt_shb[], Word16 state_lsyn_filt_dwn_shb[], Word16 state_32and48k_WB_upsample[], Word16 state_resamp_HB[] );

		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		void elliptic_bpf_48k_generic_fx(
		const Word16* input_fx, /* i : i signal Q_input_fx */
		Word16 *Q_input_fx,
		Word16 output_fx[], /* o : output signal */
		Word32 memory_fx[][4], /* i/o: 4 arrays of 4 for memory memory_fx_Q */
		Word16 memory_fx_Q[],
		const Word16 full_band_bpf[][5], /* i : filter coefficients b0,b1,b2,a0,a1,a2 Q13 */
		const Word32 *output32_fx,
		const Word16 *scale_output32_fx);
		#else
		void elliptic_bpf_48k_generic_fx(
		const Word16 input_fx[], /* i : i signal Q_input_fx */
		Word16 *Q_input_fx,
		@@ -3266,6 +3281,7 @@ void elliptic_bpf_48k_generic_fx(
		Word16 memory_fx_Q[],
		const Word16 full_band_bpf[][5] /* i : filter coefficients b0,b1,b2,a0,a1,a2 Q13 */
		);
		#endif

		void synthesise_fb_high_band_fx(
		const Word16 excitation_in[], /* i : full band excitation */

lib_com/swb_tbe_com_fx.c

+62 −8

Original line number	Diff line number	Diff line
		@@ -18,6 +18,8 @@
		#define THR_ENV_ERROR_PLOSIVE 200.0f /* threshold for envelope error used in plosive detection */
		#define THR_ENV_ERROR_PLOSIVE_FX 200 /* threshold for envelope error used in plosive detection Q0 */



		/-----------------------------------------------------------------
		* Local function prototypes
		-----------------------------------------------------------------/
		@@ -6692,9 +6694,8 @@ void wb_tbe_extras_reset_synth_fx(
		* 18th-order elliptic bandpass filter at 14.0 to 20 kHz sampled at 48 kHz
		* Implemented as 3 fourth order sections cascaded.
		-------------------------------------------------------------------/

		#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic

		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		#else
		void elliptic_bpf_48k_generic_fx(
		const Word16 input_fx[], /* i : input signal Q_input_fx*/
		Word16 *Q_input_fx,
		@@ -6703,6 +6704,7 @@ void elliptic_bpf_48k_generic_fx(
		Word16 memory_fx_Q[],
		const Word16 full_band_bpf_fx[][5] /* i : filter coefficients b0,b1,b2,a0,a1,a2 Q13 */
		)
		#endif
		{
		Word16 i, j;
		Word16 memory_fx0[4][4], memory_fx[4][4], Q_temp, Q_temp2;
		@@ -6766,14 +6768,14 @@ void elliptic_bpf_48k_generic_fx(
		L_tmpX = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[0], full_band_bpf_fx[3][3] ), 2 ) ); /Q_input_fx + 11 + 13 -15 +2/
		L_tmp[3] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( memory2_fx[1][3], full_band_bpf_fx[3][4] ), 2 ) ); /Q_input_fx + 11 + 13 -15 +2/
		move32();
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic // STAGE1
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE1
		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		Word64 W_tmpX;
		Word64 W_tmpY = W_add( 0, 0 );
		W_tmpX = W_deposit32_l( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ) );
		Word64 W_tmpY;
		W_tmpX = W_mac_16_16( 0, input_fx[i - 4], full_band_bpf_fx[0][4] );
		W_tmpX = W_mac_16_16( W_tmpX, input_fx[i - 3], full_band_bpf_fx[0][3] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 1], full_band_bpf_fx[3][1] );
		W_tmpY = W_msu_32_16( 0, L_tmp[i - 1], full_band_bpf_fx[3][1] );
		W_tmpX = W_mac_16_16( W_tmpX, input_fx[i - 2], full_band_bpf_fx[0][2] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 2], full_band_bpf_fx[3][2] );
		W_tmpX = W_mac_16_16( W_tmpX, input_fx[i - 1], full_band_bpf_fx[0][1] );
		@@ -6853,6 +6855,26 @@ void elliptic_bpf_48k_generic_fx(
		L_tmp2[3] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( memory2_fx[2][3], full_band_bpf_fx[4][4] ), 2 ) ); /Q_input_fx + 6 +13 -15 +2 / /14 + Q_input_fx - shift_flag/
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_tmp2[3] ) );

		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2
		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		Word64 W_tmpX;
		Word64 W_tmpY;
		W_tmpX = W_mac_32_16( 0, L_tmp[i - 4], full_band_bpf_fx[1][4] );
		W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i - 3], full_band_bpf_fx[1][3] );
		W_tmpY = W_msu_32_16( 0, L_tmp2[i - 1], full_band_bpf_fx[4][1] );
		W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i - 2], full_band_bpf_fx[1][2] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp2[i - 2], full_band_bpf_fx[4][2] );
		W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i - 1], full_band_bpf_fx[1][1] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp2[i - 3], full_band_bpf_fx[4][3] );
		W_tmpX = W_mac_32_16( W_tmpX, L_tmp[i], full_band_bpf_fx[1][0] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp2[i - 4], full_band_bpf_fx[4][4] );
		L_tmp2[i] = W_sat_l( W_add( W_shr( W_tmpX, 3 + 16 ), W_shl( W_tmpY, 2 - 16 ) ) );
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_tmp2[i] ) );
		}
		#else
		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		L_tmpX = L_shr( Mult_32_16( L_tmp[i - 4], full_band_bpf_fx[1][4] ), 3 ); /Q_input_fx + 11 + 13 - 15 -3/
		@@ -6867,7 +6889,7 @@ void elliptic_bpf_48k_generic_fx(
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_tmp2[i] ) );
		}

		#endif
		Q_temp = norm_l( L_tmpMax );
		Q_temp = sub( Q_temp, 4 );
		Scale_sig32( L_tmp2, 960, Q_temp );
		@@ -6936,7 +6958,29 @@ void elliptic_bpf_48k_generic_fx(
		L_output[3] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( memory2_fx_3[3], full_band_bpf_fx[5][4] ), 2 ) ); /Q_input_fx + 1 +Q_temp+13 -15 + 2/
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_output[3] ) );
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE3
		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		Word64 W_tmpX;
		Word64 W_tmpY;

		W_tmpX = W_mac_32_16( 0, L_tmp2[i - 4], full_band_bpf_fx[2][4] );
		W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i - 3], full_band_bpf_fx[2][3] );
		W_tmpY = W_msu_32_16( 0, L_output[i - 1], full_band_bpf_fx[5][1] );

		W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i - 2], full_band_bpf_fx[2][2] );
		W_tmpY = W_msu_32_16( W_tmpY, L_output[i - 2], full_band_bpf_fx[5][2] );

		W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i - 1], full_band_bpf_fx[2][1] );
		W_tmpY = W_msu_32_16( W_tmpY, L_output[i - 3], full_band_bpf_fx[5][3] );

		W_tmpX = W_mac_32_16( W_tmpX, L_tmp2[i], full_band_bpf_fx[2][0] );
		W_tmpY = W_msu_32_16( W_tmpY, L_output[i - 4], full_band_bpf_fx[5][4] );
		L_output[i] = W_sat_l( W_add( W_shr( W_tmpX, 3 + 16 ), W_shl( W_tmpY, 2 - 16 ) ) );
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_output[i] ) );
		}
		#else
		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		L_tmpX = L_shr( Mult_32_16( L_tmp2[i - 4], full_band_bpf_fx[2][4] ), 3 ); /Q_input_fx + 6 +Q_temp+13 -15 -3 /
		@@ -6954,6 +6998,7 @@ void elliptic_bpf_48k_generic_fx(
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_output[i] ) );
		}
		#endif
		memory_fx2[2][0] = L_tmp2[L_FRAME48k - 4];
		memory_fx2[2][1] = L_tmp2[L_FRAME48k - 3];
		memory_fx2[2][2] = L_tmp2[L_FRAME48k - 2];
		@@ -7047,12 +7092,21 @@ void synthesise_fb_high_band_fx(
		IF( EQ_16( L_frame, L_FRAME16k ) )
		{
		/* for 16kHz ACELP core */
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		elliptic_bpf_48k_generic_fx( excitation_in_interp3, &exp_tmp, NULL, bpf_memory, bpf_memory_Q, full_band_bpf_3_fx, tmp32, scale_tmp32 );
		#else
		elliptic_bpf_48k_generic_fx( excitation_in_interp3, &exp_tmp, tmp, bpf_memory, bpf_memory_Q, full_band_bpf_3_fx );
		#endif
		}
		ELSE
		{
		/* for 12.8kHz ACELP core */
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		elliptic_bpf_48k_generic_fx( excitation_in_interp3, &exp_tmp, NULL, bpf_memory, bpf_memory_Q, full_band_bpf_1_fx, tmp32, scale_tmp32 );
		#else
		elliptic_bpf_48k_generic_fx( excitation_in_interp3, &exp_tmp, tmp, bpf_memory, bpf_memory_Q, full_band_bpf_1_fx );
		#endif

		}
		pop_wmops(); /push_wmops( "SYNTHESISE_FB_HIGH_BAND PART A.1" );/
		pop_wmops(); /push_wmops( "SYNTHESISE_FB_HIGH_BAND PART A" )/

lib_enc/swb_tbe_enc_fx.c

+12 −1

Original line number	Diff line number	Diff line
		@@ -7324,8 +7324,11 @@ void fb_tbe_enc_fx(
		exp_temp = sub( exp_temp, 1 );

		Copy_Scale_sig( new_input, input_fhb, L_FRAME48k, exp_temp );

		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		elliptic_bpf_48k_generic_fx( input_fhb, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx, NULL, NULL );
		#else
		elliptic_bpf_48k_generic_fx( input_fhb, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );
		#endif
		Sample_Delay_HP = NS2SA( 48000, ACELP_LOOK_NS + DELAY_FD_BWE_ENC_12k8_NS + DELAY_FIR_RESAMPL_NS ) - L_FRAME48k / 2;

		IF( NE_16( st->last_extl, FB_TBE ) )
		@@ -7449,11 +7452,19 @@ void fb_tbe_enc_ivas_fx(

		IF( EQ_16( st->element_mode, IVAS_CPE_DFT ) )
		{
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		elliptic_bpf_48k_generic_fx( input_fhb_new, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx, NULL, NULL );
		#else
		elliptic_bpf_48k_generic_fx( input_fhb_new, &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );
		#endif
		}
		ELSE
		{
		#ifdef FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_OUT32
		elliptic_bpf_48k_generic_fx( input_fhb_new + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ), &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx, NULL, NULL );
		#else
		elliptic_bpf_48k_generic_fx( input_fhb_new + NS2SA( 48000, DELAY_FIR_RESAMPL_NS ), &exp_temp, tmp_vec, hBWE_TD->elliptic_bpf_2_48k_mem_fx, hBWE_TD->elliptic_bpf_2_48k_mem_fx_Q, full_band_bpf_2_fx );
		#endif
		}

		test();