delete FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE1, STAGE2 and STAGE 3 completely (82e1f875) · Commits · SA4 / Audio / IVAS BASOP

lib_com/prot_fx.h

+3 −3

Original line number	Diff line number	Diff line
		@@ -59,9 +59,9 @@
		#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_STAGE1 //deleted completely
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2 //deleted completely
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE3 //deleted completely
		//#define FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_upsampledsig // be // 2 WMOPS - deleted completely

		// Numbers ALL: 83,45

lib_com/swb_tbe_com_fx.c

+3 −65

Original line number	Diff line number	Diff line
		@@ -6765,24 +6765,7 @@ 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
		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		Word64 W_tmpX;
		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( 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] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 3], full_band_bpf_fx[3][3] );
		W_tmpX = W_mac_16_16( W_tmpX, input_fx[i], full_band_bpf_fx[0][0] );
		W_tmpY = W_msu_32_16( W_tmpY, L_tmp[i - 4], full_band_bpf_fx[3][4] );
		L_tmp[i] = W_sat_l( W_add( W_shr( W_tmpX, 3 ), W_shl( W_tmpY, 2 - 16 ) ) ); /Q_input_fx + 11/
		move32();
		}
		#else /FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE1/

		FOR( i = 4; i < L_FRAME48k; i++ )
		{
		L_tmpX = L_shr( L_mult( input_fx[i - 4], full_band_bpf_fx[0][4] ), 3 ); /Q_input_fx + 13 + 1 - 3/
		@@ -6796,8 +6779,6 @@ void elliptic_bpf_48k_generic_fx(
		L_tmp[i] = L_sub_sat( L_tmpX, L_shl_sat( Mult_32_16( L_tmp[i - 4], full_band_bpf_fx[3][4] ), 2 ) ); /Q_input_fx + 11 + 13 -15 +2/
		move32();
		}
		#endif


		memory_fx2[0][0] = input_fx[L_FRAME48k - 4];
		memory_fx2[0][1] = input_fx[L_FRAME48k - 3];
		@@ -6853,27 +6834,6 @@ void elliptic_bpf_48k_generic_fx(
		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 /FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2/
		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/
		@@ -6888,7 +6848,7 @@ void elliptic_bpf_48k_generic_fx(
		move32();
		L_tmpMax = L_max( L_tmpMax, L_abs( L_tmp2[i] ) );
		}
		#endif /FIX_1439_SPEEDUP_elliptic_bpf_48k_generic_STAGE2/

		Q_temp = norm_l( L_tmpMax );
		Q_temp = sub( Q_temp, 4 );
		Scale_sig32( L_tmp2, 960, Q_temp );
		@@ -6957,29 +6917,7 @@ 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 /
		@@ -6997,7 +6935,7 @@ 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];