more fixed related to VBR (99ee2de2) · Commits · IVAS Codec Public Collaboration / IVAS Codec BASOP

lib_com/wi_fx.c

+106 −8

Original line number	Diff line number	Diff line
		@@ -434,7 +434,9 @@ Word16 DTFS_alignment_full_fx(
		Word16 n, fshift_fx;
		Word32 corr_fx, maxcorr_fx;
		Word16 Eshift, Adiff_fx;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif

		/* Calculating the expected alignment shift */
		Eshift = mult_r(ph_offset_fx, shl(X2_DTFS_fx.lag_fx, 7)); /* confirmed I<2 by smv12.org, Q7 */
		@@ -468,8 +470,13 @@ Word16 DTFS_alignment_full_fx(

		FOR (k=0; k<=HalfLag; k++)
		{
		#ifdef BASOP_NOGLOB
		ab1[k] = round_fx_o(L_mac_o(L_mult_o(X1_DTFS_fx.a_fx[k], X2_DTFS_fx.a_fx[k], &Overflow), X1_DTFS_fx.b_fx[k], X2_DTFS_fx.b_fx[k], &Overflow), &Overflow); /* Q(-15); */
		ab2[k] = round_fx_o(L_msu_o(L_mult_o(X1_DTFS_fx.b_fx[k], X2_DTFS_fx.a_fx[k], &Overflow), X1_DTFS_fx.a_fx[k], X2_DTFS_fx.b_fx[k], &Overflow), &Overflow); /* Q(-15); */
		#else
		ab1[k] = round_fx(L_mac(L_mult(X1_DTFS_fx.a_fx[k], X2_DTFS_fx.a_fx[k]), X1_DTFS_fx.b_fx[k], X2_DTFS_fx.b_fx[k])); /* Q(-15); */
		ab2[k] = round_fx(L_msu(L_mult(X1_DTFS_fx.b_fx[k], X2_DTFS_fx.a_fx[k]), X1_DTFS_fx.a_fx[k], X2_DTFS_fx.b_fx[k])); /* Q(-15); */
		#endif
		}
		IF (FR_flag == 0)
		{
		@@ -499,9 +506,15 @@ Word16 DTFS_alignment_full_fx(
		temp1 = add(n,shl(X2_DTFS_fx.lag_fx,1)); /* add lag_fx in Q1to make positive */
		FOR (k=0; k<=HalfLag; k++)
		{
		#ifdef BASOP_NOGLOB
		corr_fx = L_mac_o(corr_fx, ab1[k], C_fx[(2temp)%(4X2_DTFS_fx.lag_fx)], &Overflow);
		corr_fx = L_mac_o(corr_fx, ab2[k], S_fx[(2temp)%(4X2_DTFS_fx.lag_fx)], &Overflow);
		temp = add_o(temp, temp1, &Overflow);
		#else
		corr_fx = L_mac(corr_fx, ab1[k], C_fx[(2temp)%(4X2_DTFS_fx.lag_fx)]);
		corr_fx = L_mac(corr_fx, ab2[k], S_fx[(2temp)%(4X2_DTFS_fx.lag_fx)]);
		temp = add(temp, temp1);
		#endif
		}

		if (GT_32(corr_fx, maxcorr_fx))
		@@ -603,7 +616,9 @@ void Q2phaseShift_fx(
		Word16 k ;
		Word16 temp, HalfLag;
		Word32 temp2;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif
		temp2 = L_deposit_l(0);

		HalfLag = s_min(shr(X_fx->lag_fx, 1),X_fx->nH_fx);
		@@ -614,9 +629,15 @@ void Q2phaseShift_fx(
		FOR (k=0; k<=HalfLag; k++)
		{
		temp = X_fx->a_fx[k];
		#ifdef BASOP_NOGLOB
		X_fx->a_fx[k] = round_fx_o(L_msu_o(L_mult_o(temp, C_fx[temp2%(4Lag)], &Overflow), X_fx->b_fx[k], S_fx[temp2%(4Lag)], &Overflow), &Overflow) ; /* X.Q */
		X_fx->b_fx[k] = round_fx_o(L_mac_o(L_mult_o(X_fx->b_fx[k], C_fx[temp2%(4Lag)], &Overflow), temp, S_fx[temp2%(4Lag)], &Overflow), &Overflow) ;
		temp2 = L_add_o(temp2, (Word32)ph, &Overflow);
		#else
		X_fx->a_fx[k] = round_fx(L_msu(L_mult(temp, C_fx[temp2 % (4 * Lag)]), X_fx->b_fx[k], S_fx[temp2 % (4 * Lag)])); /* X.Q */
		X_fx->b_fx[k] = round_fx(L_mac(L_mult(X_fx->b_fx[k], C_fx[temp2 % (4 * Lag)]), temp, S_fx[temp2 % (4 * Lag)]));
		temp2 = L_add(temp2, (Word32)ph);
		#endif
		}
		}

		@@ -627,9 +648,15 @@ void Q2phaseShift_fx(
		FOR (k=0; k<=HalfLag; k++)
		{
		temp = X_fx->a_fx[k];
		#ifdef BASOP_NOGLOB
		X_fx->a_fx[k] = round_fx_o(L_mac_o(L_mult_o(temp, C_fx[temp2%(4Lag)], &Overflow), X_fx->b_fx[k], S_fx[temp2%(4Lag)], &Overflow), &Overflow) ; /* X.Q */
		X_fx->b_fx[k] = round_fx_o(L_msu_o(L_mult_o(X_fx->b_fx[k], C_fx[temp2%(4Lag)], &Overflow), temp, S_fx[temp2%(4Lag)], &Overflow), &Overflow) ;
		temp2 = add_o((Word16)temp2, negate(ph), &Overflow);
		#else
		X_fx->a_fx[k] = round_fx(L_mac(L_mult(temp, C_fx[temp2%(4Lag)]), X_fx->b_fx[k], S_fx[temp2%(4Lag)])) ; /* X.Q */
		X_fx->b_fx[k] = round_fx(L_msu(L_mult(X_fx->b_fx[k], C_fx[temp2%(4Lag)]), temp, S_fx[temp2%(4Lag)])) ;
		temp2 = add((Word16)temp2, negate(ph));
		#endif
		}
		}
		}
		@@ -912,19 +939,33 @@ void DTFS_to_fs_fx(

		FOR ( k=1; k<=nH; k++ )
		{
		#ifdef BASOP_NOGLOB
		X_fx->a_fx[k]=round_fx_o(L_shl_o(La[k], temp, &Overflow), &Overflow); /* Q(8+temp-16)=Q(temp-8) */
		X_fx->a_fx[k]=mult_ro(X_fx->a_fx[k], inv_lag, &Overflow);
		move16(); /* Q(temp-8+19+1-16)=Q(temp-4) of a[k]2.0/N /
		X_fx->b_fx[k]=round_fx_o(L_shl_o(Lb[k], temp, &Overflow), &Overflow); /* Q(8+temp-16)=Q(temp-8) */
		X_fx->b_fx[k]=mult_ro(X_fx->b_fx[k], inv_lag, &Overflow);
		move16(); /* Q(temp-8+19+1-16)=Q(temp-4) of b[k]2.0/N /
		#else
		X_fx->a_fx[k]=round_fx(L_shl(La[k], temp)); /* Q(8+temp-16)=Q(temp-8) */
		X_fx->a_fx[k]=mult_r(X_fx->a_fx[k], inv_lag);
		move16(); /* Q(temp-8+19+1-16)=Q(temp-4) of a[k]2.0/N /
		X_fx->b_fx[k]=round_fx(L_shl(Lb[k], temp)); /* Q(8+temp-16)=Q(temp-8) */
		X_fx->b_fx[k]=mult_r(X_fx->b_fx[k], inv_lag);
		move16(); /* Q(temp-8+19+1-16)=Q(temp-4) of b[k]2.0/N /
		#endif
		}

		/* IF ( N%2 == 0 ) */
		IF ( s_and(N,1) == 0 )
		{
		#ifdef BASOP_NOGLOB
		X_fx->a_fx[k]=round_fx_o(L_shl_o(La[k], temp, &Overflow), &Overflow); /* Q(8+temp-16)=Q(temp-8) */
		X_fx->a_fx[k]=mult_ro(X_fx->a_fx[k], inv_lag, &Overflow);
		#else
		X_fx->a_fx[k]=round_fx(L_shl(La[k], temp)); /* Q(8+temp-16)=Q(temp-8) */
		X_fx->a_fx[k]=mult_r(X_fx->a_fx[k], inv_lag);
		#endif
		move16(); /* Q(temp-8+19+1-16)=Q(temp-4) of a[k]1.0/N /
		X_fx->b_fx[k]=0;
		move16();
		@@ -1558,7 +1599,9 @@ Word32 DTFS_getEngy_fx(
		Word16 k, HalfLag_fx;
		Word32 en_fx=0;
		Word16 temp_a_fx,temp_b_fx;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif
		HalfLag_fx = shr(sub(X_fx->lag_fx, 1), 1);
		HalfLag_fx = s_min(HalfLag_fx,X_fx->nH_fx);
		FOR (k=1; k<=HalfLag_fx; k++)
		@@ -1568,14 +1611,22 @@ Word32 DTFS_getEngy_fx(
		temp_b_fx = X_fx->b_fx[k];
		move16();

		#ifdef BASOP_NOGLOB
		en_fx = L_mac0_o(en_fx, temp_a_fx, temp_a_fx, &Overflow);
		en_fx = L_mac0_o(en_fx, temp_b_fx, temp_b_fx, &Overflow);
		#else
		en_fx = L_mac0(en_fx, temp_a_fx, temp_a_fx);
		en_fx = L_mac0(en_fx, temp_b_fx, temp_b_fx);
		#endif
		}

		en_fx = L_shr(en_fx, 1);
		temp_a_fx = X_fx->a_fx[0];
		#ifdef BASOP_NOGLOB
		en_fx = L_mac0_o(en_fx,temp_a_fx, temp_a_fx, &Overflow);
		#else
		en_fx = L_mac0(en_fx,temp_a_fx, temp_a_fx);

		#endif
		/* IF (X_fx->lag_fx%2 == 0) */
		IF (s_and(X_fx->lag_fx,1) == 0)
		{
		@@ -1584,8 +1635,13 @@ Word32 DTFS_getEngy_fx(
		temp_b_fx = X_fx->b_fx[k];
		move16();

		#ifdef BASOP_NOGLOB
		en_fx = L_mac0_o(en_fx, temp_a_fx, temp_a_fx, &Overflow);
		en_fx = L_mac0_o(en_fx, temp_b_fx, temp_b_fx, &Overflow);
		#else
		en_fx = L_mac0(en_fx, temp_a_fx, temp_a_fx);
		en_fx = L_mac0(en_fx, temp_b_fx, temp_b_fx);
		#endif
		}

		return en_fx ; /* 2X1.Q+1=Q13 /
		@@ -1617,21 +1673,37 @@ Word32 DTFS_getEngy_P2A_fx(
		{
		Word16 k, HalfLag_fx;
		Word32 en_fx=0;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif
		HalfLag_fx = shr(sub(X_fx->lag_fx, 1), 1);
		HalfLag_fx = s_min(HalfLag_fx,X_fx->nH_fx);
		FOR (k=1; k<=HalfLag_fx; k++)
		{
		#ifdef BASOP_NOGLOB
		en_fx = L_mac0_o(en_fx, X_fx->a_fx[k], X_fx->a_fx[k], &Overflow);
		en_fx = L_mac0_o(en_fx, X_fx->b_fx[k], X_fx->b_fx[k], &Overflow);
		#else
		en_fx = L_mac0(en_fx, X_fx->a_fx[k], X_fx->a_fx[k]);
		en_fx = L_mac0(en_fx, X_fx->b_fx[k], X_fx->b_fx[k]);
		#endif
		}
		en_fx = L_shr(en_fx, 1);
		#ifdef BASOP_NOGLOB
		en_fx = L_mac0_o(en_fx,X_fx->a_fx[0], X_fx->a_fx[0], &Overflow);
		#else
		en_fx = L_mac0(en_fx,X_fx->a_fx[0], X_fx->a_fx[0]);
		#endif
		/* IF (X_fx->lag_fx%2 == 0) */
		IF (s_and(X_fx->lag_fx,1) == 0)
		{
		#ifdef BASOP_NOGLOB
		en_fx = L_mac0_o(en_fx, X_fx->a_fx[k], X_fx->a_fx[k], &Overflow);
		en_fx = L_mac0_o(en_fx, X_fx->b_fx[k], X_fx->b_fx[k], &Overflow);
		#else
		en_fx = L_mac0(en_fx, X_fx->a_fx[k], X_fx->a_fx[k]);
		en_fx = L_mac0(en_fx, X_fx->b_fx[k], X_fx->b_fx[k]);
		#endif
		}

		return en_fx ; /* 2X1.Q /
		@@ -3483,7 +3555,9 @@ void DTFS_peaktoaverage_fx( DTFS_STRUCTURE_FX X_fx,Word32 pos_fx, Word16 Qpos,
		Word16 expa,expb,fraca,fracb,scale;
		Word16 exp,tmp;
		Word32 L_tmp;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif
		L_sum=DTFS_getEngy_P2A_fx(&X_fx); /* 2Q */
		DTFS_fast_fs_inv_fx(&X_fx,time_fx , 256, 8);

		@@ -3524,17 +3598,29 @@ void DTFS_peaktoaverage_fx( DTFS_STRUCTURE_FX X_fx,Word32 pos_fx, Word16 Qpos,
		ELSE
		{
		expa = norm_l(maxPosEn_fx);
		#ifdef BASOP_NOGLOB
		fraca = extract_h(L_shl_o(maxPosEn_fx,expa, &Overflow));
		#else
		fraca = extract_h(L_shl(maxPosEn_fx,expa));
		#endif
		expa = sub(30,add(expa,1));


		expb = norm_l(L_sum);
		#ifdef BASOP_NOGLOB
		fracb = round_fx_o(L_shl_o(L_sum,expb, &Overflow), &Overflow);
		#else
		fracb = round_fx(L_shl(L_sum,expb));
		#endif
		expb = sub(30,add(expb,shl(X_fx.Q,1)));


		scale = shr(sub(fraca,fracb),15);
		#ifdef BASOP_NOGLOB
		fracb = shl_o(fracb,scale, &Overflow);
		#else
		fracb = shl(fracb,scale);
		#endif
		expb = sub(expb,scale);

		tmp = div_s(fracb,fraca);
		@@ -3555,17 +3641,29 @@ void DTFS_peaktoaverage_fx( DTFS_STRUCTURE_FX X_fx,Word32 pos_fx, Word16 Qpos,
		{

		expa = norm_l(maxNegEn_fx);
		#ifdef BASOP_NOGLOB
		fraca = extract_h(L_shl_o(maxNegEn_fx,expa, &Overflow));
		#else
		fraca = extract_h(L_shl(maxNegEn_fx,expa));
		#endif
		expa = 30-expa- 1;


		expb = norm_l(L_sum);
		#ifdef BASOP_NOGLOB
		fracb = round_fx_o(L_shl_o(L_sum,expb, &Overflow), &Overflow);
		#else
		fracb = round_fx(L_shl(L_sum,expb));
		#endif
		expb = 30-expb - (2*X_fx.Q);


		scale = shr(sub(fraca,fracb),15);
		#ifdef BASOP_NOGLOB
		fracb = shl_o(fracb,scale, &Overflow);
		#else
		fracb = shl(fracb,scale);
		#endif
		expb = sub(expb,scale);

		tmp = div_s(fracb,fraca);

lib_enc/ppp_enc_fx.c

+24 −2

Original line number	Diff line number	Diff line
		@@ -658,11 +658,12 @@ static Word16 DTFS_alignment_fine_new_fx(DTFS_STRUCTURE_FX X1_fx, DTFS_STRUCTURE
		#ifdef BASOP_NOGLOB
		corr_fx = L_mac_o(corr_fx, ab1[k], C_fx[temp % (4 * X2_fx.lag_fx)], &Overflow);
		corr_fx = L_mac_o(corr_fx, ab2[k], S_fx[temp % (4 * X2_fx.lag_fx)], &Overflow);
		temp = add_o(temp, temp1, &Overflow);
		#else
		corr_fx = L_mac(corr_fx, ab1[k], C_fx[temp % (4 * X2_fx.lag_fx)]);
		corr_fx = L_mac(corr_fx, ab2[k], S_fx[temp % (4 * X2_fx.lag_fx)]);
		#endif
		temp = add(temp, temp1);
		#endif
		}
		temp = sub(32767, extract_l(L_shr(L_mult(82, abs_s(n)), 1))); /* Q15 */
		Qcorr = norm_l(corr_fx);
		@@ -753,7 +754,9 @@ static void LPCPowSpect_fx(
		Word32 Ltemp, Lw;
		Word32 Lacc;
		Word16 tmp, exp;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif
		FOR(k = 0; k < Nf; k++)
		{

		@@ -784,9 +787,15 @@ static void LPCPowSpect_fx(

		t1 = add(t1, (Word16)L_shr(Ltemp, 16)); /* t1 is interpolated cos(w) */
		Ltemp = L_shr(L_mult(LPC[i], t1), 1); /* Ltemp in Q27 */
		#ifdef BASOP_NOGLOB
		Re = L_add_o(Re, Ltemp, &Overflow); /* Re=1-sum(LPC[i]cos(Lw)); /
		Ltemp = L_add_o(Lw, 0x6000, &Overflow); /* add 0.75, which is 3pi/2 to convert sin to cos */
		Ltemp = L_shl_o(Ltemp, 10, &Overflow); /* Q25 */
		#else
		Re = L_add(Re, Ltemp); /* Re=1-sum(LPC[i]cos(Lw)); /
		Ltemp = L_add(Lw, 0x6000); /* add 0.75, which is 3pi/2 to convert sin to cos */
		Ltemp = L_shl(Ltemp, 10); /* Q25 */
		#endif
		w = extract_h(Ltemp); /* w is equivalent cos index */
		dl = extract_l(Ltemp); /* dl is 6 bit left-over for interpolation */
		w = s_and(w, 511);
		@@ -808,8 +817,13 @@ static void LPCPowSpect_fx(

		t1 = add(t1, (Word16)L_shr(Ltemp, 16)); /* t1 is interpolated cos(w) */
		Ltemp = L_shr(L_mult(LPC[i], t1), 1); /* Ltemp in Q27 */
		#ifdef BASOP_NOGLOB
		Im = L_sub_o(Im, Ltemp, &Overflow); /* Im=sum(LPC[i]sin(Lw)) /
		Lw = L_add_o(Lw, freq[k], &Overflow); /* Lw=(i+1)freq[k] /
		#else
		Im = L_sub(Im, Ltemp); /* Im=sum(LPC[i]sin(Lw)) /
		Lw = L_add(Lw, freq[k]); /* Lw=(i+1)freq[k] /
		#endif
		}
		/* If necessary, we can block-normalize Re and Im to improve precision */
		dh = extract_h(Re);
		@@ -824,7 +838,11 @@ static void LPCPowSpect_fx(
		ELSE
		Lacc = L_mult0(dh, dl);

		#ifdef BASOP_NOGLOB
		Lacc = L_add_o(L_shr(Lacc, 15), L_shr(L_mult_o(dh, dh, &Overflow), 1), &Overflow); /* Lacc=ReRe /
		#else
		Lacc = L_add(L_shr(Lacc, 15), L_shr(L_mult(dh, dh), 1)); /* Lacc=ReRe /
		#endif
		dh = extract_h(Im);
		dl = extract_l(Im);

		@@ -855,7 +873,11 @@ static void LPCPowSpect_fx(
		move16();
		}
		Ltemp = L_deposit_h(tmp);
		#ifdef BASOP_NOGLOB
		out[k] = round_fx_o(L_shl_o(Ltemp, negate(add(exp, 8)), &Overflow), &Overflow);
		#else
		out[k] = round_fx(L_shl(Ltemp, negate(add(exp, 8))));
		#endif

		/* out[k] = shl(tmp,-exp-8); in Q7 */

lib_enc/voiced_enc_fx.c

+23 −1

Original line number	Diff line number	Diff line
		@@ -738,7 +738,11 @@ ivas_error ppp_voiced_encoder_fx(
		Ltemp1 = L_negate(Ltemp1);
		}
		expa = norm_l(Ltemp2);
		#ifdef BASOP_NOGLOB
		fraca = extract_h(L_shl_o(Ltemp2,expa, &Overflow));
		#else
		fraca = extract_h(L_shl(Ltemp2,expa));
		#endif
		expa = sub(30,expa);

		expb = norm_l(Ltemp1);
		@@ -746,12 +750,20 @@ ivas_error ppp_voiced_encoder_fx(
		expb = sub(30, add(expb, Qadj));

		scale = shr(sub(fraca,fracb),15);
		#ifdef BASOP_NOGLOB
		fracb = shl_o(fracb,scale, &Overflow);
		#else
		fracb = shl(fracb,scale);
		#endif
		expb = sub(expb,scale);

		tmp = div_s(fracb,fraca);
		exp = sub(expb,expa);
		#ifdef BASOP_NOGLOB
		Ltemp_fx = L_shl_o(tmp, add(exp,14), &Overflow);
		#else
		Ltemp_fx = L_shl(tmp, add(exp,14));
		#endif
		/-------------------------------------------/
		sp_hb_enratio_fx = L_add(Ltemp_fx, 0); /* Q29 */
		}
		@@ -1293,7 +1305,11 @@ static void synthesis_filter_fx (Word16 b[], Word16 x[], Word16 y[], Word16 buf[
		FOR (j = P - 1; j > 0; j--)
		{
		/* acc = L_sub(acc, L_mult(memory[j], coef[j])); */
		#ifdef BASOP_NOGLOB
		acc = L_msu_o(acc, buf[j], b[j], &Overflow);/Q13 /
		#else
		acc = L_msu(acc, buf[j], b[j]);/Q13 /
		#endif
		buf[j] = buf[j - 1];
		move16();
		}
		@@ -1494,7 +1510,9 @@ static Word16 DTFS_alignment_extract_td_fx(Word16* x1, Word16* x2, Word16 lag)
		{
		Word16 j, k, idx, Adiff_fx;
		Word32 maxcorr, corr;

		#ifdef BASOP_NOGLOB
		Flag Overflow = 0;
		#endif
		maxcorr = L_add(MIN_32, 0);
		Adiff_fx = (Word16)(s_max(4, shr(lag, 3)));

		@@ -1505,7 +1523,11 @@ static Word16 DTFS_alignment_extract_td_fx(Word16* x1, Word16* x2, Word16 lag)
		corr = L_deposit_l(0);
		FOR(k = 0; k < lag; k++)
		{
		#ifdef BASOP_NOGLOB
		corr = L_mac_o(corr, x1[k], x2[(k - j + lag) % lag], &Overflow);
		#else
		corr = L_mac(corr, x1[k], x2[(k - j + lag) % lag]);
		#endif
		}
		if (GT_32(corr, maxcorr))
		{