c2t64fx.c revision d23f7ef608b3ef917a6f9b371feebc98b74ee485 
1/*
2 ** Copyright 2003-2010, VisualOn, Inc.
3 **
4 ** Licensed under the Apache License, Version 2.0 (the "License");
5 ** you may not use this file except in compliance with the License.
6 ** You may obtain a copy of the License at
7 **
8 **     http://www.apache.org/licenses/LICENSE-2.0
9 **
10 ** Unless required by applicable law or agreed to in writing, software
11 ** distributed under the License is distributed on an "AS IS" BASIS,
12 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 ** See the License for the specific language governing permissions and
14 ** limitations under the License.
15 */
16
17/************************************************************************
18*      File: c2t64fx.c                                                  *
19*                                                                       *
20*	   Description:Performs algebraic codebook search for 6.60kbits mode*
21*                                                                       *
22*************************************************************************/
23
24#include "typedef.h"
25#include "basic_op.h"
26#include "math_op.h"
27#include "acelp.h"
28#include "cnst.h"
29
30#define NB_TRACK  2
31#define STEP      2
32#define NB_POS    32
33#define MSIZE     1024
34
35/*************************************************************************
36* Function:  ACELP_2t64_fx()                                             *
37*                                                                        *
38* 12 bits algebraic codebook.                                            *
39* 2 tracks x 32 positions per track = 64 samples.                        *
40*                                                                        *
41* 12 bits --> 2 pulses in a frame of 64 samples.                         *
42*                                                                        *
43* All pulses can have two (2) possible amplitudes: +1 or -1.             *
44* Each pulse can have 32 possible positions.                             *
45**************************************************************************/
46
47void ACELP_2t64_fx(
48		Word16 dn[],                          /* (i) <12b : correlation between target x[] and H[]      */
49		Word16 cn[],                          /* (i) <12b : residual after long term prediction         */
50		Word16 H[],                           /* (i) Q12: impulse response of weighted synthesis filter */
51		Word16 code[],                        /* (o) Q9 : algebraic (fixed) codebook excitation         */
52		Word16 y[],                           /* (o) Q9 : filtered fixed codebook excitation            */
53		Word16 * index                        /* (o) : index (12): 5+1+5+1 = 11 bits.                   */
54		)
55{
56	Word32 i, j, k, i0, i1, ix, iy, pos, pos2;
57	Word16 ps, psk, ps1, ps2, alpk, alp1, alp2, sq;
58	Word16 alp, val, exp, k_cn, k_dn;
59	Word16 *p0, *p1, *p2, *psign;
60	Word16 *h, *h_inv, *ptr_h1, *ptr_h2, *ptr_hf;
61
62	Word16 sign[L_SUBFR], vec[L_SUBFR], dn2[L_SUBFR];
63	Word16 h_buf[4 * L_SUBFR] = {0};
64	Word16 rrixix[NB_TRACK][NB_POS];
65	Word16 rrixiy[MSIZE];
66	Word32 s, cor;
67
68	/*----------------------------------------------------------------*
69	 * Find sign for each pulse position.                             *
70	 *----------------------------------------------------------------*/
71	alp = 8192;                              /* alp = 2.0 (Q12) */
72
73	/* calculate energy for normalization of cn[] and dn[] */
74	/* set k_cn = 32..32767 (ener_cn = 2^30..256-0) */
75#ifdef ASM_OPT             /* asm optimization branch */
76	s = Dot_product12_asm(cn, cn, L_SUBFR, &exp);
77#else
78	s = Dot_product12(cn, cn, L_SUBFR, &exp);
79#endif
80
81	Isqrt_n(&s, &exp);
82	s = L_shl(s, add1(exp, 5));
83	if (s > INT_MAX - 0x8000) {
84		s = INT_MAX - 0x8000;
85	}
86	k_cn = vo_round(s);
87
88	/* set k_dn = 32..512 (ener_dn = 2^30..2^22) */
89#ifdef ASM_OPT                  /* asm optimization branch */
90	s = Dot_product12_asm(dn, dn, L_SUBFR, &exp);
91#else
92	s = Dot_product12(dn, dn, L_SUBFR, &exp);
93#endif
94
95	Isqrt_n(&s, &exp);
96	k_dn = vo_round(L_shl(s, (exp + 8)));    /* k_dn = 256..4096 */
97	k_dn = vo_mult_r(alp, k_dn);              /* alp in Q12 */
98
99	/* mix normalized cn[] and dn[] */
100	p0 = cn;
101	p1 = dn;
102	p2 = dn2;
103
104	for (i = 0; i < L_SUBFR/4; i++)
105	{
106		s = (k_cn* (*p0++))+(k_dn * (*p1++));
107		*p2++ = s >> 7;
108		s = (k_cn* (*p0++))+(k_dn * (*p1++));
109		*p2++ = s >> 7;
110		s = (k_cn* (*p0++))+(k_dn * (*p1++));
111		*p2++ = s >> 7;
112		s = (k_cn* (*p0++))+(k_dn * (*p1++));
113		*p2++ = s >> 7;
114	}
115
116	/* set sign according to dn2[] = k_cn*cn[] + k_dn*dn[]    */
117	for (i = 0; i < L_SUBFR; i ++)
118	{
119		val = dn[i];
120		ps = dn2[i];
121		if (ps >= 0)
122		{
123			sign[i] = 32767;             /* sign = +1 (Q12) */
124			vec[i] = -32768;
125		} else
126		{
127			sign[i] = -32768;            /* sign = -1 (Q12) */
128			vec[i] = 32767;
129			dn[i] = -val;
130		}
131	}
132	/*------------------------------------------------------------*
133	 * Compute h_inv[i].                                          *
134	 *------------------------------------------------------------*/
135	/* impulse response buffer for fast computation */
136	h = h_buf + L_SUBFR;
137	h_inv = h + (L_SUBFR<<1);
138
139	for (i = 0; i < L_SUBFR; i++)
140	{
141		h[i] = H[i];
142		h_inv[i] = vo_negate(h[i]);
143	}
144
145	/*------------------------------------------------------------*
146	 * Compute rrixix[][] needed for the codebook search.         *
147	 * Result is multiplied by 0.5                                *
148	 *------------------------------------------------------------*/
149	/* Init pointers to last position of rrixix[] */
150	p0 = &rrixix[0][NB_POS - 1];
151	p1 = &rrixix[1][NB_POS - 1];
152
153	ptr_h1 = h;
154	cor = 0x00010000L;                          /* for rounding */
155	for (i = 0; i < NB_POS; i++)
156	{
157		cor += ((*ptr_h1) * (*ptr_h1) << 1);
158		ptr_h1++;
159		*p1-- = (extract_h(cor) >> 1);
160		cor += ((*ptr_h1) * (*ptr_h1) << 1);
161		ptr_h1++;
162		*p0-- = (extract_h(cor) >> 1);
163	}
164
165	/*------------------------------------------------------------*
166	 * Compute rrixiy[][] needed for the codebook search.         *
167	 *------------------------------------------------------------*/
168	pos = MSIZE - 1;
169	pos2 = MSIZE - 2;
170	ptr_hf = h + 1;
171
172	for (k = 0; k < NB_POS; k++)
173	{
174		p1 = &rrixiy[pos];
175		p0 = &rrixiy[pos2];
176		cor = 0x00008000L;                        /* for rounding */
177		ptr_h1 = h;
178		ptr_h2 = ptr_hf;
179
180		for (i = (k + 1); i < NB_POS; i++)
181		{
182			cor += ((*ptr_h1) * (*ptr_h2))<<1;
183			ptr_h1++;
184			ptr_h2++;
185			*p1 = extract_h(cor);
186			cor += ((*ptr_h1) * (*ptr_h2))<<1;
187			ptr_h1++;
188			ptr_h2++;
189			*p0 = extract_h(cor);
190
191			p1 -= (NB_POS + 1);
192			p0 -= (NB_POS + 1);
193		}
194		cor += ((*ptr_h1) * (*ptr_h2))<<1;
195		ptr_h1++;
196		ptr_h2++;
197		*p1 = extract_h(cor);
198
199		pos -= NB_POS;
200		pos2--;
201		ptr_hf += STEP;
202	}
203
204	/*------------------------------------------------------------*
205	 * Modification of rrixiy[][] to take signs into account.     *
206	 *------------------------------------------------------------*/
207	p0 = rrixiy;
208	for (i = 0; i < L_SUBFR; i += STEP)
209	{
210		psign = sign;
211		if (psign[i] < 0)
212		{
213			psign = vec;
214		}
215		for (j = 1; j < L_SUBFR; j += STEP)
216		{
217			*p0 = vo_mult(*p0, psign[j]);
218			p0++;
219		}
220	}
221	/*-------------------------------------------------------------------*
222	 * search 2 pulses:                                                  *
223	 * ~@~~~~~~~~~~~~~~                                                  *
224	 * 32 pos x 32 pos = 1024 tests (all combinaisons is tested)         *
225	 *-------------------------------------------------------------------*/
226	p0 = rrixix[0];
227	p1 = rrixix[1];
228	p2 = rrixiy;
229
230	psk = -1;
231	alpk = 1;
232	ix = 0;
233	iy = 1;
234
235	for (i0 = 0; i0 < L_SUBFR; i0 += STEP)
236	{
237		ps1 = dn[i0];
238		alp1 = (*p0++);
239		pos = -1;
240		for (i1 = 1; i1 < L_SUBFR; i1 += STEP)
241		{
242			ps2 = add1(ps1, dn[i1]);
243			alp2 = add1(alp1, add1(*p1++, *p2++));
244			sq = vo_mult(ps2, ps2);
245			s = vo_L_mult(alpk, sq) - ((psk * alp2)<<1);
246			if (s > 0)
247			{
248				psk = sq;
249				alpk = alp2;
250				pos = i1;
251			}
252		}
253		p1 -= NB_POS;
254		if (pos >= 0)
255		{
256			ix = i0;
257			iy = pos;
258		}
259	}
260	/*-------------------------------------------------------------------*
261	 * Build the codeword, the filtered codeword and index of codevector.*
262	 *-------------------------------------------------------------------*/
263
264	for (i = 0; i < L_SUBFR; i++)
265	{
266		code[i] = 0;
267	}
268
269	i0 = (ix >> 1);                       /* pos of pulse 1 (0..31) */
270	i1 = (iy >> 1);                       /* pos of pulse 2 (0..31) */
271	if (sign[ix] > 0)
272	{
273		code[ix] = 512;                     /* codeword in Q9 format */
274		p0 = h - ix;
275	} else
276	{
277		code[ix] = -512;
278		i0 += NB_POS;
279		p0 = h_inv - ix;
280	}
281	if (sign[iy] > 0)
282	{
283		code[iy] = 512;
284		p1 = h - iy;
285	} else
286	{
287		code[iy] = -512;
288		i1 += NB_POS;
289		p1 = h_inv - iy;
290	}
291	*index = add1((i0 << 6), i1);
292	for (i = 0; i < L_SUBFR; i++)
293	{
294		y[i] = vo_shr_r(add1((*p0++), (*p1++)), 3);
295	}
296	return;
297}
298
299
300
301