1/*
2 * Copyright (c) 2001, 2002 Fabrice Bellard
3 *
4 * This file is part of FFmpeg.
5 *
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#include <stdint.h>
22
23#include "libavutil/attributes.h"
24#include "libavutil/mem_internal.h"
25#include "libavutil/thread.h"
26
27#include "dct32.h"
28#include "mathops.h"
29#include "mpegaudiodsp.h"
30#include "mpegaudio.h"
31
32#if USE_FLOATS
33#define RENAME(n) n##_float
34
35static inline float round_sample(float *sum)
36{
37    float sum1=*sum;
38    *sum = 0;
39    return sum1;
40}
41
42#define MACS(rt, ra, rb) rt+=(ra)*(rb)
43#define MULS(ra, rb) ((ra)*(rb))
44#define MULH3(x, y, s) ((s)*(y)*(x))
45#define MLSS(rt, ra, rb) rt-=(ra)*(rb)
46#define MULLx(x, y, s) ((y)*(x))
47#define FIXHR(x)        ((float)(x))
48#define FIXR(x)        ((float)(x))
49#define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
50
51#else
52
53#define RENAME(n) n##_fixed
54#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
55
56static inline int round_sample(int64_t *sum)
57{
58    int sum1;
59    sum1 = (int)((*sum) >> OUT_SHIFT);
60    *sum &= (1<<OUT_SHIFT)-1;
61    return av_clip_int16(sum1);
62}
63
64#   define MULS(ra, rb) MUL64(ra, rb)
65#   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
66#   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
67#   define MULH3(x, y, s) MULH((s)*(x), y)
68#   define MULLx(x, y, s) MULL((int)(x),(y),s)
69#   define SHR(a,b)       (((int)(a))>>(b))
70#   define FIXR(a)        ((int)((a) * FRAC_ONE + 0.5))
71#   define FIXHR(a)       ((int)((a) * (1LL<<32) + 0.5))
72#endif
73
74/** Window for MDCT. Actually only the elements in [0,17] and
75    [MDCT_BUF_SIZE/2, MDCT_BUF_SIZE/2 + 17] are actually used. The rest
76    is just to preserve alignment for SIMD implementations.
77*/
78DECLARE_ALIGNED(16, INTFLOAT, RENAME(ff_mdct_win))[8][MDCT_BUF_SIZE];
79
80DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
81
82#define SUM8(op, sum, w, p)               \
83{                                         \
84    op(sum, (w)[0 * 64], (p)[0 * 64]);    \
85    op(sum, (w)[1 * 64], (p)[1 * 64]);    \
86    op(sum, (w)[2 * 64], (p)[2 * 64]);    \
87    op(sum, (w)[3 * 64], (p)[3 * 64]);    \
88    op(sum, (w)[4 * 64], (p)[4 * 64]);    \
89    op(sum, (w)[5 * 64], (p)[5 * 64]);    \
90    op(sum, (w)[6 * 64], (p)[6 * 64]);    \
91    op(sum, (w)[7 * 64], (p)[7 * 64]);    \
92}
93
94#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
95{                                               \
96    INTFLOAT tmp;\
97    tmp = p[0 * 64];\
98    op1(sum1, (w1)[0 * 64], tmp);\
99    op2(sum2, (w2)[0 * 64], tmp);\
100    tmp = p[1 * 64];\
101    op1(sum1, (w1)[1 * 64], tmp);\
102    op2(sum2, (w2)[1 * 64], tmp);\
103    tmp = p[2 * 64];\
104    op1(sum1, (w1)[2 * 64], tmp);\
105    op2(sum2, (w2)[2 * 64], tmp);\
106    tmp = p[3 * 64];\
107    op1(sum1, (w1)[3 * 64], tmp);\
108    op2(sum2, (w2)[3 * 64], tmp);\
109    tmp = p[4 * 64];\
110    op1(sum1, (w1)[4 * 64], tmp);\
111    op2(sum2, (w2)[4 * 64], tmp);\
112    tmp = p[5 * 64];\
113    op1(sum1, (w1)[5 * 64], tmp);\
114    op2(sum2, (w2)[5 * 64], tmp);\
115    tmp = p[6 * 64];\
116    op1(sum1, (w1)[6 * 64], tmp);\
117    op2(sum2, (w2)[6 * 64], tmp);\
118    tmp = p[7 * 64];\
119    op1(sum1, (w1)[7 * 64], tmp);\
120    op2(sum2, (w2)[7 * 64], tmp);\
121}
122
123void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window,
124                                  int *dither_state, OUT_INT *samples,
125                                  ptrdiff_t incr)
126{
127    register const MPA_INT *w, *w2, *p;
128    int j;
129    OUT_INT *samples2;
130#if USE_FLOATS
131    float sum, sum2;
132#else
133    int64_t sum, sum2;
134#endif
135
136    /* copy to avoid wrap */
137    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
138
139    samples2 = samples + 31 * incr;
140    w = window;
141    w2 = window + 31;
142
143    sum = *dither_state;
144    p = synth_buf + 16;
145    SUM8(MACS, sum, w, p);
146    p = synth_buf + 48;
147    SUM8(MLSS, sum, w + 32, p);
148    *samples = round_sample(&sum);
149    samples += incr;
150    w++;
151
152    /* we calculate two samples at the same time to avoid one memory
153       access per two sample */
154    for(j=1;j<16;j++) {
155        sum2 = 0;
156        p = synth_buf + 16 + j;
157        SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
158        p = synth_buf + 48 - j;
159        SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
160
161        *samples = round_sample(&sum);
162        samples += incr;
163        sum += sum2;
164        *samples2 = round_sample(&sum);
165        samples2 -= incr;
166        w++;
167        w2--;
168    }
169
170    p = synth_buf + 32;
171    SUM8(MLSS, sum, w + 32, p);
172    *samples = round_sample(&sum);
173    *dither_state= sum;
174}
175
176/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
177   32 samples. */
178void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr,
179                                 int *synth_buf_offset,
180                                 MPA_INT *window, int *dither_state,
181                                 OUT_INT *samples, ptrdiff_t incr,
182                                 MPA_INT *sb_samples)
183{
184    MPA_INT *synth_buf;
185    int offset;
186
187    offset = *synth_buf_offset;
188    synth_buf = synth_buf_ptr + offset;
189
190    s->RENAME(dct32)(synth_buf, sb_samples);
191    s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr);
192
193    offset = (offset - 32) & 511;
194    *synth_buf_offset = offset;
195}
196
197static av_cold void mpa_synth_init(MPA_INT *window)
198{
199    int i, j;
200
201    /* max = 18760, max sum over all 16 coefs : 44736 */
202    for(i=0;i<257;i++) {
203        INTFLOAT v;
204        v = ff_mpa_enwindow[i];
205#if USE_FLOATS
206        v *= 1.0 / (1LL<<(16 + FRAC_BITS));
207#endif
208        window[i] = v;
209        if ((i & 63) != 0)
210            v = -v;
211        if (i != 0)
212            window[512 - i] = v;
213    }
214
215
216    // Needed for avoiding shuffles in ASM implementations
217    for(i=0; i < 8; i++)
218        for(j=0; j < 16; j++)
219            window[512+16*i+j] = window[64*i+32-j];
220
221    for(i=0; i < 8; i++)
222        for(j=0; j < 16; j++)
223            window[512+128+16*i+j] = window[64*i+48-j];
224}
225
226static av_cold void mpa_synth_window_init(void)
227{
228    mpa_synth_init(RENAME(ff_mpa_synth_window));
229}
230
231av_cold void RENAME(ff_mpa_synth_init)(void)
232{
233    static AVOnce init_static_once = AV_ONCE_INIT;
234    ff_thread_once(&init_static_once, mpa_synth_window_init);
235}
236
237/* cos(pi*i/18) */
238#define C1 FIXHR(0.98480775301220805936/2)
239#define C2 FIXHR(0.93969262078590838405/2)
240#define C3 FIXHR(0.86602540378443864676/2)
241#define C4 FIXHR(0.76604444311897803520/2)
242#define C5 FIXHR(0.64278760968653932632/2)
243#define C6 FIXHR(0.5/2)
244#define C7 FIXHR(0.34202014332566873304/2)
245#define C8 FIXHR(0.17364817766693034885/2)
246
247/* 0.5 / cos(pi*(2*i+1)/36) */
248static const INTFLOAT icos36[9] = {
249    FIXR(0.50190991877167369479),
250    FIXR(0.51763809020504152469), //0
251    FIXR(0.55168895948124587824),
252    FIXR(0.61038729438072803416),
253    FIXR(0.70710678118654752439), //1
254    FIXR(0.87172339781054900991),
255    FIXR(1.18310079157624925896),
256    FIXR(1.93185165257813657349), //2
257    FIXR(5.73685662283492756461),
258};
259
260/* 0.5 / cos(pi*(2*i+1)/36) */
261static const INTFLOAT icos36h[9] = {
262    FIXHR(0.50190991877167369479/2),
263    FIXHR(0.51763809020504152469/2), //0
264    FIXHR(0.55168895948124587824/2),
265    FIXHR(0.61038729438072803416/2),
266    FIXHR(0.70710678118654752439/2), //1
267    FIXHR(0.87172339781054900991/2),
268    FIXHR(1.18310079157624925896/4),
269    FIXHR(1.93185165257813657349/4), //2
270//    FIXHR(5.73685662283492756461),
271};
272
273/* using Lee like decomposition followed by hand coded 9 points DCT */
274static void imdct36(INTFLOAT *out, INTFLOAT *buf, SUINTFLOAT *in, INTFLOAT *win)
275{
276    int i, j;
277    SUINTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
278    SUINTFLOAT tmp[18], *tmp1, *in1;
279
280    for (i = 17; i >= 1; i--)
281        in[i] += in[i-1];
282    for (i = 17; i >= 3; i -= 2)
283        in[i] += in[i-2];
284
285    for (j = 0; j < 2; j++) {
286        tmp1 = tmp + j;
287        in1 = in + j;
288
289        t2 = in1[2*4] + in1[2*8] - in1[2*2];
290
291        t3 = in1[2*0] + SHR(in1[2*6],1);
292        t1 = in1[2*0] - in1[2*6];
293        tmp1[ 6] = t1 - SHR(t2,1);
294        tmp1[16] = t1 + t2;
295
296        t0 = MULH3(in1[2*2] + in1[2*4] ,    C2, 2);
297        t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
298        t2 = MULH3(in1[2*2] + in1[2*8] ,   -C4, 2);
299
300        tmp1[10] = t3 - t0 - t2;
301        tmp1[ 2] = t3 + t0 + t1;
302        tmp1[14] = t3 + t2 - t1;
303
304        tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
305        t2 = MULH3(in1[2*1] + in1[2*5],    C1, 2);
306        t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
307        t0 = MULH3(in1[2*3], C3, 2);
308
309        t1 = MULH3(in1[2*1] + in1[2*7],   -C5, 2);
310
311        tmp1[ 0] = t2 + t3 + t0;
312        tmp1[12] = t2 + t1 - t0;
313        tmp1[ 8] = t3 - t1 - t0;
314    }
315
316    i = 0;
317    for (j = 0; j < 4; j++) {
318        t0 = tmp[i];
319        t1 = tmp[i + 2];
320        s0 = t1 + t0;
321        s2 = t1 - t0;
322
323        t2 = tmp[i + 1];
324        t3 = tmp[i + 3];
325        s1 = MULH3(t3 + t2, icos36h[    j], 2);
326        s3 = MULLx(t3 - t2, icos36 [8 - j], FRAC_BITS);
327
328        t0 = s0 + s1;
329        t1 = s0 - s1;
330        out[(9 + j) * SBLIMIT] = MULH3(t1, win[     9 + j], 1) + buf[4*(9 + j)];
331        out[(8 - j) * SBLIMIT] = MULH3(t1, win[     8 - j], 1) + buf[4*(8 - j)];
332        buf[4 * ( 9 + j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + j], 1);
333        buf[4 * ( 8 - j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - j], 1);
334
335        t0 = s2 + s3;
336        t1 = s2 - s3;
337        out[(9 + 8 - j) * SBLIMIT] = MULH3(t1, win[     9 + 8 - j], 1) + buf[4*(9 + 8 - j)];
338        out[         j  * SBLIMIT] = MULH3(t1, win[             j], 1) + buf[4*(        j)];
339        buf[4 * ( 9 + 8 - j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 8 - j], 1);
340        buf[4 * (         j     )] = MULH3(t0, win[MDCT_BUF_SIZE/2         + j], 1);
341        i += 4;
342    }
343
344    s0 = tmp[16];
345    s1 = MULH3(tmp[17], icos36h[4], 2);
346    t0 = s0 + s1;
347    t1 = s0 - s1;
348    out[(9 + 4) * SBLIMIT] = MULH3(t1, win[     9 + 4], 1) + buf[4*(9 + 4)];
349    out[(8 - 4) * SBLIMIT] = MULH3(t1, win[     8 - 4], 1) + buf[4*(8 - 4)];
350    buf[4 * ( 9 + 4     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 4], 1);
351    buf[4 * ( 8 - 4     )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - 4], 1);
352}
353
354void RENAME(ff_imdct36_blocks)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,
355                               int count, int switch_point, int block_type)
356{
357    int j;
358    for (j=0 ; j < count; j++) {
359        /* apply window & overlap with previous buffer */
360
361        /* select window */
362        int win_idx = (switch_point && j < 2) ? 0 : block_type;
363        INTFLOAT *win = RENAME(ff_mdct_win)[win_idx + (4 & -(j & 1))];
364
365        imdct36(out, buf, in, win);
366
367        in  += 18;
368        buf += ((j&3) != 3 ? 1 : (72-3));
369        out++;
370    }
371}
372
373