Mercurial > libavcodec.hg
annotate wmadec.c @ 11679:fe1e6103b68d libavcodec
Align plane size to word-boundary.
Patch by Sebastian Vater, cdgs D basty A googlemail
author | cehoyos |
---|---|
date | Wed, 05 May 2010 17:02:55 +0000 |
parents | 7dd2a45249a9 |
children | ab330d7112ad |
rev | line source |
---|---|
783 | 1 /* |
2 * WMA compatible decoder | |
8629
04423b2f6e0b
cosmetics: Remove pointless period after copyright statement non-sentences.
diego
parents:
7712
diff
changeset
|
3 * Copyright (c) 2002 The FFmpeg Project |
783 | 4 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
5 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
6 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
7 * FFmpeg is free software; you can redistribute it and/or |
783 | 8 * modify it under the terms of the GNU Lesser General Public |
9 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
10 * version 2.1 of the License, or (at your option) any later version. |
783 | 11 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
12 * FFmpeg is distributed in the hope that it will be useful, |
783 | 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 * Lesser General Public License for more details. | |
16 * | |
17 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3776
diff
changeset
|
18 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
3022
diff
changeset
|
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
783 | 20 */ |
1106 | 21 |
22 /** | |
11644
7dd2a45249a9
Remove explicit filename from Doxygen @file commands.
diego
parents:
11560
diff
changeset
|
23 * @file |
1106 | 24 * WMA compatible decoder. |
1967
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
25 * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2. |
2967 | 26 * WMA v1 is identified by audio format 0x160 in Microsoft media files |
1967
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
27 * (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161. |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
28 * |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
29 * To use this decoder, a calling application must supply the extra data |
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
30 * bytes provided with the WMA data. These are the extra, codec-specific |
2967 | 31 * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes |
32 * to the decoder using the extradata[_size] fields in AVCodecContext. There | |
1967
2b0fc6b25ab8
add the minimal documentation to make this decoder useful
melanson
parents:
1750
diff
changeset
|
33 * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data. |
1106 | 34 */ |
35 | |
783 | 36 #include "avcodec.h" |
4490 | 37 #include "wma.h" |
783 | 38 |
4490 | 39 #undef NDEBUG |
40 #include <assert.h> | |
3113 | 41 |
42 #define EXPVLCBITS 8 | |
43 #define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS) | |
44 | |
45 #define HGAINVLCBITS 9 | |
46 #define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS) | |
3022 | 47 |
4601 | 48 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len); |
783 | 49 |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
50 #ifdef TRACE |
5517 | 51 static void dump_shorts(WMACodecContext *s, const char *name, const short *tab, int n) |
783 | 52 { |
53 int i; | |
54 | |
4600 | 55 tprintf(s->avctx, "%s[%d]:\n", name, n); |
783 | 56 for(i=0;i<n;i++) { |
57 if ((i & 7) == 0) | |
4600 | 58 tprintf(s->avctx, "%4d: ", i); |
59 tprintf(s->avctx, " %5d.0", tab[i]); | |
783 | 60 if ((i & 7) == 7) |
4600 | 61 tprintf(s->avctx, "\n"); |
783 | 62 } |
63 } | |
64 | |
5517 | 65 static void dump_floats(WMACodecContext *s, const char *name, int prec, const float *tab, int n) |
783 | 66 { |
67 int i; | |
68 | |
4600 | 69 tprintf(s->avctx, "%s[%d]:\n", name, n); |
783 | 70 for(i=0;i<n;i++) { |
71 if ((i & 7) == 0) | |
4600 | 72 tprintf(s->avctx, "%4d: ", i); |
73 tprintf(s->avctx, " %8.*f", prec, tab[i]); | |
783 | 74 if ((i & 7) == 7) |
4600 | 75 tprintf(s->avctx, "\n"); |
783 | 76 } |
77 if ((i & 7) != 0) | |
4600 | 78 tprintf(s->avctx, "\n"); |
783 | 79 } |
80 #endif | |
81 | |
82 static int wma_decode_init(AVCodecContext * avctx) | |
83 { | |
4601 | 84 WMACodecContext *s = avctx->priv_data; |
10275 | 85 int i, flags2; |
783 | 86 uint8_t *extradata; |
2967 | 87 |
4600 | 88 s->avctx = avctx; |
89 | |
783 | 90 /* extract flag infos */ |
91 flags2 = 0; | |
92 extradata = avctx->extradata; | |
4490 | 93 if (avctx->codec->id == CODEC_ID_WMAV1 && avctx->extradata_size >= 4) { |
5089 | 94 flags2 = AV_RL16(extradata+2); |
4490 | 95 } else if (avctx->codec->id == CODEC_ID_WMAV2 && avctx->extradata_size >= 6) { |
5089 | 96 flags2 = AV_RL16(extradata+4); |
783 | 97 } |
4491 | 98 // for(i=0; i<avctx->extradata_size; i++) |
99 // av_log(NULL, AV_LOG_ERROR, "%02X ", extradata[i]); | |
4490 | 100 |
783 | 101 s->use_exp_vlc = flags2 & 0x0001; |
102 s->use_bit_reservoir = flags2 & 0x0002; | |
103 s->use_variable_block_len = flags2 & 0x0004; | |
104 | |
5086
a10ebd496bd9
sanity checks (should prevent hypothetical div by zero issue)
michael
parents:
4785
diff
changeset
|
105 if(ff_wma_init(avctx, flags2)<0) |
a10ebd496bd9
sanity checks (should prevent hypothetical div by zero issue)
michael
parents:
4785
diff
changeset
|
106 return -1; |
783 | 107 |
108 /* init MDCT */ | |
109 for(i = 0; i < s->nb_block_sizes; i++) | |
9658
67a20f0eb42c
Support for getting (i)MDCT output multiplied by a constant scaling factor.
serge
parents:
9472
diff
changeset
|
110 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1, 1.0); |
2967 | 111 |
783 | 112 if (s->use_noise_coding) { |
4490 | 113 init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits), |
114 ff_wma_hgain_huffbits, 1, 1, | |
115 ff_wma_hgain_huffcodes, 2, 2, 0); | |
783 | 116 } |
117 | |
118 if (s->use_exp_vlc) { | |
11215
964d01b50f17
remove a Huffman table from WMA which also exists in AAC
stefang
parents:
11062
diff
changeset
|
119 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_aac_scalefactor_bits), //FIXME move out of context |
964d01b50f17
remove a Huffman table from WMA which also exists in AAC
stefang
parents:
11062
diff
changeset
|
120 ff_aac_scalefactor_bits, 1, 1, |
964d01b50f17
remove a Huffman table from WMA which also exists in AAC
stefang
parents:
11062
diff
changeset
|
121 ff_aac_scalefactor_code, 4, 4, 0); |
783 | 122 } else { |
123 wma_lsp_to_curve_init(s, s->frame_len); | |
124 } | |
125 | |
7451
85ab7655ad4d
Modify all codecs to report their supported input and output sample format(s).
pross
parents:
7244
diff
changeset
|
126 avctx->sample_fmt = SAMPLE_FMT_S16; |
783 | 127 return 0; |
128 } | |
129 | |
4497 | 130 /** |
131 * compute x^-0.25 with an exponent and mantissa table. We use linear | |
132 * interpolation to reduce the mantissa table size at a small speed | |
133 * expense (linear interpolation approximately doubles the number of | |
134 * bits of precision). | |
135 */ | |
4601 | 136 static inline float pow_m1_4(WMACodecContext *s, float x) |
783 | 137 { |
138 union { | |
139 float f; | |
140 unsigned int v; | |
141 } u, t; | |
142 unsigned int e, m; | |
143 float a, b; | |
144 | |
145 u.f = x; | |
146 e = u.v >> 23; | |
147 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1); | |
148 /* build interpolation scale: 1 <= t < 2. */ | |
149 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23); | |
150 a = s->lsp_pow_m_table1[m]; | |
151 b = s->lsp_pow_m_table2[m]; | |
152 return s->lsp_pow_e_table[e] * (a + b * t.f); | |
153 } | |
154 | |
4601 | 155 static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len) |
2967 | 156 { |
783 | 157 float wdel, a, b; |
158 int i, e, m; | |
159 | |
160 wdel = M_PI / frame_len; | |
161 for(i=0;i<frame_len;i++) | |
162 s->lsp_cos_table[i] = 2.0f * cos(wdel * i); | |
163 | |
164 /* tables for x^-0.25 computation */ | |
165 for(i=0;i<256;i++) { | |
166 e = i - 126; | |
167 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25); | |
168 } | |
169 | |
170 /* NOTE: these two tables are needed to avoid two operations in | |
171 pow_m1_4 */ | |
172 b = 1.0; | |
173 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) { | |
174 m = (1 << LSP_POW_BITS) + i; | |
175 a = (float)m * (0.5 / (1 << LSP_POW_BITS)); | |
176 a = pow(a, -0.25); | |
177 s->lsp_pow_m_table1[i] = 2 * a - b; | |
178 s->lsp_pow_m_table2[i] = b - a; | |
179 b = a; | |
180 } | |
181 #if 0 | |
182 for(i=1;i<20;i++) { | |
183 float v, r1, r2; | |
184 v = 5.0 / i; | |
185 r1 = pow_m1_4(s, v); | |
186 r2 = pow(v,-0.25); | |
187 printf("%f^-0.25=%f e=%f\n", v, r1, r2 - r1); | |
188 } | |
189 #endif | |
190 } | |
191 | |
4497 | 192 /** |
193 * NOTE: We use the same code as Vorbis here | |
194 * @todo optimize it further with SSE/3Dnow | |
195 */ | |
4601 | 196 static void wma_lsp_to_curve(WMACodecContext *s, |
2967 | 197 float *out, float *val_max_ptr, |
783 | 198 int n, float *lsp) |
199 { | |
200 int i, j; | |
201 float p, q, w, v, val_max; | |
202 | |
203 val_max = 0; | |
204 for(i=0;i<n;i++) { | |
205 p = 0.5f; | |
206 q = 0.5f; | |
207 w = s->lsp_cos_table[i]; | |
208 for(j=1;j<NB_LSP_COEFS;j+=2){ | |
209 q *= w - lsp[j - 1]; | |
210 p *= w - lsp[j]; | |
211 } | |
212 p *= p * (2.0f - w); | |
213 q *= q * (2.0f + w); | |
214 v = p + q; | |
215 v = pow_m1_4(s, v); | |
216 if (v > val_max) | |
217 val_max = v; | |
218 out[i] = v; | |
219 } | |
220 *val_max_ptr = val_max; | |
221 } | |
222 | |
4497 | 223 /** |
224 * decode exponents coded with LSP coefficients (same idea as Vorbis) | |
225 */ | |
4601 | 226 static void decode_exp_lsp(WMACodecContext *s, int ch) |
783 | 227 { |
228 float lsp_coefs[NB_LSP_COEFS]; | |
229 int val, i; | |
230 | |
231 for(i = 0; i < NB_LSP_COEFS; i++) { | |
232 if (i == 0 || i >= 8) | |
233 val = get_bits(&s->gb, 3); | |
234 else | |
235 val = get_bits(&s->gb, 4); | |
4490 | 236 lsp_coefs[i] = ff_wma_lsp_codebook[i][val]; |
783 | 237 } |
238 | |
239 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch], | |
240 s->block_len, lsp_coefs); | |
241 } | |
242 | |
10564 | 243 /** pow(10, i / 16.0) for i in -60..95 */ |
10561 | 244 static const float pow_tab[] = { |
10312 | 245 1.7782794100389e-04, 2.0535250264571e-04, |
246 2.3713737056617e-04, 2.7384196342644e-04, | |
247 3.1622776601684e-04, 3.6517412725484e-04, | |
248 4.2169650342858e-04, 4.8696752516586e-04, | |
249 5.6234132519035e-04, 6.4938163157621e-04, | |
250 7.4989420933246e-04, 8.6596432336006e-04, | |
251 1.0000000000000e-03, 1.1547819846895e-03, | |
252 1.3335214321633e-03, 1.5399265260595e-03, | |
253 1.7782794100389e-03, 2.0535250264571e-03, | |
254 2.3713737056617e-03, 2.7384196342644e-03, | |
255 3.1622776601684e-03, 3.6517412725484e-03, | |
256 4.2169650342858e-03, 4.8696752516586e-03, | |
257 5.6234132519035e-03, 6.4938163157621e-03, | |
258 7.4989420933246e-03, 8.6596432336006e-03, | |
259 1.0000000000000e-02, 1.1547819846895e-02, | |
260 1.3335214321633e-02, 1.5399265260595e-02, | |
261 1.7782794100389e-02, 2.0535250264571e-02, | |
262 2.3713737056617e-02, 2.7384196342644e-02, | |
263 3.1622776601684e-02, 3.6517412725484e-02, | |
264 4.2169650342858e-02, 4.8696752516586e-02, | |
265 5.6234132519035e-02, 6.4938163157621e-02, | |
266 7.4989420933246e-02, 8.6596432336007e-02, | |
267 1.0000000000000e-01, 1.1547819846895e-01, | |
268 1.3335214321633e-01, 1.5399265260595e-01, | |
269 1.7782794100389e-01, 2.0535250264571e-01, | |
270 2.3713737056617e-01, 2.7384196342644e-01, | |
271 3.1622776601684e-01, 3.6517412725484e-01, | |
272 4.2169650342858e-01, 4.8696752516586e-01, | |
273 5.6234132519035e-01, 6.4938163157621e-01, | |
274 7.4989420933246e-01, 8.6596432336007e-01, | |
275 1.0000000000000e+00, 1.1547819846895e+00, | |
276 1.3335214321633e+00, 1.5399265260595e+00, | |
277 1.7782794100389e+00, 2.0535250264571e+00, | |
278 2.3713737056617e+00, 2.7384196342644e+00, | |
279 3.1622776601684e+00, 3.6517412725484e+00, | |
280 4.2169650342858e+00, 4.8696752516586e+00, | |
281 5.6234132519035e+00, 6.4938163157621e+00, | |
282 7.4989420933246e+00, 8.6596432336007e+00, | |
283 1.0000000000000e+01, 1.1547819846895e+01, | |
284 1.3335214321633e+01, 1.5399265260595e+01, | |
285 1.7782794100389e+01, 2.0535250264571e+01, | |
286 2.3713737056617e+01, 2.7384196342644e+01, | |
287 3.1622776601684e+01, 3.6517412725484e+01, | |
288 4.2169650342858e+01, 4.8696752516586e+01, | |
289 5.6234132519035e+01, 6.4938163157621e+01, | |
290 7.4989420933246e+01, 8.6596432336007e+01, | |
291 1.0000000000000e+02, 1.1547819846895e+02, | |
292 1.3335214321633e+02, 1.5399265260595e+02, | |
293 1.7782794100389e+02, 2.0535250264571e+02, | |
294 2.3713737056617e+02, 2.7384196342644e+02, | |
295 3.1622776601684e+02, 3.6517412725484e+02, | |
296 4.2169650342858e+02, 4.8696752516586e+02, | |
297 5.6234132519035e+02, 6.4938163157621e+02, | |
298 7.4989420933246e+02, 8.6596432336007e+02, | |
299 1.0000000000000e+03, 1.1547819846895e+03, | |
300 1.3335214321633e+03, 1.5399265260595e+03, | |
301 1.7782794100389e+03, 2.0535250264571e+03, | |
302 2.3713737056617e+03, 2.7384196342644e+03, | |
303 3.1622776601684e+03, 3.6517412725484e+03, | |
304 4.2169650342858e+03, 4.8696752516586e+03, | |
305 5.6234132519035e+03, 6.4938163157621e+03, | |
306 7.4989420933246e+03, 8.6596432336007e+03, | |
307 1.0000000000000e+04, 1.1547819846895e+04, | |
308 1.3335214321633e+04, 1.5399265260595e+04, | |
10561 | 309 1.7782794100389e+04, 2.0535250264571e+04, |
310 2.3713737056617e+04, 2.7384196342644e+04, | |
311 3.1622776601684e+04, 3.6517412725484e+04, | |
312 4.2169650342858e+04, 4.8696752516586e+04, | |
10564 | 313 5.6234132519035e+04, 6.4938163157621e+04, |
314 7.4989420933246e+04, 8.6596432336007e+04, | |
315 1.0000000000000e+05, 1.1547819846895e+05, | |
316 1.3335214321633e+05, 1.5399265260595e+05, | |
317 1.7782794100389e+05, 2.0535250264571e+05, | |
318 2.3713737056617e+05, 2.7384196342644e+05, | |
319 3.1622776601684e+05, 3.6517412725484e+05, | |
320 4.2169650342858e+05, 4.8696752516586e+05, | |
321 5.6234132519035e+05, 6.4938163157621e+05, | |
322 7.4989420933246e+05, 8.6596432336007e+05, | |
10312 | 323 }; |
324 | |
4497 | 325 /** |
326 * decode exponents coded with VLC codes | |
327 */ | |
4601 | 328 static int decode_exp_vlc(WMACodecContext *s, int ch) |
783 | 329 { |
330 int last_exp, n, code; | |
10275 | 331 const uint16_t *ptr; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
332 float v, max_scale; |
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
333 uint32_t *q, *q_end, iv; |
10312 | 334 const float *ptab = pow_tab + 60; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
335 const uint32_t *iptab = (const uint32_t*)ptab; |
2967 | 336 |
10275 | 337 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
338 q = (uint32_t *)s->exponents[ch]; |
783 | 339 q_end = q + s->block_len; |
340 max_scale = 0; | |
341 if (s->version == 1) { | |
342 last_exp = get_bits(&s->gb, 5) + 10; | |
10312 | 343 v = ptab[last_exp]; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
344 iv = iptab[last_exp]; |
783 | 345 max_scale = v; |
346 n = *ptr++; | |
10317 | 347 switch (n & 3) do { |
348 case 0: *q++ = iv; | |
349 case 3: *q++ = iv; | |
350 case 2: *q++ = iv; | |
351 case 1: *q++ = iv; | |
352 } while ((n -= 4) > 0); | |
4490 | 353 }else |
354 last_exp = 36; | |
355 | |
783 | 356 while (q < q_end) { |
3113 | 357 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX); |
10654 | 358 if (code < 0){ |
359 av_log(s->avctx, AV_LOG_ERROR, "Exponent vlc invalid\n"); | |
783 | 360 return -1; |
10654 | 361 } |
783 | 362 /* NOTE: this offset is the same as MPEG4 AAC ! */ |
363 last_exp += code - 60; | |
10561 | 364 if ((unsigned)last_exp + 60 > FF_ARRAY_ELEMS(pow_tab)) { |
365 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n", | |
366 last_exp); | |
10312 | 367 return -1; |
10561 | 368 } |
10312 | 369 v = ptab[last_exp]; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
370 iv = iptab[last_exp]; |
783 | 371 if (v > max_scale) |
372 max_scale = v; | |
373 n = *ptr++; | |
10317 | 374 switch (n & 3) do { |
375 case 0: *q++ = iv; | |
376 case 3: *q++ = iv; | |
377 case 2: *q++ = iv; | |
378 case 1: *q++ = iv; | |
379 } while ((n -= 4) > 0); | |
783 | 380 } |
381 s->max_exponent[ch] = max_scale; | |
382 return 0; | |
383 } | |
384 | |
4737
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
385 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
386 /** |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
387 * Apply MDCT window and add into output. |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
388 * |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
389 * We ensure that when the windows overlap their squared sum |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
390 * is always 1 (MDCT reconstruction rule). |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
391 */ |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
392 static void wma_window(WMACodecContext *s, float *out) |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
393 { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
394 float *in = s->output; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
395 int block_len, bsize, n; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
396 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
397 /* left part */ |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
398 if (s->block_len_bits <= s->prev_block_len_bits) { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
399 block_len = s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
400 bsize = s->frame_len_bits - s->block_len_bits; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
401 |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10275
diff
changeset
|
402 s->dsp.vector_fmul_add(out, in, s->windows[bsize], |
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10275
diff
changeset
|
403 out, block_len); |
4737
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
404 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
405 } else { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
406 block_len = 1 << s->prev_block_len_bits; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
407 n = (s->block_len - block_len) / 2; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
408 bsize = s->frame_len_bits - s->prev_block_len_bits; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
409 |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10275
diff
changeset
|
410 s->dsp.vector_fmul_add(out+n, in+n, s->windows[bsize], |
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10275
diff
changeset
|
411 out+n, block_len); |
4737
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
412 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
413 memcpy(out+n+block_len, in+n+block_len, n*sizeof(float)); |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
414 } |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
415 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
416 out += s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
417 in += s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
418 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
419 /* right part */ |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
420 if (s->block_len_bits <= s->next_block_len_bits) { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
421 block_len = s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
422 bsize = s->frame_len_bits - s->block_len_bits; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
423 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
424 s->dsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len); |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
425 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
426 } else { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
427 block_len = 1 << s->next_block_len_bits; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
428 n = (s->block_len - block_len) / 2; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
429 bsize = s->frame_len_bits - s->next_block_len_bits; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
430 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
431 memcpy(out, in, n*sizeof(float)); |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
432 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
433 s->dsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len); |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
434 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
435 memset(out+n+block_len, 0, n*sizeof(float)); |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
436 } |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
437 } |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
438 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
439 |
4497 | 440 /** |
441 * @return 0 if OK. 1 if last block of frame. return -1 if | |
442 * unrecorrable error. | |
443 */ | |
4601 | 444 static int wma_decode_block(WMACodecContext *s) |
783 | 445 { |
9844
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
446 int n, v, a, ch, bsize; |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
447 int coef_nb_bits, total_gain; |
783 | 448 int nb_coefs[MAX_CHANNELS]; |
449 float mdct_norm; | |
450 | |
1343 | 451 #ifdef TRACE |
4600 | 452 tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); |
1343 | 453 #endif |
783 | 454 |
455 /* compute current block length */ | |
456 if (s->use_variable_block_len) { | |
457 n = av_log2(s->nb_block_sizes - 1) + 1; | |
2967 | 458 |
783 | 459 if (s->reset_block_lengths) { |
460 s->reset_block_lengths = 0; | |
461 v = get_bits(&s->gb, n); | |
10654 | 462 if (v >= s->nb_block_sizes){ |
463 av_log(s->avctx, AV_LOG_ERROR, "prev_block_len_bits %d out of range\n", s->frame_len_bits - v); | |
783 | 464 return -1; |
10654 | 465 } |
783 | 466 s->prev_block_len_bits = s->frame_len_bits - v; |
467 v = get_bits(&s->gb, n); | |
10654 | 468 if (v >= s->nb_block_sizes){ |
469 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits %d out of range\n", s->frame_len_bits - v); | |
783 | 470 return -1; |
10654 | 471 } |
783 | 472 s->block_len_bits = s->frame_len_bits - v; |
473 } else { | |
474 /* update block lengths */ | |
475 s->prev_block_len_bits = s->block_len_bits; | |
476 s->block_len_bits = s->next_block_len_bits; | |
477 } | |
478 v = get_bits(&s->gb, n); | |
10654 | 479 if (v >= s->nb_block_sizes){ |
480 av_log(s->avctx, AV_LOG_ERROR, "next_block_len_bits %d out of range\n", s->frame_len_bits - v); | |
783 | 481 return -1; |
10654 | 482 } |
783 | 483 s->next_block_len_bits = s->frame_len_bits - v; |
484 } else { | |
485 /* fixed block len */ | |
486 s->next_block_len_bits = s->frame_len_bits; | |
487 s->prev_block_len_bits = s->frame_len_bits; | |
488 s->block_len_bits = s->frame_len_bits; | |
489 } | |
490 | |
491 /* now check if the block length is coherent with the frame length */ | |
492 s->block_len = 1 << s->block_len_bits; | |
10654 | 493 if ((s->block_pos + s->block_len) > s->frame_len){ |
494 av_log(s->avctx, AV_LOG_ERROR, "frame_len overflow\n"); | |
783 | 495 return -1; |
10654 | 496 } |
783 | 497 |
498 if (s->nb_channels == 2) { | |
5513 | 499 s->ms_stereo = get_bits1(&s->gb); |
783 | 500 } |
501 v = 0; | |
502 for(ch = 0; ch < s->nb_channels; ch++) { | |
5513 | 503 a = get_bits1(&s->gb); |
783 | 504 s->channel_coded[ch] = a; |
505 v |= a; | |
506 } | |
7243 | 507 |
508 bsize = s->frame_len_bits - s->block_len_bits; | |
509 | |
783 | 510 /* if no channel coded, no need to go further */ |
511 /* XXX: fix potential framing problems */ | |
512 if (!v) | |
513 goto next; | |
514 | |
515 /* read total gain and extract corresponding number of bits for | |
516 coef escape coding */ | |
517 total_gain = 1; | |
518 for(;;) { | |
519 a = get_bits(&s->gb, 7); | |
520 total_gain += a; | |
521 if (a != 127) | |
522 break; | |
523 } | |
2967 | 524 |
4490 | 525 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain); |
783 | 526 |
527 /* compute number of coefficients */ | |
528 n = s->coefs_end[bsize] - s->coefs_start; | |
529 for(ch = 0; ch < s->nb_channels; ch++) | |
530 nb_coefs[ch] = n; | |
531 | |
532 /* complex coding */ | |
533 if (s->use_noise_coding) { | |
534 | |
535 for(ch = 0; ch < s->nb_channels; ch++) { | |
536 if (s->channel_coded[ch]) { | |
537 int i, n, a; | |
538 n = s->exponent_high_sizes[bsize]; | |
539 for(i=0;i<n;i++) { | |
5513 | 540 a = get_bits1(&s->gb); |
783 | 541 s->high_band_coded[ch][i] = a; |
542 /* if noise coding, the coefficients are not transmitted */ | |
543 if (a) | |
544 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; | |
545 } | |
546 } | |
547 } | |
548 for(ch = 0; ch < s->nb_channels; ch++) { | |
549 if (s->channel_coded[ch]) { | |
550 int i, n, val, code; | |
551 | |
552 n = s->exponent_high_sizes[bsize]; | |
553 val = (int)0x80000000; | |
554 for(i=0;i<n;i++) { | |
555 if (s->high_band_coded[ch][i]) { | |
556 if (val == (int)0x80000000) { | |
557 val = get_bits(&s->gb, 7) - 19; | |
558 } else { | |
3113 | 559 code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX); |
10654 | 560 if (code < 0){ |
561 av_log(s->avctx, AV_LOG_ERROR, "hgain vlc invalid\n"); | |
783 | 562 return -1; |
10654 | 563 } |
783 | 564 val += code - 18; |
565 } | |
566 s->high_band_values[ch][i] = val; | |
567 } | |
568 } | |
569 } | |
570 } | |
571 } | |
2967 | 572 |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
573 /* exponents can be reused in short blocks. */ |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
574 if ((s->block_len_bits == s->frame_len_bits) || |
5513 | 575 get_bits1(&s->gb)) { |
783 | 576 for(ch = 0; ch < s->nb_channels; ch++) { |
577 if (s->channel_coded[ch]) { | |
578 if (s->use_exp_vlc) { | |
579 if (decode_exp_vlc(s, ch) < 0) | |
580 return -1; | |
581 } else { | |
582 decode_exp_lsp(s, ch); | |
583 } | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
584 s->exponents_bsize[ch] = bsize; |
783 | 585 } |
586 } | |
587 } | |
588 | |
589 /* parse spectral coefficients : just RLE encoding */ | |
590 for(ch = 0; ch < s->nb_channels; ch++) { | |
591 if (s->channel_coded[ch]) { | |
9844
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
592 int tindex; |
9868 | 593 WMACoef* ptr = &s->coefs1[ch][0]; |
783 | 594 |
595 /* special VLC tables are used for ms stereo because | |
596 there is potentially less energy there */ | |
597 tindex = (ch == 1 && s->ms_stereo); | |
9868 | 598 memset(ptr, 0, s->block_len * sizeof(WMACoef)); |
9844
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
599 ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex], |
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
600 s->level_table[tindex], s->run_table[tindex], |
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
601 0, ptr, 0, nb_coefs[ch], |
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
602 s->block_len, s->frame_len_bits, coef_nb_bits); |
783 | 603 } |
604 if (s->version == 1 && s->nb_channels >= 2) { | |
605 align_get_bits(&s->gb); | |
606 } | |
607 } | |
2967 | 608 |
783 | 609 /* normalize */ |
610 { | |
611 int n4 = s->block_len / 2; | |
612 mdct_norm = 1.0 / (float)n4; | |
613 if (s->version == 1) { | |
614 mdct_norm *= sqrt(n4); | |
615 } | |
616 } | |
617 | |
618 /* finally compute the MDCT coefficients */ | |
619 for(ch = 0; ch < s->nb_channels; ch++) { | |
620 if (s->channel_coded[ch]) { | |
9868 | 621 WMACoef *coefs1; |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
622 float *coefs, *exponents, mult, mult1, noise; |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
623 int i, j, n, n1, last_high_band, esize; |
783 | 624 float exp_power[HIGH_BAND_MAX_SIZE]; |
625 | |
626 coefs1 = s->coefs1[ch]; | |
627 exponents = s->exponents[ch]; | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
628 esize = s->exponents_bsize[ch]; |
3235 | 629 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; |
783 | 630 mult *= mdct_norm; |
631 coefs = s->coefs[ch]; | |
632 if (s->use_noise_coding) { | |
633 mult1 = mult; | |
634 /* very low freqs : noise */ | |
635 for(i = 0;i < s->coefs_start; i++) { | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
636 *coefs++ = s->noise_table[s->noise_index] * |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
637 exponents[i<<bsize>>esize] * mult1; |
783 | 638 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); |
639 } | |
2967 | 640 |
783 | 641 n1 = s->exponent_high_sizes[bsize]; |
642 | |
643 /* compute power of high bands */ | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
644 exponents = s->exponents[ch] + |
10653 | 645 (s->high_band_start[bsize]<<bsize>>esize); |
783 | 646 last_high_band = 0; /* avoid warning */ |
647 for(j=0;j<n1;j++) { | |
2967 | 648 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 649 s->block_len_bits][j]; |
650 if (s->high_band_coded[ch][j]) { | |
651 float e2, v; | |
652 e2 = 0; | |
653 for(i = 0;i < n; i++) { | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
654 v = exponents[i<<bsize>>esize]; |
783 | 655 e2 += v * v; |
656 } | |
657 exp_power[j] = e2 / n; | |
658 last_high_band = j; | |
4600 | 659 tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n); |
783 | 660 } |
10653 | 661 exponents += n<<bsize>>esize; |
783 | 662 } |
663 | |
664 /* main freqs and high freqs */ | |
10653 | 665 exponents = s->exponents[ch] + (s->coefs_start<<bsize>>esize); |
783 | 666 for(j=-1;j<n1;j++) { |
667 if (j < 0) { | |
2967 | 668 n = s->high_band_start[bsize] - |
783 | 669 s->coefs_start; |
670 } else { | |
2967 | 671 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 672 s->block_len_bits][j]; |
673 } | |
674 if (j >= 0 && s->high_band_coded[ch][j]) { | |
675 /* use noise with specified power */ | |
676 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); | |
3235 | 677 /* XXX: use a table */ |
678 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); | |
783 | 679 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); |
680 mult1 *= mdct_norm; | |
681 for(i = 0;i < n; i++) { | |
682 noise = s->noise_table[s->noise_index]; | |
683 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
684 *coefs++ = noise * |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
685 exponents[i<<bsize>>esize] * mult1; |
783 | 686 } |
10653 | 687 exponents += n<<bsize>>esize; |
783 | 688 } else { |
689 /* coded values + small noise */ | |
690 for(i = 0;i < n; i++) { | |
691 noise = s->noise_table[s->noise_index]; | |
692 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
693 *coefs++ = ((*coefs1++) + noise) * |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
694 exponents[i<<bsize>>esize] * mult; |
783 | 695 } |
10653 | 696 exponents += n<<bsize>>esize; |
783 | 697 } |
698 } | |
699 | |
700 /* very high freqs : noise */ | |
701 n = s->block_len - s->coefs_end[bsize]; | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
702 mult1 = mult * exponents[((-1<<bsize))>>esize]; |
783 | 703 for(i = 0; i < n; i++) { |
704 *coefs++ = s->noise_table[s->noise_index] * mult1; | |
705 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
706 } | |
707 } else { | |
708 /* XXX: optimize more */ | |
709 for(i = 0;i < s->coefs_start; i++) | |
710 *coefs++ = 0.0; | |
711 n = nb_coefs[ch]; | |
712 for(i = 0;i < n; i++) { | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
713 *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult; |
783 | 714 } |
715 n = s->block_len - s->coefs_end[bsize]; | |
716 for(i = 0;i < n; i++) | |
717 *coefs++ = 0.0; | |
718 } | |
719 } | |
720 } | |
721 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
722 #ifdef TRACE |
783 | 723 for(ch = 0; ch < s->nb_channels; ch++) { |
724 if (s->channel_coded[ch]) { | |
4600 | 725 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len); |
726 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len); | |
783 | 727 } |
728 } | |
729 #endif | |
2967 | 730 |
783 | 731 if (s->ms_stereo && s->channel_coded[1]) { |
732 /* nominal case for ms stereo: we do it before mdct */ | |
733 /* no need to optimize this case because it should almost | |
734 never happen */ | |
735 if (!s->channel_coded[0]) { | |
4600 | 736 tprintf(s->avctx, "rare ms-stereo case happened\n"); |
783 | 737 memset(s->coefs[0], 0, sizeof(float) * s->block_len); |
738 s->channel_coded[0] = 1; | |
739 } | |
2967 | 740 |
10236 | 741 s->dsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len); |
783 | 742 } |
743 | |
7243 | 744 next: |
783 | 745 for(ch = 0; ch < s->nb_channels; ch++) { |
9472
05663d250d5b
Remove unused variable from wma_decode_block() found by CSA.
michael
parents:
9355
diff
changeset
|
746 int n4, index; |
783 | 747 |
7244 | 748 n4 = s->block_len / 2; |
749 if(s->channel_coded[ch]){ | |
7547 | 750 ff_imdct_calc(&s->mdct_ctx[bsize], s->output, s->coefs[ch]); |
7712
7c22b99dbf5e
Fix mid/side stereo buggy output zeroing, fixes issue264 part 2.
michael
parents:
7547
diff
changeset
|
751 }else if(!(s->ms_stereo && ch==1)) |
7244 | 752 memset(s->output, 0, sizeof(s->output)); |
783 | 753 |
7244 | 754 /* multiply by the window and add in the frame */ |
755 index = (s->frame_len / 2) + s->block_pos - n4; | |
756 wma_window(s, &s->frame_out[ch][index]); | |
783 | 757 } |
7243 | 758 |
783 | 759 /* update block number */ |
760 s->block_num++; | |
761 s->block_pos += s->block_len; | |
762 if (s->block_pos >= s->frame_len) | |
763 return 1; | |
764 else | |
765 return 0; | |
766 } | |
767 | |
768 /* decode a frame of frame_len samples */ | |
4601 | 769 static int wma_decode_frame(WMACodecContext *s, int16_t *samples) |
783 | 770 { |
5525
bc4791868c52
various simplifications around recent av_clip_int16() usage
aurel
parents:
5523
diff
changeset
|
771 int ret, i, n, ch, incr; |
783 | 772 int16_t *ptr; |
773 float *iptr; | |
774 | |
1343 | 775 #ifdef TRACE |
4600 | 776 tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len); |
1343 | 777 #endif |
783 | 778 |
779 /* read each block */ | |
780 s->block_num = 0; | |
781 s->block_pos = 0; | |
782 for(;;) { | |
783 ret = wma_decode_block(s); | |
2967 | 784 if (ret < 0) |
783 | 785 return -1; |
786 if (ret) | |
787 break; | |
788 } | |
789 | |
790 /* convert frame to integer */ | |
791 n = s->frame_len; | |
792 incr = s->nb_channels; | |
11451
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
793 if (s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) { |
11452 | 794 for(ch = 0; ch < s->nb_channels; ch++) { |
795 ptr = samples + ch; | |
796 iptr = s->frame_out[ch]; | |
783 | 797 |
11452 | 798 for(i=0;i<n;i++) { |
799 *ptr = av_clip_int16(lrintf(*iptr++)); | |
800 ptr += incr; | |
801 } | |
802 /* prepare for next block */ | |
803 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len], | |
804 s->frame_len * sizeof(float)); | |
783 | 805 } |
11451
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
806 } else { |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
807 float *output[MAX_CHANNELS]; |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
808 for (ch = 0; ch < MAX_CHANNELS; ch++) |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
809 output[ch] = s->frame_out[ch]; |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
810 s->dsp.float_to_int16_interleave(samples, (const float **)output, n, incr); |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
811 for(ch = 0; ch < incr; ch++) { |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
812 /* prepare for next block */ |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
813 memmove(&s->frame_out[ch][0], &s->frame_out[ch][n], n * sizeof(float)); |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
814 } |
01559518729e
SIMD optimization using float_to_int16_interleave.
cehoyos
parents:
11215
diff
changeset
|
815 } |
783 | 816 |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
817 #ifdef TRACE |
4600 | 818 dump_shorts(s, "samples", samples, n * s->nb_channels); |
783 | 819 #endif |
820 return 0; | |
821 } | |
822 | |
2967 | 823 static int wma_decode_superframe(AVCodecContext *avctx, |
783 | 824 void *data, int *data_size, |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
825 AVPacket *avpkt) |
783 | 826 { |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
827 const uint8_t *buf = avpkt->data; |
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
828 int buf_size = avpkt->size; |
4601 | 829 WMACodecContext *s = avctx->priv_data; |
783 | 830 int nb_frames, bit_offset, i, pos, len; |
831 uint8_t *q; | |
832 int16_t *samples; | |
2967 | 833 |
4600 | 834 tprintf(avctx, "***decode_superframe:\n"); |
783 | 835 |
1750 | 836 if(buf_size==0){ |
837 s->last_superframe_len = 0; | |
838 return 0; | |
839 } | |
5957
0f2c4fa2c4f2
wma_decode_superframe always returns s->block_align, so make
reimar
parents:
5525
diff
changeset
|
840 if (buf_size < s->block_align) |
0f2c4fa2c4f2
wma_decode_superframe always returns s->block_align, so make
reimar
parents:
5525
diff
changeset
|
841 return 0; |
0f2c4fa2c4f2
wma_decode_superframe always returns s->block_align, so make
reimar
parents:
5525
diff
changeset
|
842 buf_size = s->block_align; |
2967 | 843 |
783 | 844 samples = data; |
845 | |
1025
1f9afd8b9131
GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops
michaelni
parents:
972
diff
changeset
|
846 init_get_bits(&s->gb, buf, buf_size*8); |
2967 | 847 |
783 | 848 if (s->use_bit_reservoir) { |
849 /* read super frame header */ | |
5518 | 850 skip_bits(&s->gb, 4); /* super frame index */ |
783 | 851 nb_frames = get_bits(&s->gb, 4) - 1; |
852 | |
7242 | 853 if((nb_frames+1) * s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){ |
854 av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n"); | |
855 goto fail; | |
856 } | |
857 | |
783 | 858 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3); |
859 | |
860 if (s->last_superframe_len > 0) { | |
861 // printf("skip=%d\n", s->last_bitoffset); | |
862 /* add bit_offset bits to last frame */ | |
2967 | 863 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) > |
783 | 864 MAX_CODED_SUPERFRAME_SIZE) |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
865 goto fail; |
783 | 866 q = s->last_superframe + s->last_superframe_len; |
867 len = bit_offset; | |
3362
c43fcf831f7c
Do not read full byte when less than 8 bits are still to be read.
reimar
parents:
3361
diff
changeset
|
868 while (len > 7) { |
783 | 869 *q++ = (get_bits)(&s->gb, 8); |
870 len -= 8; | |
871 } | |
872 if (len > 0) { | |
873 *q++ = (get_bits)(&s->gb, len) << (8 - len); | |
874 } | |
2967 | 875 |
783 | 876 /* XXX: bit_offset bits into last frame */ |
1025
1f9afd8b9131
GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops
michaelni
parents:
972
diff
changeset
|
877 init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8); |
783 | 878 /* skip unused bits */ |
879 if (s->last_bitoffset > 0) | |
880 skip_bits(&s->gb, s->last_bitoffset); | |
881 /* this frame is stored in the last superframe and in the | |
882 current one */ | |
883 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
884 goto fail; |
783 | 885 samples += s->nb_channels * s->frame_len; |
886 } | |
887 | |
888 /* read each frame starting from bit_offset */ | |
889 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3; | |
1025
1f9afd8b9131
GetBitContext.size is allways multiplied by 8 -> use size_in_bits to avoid useless *8 in a few inner loops
michaelni
parents:
972
diff
changeset
|
890 init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8); |
783 | 891 len = pos & 7; |
892 if (len > 0) | |
893 skip_bits(&s->gb, len); | |
2967 | 894 |
783 | 895 s->reset_block_lengths = 1; |
896 for(i=0;i<nb_frames;i++) { | |
897 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
898 goto fail; |
783 | 899 samples += s->nb_channels * s->frame_len; |
900 } | |
901 | |
902 /* we copy the end of the frame in the last frame buffer */ | |
903 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7); | |
904 s->last_bitoffset = pos & 7; | |
905 pos >>= 3; | |
906 len = buf_size - pos; | |
819 | 907 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) { |
10654 | 908 av_log(s->avctx, AV_LOG_ERROR, "len %d invalid\n", len); |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
909 goto fail; |
783 | 910 } |
911 s->last_superframe_len = len; | |
912 memcpy(s->last_superframe, buf + pos, len); | |
913 } else { | |
7242 | 914 if(s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){ |
915 av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n"); | |
916 goto fail; | |
917 } | |
783 | 918 /* single frame decode */ |
919 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
920 goto fail; |
783 | 921 samples += s->nb_channels * s->frame_len; |
922 } | |
4490 | 923 |
924 //av_log(NULL, AV_LOG_ERROR, "%d %d %d %d outbytes:%d eaten:%d\n", s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len, (int8_t *)samples - (int8_t *)data, s->block_align); | |
925 | |
783 | 926 *data_size = (int8_t *)samples - (int8_t *)data; |
927 return s->block_align; | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
928 fail: |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
929 /* when error, we reset the bit reservoir */ |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
930 s->last_superframe_len = 0; |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
931 return -1; |
783 | 932 } |
933 | |
11062 | 934 static av_cold void flush(AVCodecContext *avctx) |
935 { | |
936 WMACodecContext *s = avctx->priv_data; | |
937 | |
938 s->last_bitoffset= | |
939 s->last_superframe_len= 0; | |
940 } | |
941 | |
783 | 942 AVCodec wmav1_decoder = |
943 { | |
944 "wmav1", | |
11560
8a4984c5cacc
Define AVMediaType enum, and use it instead of enum CodecType, which
stefano
parents:
11452
diff
changeset
|
945 AVMEDIA_TYPE_AUDIO, |
783 | 946 CODEC_ID_WMAV1, |
4601 | 947 sizeof(WMACodecContext), |
783 | 948 wma_decode_init, |
949 NULL, | |
4490 | 950 ff_wma_end, |
783 | 951 wma_decode_superframe, |
11062 | 952 .flush=flush, |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6710
diff
changeset
|
953 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"), |
783 | 954 }; |
955 | |
956 AVCodec wmav2_decoder = | |
957 { | |
958 "wmav2", | |
11560
8a4984c5cacc
Define AVMediaType enum, and use it instead of enum CodecType, which
stefano
parents:
11452
diff
changeset
|
959 AVMEDIA_TYPE_AUDIO, |
783 | 960 CODEC_ID_WMAV2, |
4601 | 961 sizeof(WMACodecContext), |
783 | 962 wma_decode_init, |
963 NULL, | |
4490 | 964 ff_wma_end, |
783 | 965 wma_decode_superframe, |
11062 | 966 .flush=flush, |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6710
diff
changeset
|
967 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"), |
783 | 968 }; |