Mercurial > libavcodec.hg
annotate wmadec.c @ 10561:d0e17b0a6845 libavcodec
WMA: extend exponent table up to 75
Fixes issue 1565.
author | mru |
---|---|
date | Mon, 23 Nov 2009 22:22:16 +0000 |
parents | 88e1eb862abd |
children | 559ba9eabaff |
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 /** | |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8629
diff
changeset
|
23 * @file libavcodec/wmadec.c |
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) { | |
4490 | 119 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_wma_scale_huffbits), //FIXME move out of context |
120 ff_wma_scale_huffbits, 1, 1, | |
121 ff_wma_scale_huffcodes, 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 | |
10561 | 243 /** pow(10, i / 16.0) for i in -60..75 */ |
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, | |
10312 | 313 }; |
314 | |
4497 | 315 /** |
316 * decode exponents coded with VLC codes | |
317 */ | |
4601 | 318 static int decode_exp_vlc(WMACodecContext *s, int ch) |
783 | 319 { |
320 int last_exp, n, code; | |
10275 | 321 const uint16_t *ptr; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
322 float v, max_scale; |
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
323 uint32_t *q, *q_end, iv; |
10312 | 324 const float *ptab = pow_tab + 60; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
325 const uint32_t *iptab = (const uint32_t*)ptab; |
2967 | 326 |
10275 | 327 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
|
328 q = (uint32_t *)s->exponents[ch]; |
783 | 329 q_end = q + s->block_len; |
330 max_scale = 0; | |
331 if (s->version == 1) { | |
332 last_exp = get_bits(&s->gb, 5) + 10; | |
10312 | 333 v = ptab[last_exp]; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
334 iv = iptab[last_exp]; |
783 | 335 max_scale = v; |
336 n = *ptr++; | |
10317 | 337 switch (n & 3) do { |
338 case 0: *q++ = iv; | |
339 case 3: *q++ = iv; | |
340 case 2: *q++ = iv; | |
341 case 1: *q++ = iv; | |
342 } while ((n -= 4) > 0); | |
4490 | 343 }else |
344 last_exp = 36; | |
345 | |
783 | 346 while (q < q_end) { |
3113 | 347 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX); |
783 | 348 if (code < 0) |
349 return -1; | |
350 /* NOTE: this offset is the same as MPEG4 AAC ! */ | |
351 last_exp += code - 60; | |
10561 | 352 if ((unsigned)last_exp + 60 > FF_ARRAY_ELEMS(pow_tab)) { |
353 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n", | |
354 last_exp); | |
10312 | 355 return -1; |
10561 | 356 } |
10312 | 357 v = ptab[last_exp]; |
10313
f7376a522a7e
WMA: use type punning and unroll loops in decode_exp_vlc()
mru
parents:
10312
diff
changeset
|
358 iv = iptab[last_exp]; |
783 | 359 if (v > max_scale) |
360 max_scale = v; | |
361 n = *ptr++; | |
10317 | 362 switch (n & 3) do { |
363 case 0: *q++ = iv; | |
364 case 3: *q++ = iv; | |
365 case 2: *q++ = iv; | |
366 case 1: *q++ = iv; | |
367 } while ((n -= 4) > 0); | |
783 | 368 } |
369 s->max_exponent[ch] = max_scale; | |
370 return 0; | |
371 } | |
372 | |
4737
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
373 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
374 /** |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
375 * 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
|
376 * |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
377 * 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
|
378 * 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
|
379 */ |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
380 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
|
381 { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
382 float *in = s->output; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
383 int block_len, bsize, n; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
384 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
385 /* left part */ |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
386 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
|
387 block_len = s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
388 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
|
389 |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10275
diff
changeset
|
390 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
|
391 out, block_len); |
4737
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
392 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
393 } else { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
394 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
|
395 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
|
396 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
|
397 |
10300
4d1b9ca628fc
Drop unused args from vector_fmul_add_add, simpify code, and rename
mru
parents:
10275
diff
changeset
|
398 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
|
399 out+n, block_len); |
4737
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
400 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
401 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
|
402 } |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
403 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
404 out += s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
405 in += s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
406 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
407 /* right part */ |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
408 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
|
409 block_len = s->block_len; |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
410 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
|
411 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
412 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
|
413 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
414 } else { |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
415 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
|
416 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
|
417 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
|
418 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
419 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
|
420 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
421 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
|
422 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
423 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
|
424 } |
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 |
99d9dd34903b
Optimize by building the mdct window and multipying/adding at the same time.
banan
parents:
4601
diff
changeset
|
427 |
4497 | 428 /** |
429 * @return 0 if OK. 1 if last block of frame. return -1 if | |
430 * unrecorrable error. | |
431 */ | |
4601 | 432 static int wma_decode_block(WMACodecContext *s) |
783 | 433 { |
9844
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
434 int n, v, a, ch, bsize; |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
435 int coef_nb_bits, total_gain; |
783 | 436 int nb_coefs[MAX_CHANNELS]; |
437 float mdct_norm; | |
438 | |
1343 | 439 #ifdef TRACE |
4600 | 440 tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); |
1343 | 441 #endif |
783 | 442 |
443 /* compute current block length */ | |
444 if (s->use_variable_block_len) { | |
445 n = av_log2(s->nb_block_sizes - 1) + 1; | |
2967 | 446 |
783 | 447 if (s->reset_block_lengths) { |
448 s->reset_block_lengths = 0; | |
449 v = get_bits(&s->gb, n); | |
450 if (v >= s->nb_block_sizes) | |
451 return -1; | |
452 s->prev_block_len_bits = s->frame_len_bits - v; | |
453 v = get_bits(&s->gb, n); | |
454 if (v >= s->nb_block_sizes) | |
455 return -1; | |
456 s->block_len_bits = s->frame_len_bits - v; | |
457 } else { | |
458 /* update block lengths */ | |
459 s->prev_block_len_bits = s->block_len_bits; | |
460 s->block_len_bits = s->next_block_len_bits; | |
461 } | |
462 v = get_bits(&s->gb, n); | |
463 if (v >= s->nb_block_sizes) | |
464 return -1; | |
465 s->next_block_len_bits = s->frame_len_bits - v; | |
466 } else { | |
467 /* fixed block len */ | |
468 s->next_block_len_bits = s->frame_len_bits; | |
469 s->prev_block_len_bits = s->frame_len_bits; | |
470 s->block_len_bits = s->frame_len_bits; | |
471 } | |
472 | |
473 /* now check if the block length is coherent with the frame length */ | |
474 s->block_len = 1 << s->block_len_bits; | |
475 if ((s->block_pos + s->block_len) > s->frame_len) | |
476 return -1; | |
477 | |
478 if (s->nb_channels == 2) { | |
5513 | 479 s->ms_stereo = get_bits1(&s->gb); |
783 | 480 } |
481 v = 0; | |
482 for(ch = 0; ch < s->nb_channels; ch++) { | |
5513 | 483 a = get_bits1(&s->gb); |
783 | 484 s->channel_coded[ch] = a; |
485 v |= a; | |
486 } | |
7243 | 487 |
488 bsize = s->frame_len_bits - s->block_len_bits; | |
489 | |
783 | 490 /* if no channel coded, no need to go further */ |
491 /* XXX: fix potential framing problems */ | |
492 if (!v) | |
493 goto next; | |
494 | |
495 /* read total gain and extract corresponding number of bits for | |
496 coef escape coding */ | |
497 total_gain = 1; | |
498 for(;;) { | |
499 a = get_bits(&s->gb, 7); | |
500 total_gain += a; | |
501 if (a != 127) | |
502 break; | |
503 } | |
2967 | 504 |
4490 | 505 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain); |
783 | 506 |
507 /* compute number of coefficients */ | |
508 n = s->coefs_end[bsize] - s->coefs_start; | |
509 for(ch = 0; ch < s->nb_channels; ch++) | |
510 nb_coefs[ch] = n; | |
511 | |
512 /* complex coding */ | |
513 if (s->use_noise_coding) { | |
514 | |
515 for(ch = 0; ch < s->nb_channels; ch++) { | |
516 if (s->channel_coded[ch]) { | |
517 int i, n, a; | |
518 n = s->exponent_high_sizes[bsize]; | |
519 for(i=0;i<n;i++) { | |
5513 | 520 a = get_bits1(&s->gb); |
783 | 521 s->high_band_coded[ch][i] = a; |
522 /* if noise coding, the coefficients are not transmitted */ | |
523 if (a) | |
524 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; | |
525 } | |
526 } | |
527 } | |
528 for(ch = 0; ch < s->nb_channels; ch++) { | |
529 if (s->channel_coded[ch]) { | |
530 int i, n, val, code; | |
531 | |
532 n = s->exponent_high_sizes[bsize]; | |
533 val = (int)0x80000000; | |
534 for(i=0;i<n;i++) { | |
535 if (s->high_band_coded[ch][i]) { | |
536 if (val == (int)0x80000000) { | |
537 val = get_bits(&s->gb, 7) - 19; | |
538 } else { | |
3113 | 539 code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX); |
783 | 540 if (code < 0) |
541 return -1; | |
542 val += code - 18; | |
543 } | |
544 s->high_band_values[ch][i] = val; | |
545 } | |
546 } | |
547 } | |
548 } | |
549 } | |
2967 | 550 |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
551 /* exponents can be reused in short blocks. */ |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
552 if ((s->block_len_bits == s->frame_len_bits) || |
5513 | 553 get_bits1(&s->gb)) { |
783 | 554 for(ch = 0; ch < s->nb_channels; ch++) { |
555 if (s->channel_coded[ch]) { | |
556 if (s->use_exp_vlc) { | |
557 if (decode_exp_vlc(s, ch) < 0) | |
558 return -1; | |
559 } else { | |
560 decode_exp_lsp(s, ch); | |
561 } | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
562 s->exponents_bsize[ch] = bsize; |
783 | 563 } |
564 } | |
565 } | |
566 | |
567 /* parse spectral coefficients : just RLE encoding */ | |
568 for(ch = 0; ch < s->nb_channels; ch++) { | |
569 if (s->channel_coded[ch]) { | |
9844
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
570 int tindex; |
9868 | 571 WMACoef* ptr = &s->coefs1[ch][0]; |
783 | 572 |
573 /* special VLC tables are used for ms stereo because | |
574 there is potentially less energy there */ | |
575 tindex = (ch == 1 && s->ms_stereo); | |
9868 | 576 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
|
577 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
|
578 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
|
579 0, ptr, 0, nb_coefs[ch], |
eb5916527064
Move run level decode functionality to ff_wma_run_level_decode
faust3
parents:
9658
diff
changeset
|
580 s->block_len, s->frame_len_bits, coef_nb_bits); |
783 | 581 } |
582 if (s->version == 1 && s->nb_channels >= 2) { | |
583 align_get_bits(&s->gb); | |
584 } | |
585 } | |
2967 | 586 |
783 | 587 /* normalize */ |
588 { | |
589 int n4 = s->block_len / 2; | |
590 mdct_norm = 1.0 / (float)n4; | |
591 if (s->version == 1) { | |
592 mdct_norm *= sqrt(n4); | |
593 } | |
594 } | |
595 | |
596 /* finally compute the MDCT coefficients */ | |
597 for(ch = 0; ch < s->nb_channels; ch++) { | |
598 if (s->channel_coded[ch]) { | |
9868 | 599 WMACoef *coefs1; |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
600 float *coefs, *exponents, mult, mult1, noise; |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
601 int i, j, n, n1, last_high_band, esize; |
783 | 602 float exp_power[HIGH_BAND_MAX_SIZE]; |
603 | |
604 coefs1 = s->coefs1[ch]; | |
605 exponents = s->exponents[ch]; | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
606 esize = s->exponents_bsize[ch]; |
3235 | 607 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; |
783 | 608 mult *= mdct_norm; |
609 coefs = s->coefs[ch]; | |
610 if (s->use_noise_coding) { | |
611 mult1 = mult; | |
612 /* very low freqs : noise */ | |
613 for(i = 0;i < s->coefs_start; i++) { | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
614 *coefs++ = s->noise_table[s->noise_index] * |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
615 exponents[i<<bsize>>esize] * mult1; |
783 | 616 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); |
617 } | |
2967 | 618 |
783 | 619 n1 = s->exponent_high_sizes[bsize]; |
620 | |
621 /* compute power of high bands */ | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
622 exponents = s->exponents[ch] + |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
623 (s->high_band_start[bsize]<<bsize); |
783 | 624 last_high_band = 0; /* avoid warning */ |
625 for(j=0;j<n1;j++) { | |
2967 | 626 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 627 s->block_len_bits][j]; |
628 if (s->high_band_coded[ch][j]) { | |
629 float e2, v; | |
630 e2 = 0; | |
631 for(i = 0;i < n; i++) { | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
632 v = exponents[i<<bsize>>esize]; |
783 | 633 e2 += v * v; |
634 } | |
635 exp_power[j] = e2 / n; | |
636 last_high_band = j; | |
4600 | 637 tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n); |
783 | 638 } |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
639 exponents += n<<bsize; |
783 | 640 } |
641 | |
642 /* main freqs and high freqs */ | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
643 exponents = s->exponents[ch] + (s->coefs_start<<bsize); |
783 | 644 for(j=-1;j<n1;j++) { |
645 if (j < 0) { | |
2967 | 646 n = s->high_band_start[bsize] - |
783 | 647 s->coefs_start; |
648 } else { | |
2967 | 649 n = s->exponent_high_bands[s->frame_len_bits - |
783 | 650 s->block_len_bits][j]; |
651 } | |
652 if (j >= 0 && s->high_band_coded[ch][j]) { | |
653 /* use noise with specified power */ | |
654 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); | |
3235 | 655 /* XXX: use a table */ |
656 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); | |
783 | 657 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); |
658 mult1 *= mdct_norm; | |
659 for(i = 0;i < n; i++) { | |
660 noise = s->noise_table[s->noise_index]; | |
661 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
|
662 *coefs++ = noise * |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
663 exponents[i<<bsize>>esize] * mult1; |
783 | 664 } |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
665 exponents += n<<bsize; |
783 | 666 } else { |
667 /* coded values + small noise */ | |
668 for(i = 0;i < n; i++) { | |
669 noise = s->noise_table[s->noise_index]; | |
670 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
|
671 *coefs++ = ((*coefs1++) + noise) * |
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
672 exponents[i<<bsize>>esize] * mult; |
783 | 673 } |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
674 exponents += n<<bsize; |
783 | 675 } |
676 } | |
677 | |
678 /* very high freqs : noise */ | |
679 n = s->block_len - s->coefs_end[bsize]; | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
680 mult1 = mult * exponents[((-1<<bsize))>>esize]; |
783 | 681 for(i = 0; i < n; i++) { |
682 *coefs++ = s->noise_table[s->noise_index] * mult1; | |
683 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); | |
684 } | |
685 } else { | |
686 /* XXX: optimize more */ | |
687 for(i = 0;i < s->coefs_start; i++) | |
688 *coefs++ = 0.0; | |
689 n = nb_coefs[ch]; | |
690 for(i = 0;i < n; i++) { | |
4785
4ae9ab738aec
WMA decoder improvement, output closer to the dmo binary.
banan
parents:
4737
diff
changeset
|
691 *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult; |
783 | 692 } |
693 n = s->block_len - s->coefs_end[bsize]; | |
694 for(i = 0;i < n; i++) | |
695 *coefs++ = 0.0; | |
696 } | |
697 } | |
698 } | |
699 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
700 #ifdef TRACE |
783 | 701 for(ch = 0; ch < s->nb_channels; ch++) { |
702 if (s->channel_coded[ch]) { | |
4600 | 703 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len); |
704 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len); | |
783 | 705 } |
706 } | |
707 #endif | |
2967 | 708 |
783 | 709 if (s->ms_stereo && s->channel_coded[1]) { |
710 /* nominal case for ms stereo: we do it before mdct */ | |
711 /* no need to optimize this case because it should almost | |
712 never happen */ | |
713 if (!s->channel_coded[0]) { | |
4600 | 714 tprintf(s->avctx, "rare ms-stereo case happened\n"); |
783 | 715 memset(s->coefs[0], 0, sizeof(float) * s->block_len); |
716 s->channel_coded[0] = 1; | |
717 } | |
2967 | 718 |
10236 | 719 s->dsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len); |
783 | 720 } |
721 | |
7243 | 722 next: |
783 | 723 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
|
724 int n4, index; |
783 | 725 |
7244 | 726 n4 = s->block_len / 2; |
727 if(s->channel_coded[ch]){ | |
7547 | 728 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
|
729 }else if(!(s->ms_stereo && ch==1)) |
7244 | 730 memset(s->output, 0, sizeof(s->output)); |
783 | 731 |
7244 | 732 /* multiply by the window and add in the frame */ |
733 index = (s->frame_len / 2) + s->block_pos - n4; | |
734 wma_window(s, &s->frame_out[ch][index]); | |
783 | 735 } |
7243 | 736 |
783 | 737 /* update block number */ |
738 s->block_num++; | |
739 s->block_pos += s->block_len; | |
740 if (s->block_pos >= s->frame_len) | |
741 return 1; | |
742 else | |
743 return 0; | |
744 } | |
745 | |
746 /* decode a frame of frame_len samples */ | |
4601 | 747 static int wma_decode_frame(WMACodecContext *s, int16_t *samples) |
783 | 748 { |
5525
bc4791868c52
various simplifications around recent av_clip_int16() usage
aurel
parents:
5523
diff
changeset
|
749 int ret, i, n, ch, incr; |
783 | 750 int16_t *ptr; |
751 float *iptr; | |
752 | |
1343 | 753 #ifdef TRACE |
4600 | 754 tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len); |
1343 | 755 #endif |
783 | 756 |
757 /* read each block */ | |
758 s->block_num = 0; | |
759 s->block_pos = 0; | |
760 for(;;) { | |
761 ret = wma_decode_block(s); | |
2967 | 762 if (ret < 0) |
783 | 763 return -1; |
764 if (ret) | |
765 break; | |
766 } | |
767 | |
768 /* convert frame to integer */ | |
769 n = s->frame_len; | |
770 incr = s->nb_channels; | |
771 for(ch = 0; ch < s->nb_channels; ch++) { | |
772 ptr = samples + ch; | |
773 iptr = s->frame_out[ch]; | |
774 | |
775 for(i=0;i<n;i++) { | |
5525
bc4791868c52
various simplifications around recent av_clip_int16() usage
aurel
parents:
5523
diff
changeset
|
776 *ptr = av_clip_int16(lrintf(*iptr++)); |
783 | 777 ptr += incr; |
778 } | |
779 /* prepare for next block */ | |
780 memmove(&s->frame_out[ch][0], &s->frame_out[ch][s->frame_len], | |
781 s->frame_len * sizeof(float)); | |
782 } | |
783 | |
1342
f574934c4219
uniformization (now it uses the same trace functions as h264, defined in common.h)
al3x
parents:
1303
diff
changeset
|
784 #ifdef TRACE |
4600 | 785 dump_shorts(s, "samples", samples, n * s->nb_channels); |
783 | 786 #endif |
787 return 0; | |
788 } | |
789 | |
2967 | 790 static int wma_decode_superframe(AVCodecContext *avctx, |
783 | 791 void *data, int *data_size, |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
792 AVPacket *avpkt) |
783 | 793 { |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
794 const uint8_t *buf = avpkt->data; |
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
795 int buf_size = avpkt->size; |
4601 | 796 WMACodecContext *s = avctx->priv_data; |
783 | 797 int nb_frames, bit_offset, i, pos, len; |
798 uint8_t *q; | |
799 int16_t *samples; | |
2967 | 800 |
4600 | 801 tprintf(avctx, "***decode_superframe:\n"); |
783 | 802 |
1750 | 803 if(buf_size==0){ |
804 s->last_superframe_len = 0; | |
805 return 0; | |
806 } | |
5957
0f2c4fa2c4f2
wma_decode_superframe always returns s->block_align, so make
reimar
parents:
5525
diff
changeset
|
807 if (buf_size < s->block_align) |
0f2c4fa2c4f2
wma_decode_superframe always returns s->block_align, so make
reimar
parents:
5525
diff
changeset
|
808 return 0; |
0f2c4fa2c4f2
wma_decode_superframe always returns s->block_align, so make
reimar
parents:
5525
diff
changeset
|
809 buf_size = s->block_align; |
2967 | 810 |
783 | 811 samples = data; |
812 | |
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
|
813 init_get_bits(&s->gb, buf, buf_size*8); |
2967 | 814 |
783 | 815 if (s->use_bit_reservoir) { |
816 /* read super frame header */ | |
5518 | 817 skip_bits(&s->gb, 4); /* super frame index */ |
783 | 818 nb_frames = get_bits(&s->gb, 4) - 1; |
819 | |
7242 | 820 if((nb_frames+1) * s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){ |
821 av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n"); | |
822 goto fail; | |
823 } | |
824 | |
783 | 825 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3); |
826 | |
827 if (s->last_superframe_len > 0) { | |
828 // printf("skip=%d\n", s->last_bitoffset); | |
829 /* add bit_offset bits to last frame */ | |
2967 | 830 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) > |
783 | 831 MAX_CODED_SUPERFRAME_SIZE) |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
832 goto fail; |
783 | 833 q = s->last_superframe + s->last_superframe_len; |
834 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
|
835 while (len > 7) { |
783 | 836 *q++ = (get_bits)(&s->gb, 8); |
837 len -= 8; | |
838 } | |
839 if (len > 0) { | |
840 *q++ = (get_bits)(&s->gb, len) << (8 - len); | |
841 } | |
2967 | 842 |
783 | 843 /* 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
|
844 init_get_bits(&s->gb, s->last_superframe, MAX_CODED_SUPERFRAME_SIZE*8); |
783 | 845 /* skip unused bits */ |
846 if (s->last_bitoffset > 0) | |
847 skip_bits(&s->gb, s->last_bitoffset); | |
848 /* this frame is stored in the last superframe and in the | |
849 current one */ | |
850 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
851 goto fail; |
783 | 852 samples += s->nb_channels * s->frame_len; |
853 } | |
854 | |
855 /* read each frame starting from bit_offset */ | |
856 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
|
857 init_get_bits(&s->gb, buf + (pos >> 3), (MAX_CODED_SUPERFRAME_SIZE - (pos >> 3))*8); |
783 | 858 len = pos & 7; |
859 if (len > 0) | |
860 skip_bits(&s->gb, len); | |
2967 | 861 |
783 | 862 s->reset_block_lengths = 1; |
863 for(i=0;i<nb_frames;i++) { | |
864 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
865 goto fail; |
783 | 866 samples += s->nb_channels * s->frame_len; |
867 } | |
868 | |
869 /* we copy the end of the frame in the last frame buffer */ | |
870 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7); | |
871 s->last_bitoffset = pos & 7; | |
872 pos >>= 3; | |
873 len = buf_size - pos; | |
819 | 874 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) { |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
875 goto fail; |
783 | 876 } |
877 s->last_superframe_len = len; | |
878 memcpy(s->last_superframe, buf + pos, len); | |
879 } else { | |
7242 | 880 if(s->nb_channels * s->frame_len * sizeof(int16_t) > *data_size){ |
881 av_log(s->avctx, AV_LOG_ERROR, "Insufficient output space\n"); | |
882 goto fail; | |
883 } | |
783 | 884 /* single frame decode */ |
885 if (wma_decode_frame(s, samples) < 0) | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
886 goto fail; |
783 | 887 samples += s->nb_channels * s->frame_len; |
888 } | |
4490 | 889 |
890 //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); | |
891 | |
783 | 892 *data_size = (int8_t *)samples - (int8_t *)data; |
893 return s->block_align; | |
964
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
894 fail: |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
895 /* when error, we reset the bit reservoir */ |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
896 s->last_superframe_len = 0; |
6e6773512288
oops : better error resilience - should fix most wma decoding problems
bellard
parents:
819
diff
changeset
|
897 return -1; |
783 | 898 } |
899 | |
900 AVCodec wmav1_decoder = | |
901 { | |
902 "wmav1", | |
903 CODEC_TYPE_AUDIO, | |
904 CODEC_ID_WMAV1, | |
4601 | 905 sizeof(WMACodecContext), |
783 | 906 wma_decode_init, |
907 NULL, | |
4490 | 908 ff_wma_end, |
783 | 909 wma_decode_superframe, |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6710
diff
changeset
|
910 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"), |
783 | 911 }; |
912 | |
913 AVCodec wmav2_decoder = | |
914 { | |
915 "wmav2", | |
916 CODEC_TYPE_AUDIO, | |
917 CODEC_ID_WMAV2, | |
4601 | 918 sizeof(WMACodecContext), |
783 | 919 wma_decode_init, |
920 NULL, | |
4490 | 921 ff_wma_end, |
783 | 922 wma_decode_superframe, |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6710
diff
changeset
|
923 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"), |
783 | 924 }; |