| File: | c:\siege\siege/src/siege/internal/stb/stb_vorbis.c |
| Location: | line 4577, column 13 |
| Description: | Value stored to 'previous' is never read |
| 1 | // Ogg Vorbis I audio decoder -- version 0.99996 |
| 2 | // |
| 3 | // Written in April 2007 by Sean Barrett, sponsored by RAD Game Tools. |
| 4 | // |
| 5 | // Placed in the public domain April 2007 by the author: no copyright is |
| 6 | // claimed, and you may use it for any purpose you like. |
| 7 | // |
| 8 | // No warranty for any purpose is expressed or implied by the author (nor |
| 9 | // by RAD Game Tools). Report bugs and send enhancements to the author. |
| 10 | // |
| 11 | // Get the latest version and other information at: |
| 12 | // http://nothings.org/stb_vorbis/ |
| 13 | |
| 14 | |
| 15 | // Todo: |
| 16 | // |
| 17 | // - seeking (note you can seek yourself using the pushdata API) |
| 18 | // |
| 19 | // Limitations: |
| 20 | // |
| 21 | // - floor 0 not supported (used in old ogg vorbis files) |
| 22 | // - lossless sample-truncation at beginning ignored |
| 23 | // - cannot concatenate multiple vorbis streams |
| 24 | // - sample positions are 32-bit, limiting seekable 192Khz |
| 25 | // files to around 6 hours (Ogg supports 64-bit) |
| 26 | // |
| 27 | // All of these limitations may be removed in future versions. |
| 28 | |
| 29 | |
| 30 | ////////////////////////////////////////////////////////////////////////////// |
| 31 | // |
| 32 | // HEADER BEGINS HERE |
| 33 | // |
| 34 | |
| 35 | #ifndef STB_VORBIS_INCLUDE_STB_VORBIS_H |
| 36 | #define STB_VORBIS_INCLUDE_STB_VORBIS_H |
| 37 | |
| 38 | #if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO) |
| 39 | #define STB_VORBIS_NO_STDIO 1 |
| 40 | #endif |
| 41 | |
| 42 | #ifndef STB_VORBIS_NO_STDIO |
| 43 | #include <stdio.h> |
| 44 | #endif |
| 45 | |
| 46 | #ifdef __cplusplus |
| 47 | extern "C" { |
| 48 | #endif |
| 49 | |
| 50 | /////////// THREAD SAFETY |
| 51 | |
| 52 | // Individual stb_vorbis* handles are not thread-safe; you cannot decode from |
| 53 | // them from multiple threads at the same time. However, you can have multiple |
| 54 | // stb_vorbis* handles and decode from them independently in multiple thrads. |
| 55 | |
| 56 | |
| 57 | /////////// MEMORY ALLOCATION |
| 58 | |
| 59 | // normally stb_vorbis uses malloc() to allocate memory at startup, |
| 60 | // and alloca() to allocate temporary memory during a frame on the |
| 61 | // stack. (Memory consumption will depend on the amount of setup |
| 62 | // data in the file and how you set the compile flags for speed |
| 63 | // vs. size. In my test files the maximal-size usage is ~150KB.) |
| 64 | // |
| 65 | // You can modify the wrapper functions in the source (setup_malloc, |
| 66 | // setup_temp_malloc, temp_malloc) to change this behavior, or you |
| 67 | // can use a simpler allocation model: you pass in a buffer from |
| 68 | // which stb_vorbis will allocate _all_ its memory (including the |
| 69 | // temp memory). "open" may fail with a VORBIS_outofmem if you |
| 70 | // do not pass in enough data; there is no way to determine how |
| 71 | // much you do need except to succeed (at which point you can |
| 72 | // query get_info to find the exact amount required. yes I know |
| 73 | // this is lame). |
| 74 | // |
| 75 | // If you pass in a non-NULL buffer of the type below, allocation |
| 76 | // will occur from it as described above. Otherwise just pass NULL |
| 77 | // to use malloc()/alloca() |
| 78 | |
| 79 | typedef struct |
| 80 | { |
| 81 | char *alloc_buffer; |
| 82 | int alloc_buffer_length_in_bytes; |
| 83 | } stb_vorbis_alloc; |
| 84 | |
| 85 | |
| 86 | /////////// FUNCTIONS USEABLE WITH ALL INPUT MODES |
| 87 | |
| 88 | typedef struct stb_vorbis stb_vorbis; |
| 89 | |
| 90 | typedef struct |
| 91 | { |
| 92 | unsigned int sample_rate; |
| 93 | int channels; |
| 94 | |
| 95 | unsigned int setup_memory_required; |
| 96 | unsigned int setup_temp_memory_required; |
| 97 | unsigned int temp_memory_required; |
| 98 | |
| 99 | int max_frame_size; |
| 100 | } stb_vorbis_info; |
| 101 | |
| 102 | // get general information about the file |
| 103 | extern stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f); |
| 104 | |
| 105 | // get the last error detected (clears it, too) |
| 106 | extern int stb_vorbis_get_error(stb_vorbis *f); |
| 107 | |
| 108 | // close an ogg vorbis file and free all memory in use |
| 109 | extern void stb_vorbis_close(stb_vorbis *f); |
| 110 | |
| 111 | // this function returns the offset (in samples) from the beginning of the |
| 112 | // file that will be returned by the next decode, if it is known, or -1 |
| 113 | // otherwise. after a flush_pushdata() call, this may take a while before |
| 114 | // it becomes valid again. |
| 115 | // NOT WORKING YET after a seek with PULLDATA API |
| 116 | extern int stb_vorbis_get_sample_offset(stb_vorbis *f); |
| 117 | |
| 118 | // returns the current seek point within the file, or offset from the beginning |
| 119 | // of the memory buffer. In pushdata mode it returns 0. |
| 120 | extern unsigned int stb_vorbis_get_file_offset(stb_vorbis *f); |
| 121 | |
| 122 | /////////// PUSHDATA API |
| 123 | |
| 124 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 125 | |
| 126 | // this API allows you to get blocks of data from any source and hand |
| 127 | // them to stb_vorbis. you have to buffer them; stb_vorbis will tell |
| 128 | // you how much it used, and you have to give it the rest next time; |
| 129 | // and stb_vorbis may not have enough data to work with and you will |
| 130 | // need to give it the same data again PLUS more. Note that the Vorbis |
| 131 | // specification does not bound the size of an individual frame. |
| 132 | |
| 133 | extern stb_vorbis *stb_vorbis_open_pushdata( |
| 134 | unsigned char *datablock, int datablock_length_in_bytes, |
| 135 | int *datablock_memory_consumed_in_bytes, |
| 136 | int *error, |
| 137 | stb_vorbis_alloc *alloc_buffer); |
| 138 | // create a vorbis decoder by passing in the initial data block containing |
| 139 | // the ogg&vorbis headers (you don't need to do parse them, just provide |
| 140 | // the first N bytes of the file--you're told if it's not enough, see below) |
| 141 | // on success, returns an stb_vorbis *, does not set error, returns the amount of |
| 142 | // data parsed/consumed on this call in *datablock_memory_consumed_in_bytes; |
| 143 | // on failure, returns NULL on error and sets *error, does not change *datablock_memory_consumed |
| 144 | // if returns NULL and *error is VORBIS_need_more_data, then the input block was |
| 145 | // incomplete and you need to pass in a larger block from the start of the file |
| 146 | |
| 147 | extern int stb_vorbis_decode_frame_pushdata( |
| 148 | stb_vorbis *f, unsigned char *datablock, int datablock_length_in_bytes, |
| 149 | int *channels, // place to write number of float * buffers |
| 150 | float ***output, // place to write float ** array of float * buffers |
| 151 | int *samples // place to write number of output samples |
| 152 | ); |
| 153 | // decode a frame of audio sample data if possible from the passed-in data block |
| 154 | // |
| 155 | // return value: number of bytes we used from datablock |
| 156 | // possible cases: |
| 157 | // 0 bytes used, 0 samples output (need more data) |
| 158 | // N bytes used, 0 samples output (resynching the stream, keep going) |
| 159 | // N bytes used, M samples output (one frame of data) |
| 160 | // note that after opening a file, you will ALWAYS get one N-bytes,0-sample |
| 161 | // frame, because Vorbis always "discards" the first frame. |
| 162 | // |
| 163 | // Note that on resynch, stb_vorbis will rarely consume all of the buffer, |
| 164 | // instead only datablock_length_in_bytes-3 or less. This is because it wants |
| 165 | // to avoid missing parts of a page header if they cross a datablock boundary, |
| 166 | // without writing state-machiney code to record a partial detection. |
| 167 | // |
| 168 | // The number of channels returned are stored in *channels (which can be |
| 169 | // NULL--it is always the same as the number of channels reported by |
| 170 | // get_info). *output will contain an array of float* buffers, one per |
| 171 | // channel. In other words, (*output)[0][0] contains the first sample from |
| 172 | // the first channel, and (*output)[1][0] contains the first sample from |
| 173 | // the second channel. |
| 174 | |
| 175 | extern void stb_vorbis_flush_pushdata(stb_vorbis *f); |
| 176 | // inform stb_vorbis that your next datablock will not be contiguous with |
| 177 | // previous ones (e.g. you've seeked in the data); future attempts to decode |
| 178 | // frames will cause stb_vorbis to resynchronize (as noted above), and |
| 179 | // once it sees a valid Ogg page (typically 4-8KB, as large as 64KB), it |
| 180 | // will begin decoding the _next_ frame. |
| 181 | // |
| 182 | // if you want to seek using pushdata, you need to seek in your file, then |
| 183 | // call stb_vorbis_flush_pushdata(), then start calling decoding, then once |
| 184 | // decoding is returning you data, call stb_vorbis_get_sample_offset, and |
| 185 | // if you don't like the result, seek your file again and repeat. |
| 186 | #endif |
| 187 | |
| 188 | |
| 189 | ////////// PULLING INPUT API |
| 190 | |
| 191 | #ifndef STB_VORBIS_NO_PULLDATA_API |
| 192 | // This API assumes stb_vorbis is allowed to pull data from a source-- |
| 193 | // either a block of memory containing the _entire_ vorbis stream, or a |
| 194 | // FILE * that you or it create, or possibly some other reading mechanism |
| 195 | // if you go modify the source to replace the FILE * case with some kind |
| 196 | // of callback to your code. (But if you don't support seeking, you may |
| 197 | // just want to go ahead and use pushdata.) |
| 198 | |
| 199 | #if !defined(STB_VORBIS_NO_STDIO) && !defined(STB_VORBIS_NO_INTEGER_CONVERSION) |
| 200 | extern int stb_vorbis_decode_filename(char *filename, int *channels, short **output); |
| 201 | #endif |
| 202 | extern int stb_vorbis_decode_memory(unsigned char *mem, int len, int *channels, short **output); |
| 203 | // decode an entire file and output the data interleaved into a malloc()ed |
| 204 | // buffer stored in *output. The return value is the number of samples |
| 205 | // decoded, or -1 if the file could not be opened or was not an ogg vorbis file. |
| 206 | // When you're done with it, just free() the pointer returned in *output. |
| 207 | |
| 208 | extern stb_vorbis * stb_vorbis_open_memory(unsigned char *data, int len, |
| 209 | int *error, stb_vorbis_alloc *alloc_buffer); |
| 210 | // create an ogg vorbis decoder from an ogg vorbis stream in memory (note |
| 211 | // this must be the entire stream!). on failure, returns NULL and sets *error |
| 212 | |
| 213 | #ifndef STB_VORBIS_NO_STDIO |
| 214 | extern stb_vorbis * stb_vorbis_open_filename(char *filename, |
| 215 | int *error, stb_vorbis_alloc *alloc_buffer); |
| 216 | // create an ogg vorbis decoder from a filename via fopen(). on failure, |
| 217 | // returns NULL and sets *error (possibly to VORBIS_file_open_failure). |
| 218 | |
| 219 | extern stb_vorbis * stb_vorbis_open_file(FILE *f, int close_handle_on_close, |
| 220 | int *error, stb_vorbis_alloc *alloc_buffer); |
| 221 | // create an ogg vorbis decoder from an open FILE *, looking for a stream at |
| 222 | // the _current_ seek point (ftell). on failure, returns NULL and sets *error. |
| 223 | // note that stb_vorbis must "own" this stream; if you seek it in between |
| 224 | // calls to stb_vorbis, it will become confused. Morever, if you attempt to |
| 225 | // perform stb_vorbis_seek_*() operations on this file, it will assume it |
| 226 | // owns the _entire_ rest of the file after the start point. Use the next |
| 227 | // function, stb_vorbis_open_file_section(), to limit it. |
| 228 | |
| 229 | extern stb_vorbis * stb_vorbis_open_file_section(FILE *f, int close_handle_on_close, |
| 230 | int *error, stb_vorbis_alloc *alloc_buffer, unsigned int len); |
| 231 | // create an ogg vorbis decoder from an open FILE *, looking for a stream at |
| 232 | // the _current_ seek point (ftell); the stream will be of length 'len' bytes. |
| 233 | // on failure, returns NULL and sets *error. note that stb_vorbis must "own" |
| 234 | // this stream; if you seek it in between calls to stb_vorbis, it will become |
| 235 | // confused. |
| 236 | #endif |
| 237 | |
| 238 | extern int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number); |
| 239 | extern int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number); |
| 240 | // NOT WORKING YET |
| 241 | // these functions seek in the Vorbis file to (approximately) 'sample_number'. |
| 242 | // after calling seek_frame(), the next call to get_frame_*() will include |
| 243 | // the specified sample. after calling stb_vorbis_seek(), the next call to |
| 244 | // stb_vorbis_get_samples_* will start with the specified sample. If you |
| 245 | // do not need to seek to EXACTLY the target sample when using get_samples_*, |
| 246 | // you can also use seek_frame(). |
| 247 | |
| 248 | extern void stb_vorbis_seek_start(stb_vorbis *f); |
| 249 | // this function is equivalent to stb_vorbis_seek(f,0), but it |
| 250 | // actually works |
| 251 | |
| 252 | extern unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f); |
| 253 | extern float stb_vorbis_stream_length_in_seconds(stb_vorbis *f); |
| 254 | // these functions return the total length of the vorbis stream |
| 255 | |
| 256 | extern int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output); |
| 257 | // decode the next frame and return the number of samples. the number of |
| 258 | // channels returned are stored in *channels (which can be NULL--it is always |
| 259 | // the same as the number of channels reported by get_info). *output will |
| 260 | // contain an array of float* buffers, one per channel. These outputs will |
| 261 | // be overwritten on the next call to stb_vorbis_get_frame_*. |
| 262 | // |
| 263 | // You generally should not intermix calls to stb_vorbis_get_frame_*() |
| 264 | // and stb_vorbis_get_samples_*(), since the latter calls the former. |
| 265 | |
| 266 | #ifndef STB_VORBIS_NO_INTEGER_CONVERSION |
| 267 | extern int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts); |
| 268 | extern int stb_vorbis_get_frame_short (stb_vorbis *f, int num_c, short **buffer, int num_samples); |
| 269 | #endif |
| 270 | // decode the next frame and return the number of samples per channel. the |
| 271 | // data is coerced to the number of channels you request according to the |
| 272 | // channel coercion rules (see below). You must pass in the size of your |
| 273 | // buffer(s) so that stb_vorbis will not overwrite the end of the buffer. |
| 274 | // The maximum buffer size needed can be gotten from get_info(); however, |
| 275 | // the Vorbis I specification implies an absolute maximum of 4096 samples |
| 276 | // per channel. Note that for interleaved data, you pass in the number of |
| 277 | // shorts (the size of your array), but the return value is the number of |
| 278 | // samples per channel, not the total number of samples. |
| 279 | |
| 280 | // Channel coercion rules: |
| 281 | // Let M be the number of channels requested, and N the number of channels present, |
| 282 | // and Cn be the nth channel; let stereo L be the sum of all L and center channels, |
| 283 | // and stereo R be the sum of all R and center channels (channel assignment from the |
| 284 | // vorbis spec). |
| 285 | // M N output |
| 286 | // 1 k sum(Ck) for all k |
| 287 | // 2 * stereo L, stereo R |
| 288 | // k l k > l, the first l channels, then 0s |
| 289 | // k l k <= l, the first k channels |
| 290 | // Note that this is not _good_ surround etc. mixing at all! It's just so |
| 291 | // you get something useful. |
| 292 | |
| 293 | extern int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats); |
| 294 | extern int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples); |
| 295 | // gets num_samples samples, not necessarily on a frame boundary--this requires |
| 296 | // buffering so you have to supply the buffers. DOES NOT APPLY THE COERCION RULES. |
| 297 | // Returns the number of samples stored per channel; it may be less than requested |
| 298 | // at the end of the file. If there are no more samples in the file, returns 0. |
| 299 | |
| 300 | #ifndef STB_VORBIS_NO_INTEGER_CONVERSION |
| 301 | extern int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts); |
| 302 | extern int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int num_samples); |
| 303 | #endif |
| 304 | // gets num_samples samples, not necessarily on a frame boundary--this requires |
| 305 | // buffering so you have to supply the buffers. Applies the coercion rules above |
| 306 | // to produce 'channels' channels. Returns the number of samples stored per channel; |
| 307 | // it may be less than requested at the end of the file. If there are no more |
| 308 | // samples in the file, returns 0. |
| 309 | |
| 310 | #endif |
| 311 | |
| 312 | //////// ERROR CODES |
| 313 | |
| 314 | enum STBVorbisError |
| 315 | { |
| 316 | VORBIS__no_error, |
| 317 | |
| 318 | VORBIS_need_more_data=1, // not a real error |
| 319 | |
| 320 | VORBIS_invalid_api_mixing, // can't mix API modes |
| 321 | VORBIS_outofmem, // not enough memory |
| 322 | VORBIS_feature_not_supported, // uses floor 0 |
| 323 | VORBIS_too_many_channels, // STB_VORBIS_MAX_CHANNELS is too small |
| 324 | VORBIS_file_open_failure, // fopen() failed |
| 325 | VORBIS_seek_without_length, // can't seek in unknown-length file |
| 326 | |
| 327 | VORBIS_unexpected_eof=10, // file is truncated? |
| 328 | VORBIS_seek_invalid, // seek past EOF |
| 329 | |
| 330 | // decoding errors (corrupt/invalid stream) -- you probably |
| 331 | // don't care about the exact details of these |
| 332 | |
| 333 | // vorbis errors: |
| 334 | VORBIS_invalid_setup=20, |
| 335 | VORBIS_invalid_stream, |
| 336 | |
| 337 | // ogg errors: |
| 338 | VORBIS_missing_capture_pattern=30, |
| 339 | VORBIS_invalid_stream_structure_version, |
| 340 | VORBIS_continued_packet_flag_invalid, |
| 341 | VORBIS_incorrect_stream_serial_number, |
| 342 | VORBIS_invalid_first_page, |
| 343 | VORBIS_bad_packet_type, |
| 344 | VORBIS_cant_find_last_page, |
| 345 | VORBIS_seek_failed, |
| 346 | }; |
| 347 | |
| 348 | |
| 349 | #ifdef __cplusplus |
| 350 | } |
| 351 | #endif |
| 352 | |
| 353 | #endif // STB_VORBIS_INCLUDE_STB_VORBIS_H |
| 354 | // |
| 355 | // HEADER ENDS HERE |
| 356 | // |
| 357 | ////////////////////////////////////////////////////////////////////////////// |
| 358 | |
| 359 | #ifndef STB_VORBIS_HEADER_ONLY |
| 360 | |
| 361 | // global configuration settings (e.g. set these in the project/makefile), |
| 362 | // or just set them in this file at the top (although ideally the first few |
| 363 | // should be visible when the header file is compiled too, although it's not |
| 364 | // crucial) |
| 365 | |
| 366 | // STB_VORBIS_NO_PUSHDATA_API |
| 367 | // does not compile the code for the various stb_vorbis_*_pushdata() |
| 368 | // functions |
| 369 | // #define STB_VORBIS_NO_PUSHDATA_API |
| 370 | |
| 371 | // STB_VORBIS_NO_PULLDATA_API |
| 372 | // does not compile the code for the non-pushdata APIs |
| 373 | // #define STB_VORBIS_NO_PULLDATA_API |
| 374 | |
| 375 | // STB_VORBIS_NO_STDIO |
| 376 | // does not compile the code for the APIs that use FILE *s internally |
| 377 | // or externally (implied by STB_VORBIS_NO_PULLDATA_API) |
| 378 | // #define STB_VORBIS_NO_STDIO |
| 379 | |
| 380 | // STB_VORBIS_NO_INTEGER_CONVERSION |
| 381 | // does not compile the code for converting audio sample data from |
| 382 | // float to integer (implied by STB_VORBIS_NO_PULLDATA_API) |
| 383 | // #define STB_VORBIS_NO_INTEGER_CONVERSION |
| 384 | |
| 385 | // STB_VORBIS_NO_FAST_SCALED_FLOAT |
| 386 | // does not use a fast float-to-int trick to accelerate float-to-int on |
| 387 | // most platforms which requires endianness be defined correctly. |
| 388 | //#define STB_VORBIS_NO_FAST_SCALED_FLOAT |
| 389 | |
| 390 | |
| 391 | // STB_VORBIS_MAX_CHANNELS [number] |
| 392 | // globally define this to the maximum number of channels you need. |
| 393 | // The spec does not put a restriction on channels except that |
| 394 | // the count is stored in a byte, so 255 is the hard limit. |
| 395 | // Reducing this saves about 16 bytes per value, so using 16 saves |
| 396 | // (255-16)*16 or around 4KB. Plus anything other memory usage |
| 397 | // I forgot to account for. Can probably go as low as 8 (7.1 audio), |
| 398 | // 6 (5.1 audio), or 2 (stereo only). |
| 399 | #ifndef STB_VORBIS_MAX_CHANNELS16 |
| 400 | #define STB_VORBIS_MAX_CHANNELS16 16 // enough for anyone? |
| 401 | #endif |
| 402 | |
| 403 | // STB_VORBIS_PUSHDATA_CRC_COUNT [number] |
| 404 | // after a flush_pushdata(), stb_vorbis begins scanning for the |
| 405 | // next valid page, without backtracking. when it finds something |
| 406 | // that looks like a page, it streams through it and verifies its |
| 407 | // CRC32. Should that validation fail, it keeps scanning. But it's |
| 408 | // possible that _while_ streaming through to check the CRC32 of |
| 409 | // one candidate page, it sees another candidate page. This #define |
| 410 | // determines how many "overlapping" candidate pages it can search |
| 411 | // at once. Note that "real" pages are typically ~4KB to ~8KB, whereas |
| 412 | // garbage pages could be as big as 64KB, but probably average ~16KB. |
| 413 | // So don't hose ourselves by scanning an apparent 64KB page and |
| 414 | // missing a ton of real ones in the interim; so minimum of 2 |
| 415 | #ifndef STB_VORBIS_PUSHDATA_CRC_COUNT4 |
| 416 | #define STB_VORBIS_PUSHDATA_CRC_COUNT4 4 |
| 417 | #endif |
| 418 | |
| 419 | // STB_VORBIS_FAST_HUFFMAN_LENGTH [number] |
| 420 | // sets the log size of the huffman-acceleration table. Maximum |
| 421 | // supported value is 24. with larger numbers, more decodings are O(1), |
| 422 | // but the table size is larger so worse cache missing, so you'll have |
| 423 | // to probe (and try multiple ogg vorbis files) to find the sweet spot. |
| 424 | #ifndef STB_VORBIS_FAST_HUFFMAN_LENGTH10 |
| 425 | #define STB_VORBIS_FAST_HUFFMAN_LENGTH10 10 |
| 426 | #endif |
| 427 | |
| 428 | // STB_VORBIS_FAST_BINARY_LENGTH [number] |
| 429 | // sets the log size of the binary-search acceleration table. this |
| 430 | // is used in similar fashion to the fast-huffman size to set initial |
| 431 | // parameters for the binary search |
| 432 | |
| 433 | // STB_VORBIS_FAST_HUFFMAN_INT |
| 434 | // The fast huffman tables are much more efficient if they can be |
| 435 | // stored as 16-bit results instead of 32-bit results. This restricts |
| 436 | // the codebooks to having only 65535 possible outcomes, though. |
| 437 | // (At least, accelerated by the huffman table.) |
| 438 | #ifndef STB_VORBIS_FAST_HUFFMAN_INT |
| 439 | #define STB_VORBIS_FAST_HUFFMAN_SHORT |
| 440 | #endif |
| 441 | |
| 442 | // STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH |
| 443 | // If the 'fast huffman' search doesn't succeed, then stb_vorbis falls |
| 444 | // back on binary searching for the correct one. This requires storing |
| 445 | // extra tables with the huffman codes in sorted order. Defining this |
| 446 | // symbol trades off space for speed by forcing a linear search in the |
| 447 | // non-fast case, except for "sparse" codebooks. |
| 448 | // #define STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH |
| 449 | |
| 450 | // STB_VORBIS_DIVIDES_IN_RESIDUE |
| 451 | // stb_vorbis precomputes the result of the scalar residue decoding |
| 452 | // that would otherwise require a divide per chunk. you can trade off |
| 453 | // space for time by defining this symbol. |
| 454 | // #define STB_VORBIS_DIVIDES_IN_RESIDUE |
| 455 | |
| 456 | // STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 457 | // vorbis VQ codebooks can be encoded two ways: with every case explicitly |
| 458 | // stored, or with all elements being chosen from a small range of values, |
| 459 | // and all values possible in all elements. By default, stb_vorbis expands |
| 460 | // this latter kind out to look like the former kind for ease of decoding, |
| 461 | // because otherwise an integer divide-per-vector-element is required to |
| 462 | // unpack the index. If you define STB_VORBIS_DIVIDES_IN_CODEBOOK, you can |
| 463 | // trade off storage for speed. |
| 464 | //#define STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 465 | |
| 466 | // STB_VORBIS_CODEBOOK_SHORTS |
| 467 | // The vorbis file format encodes VQ codebook floats as ax+b where a and |
| 468 | // b are floating point per-codebook constants, and x is a 16-bit int. |
| 469 | // Normally, stb_vorbis decodes them to floats rather than leaving them |
| 470 | // as 16-bit ints and computing ax+b while decoding. This is a speed/space |
| 471 | // tradeoff; you can save space by defining this flag. |
| 472 | #ifndef STB_VORBIS_CODEBOOK_SHORTS |
| 473 | #define STB_VORBIS_CODEBOOK_FLOATS |
| 474 | #endif |
| 475 | |
| 476 | // STB_VORBIS_DIVIDE_TABLE |
| 477 | // this replaces small integer divides in the floor decode loop with |
| 478 | // table lookups. made less than 1% difference, so disabled by default. |
| 479 | |
| 480 | // STB_VORBIS_NO_INLINE_DECODE |
| 481 | // disables the inlining of the scalar codebook fast-huffman decode. |
| 482 | // might save a little codespace; useful for debugging |
| 483 | // #define STB_VORBIS_NO_INLINE_DECODE |
| 484 | |
| 485 | // STB_VORBIS_NO_DEFER_FLOOR |
| 486 | // Normally we only decode the floor without synthesizing the actual |
| 487 | // full curve. We can instead synthesize the curve immediately. This |
| 488 | // requires more memory and is very likely slower, so I don't think |
| 489 | // you'd ever want to do it except for debugging. |
| 490 | // #define STB_VORBIS_NO_DEFER_FLOOR |
| 491 | |
| 492 | |
| 493 | |
| 494 | |
| 495 | ////////////////////////////////////////////////////////////////////////////// |
| 496 | |
| 497 | #ifdef STB_VORBIS_NO_PULLDATA_API |
| 498 | #define STB_VORBIS_NO_INTEGER_CONVERSION |
| 499 | #define STB_VORBIS_NO_STDIO |
| 500 | #endif |
| 501 | |
| 502 | #if defined(STB_VORBIS_NO_CRT) && !defined(STB_VORBIS_NO_STDIO) |
| 503 | #define STB_VORBIS_NO_STDIO 1 |
| 504 | #endif |
| 505 | |
| 506 | #ifndef STB_VORBIS_NO_INTEGER_CONVERSION |
| 507 | #ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT |
| 508 | |
| 509 | // only need endianness for fast-float-to-int, which we don't |
| 510 | // use for pushdata |
| 511 | |
| 512 | #ifndef STB_VORBIS_BIG_ENDIAN |
| 513 | #define STB_VORBIS_ENDIAN0 0 |
| 514 | #else |
| 515 | #define STB_VORBIS_ENDIAN0 1 |
| 516 | #endif |
| 517 | |
| 518 | #endif |
| 519 | #endif |
| 520 | |
| 521 | |
| 522 | #ifndef STB_VORBIS_NO_STDIO |
| 523 | #include <stdio.h> |
| 524 | #endif |
| 525 | |
| 526 | #ifndef STB_VORBIS_NO_CRT |
| 527 | #include <stdlib.h> |
| 528 | #include <string.h> |
| 529 | #include <assert.h> |
| 530 | #include <math.h> |
| 531 | #else |
| 532 | #define NULL((void*)0) 0 |
| 533 | #endif |
| 534 | |
| 535 | #ifndef _MSC_VER |
| 536 | #if __GNUC__4 |
| 537 | #define __forceinlineinline inline |
| 538 | #else |
| 539 | #define __forceinlineinline |
| 540 | #endif |
| 541 | #endif |
| 542 | |
| 543 | #if STB_VORBIS_MAX_CHANNELS16 > 256 |
| 544 | #error "Value of STB_VORBIS_MAX_CHANNELS outside of allowed range" |
| 545 | #endif |
| 546 | |
| 547 | #if STB_VORBIS_FAST_HUFFMAN_LENGTH10 > 24 |
| 548 | #error "Value of STB_VORBIS_FAST_HUFFMAN_LENGTH outside of allowed range" |
| 549 | #endif |
| 550 | |
| 551 | |
| 552 | #define MAX_BLOCKSIZE_LOG13 13 // from specification |
| 553 | #define MAX_BLOCKSIZE(1 << 13) (1 << MAX_BLOCKSIZE_LOG13) |
| 554 | |
| 555 | |
| 556 | typedef unsigned char uint8; |
| 557 | typedef signed char int8; |
| 558 | typedef unsigned short uint16; |
| 559 | typedef signed short int16; |
| 560 | typedef unsigned int uint32; |
| 561 | typedef signed int int32; |
| 562 | |
| 563 | #ifndef TRUE1 |
| 564 | #define TRUE1 1 |
| 565 | #define FALSE0 0 |
| 566 | #endif |
| 567 | |
| 568 | #ifdef STB_VORBIS_CODEBOOK_FLOATS |
| 569 | typedef float codetype; |
| 570 | #else |
| 571 | typedef uint16 codetype; |
| 572 | #endif |
| 573 | |
| 574 | // @NOTE |
| 575 | // |
| 576 | // Some arrays below are tagged "//varies", which means it's actually |
| 577 | // a variable-sized piece of data, but rather than malloc I assume it's |
| 578 | // small enough it's better to just allocate it all together with the |
| 579 | // main thing |
| 580 | // |
| 581 | // Most of the variables are specified with the smallest size I could pack |
| 582 | // them into. It might give better performance to make them all full-sized |
| 583 | // integers. It should be safe to freely rearrange the structures or change |
| 584 | // the sizes larger--nothing relies on silently truncating etc., nor the |
| 585 | // order of variables. |
| 586 | |
| 587 | #define FAST_HUFFMAN_TABLE_SIZE(1 << 10) (1 << STB_VORBIS_FAST_HUFFMAN_LENGTH10) |
| 588 | #define FAST_HUFFMAN_TABLE_MASK((1 << 10) - 1) (FAST_HUFFMAN_TABLE_SIZE(1 << 10) - 1) |
| 589 | |
| 590 | typedef struct |
| 591 | { |
| 592 | int dimensions, entries; |
| 593 | uint8 *codeword_lengths; |
| 594 | float minimum_value; |
| 595 | float delta_value; |
| 596 | uint8 value_bits; |
| 597 | uint8 lookup_type; |
| 598 | uint8 sequence_p; |
| 599 | uint8 sparse; |
| 600 | uint32 lookup_values; |
| 601 | codetype *multiplicands; |
| 602 | uint32 *codewords; |
| 603 | #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT |
| 604 | int16 fast_huffman[FAST_HUFFMAN_TABLE_SIZE(1 << 10)]; |
| 605 | #else |
| 606 | int32 fast_huffman[FAST_HUFFMAN_TABLE_SIZE(1 << 10)]; |
| 607 | #endif |
| 608 | uint32 *sorted_codewords; |
| 609 | int *sorted_values; |
| 610 | int sorted_entries; |
| 611 | } Codebook; |
| 612 | |
| 613 | typedef struct |
| 614 | { |
| 615 | uint8 order; |
| 616 | uint16 rate; |
| 617 | uint16 bark_map_size; |
| 618 | uint8 amplitude_bits; |
| 619 | uint8 amplitude_offset; |
| 620 | uint8 number_of_books; |
| 621 | uint8 book_list[16]; // varies |
| 622 | } Floor0; |
| 623 | |
| 624 | typedef struct |
| 625 | { |
| 626 | uint8 partitions; |
| 627 | uint8 partition_class_list[32]; // varies |
| 628 | uint8 class_dimensions[16]; // varies |
| 629 | uint8 class_subclasses[16]; // varies |
| 630 | uint8 class_masterbooks[16]; // varies |
| 631 | int16 subclass_books[16][8]; // varies |
| 632 | uint16 Xlist[31*8+2]; // varies |
| 633 | uint8 sorted_order[31*8+2]; |
| 634 | uint8 neighbors[31*8+2][2]; |
| 635 | uint8 floor1_multiplier; |
| 636 | uint8 rangebits; |
| 637 | int values; |
| 638 | } Floor1; |
| 639 | |
| 640 | typedef union |
| 641 | { |
| 642 | Floor0 floor0; |
| 643 | Floor1 floor1; |
| 644 | } Floor; |
| 645 | |
| 646 | typedef struct |
| 647 | { |
| 648 | uint32 begin, end; |
| 649 | uint32 part_size; |
| 650 | uint8 classifications; |
| 651 | uint8 classbook; |
| 652 | uint8 **classdata; |
| 653 | int16 (*residue_books)[8]; |
| 654 | } Residue; |
| 655 | |
| 656 | typedef struct |
| 657 | { |
| 658 | uint8 magnitude; |
| 659 | uint8 angle; |
| 660 | uint8 mux; |
| 661 | } MappingChannel; |
| 662 | |
| 663 | typedef struct |
| 664 | { |
| 665 | uint16 coupling_steps; |
| 666 | MappingChannel *chan; |
| 667 | uint8 submaps; |
| 668 | uint8 submap_floor[15]; // varies |
| 669 | uint8 submap_residue[15]; // varies |
| 670 | } Mapping; |
| 671 | |
| 672 | typedef struct |
| 673 | { |
| 674 | uint8 blockflag; |
| 675 | uint8 mapping; |
| 676 | uint16 windowtype; |
| 677 | uint16 transformtype; |
| 678 | } Mode; |
| 679 | |
| 680 | typedef struct |
| 681 | { |
| 682 | uint32 goal_crc; // expected crc if match |
| 683 | int bytes_left; // bytes left in packet |
| 684 | uint32 crc_so_far; // running crc |
| 685 | int bytes_done; // bytes processed in _current_ chunk |
| 686 | uint32 sample_loc; // granule pos encoded in page |
| 687 | } CRCscan; |
| 688 | |
| 689 | typedef struct |
| 690 | { |
| 691 | uint32 page_start, page_end; |
| 692 | uint32 after_previous_page_start; |
| 693 | uint32 first_decoded_sample; |
| 694 | uint32 last_decoded_sample; |
| 695 | } ProbedPage; |
| 696 | |
| 697 | struct stb_vorbis |
| 698 | { |
| 699 | // user-accessible info |
| 700 | unsigned int sample_rate; |
| 701 | int channels; |
| 702 | |
| 703 | unsigned int setup_memory_required; |
| 704 | unsigned int temp_memory_required; |
| 705 | unsigned int setup_temp_memory_required; |
| 706 | |
| 707 | // input config |
| 708 | #ifndef STB_VORBIS_NO_STDIO |
| 709 | FILE *f; |
| 710 | uint32 f_start; |
| 711 | int close_on_free; |
| 712 | #endif |
| 713 | |
| 714 | uint8 *stream; |
| 715 | uint8 *stream_start; |
| 716 | uint8 *stream_end; |
| 717 | |
| 718 | uint32 stream_len; |
| 719 | |
| 720 | uint8 push_mode; |
| 721 | |
| 722 | uint32 first_audio_page_offset; |
| 723 | |
| 724 | ProbedPage p_first, p_last; |
| 725 | |
| 726 | // memory management |
| 727 | stb_vorbis_alloc alloc; |
| 728 | int setup_offset; |
| 729 | int temp_offset; |
| 730 | |
| 731 | // run-time results |
| 732 | int eof; |
| 733 | enum STBVorbisError error; |
| 734 | |
| 735 | // user-useful data |
| 736 | |
| 737 | // header info |
| 738 | int blocksize[2]; |
| 739 | int blocksize_0, blocksize_1; |
| 740 | int codebook_count; |
| 741 | Codebook *codebooks; |
| 742 | int floor_count; |
| 743 | uint16 floor_types[64]; // varies |
| 744 | Floor *floor_config; |
| 745 | int residue_count; |
| 746 | uint16 residue_types[64]; // varies |
| 747 | Residue *residue_config; |
| 748 | int mapping_count; |
| 749 | Mapping *mapping; |
| 750 | int mode_count; |
| 751 | Mode mode_config[64]; // varies |
| 752 | |
| 753 | uint32 total_samples; |
| 754 | |
| 755 | // decode buffer |
| 756 | float *channel_buffers[STB_VORBIS_MAX_CHANNELS16]; |
| 757 | float *outputs [STB_VORBIS_MAX_CHANNELS16]; |
| 758 | |
| 759 | float *previous_window[STB_VORBIS_MAX_CHANNELS16]; |
| 760 | int previous_length; |
| 761 | |
| 762 | #ifndef STB_VORBIS_NO_DEFER_FLOOR |
| 763 | int16 *finalY[STB_VORBIS_MAX_CHANNELS16]; |
| 764 | #else |
| 765 | float *floor_buffers[STB_VORBIS_MAX_CHANNELS16]; |
| 766 | #endif |
| 767 | |
| 768 | uint32 current_loc; // sample location of next frame to decode |
| 769 | int current_loc_valid; |
| 770 | |
| 771 | // per-blocksize precomputed data |
| 772 | |
| 773 | // twiddle factors |
| 774 | float *A[2],*B[2],*C(2 | 4 | 1)[2]; |
| 775 | float *window[2]; |
| 776 | uint16 *bit_reverse[2]; |
| 777 | |
| 778 | // current page/packet/segment streaming info |
| 779 | uint32 serial; // stream serial number for verification |
| 780 | int last_page; |
| 781 | int segment_count; |
| 782 | uint8 segments[255]; |
| 783 | uint8 page_flag; |
| 784 | uint8 bytes_in_seg; |
| 785 | uint8 first_decode; |
| 786 | int next_seg; |
| 787 | int last_seg; // flag that we're on the last segment |
| 788 | int last_seg_which; // what was the segment number of the last seg? |
| 789 | uint32 acc; |
| 790 | int valid_bits; |
| 791 | int packet_bytes; |
| 792 | int end_seg_with_known_loc; |
| 793 | uint32 known_loc_for_packet; |
| 794 | int discard_samples_deferred; |
| 795 | uint32 samples_output; |
| 796 | |
| 797 | // push mode scanning |
| 798 | int page_crc_tests; // only in push_mode: number of tests active; -1 if not searching |
| 799 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 800 | CRCscan scan[STB_VORBIS_PUSHDATA_CRC_COUNT4]; |
| 801 | #endif |
| 802 | |
| 803 | // sample-access |
| 804 | int channel_buffer_start; |
| 805 | int channel_buffer_end; |
| 806 | }; |
| 807 | |
| 808 | extern int my_prof(int slot); |
| 809 | //#define stb_prof my_prof |
| 810 | |
| 811 | #ifndef stb_prof |
| 812 | #define stb_prof(x)(void)0 (void)0 |
| 813 | #endif |
| 814 | |
| 815 | #if defined(STB_VORBIS_NO_PUSHDATA_API) |
| 816 | #define IS_PUSH_MODE(f)((f)->push_mode) FALSE0 |
| 817 | #elif defined(STB_VORBIS_NO_PULLDATA_API) |
| 818 | #define IS_PUSH_MODE(f)((f)->push_mode) TRUE1 |
| 819 | #else |
| 820 | #define IS_PUSH_MODE(f)((f)->push_mode) ((f)->push_mode) |
| 821 | #endif |
| 822 | |
| 823 | typedef struct stb_vorbis vorb; |
| 824 | |
| 825 | static int error(vorb *f, enum STBVorbisError e) |
| 826 | { |
| 827 | f->error = e; |
| 828 | if (!f->eof && e != VORBIS_need_more_data) { |
| 829 | f->error=e; // breakpoint for debugging |
| 830 | } |
| 831 | return 0; |
| 832 | } |
| 833 | |
| 834 | |
| 835 | // these functions are used for allocating temporary memory |
| 836 | // while decoding. if you can afford the stack space, use |
| 837 | // alloca(); otherwise, provide a temp buffer and it will |
| 838 | // allocate out of those. |
| 839 | |
| 840 | #define array_size_required(count,size)(count*(sizeof(void *)+(size))) (count*(sizeof(void *)+(size))) |
| 841 | |
| 842 | #define temp_alloc(f,size)(f->alloc.alloc_buffer ? setup_temp_malloc(f,size) : alloca (size)) (f->alloc.alloc_buffer ? setup_temp_malloc(f,size) : alloca(size)) |
| 843 | #ifdef dealloca |
| 844 | #define temp_free(f,p)0 (f->alloc.alloc_buffer ? 0 : dealloca(size)) |
| 845 | #else |
| 846 | #define temp_free(f,p)0 0 |
| 847 | #endif |
| 848 | #define temp_alloc_save(f)((f)->temp_offset) ((f)->temp_offset) |
| 849 | #define temp_alloc_restore(f,p)((f)->temp_offset = (p)) ((f)->temp_offset = (p)) |
| 850 | |
| 851 | #define temp_block_array(f,count,size)make_block_array((f->alloc.alloc_buffer ? setup_temp_malloc (f,(count*(sizeof(void *)+(size)))) : alloca((count*(sizeof(void *)+(size))))), count, size) make_block_array(temp_alloc(f,array_size_required(count,size))(f->alloc.alloc_buffer ? setup_temp_malloc(f,(count*(sizeof (void *)+(size)))) : alloca((count*(sizeof(void *)+(size))))), count, size) |
| 852 | |
| 853 | // given a sufficiently large block of memory, make an array of pointers to subblocks of it |
| 854 | static void *make_block_array(void *mem, int count, int size) |
| 855 | { |
| 856 | int i; |
| 857 | void ** p = (void **) mem; |
| 858 | char *q = (char *) (p + count); |
| 859 | for (i=0; i < count; ++i) { |
| 860 | p[i] = q; |
| 861 | q += size; |
| 862 | } |
| 863 | return p; |
| 864 | } |
| 865 | |
| 866 | static void *setup_malloc(vorb *f, int sz) |
| 867 | { |
| 868 | sz = (sz+3) & ~3; |
| 869 | f->setup_memory_required += sz; |
| 870 | if (f->alloc.alloc_buffer) { |
| 871 | void *p = (char *) f->alloc.alloc_buffer + f->setup_offset; |
| 872 | if (f->setup_offset + sz > f->temp_offset) return NULL((void*)0); |
| 873 | f->setup_offset += sz; |
| 874 | return p; |
| 875 | } |
| 876 | return sz ? malloc(sz) : NULL((void*)0); |
| 877 | } |
| 878 | |
| 879 | static void setup_free(vorb *f, void *p) |
| 880 | { |
| 881 | if (f->alloc.alloc_buffer) return; // do nothing; setup mem is not a stack |
| 882 | free(p); |
| 883 | } |
| 884 | |
| 885 | static void *setup_temp_malloc(vorb *f, int sz) |
| 886 | { |
| 887 | sz = (sz+3) & ~3; |
| 888 | if (f->alloc.alloc_buffer) { |
| 889 | if (f->temp_offset - sz < f->setup_offset) return NULL((void*)0); |
| 890 | f->temp_offset -= sz; |
| 891 | return (char *) f->alloc.alloc_buffer + f->temp_offset; |
| 892 | } |
| 893 | return malloc(sz); |
| 894 | } |
| 895 | |
| 896 | static void setup_temp_free(vorb *f, void *p, size_t sz) |
| 897 | { |
| 898 | if (f->alloc.alloc_buffer) { |
| 899 | f->temp_offset += (sz+3)&~3; |
| 900 | return; |
| 901 | } |
| 902 | free(p); |
| 903 | } |
| 904 | |
| 905 | #define CRC32_POLY0x04c11db7 0x04c11db7 // from spec |
| 906 | |
| 907 | static uint32 crc_table[256]; |
| 908 | static void crc32_init(void) |
| 909 | { |
| 910 | int i,j; |
| 911 | uint32 s; |
| 912 | for(i=0; i < 256; i++) { |
| 913 | for (s=i<<24, j=0; j < 8; ++j) |
| 914 | s = (s << 1) ^ (s >= (1<<31) ? CRC32_POLY0x04c11db7 : 0); |
| 915 | crc_table[i] = s; |
| 916 | } |
| 917 | } |
| 918 | |
| 919 | static __forceinlineinline uint32 crc32_update(uint32 crc, uint8 byte) |
| 920 | { |
| 921 | return (crc << 8) ^ crc_table[byte ^ (crc >> 24)]; |
| 922 | } |
| 923 | |
| 924 | |
| 925 | // used in setup, and for huffman that doesn't go fast path |
| 926 | static unsigned int bit_reverse(unsigned int n) |
| 927 | { |
| 928 | n = ((n & 0xAAAAAAAA) >> 1) | ((n & 0x55555555) << 1); |
| 929 | n = ((n & 0xCCCCCCCC) >> 2) | ((n & 0x33333333) << 2); |
| 930 | n = ((n & 0xF0F0F0F0) >> 4) | ((n & 0x0F0F0F0F) << 4); |
| 931 | n = ((n & 0xFF00FF00) >> 8) | ((n & 0x00FF00FF) << 8); |
| 932 | return (n >> 16) | (n << 16); |
| 933 | } |
| 934 | |
| 935 | static float square(float x) |
| 936 | { |
| 937 | return x*x; |
| 938 | } |
| 939 | |
| 940 | // this is a weird definition of log2() for which log2(1) = 1, log2(2) = 2, log2(4) = 3 |
| 941 | // as required by the specification. fast(?) implementation from stb.h |
| 942 | // @OPTIMIZE: called multiple times per-packet with "constants"; move to setup |
| 943 | static int ilog(int32 n) |
| 944 | { |
| 945 | static signed char log2_4[16] = { 0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4 }; |
| 946 | |
| 947 | // 2 compares if n < 16, 3 compares otherwise (4 if signed or n > 1<<29) |
| 948 | if (n < (1U << 14)) |
| 949 | if (n < (1U << 4)) return 0 + log2_4[n ]; |
| 950 | else if (n < (1U << 9)) return 5 + log2_4[n >> 5]; |
| 951 | else return 10 + log2_4[n >> 10]; |
| 952 | else if (n < (1U << 24)) |
| 953 | if (n < (1U << 19)) return 15 + log2_4[n >> 15]; |
| 954 | else return 20 + log2_4[n >> 20]; |
| 955 | else if (n < (1U << 29)) return 25 + log2_4[n >> 25]; |
| 956 | else if (n < (1U << 31)) return 30 + log2_4[n >> 30]; |
| 957 | else return 0; // signed n returns 0 |
| 958 | } |
| 959 | |
| 960 | #ifndef M_PI3.14159265358979323846 |
| 961 | #define M_PI3.14159265358979323846 3.14159265358979323846264f // from CRC |
| 962 | #endif |
| 963 | |
| 964 | // code length assigned to a value with no huffman encoding |
| 965 | #define NO_CODE255 255 |
| 966 | |
| 967 | /////////////////////// LEAF SETUP FUNCTIONS ////////////////////////// |
| 968 | // |
| 969 | // these functions are only called at setup, and only a few times |
| 970 | // per file |
| 971 | |
| 972 | static float float32_unpack(uint32 x) |
| 973 | { |
| 974 | // from the specification |
| 975 | uint32 mantissa = x & 0x1fffff; |
| 976 | uint32 sign = x & 0x80000000; |
| 977 | uint32 exp = (x & 0x7fe00000) >> 21; |
| 978 | double res = sign ? -(double)mantissa : (double)mantissa; |
| 979 | return (float) ldexp((float)res, exp-788); |
| 980 | } |
| 981 | |
| 982 | |
| 983 | // zlib & jpeg huffman tables assume that the output symbols |
| 984 | // can either be arbitrarily arranged, or have monotonically |
| 985 | // increasing frequencies--they rely on the lengths being sorted; |
| 986 | // this makes for a very simple generation algorithm. |
| 987 | // vorbis allows a huffman table with non-sorted lengths. This |
| 988 | // requires a more sophisticated construction, since symbols in |
| 989 | // order do not map to huffman codes "in order". |
| 990 | static void add_entry(Codebook *c, uint32 huff_code, int symbol, int count, int len, uint32 *values) |
| 991 | { |
| 992 | if (!c->sparse) { |
| 993 | c->codewords [symbol] = huff_code; |
| 994 | } else { |
| 995 | c->codewords [count] = huff_code; |
| 996 | c->codeword_lengths[count] = len; |
| 997 | values [count] = symbol; |
| 998 | } |
| 999 | } |
| 1000 | |
| 1001 | static int compute_codewords(Codebook *c, uint8 *len, int n, uint32 *values) |
| 1002 | { |
| 1003 | int i,k,m=0; |
| 1004 | uint32 available[32]; |
| 1005 | |
| 1006 | memset(available, 0, sizeof(available)); |
| 1007 | // find the first entry |
| 1008 | for (k=0; k < n; ++k) if (len[k] < NO_CODE255) break; |
| 1009 | if (k == n) { assert(c->sorted_entries == 0)((c->sorted_entries == 0) ? (void)0 : _assert("c->sorted_entries == 0" , "src/siege/internal/stb/stb_vorbis.c", 1009)); return TRUE1; } |
| 1010 | // add to the list |
| 1011 | add_entry(c, 0, k, m++, len[k], values); |
| 1012 | // add all available leaves |
| 1013 | for (i=1; i <= len[k]; ++i) |
| 1014 | available[i] = 1 << (32-i); |
| 1015 | // note that the above code treats the first case specially, |
| 1016 | // but it's really the same as the following code, so they |
| 1017 | // could probably be combined (except the initial code is 0, |
| 1018 | // and I use 0 in available[] to mean 'empty') |
| 1019 | for (i=k+1; i < n; ++i) { |
| 1020 | uint32 res; |
| 1021 | int z = len[i], y; |
| 1022 | if (z == NO_CODE255) continue; |
| 1023 | // find lowest available leaf (should always be earliest, |
| 1024 | // which is what the specification calls for) |
| 1025 | // note that this property, and the fact we can never have |
| 1026 | // more than one free leaf at a given level, isn't totally |
| 1027 | // trivial to prove, but it seems true and the assert never |
| 1028 | // fires, so! |
| 1029 | while (z > 0 && !available[z]) --z; |
| 1030 | if (z == 0) { assert(0)((0) ? (void)0 : _assert("0", "src/siege/internal/stb/stb_vorbis.c" , 1030)); return FALSE0; } |
| 1031 | res = available[z]; |
| 1032 | available[z] = 0; |
| 1033 | add_entry(c, bit_reverse(res), i, m++, len[i], values); |
| 1034 | // propogate availability up the tree |
| 1035 | if (z != len[i]) { |
| 1036 | for (y=len[i]; y > z; --y) { |
| 1037 | assert(available[y] == 0)((available[y] == 0) ? (void)0 : _assert("available[y] == 0", "src/siege/internal/stb/stb_vorbis.c", 1037)); |
| 1038 | available[y] = res + (1 << (32-y)); |
| 1039 | } |
| 1040 | } |
| 1041 | } |
| 1042 | return TRUE1; |
| 1043 | } |
| 1044 | |
| 1045 | // accelerated huffman table allows fast O(1) match of all symbols |
| 1046 | // of length <= STB_VORBIS_FAST_HUFFMAN_LENGTH |
| 1047 | static void compute_accelerated_huffman(Codebook *c) |
| 1048 | { |
| 1049 | int i, len; |
| 1050 | for (i=0; i < FAST_HUFFMAN_TABLE_SIZE(1 << 10); ++i) |
| 1051 | c->fast_huffman[i] = -1; |
| 1052 | |
| 1053 | len = c->sparse ? c->sorted_entries : c->entries; |
| 1054 | #ifdef STB_VORBIS_FAST_HUFFMAN_SHORT |
| 1055 | if (len > 32767) len = 32767; // largest possible value we can encode! |
| 1056 | #endif |
| 1057 | for (i=0; i < len; ++i) { |
| 1058 | if (c->codeword_lengths[i] <= STB_VORBIS_FAST_HUFFMAN_LENGTH10) { |
| 1059 | uint32 z = c->sparse ? bit_reverse(c->sorted_codewords[i]) : c->codewords[i]; |
| 1060 | // set table entries for all bit combinations in the higher bits |
| 1061 | while (z < FAST_HUFFMAN_TABLE_SIZE(1 << 10)) { |
| 1062 | c->fast_huffman[z] = i; |
| 1063 | z += 1 << c->codeword_lengths[i]; |
| 1064 | } |
| 1065 | } |
| 1066 | } |
| 1067 | } |
| 1068 | |
| 1069 | static int uint32_compare(const void *p, const void *q) |
| 1070 | { |
| 1071 | uint32 x = * (uint32 *) p; |
| 1072 | uint32 y = * (uint32 *) q; |
| 1073 | return x < y ? -1 : x > y; |
| 1074 | } |
| 1075 | |
| 1076 | static int include_in_sort(Codebook *c, uint8 len) |
| 1077 | { |
| 1078 | if (c->sparse) { assert(len != NO_CODE)((len != 255) ? (void)0 : _assert("len != NO_CODE", "src/siege/internal/stb/stb_vorbis.c" , 1078)); return TRUE1; } |
| 1079 | if (len == NO_CODE255) return FALSE0; |
| 1080 | if (len > STB_VORBIS_FAST_HUFFMAN_LENGTH10) return TRUE1; |
| 1081 | return FALSE0; |
| 1082 | } |
| 1083 | |
| 1084 | // if the fast table above doesn't work, we want to binary |
| 1085 | // search them... need to reverse the bits |
| 1086 | static void compute_sorted_huffman(Codebook *c, uint8 *lengths, uint32 *values) |
| 1087 | { |
| 1088 | int i, len; |
| 1089 | // build a list of all the entries |
| 1090 | // OPTIMIZATION: don't include the short ones, since they'll be caught by FAST_HUFFMAN. |
| 1091 | // this is kind of a frivolous optimization--I don't see any performance improvement, |
| 1092 | // but it's like 4 extra lines of code, so. |
| 1093 | if (!c->sparse) { |
| 1094 | int k = 0; |
| 1095 | for (i=0; i < c->entries; ++i) |
| 1096 | if (include_in_sort(c, lengths[i])) |
| 1097 | c->sorted_codewords[k++] = bit_reverse(c->codewords[i]); |
| 1098 | assert(k == c->sorted_entries)((k == c->sorted_entries) ? (void)0 : _assert("k == c->sorted_entries" , "src/siege/internal/stb/stb_vorbis.c", 1098)); |
| 1099 | } else { |
| 1100 | for (i=0; i < c->sorted_entries; ++i) |
| 1101 | c->sorted_codewords[i] = bit_reverse(c->codewords[i]); |
| 1102 | } |
| 1103 | |
| 1104 | qsort(c->sorted_codewords, c->sorted_entries, sizeof(c->sorted_codewords[0]), uint32_compare); |
| 1105 | c->sorted_codewords[c->sorted_entries] = 0xffffffff; |
| 1106 | |
| 1107 | len = c->sparse ? c->sorted_entries : c->entries; |
| 1108 | // now we need to indicate how they correspond; we could either |
| 1109 | // #1: sort a different data structure that says who they correspond to |
| 1110 | // #2: for each sorted entry, search the original list to find who corresponds |
| 1111 | // #3: for each original entry, find the sorted entry |
| 1112 | // #1 requires extra storage, #2 is slow, #3 can use binary search! |
| 1113 | for (i=0; i < len; ++i) { |
| 1114 | int huff_len = c->sparse ? lengths[values[i]] : lengths[i]; |
| 1115 | if (include_in_sort(c,huff_len)) { |
| 1116 | uint32 code = bit_reverse(c->codewords[i]); |
| 1117 | int x=0, n=c->sorted_entries; |
| 1118 | while (n > 1) { |
| 1119 | // invariant: sc[x] <= code < sc[x+n] |
| 1120 | int m = x + (n >> 1); |
| 1121 | if (c->sorted_codewords[m] <= code) { |
| 1122 | x = m; |
| 1123 | n -= (n>>1); |
| 1124 | } else { |
| 1125 | n >>= 1; |
| 1126 | } |
| 1127 | } |
| 1128 | assert(c->sorted_codewords[x] == code)((c->sorted_codewords[x] == code) ? (void)0 : _assert("c->sorted_codewords[x] == code" , "src/siege/internal/stb/stb_vorbis.c", 1128)); |
| 1129 | if (c->sparse) { |
| 1130 | c->sorted_values[x] = values[i]; |
| 1131 | c->codeword_lengths[x] = huff_len; |
| 1132 | } else { |
| 1133 | c->sorted_values[x] = i; |
| 1134 | } |
| 1135 | } |
| 1136 | } |
| 1137 | } |
| 1138 | |
| 1139 | // only run while parsing the header (3 times) |
| 1140 | static int vorbis_validate(uint8 *data) |
| 1141 | { |
| 1142 | static uint8 vorbis[6] = { 'v', 'o', 'r', 'b', 'i', 's' }; |
| 1143 | return memcmp(data, vorbis, 6) == 0; |
| 1144 | } |
| 1145 | |
| 1146 | // called from setup only, once per code book |
| 1147 | // (formula implied by specification) |
| 1148 | static int lookup1_values(int entries, int dim) |
| 1149 | { |
| 1150 | int r = (int) floor(exp((float) log((float) entries) / dim)); |
| 1151 | if ((int) floor(pow((float) r+1, dim)) <= entries) // (int) cast for MinGW warning; |
| 1152 | ++r; // floor() to avoid _ftol() when non-CRT |
| 1153 | assert(pow((float) r+1, dim) > entries)((pow((float) r+1, dim) > entries) ? (void)0 : _assert("pow((float) r+1, dim) > entries" , "src/siege/internal/stb/stb_vorbis.c", 1153)); |
| 1154 | assert((int) floor(pow((float) r, dim)) <= entries)(((int) floor(pow((float) r, dim)) <= entries) ? (void)0 : _assert("(int) floor(pow((float) r, dim)) <= entries", "src/siege/internal/stb/stb_vorbis.c" , 1154)); // (int),floor() as above |
| 1155 | return r; |
| 1156 | } |
| 1157 | |
| 1158 | // called twice per file |
| 1159 | static void compute_twiddle_factors(int n, float *A, float *B, float *C(2 | 4 | 1)) |
| 1160 | { |
| 1161 | int n4 = n >> 2, n8 = n >> 3; |
| 1162 | int k,k2; |
| 1163 | |
| 1164 | for (k=k2=0; k < n4; ++k,k2+=2) { |
| 1165 | A[k2 ] = (float) cos(4*k*M_PI3.14159265358979323846/n); |
| 1166 | A[k2+1] = (float) -sin(4*k*M_PI3.14159265358979323846/n); |
| 1167 | B[k2 ] = (float) cos((k2+1)*M_PI3.14159265358979323846/n/2) * 0.5f; |
| 1168 | B[k2+1] = (float) sin((k2+1)*M_PI3.14159265358979323846/n/2) * 0.5f; |
| 1169 | } |
| 1170 | for (k=k2=0; k < n8; ++k,k2+=2) { |
| 1171 | C(2 | 4 | 1)[k2 ] = (float) cos(2*(k2+1)*M_PI3.14159265358979323846/n); |
| 1172 | C(2 | 4 | 1)[k2+1] = (float) -sin(2*(k2+1)*M_PI3.14159265358979323846/n); |
| 1173 | } |
| 1174 | } |
| 1175 | |
| 1176 | static void compute_window(int n, float *window) |
| 1177 | { |
| 1178 | int n2 = n >> 1, i; |
| 1179 | for (i=0; i < n2; ++i) |
| 1180 | window[i] = (float) sin(0.5 * M_PI3.14159265358979323846 * square((float) sin((i - 0 + 0.5) / n2 * 0.5 * M_PI3.14159265358979323846))); |
| 1181 | } |
| 1182 | |
| 1183 | static void compute_bitreverse(int n, uint16 *rev) |
| 1184 | { |
| 1185 | int ld = ilog(n) - 1; // ilog is off-by-one from normal definitions |
| 1186 | int i, n8 = n >> 3; |
| 1187 | for (i=0; i < n8; ++i) |
| 1188 | rev[i] = (bit_reverse(i) >> (32-ld+3)) << 2; |
| 1189 | } |
| 1190 | |
| 1191 | static int init_blocksize(vorb *f, int b, int n) |
| 1192 | { |
| 1193 | int n2 = n >> 1, n4 = n >> 2, n8 = n >> 3; |
| 1194 | f->A[b] = (float *) setup_malloc(f, sizeof(float) * n2); |
| 1195 | f->B[b] = (float *) setup_malloc(f, sizeof(float) * n2); |
| 1196 | f->C(2 | 4 | 1)[b] = (float *) setup_malloc(f, sizeof(float) * n4); |
| 1197 | if (!f->A[b] || !f->B[b] || !f->C(2 | 4 | 1)[b]) return error(f, VORBIS_outofmem); |
| 1198 | compute_twiddle_factors(n, f->A[b], f->B[b], f->C(2 | 4 | 1)[b]); |
| 1199 | f->window[b] = (float *) setup_malloc(f, sizeof(float) * n2); |
| 1200 | if (!f->window[b]) return error(f, VORBIS_outofmem); |
| 1201 | compute_window(n, f->window[b]); |
| 1202 | f->bit_reverse[b] = (uint16 *) setup_malloc(f, sizeof(uint16) * n8); |
| 1203 | if (!f->bit_reverse[b]) return error(f, VORBIS_outofmem); |
| 1204 | compute_bitreverse(n, f->bit_reverse[b]); |
| 1205 | return TRUE1; |
| 1206 | } |
| 1207 | |
| 1208 | static void neighbors(uint16 *x, int n, int *plow, int *phigh) |
| 1209 | { |
| 1210 | int low = -1; |
| 1211 | int high = 65536; |
| 1212 | int i; |
| 1213 | for (i=0; i < n; ++i) { |
| 1214 | if (x[i] > low && x[i] < x[n]) { *plow = i; low = x[i]; } |
| 1215 | if (x[i] < high && x[i] > x[n]) { *phigh = i; high = x[i]; } |
| 1216 | } |
| 1217 | } |
| 1218 | |
| 1219 | // this has been repurposed so y is now the original index instead of y |
| 1220 | typedef struct |
| 1221 | { |
| 1222 | uint16 x,y; |
| 1223 | } Point; |
| 1224 | |
| 1225 | int point_compare(const void *p, const void *q) |
| 1226 | { |
| 1227 | Point *a = (Point *) p; |
| 1228 | Point *b = (Point *) q; |
| 1229 | return a->x < b->x ? -1 : a->x > b->x; |
| 1230 | } |
| 1231 | |
| 1232 | // |
| 1233 | /////////////////////// END LEAF SETUP FUNCTIONS ////////////////////////// |
| 1234 | |
| 1235 | |
| 1236 | #if defined(STB_VORBIS_NO_STDIO) |
| 1237 | #define USE_MEMORY(z)((z)->stream) TRUE1 |
| 1238 | #else |
| 1239 | #define USE_MEMORY(z)((z)->stream) ((z)->stream) |
| 1240 | #endif |
| 1241 | |
| 1242 | static uint8 get8(vorb *z) |
| 1243 | { |
| 1244 | if (USE_MEMORY(z)((z)->stream)) { |
| 1245 | if (z->stream >= z->stream_end) { z->eof = TRUE1; return 0; } |
| 1246 | return *z->stream++; |
| 1247 | } |
| 1248 | |
| 1249 | #ifndef STB_VORBIS_NO_STDIO |
| 1250 | { |
| 1251 | int c = fgetc(z->f); |
| 1252 | if (c == EOF(-1)) { z->eof = TRUE1; return 0; } |
| 1253 | return c; |
| 1254 | } |
| 1255 | #endif |
| 1256 | } |
| 1257 | |
| 1258 | static uint32 get32(vorb *f) |
| 1259 | { |
| 1260 | uint32 x; |
| 1261 | x = get8(f); |
| 1262 | x += get8(f) << 8; |
| 1263 | x += get8(f) << 16; |
| 1264 | x += get8(f) << 24; |
| 1265 | return x; |
| 1266 | } |
| 1267 | |
| 1268 | static int getn(vorb *z, uint8 *data, int n) |
| 1269 | { |
| 1270 | if (USE_MEMORY(z)((z)->stream)) { |
| 1271 | if (z->stream+n > z->stream_end) { z->eof = 1; return 0; } |
| 1272 | memcpy(data, z->stream, n); |
| 1273 | z->stream += n; |
| 1274 | return 1; |
| 1275 | } |
| 1276 | |
| 1277 | #ifndef STB_VORBIS_NO_STDIO |
| 1278 | if (fread(data, n, 1, z->f) == 1) |
| 1279 | return 1; |
| 1280 | else { |
| 1281 | z->eof = 1; |
| 1282 | return 0; |
| 1283 | } |
| 1284 | #endif |
| 1285 | } |
| 1286 | |
| 1287 | static void skip(vorb *z, int n) |
| 1288 | { |
| 1289 | if (USE_MEMORY(z)((z)->stream)) { |
| 1290 | z->stream += n; |
| 1291 | if (z->stream >= z->stream_end) z->eof = 1; |
| 1292 | return; |
| 1293 | } |
| 1294 | #ifndef STB_VORBIS_NO_STDIO |
| 1295 | { |
| 1296 | long x = ftell(z->f); |
| 1297 | fseek(z->f, x+n, SEEK_SET0); |
| 1298 | } |
| 1299 | #endif |
| 1300 | } |
| 1301 | |
| 1302 | static int set_file_offset(stb_vorbis *f, unsigned int loc) |
| 1303 | { |
| 1304 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 1305 | if (f->push_mode) return 0; |
| 1306 | #endif |
| 1307 | f->eof = 0; |
| 1308 | if (USE_MEMORY(f)((f)->stream)) { |
| 1309 | if (f->stream_start + loc >= f->stream_end || f->stream_start + loc < f->stream_start) { |
| 1310 | f->stream = f->stream_end; |
| 1311 | f->eof = 1; |
| 1312 | return 0; |
| 1313 | } else { |
| 1314 | f->stream = f->stream_start + loc; |
| 1315 | return 1; |
| 1316 | } |
| 1317 | } |
| 1318 | #ifndef STB_VORBIS_NO_STDIO |
| 1319 | if (loc + f->f_start < loc || loc >= 0x80000000) { |
| 1320 | loc = 0x7fffffff; |
| 1321 | f->eof = 1; |
| 1322 | } else { |
| 1323 | loc += f->f_start; |
| 1324 | } |
| 1325 | if (!fseek(f->f, loc, SEEK_SET0)) |
| 1326 | return 1; |
| 1327 | f->eof = 1; |
| 1328 | fseek(f->f, f->f_start, SEEK_END2); |
| 1329 | return 0; |
| 1330 | #endif |
| 1331 | } |
| 1332 | |
| 1333 | |
| 1334 | static uint8 ogg_page_header[4] = { 0x4f, 0x67, 0x67, 0x53 }; |
| 1335 | |
| 1336 | static int capture_pattern(vorb *f) |
| 1337 | { |
| 1338 | if (0x4f != get8(f)) return FALSE0; |
| 1339 | if (0x67 != get8(f)) return FALSE0; |
| 1340 | if (0x67 != get8(f)) return FALSE0; |
| 1341 | if (0x53 != get8(f)) return FALSE0; |
| 1342 | return TRUE1; |
| 1343 | } |
| 1344 | |
| 1345 | #define PAGEFLAG_continued_packet1 1 |
| 1346 | #define PAGEFLAG_first_page2 2 |
| 1347 | #define PAGEFLAG_last_page4 4 |
| 1348 | |
| 1349 | static int start_page_no_capturepattern(vorb *f) |
| 1350 | { |
| 1351 | uint32 loc0,loc1,n; |
| 1352 | int32 i; |
| 1353 | // stream structure version |
| 1354 | if (0 != get8(f)) return error(f, VORBIS_invalid_stream_structure_version); |
| 1355 | // header flag |
| 1356 | f->page_flag = get8(f); |
| 1357 | // absolute granule position |
| 1358 | loc0 = get32(f); |
| 1359 | loc1 = get32(f); |
| 1360 | // @TODO: validate loc0,loc1 as valid positions? |
| 1361 | // stream serial number -- vorbis doesn't interleave, so discard |
| 1362 | get32(f); |
| 1363 | //if (f->serial != get32(f)) return error(f, VORBIS_incorrect_stream_serial_number); |
| 1364 | // page sequence number |
| 1365 | n = get32(f); |
| 1366 | f->last_page = n; |
| 1367 | // CRC32 |
| 1368 | get32(f); |
| 1369 | // page_segments |
| 1370 | f->segment_count = get8(f); |
| 1371 | if (!getn(f, f->segments, f->segment_count)) |
| 1372 | return error(f, VORBIS_unexpected_eof); |
| 1373 | // assume we _don't_ know any the sample position of any segments |
| 1374 | f->end_seg_with_known_loc = -2; |
| 1375 | if (loc0 != ~0 || loc1 != ~0) { |
| 1376 | // determine which packet is the last one that will complete |
| 1377 | for (i=f->segment_count-1; i >= 0; --i) |
| 1378 | if (f->segments[i] < 255) |
| 1379 | break; |
| 1380 | // 'i' is now the index of the _last_ segment of a packet that ends |
| 1381 | if (i >= 0) { |
| 1382 | f->end_seg_with_known_loc = i; |
| 1383 | f->known_loc_for_packet = loc0; |
| 1384 | } |
| 1385 | } |
| 1386 | if (f->first_decode) { |
| 1387 | int i,len; |
| 1388 | ProbedPage p; |
| 1389 | len = 0; |
| 1390 | for (i=0; i < f->segment_count; ++i) |
| 1391 | len += f->segments[i]; |
| 1392 | len += 27 + f->segment_count; |
| 1393 | p.page_start = f->first_audio_page_offset; |
| 1394 | p.page_end = p.page_start + len; |
| 1395 | p.after_previous_page_start = p.page_start; |
| 1396 | p.first_decoded_sample = 0; |
| 1397 | p.last_decoded_sample = loc0; |
| 1398 | f->p_first = p; |
| 1399 | } |
| 1400 | f->next_seg = 0; |
| 1401 | return TRUE1; |
| 1402 | } |
| 1403 | |
| 1404 | static int start_page(vorb *f) |
| 1405 | { |
| 1406 | if (!capture_pattern(f)) return error(f, VORBIS_missing_capture_pattern); |
| 1407 | return start_page_no_capturepattern(f); |
| 1408 | } |
| 1409 | |
| 1410 | static int start_packet(vorb *f) |
| 1411 | { |
| 1412 | while (f->next_seg == -1) { |
| 1413 | if (!start_page(f)) return FALSE0; |
| 1414 | if (f->page_flag & PAGEFLAG_continued_packet1) |
| 1415 | return error(f, VORBIS_continued_packet_flag_invalid); |
| 1416 | } |
| 1417 | f->last_seg = FALSE0; |
| 1418 | f->valid_bits = 0; |
| 1419 | f->packet_bytes = 0; |
| 1420 | f->bytes_in_seg = 0; |
| 1421 | // f->next_seg is now valid |
| 1422 | return TRUE1; |
| 1423 | } |
| 1424 | |
| 1425 | static int maybe_start_packet(vorb *f) |
| 1426 | { |
| 1427 | if (f->next_seg == -1) { |
| 1428 | int x = get8(f); |
| 1429 | if (f->eof) return FALSE0; // EOF at page boundary is not an error! |
| 1430 | if (0x4f != x ) return error(f, VORBIS_missing_capture_pattern); |
| 1431 | if (0x67 != get8(f)) return error(f, VORBIS_missing_capture_pattern); |
| 1432 | if (0x67 != get8(f)) return error(f, VORBIS_missing_capture_pattern); |
| 1433 | if (0x53 != get8(f)) return error(f, VORBIS_missing_capture_pattern); |
| 1434 | if (!start_page_no_capturepattern(f)) return FALSE0; |
| 1435 | if (f->page_flag & PAGEFLAG_continued_packet1) { |
| 1436 | // set up enough state that we can read this packet if we want, |
| 1437 | // e.g. during recovery |
| 1438 | f->last_seg = FALSE0; |
| 1439 | f->bytes_in_seg = 0; |
| 1440 | return error(f, VORBIS_continued_packet_flag_invalid); |
| 1441 | } |
| 1442 | } |
| 1443 | return start_packet(f); |
| 1444 | } |
| 1445 | |
| 1446 | static int next_segment(vorb *f) |
| 1447 | { |
| 1448 | int len; |
| 1449 | if (f->last_seg) return 0; |
| 1450 | if (f->next_seg == -1) { |
| 1451 | f->last_seg_which = f->segment_count-1; // in case start_page fails |
| 1452 | if (!start_page(f)) { f->last_seg = 1; return 0; } |
| 1453 | if (!(f->page_flag & PAGEFLAG_continued_packet1)) return error(f, VORBIS_continued_packet_flag_invalid); |
| 1454 | } |
| 1455 | len = f->segments[f->next_seg++]; |
| 1456 | if (len < 255) { |
| 1457 | f->last_seg = TRUE1; |
| 1458 | f->last_seg_which = f->next_seg-1; |
| 1459 | } |
| 1460 | if (f->next_seg >= f->segment_count) |
| 1461 | f->next_seg = -1; |
| 1462 | assert(f->bytes_in_seg == 0)((f->bytes_in_seg == 0) ? (void)0 : _assert("f->bytes_in_seg == 0" , "src/siege/internal/stb/stb_vorbis.c", 1462)); |
| 1463 | f->bytes_in_seg = len; |
| 1464 | return len; |
| 1465 | } |
| 1466 | |
| 1467 | #define EOP(-1) (-1) |
| 1468 | #define INVALID_BITS(-1) (-1) |
| 1469 | |
| 1470 | static int get8_packet_raw(vorb *f) |
| 1471 | { |
| 1472 | if (!f->bytes_in_seg) |
| 1473 | { |
| 1474 | if (f->last_seg) return EOP(-1); |
| 1475 | else if (!next_segment(f)) return EOP(-1); |
| 1476 | } |
| 1477 | assert(f->bytes_in_seg > 0)((f->bytes_in_seg > 0) ? (void)0 : _assert("f->bytes_in_seg > 0" , "src/siege/internal/stb/stb_vorbis.c", 1477)); |
| 1478 | --f->bytes_in_seg; |
| 1479 | ++f->packet_bytes; |
| 1480 | return get8(f); |
| 1481 | } |
| 1482 | |
| 1483 | static int get8_packet(vorb *f) |
| 1484 | { |
| 1485 | int x = get8_packet_raw(f); |
| 1486 | f->valid_bits = 0; |
| 1487 | return x; |
| 1488 | } |
| 1489 | |
| 1490 | static void flush_packet(vorb *f) |
| 1491 | { |
| 1492 | while (get8_packet_raw(f) != EOP(-1)); |
| 1493 | } |
| 1494 | |
| 1495 | // @OPTIMIZE: this is the secondary bit decoder, so it's probably not as important |
| 1496 | // as the huffman decoder? |
| 1497 | static uint32 get_bits(vorb *f, int n) |
| 1498 | { |
| 1499 | uint32 z; |
| 1500 | |
| 1501 | if (f->valid_bits < 0) return 0; |
| 1502 | if (f->valid_bits < n) { |
| 1503 | if (n > 24) { |
| 1504 | // the accumulator technique below would not work correctly in this case |
| 1505 | z = get_bits(f, 24); |
| 1506 | z += get_bits(f, n-24) << 24; |
| 1507 | return z; |
| 1508 | } |
| 1509 | if (f->valid_bits == 0) f->acc = 0; |
| 1510 | while (f->valid_bits < n) { |
| 1511 | int z = get8_packet_raw(f); |
| 1512 | if (z == EOP(-1)) { |
| 1513 | f->valid_bits = INVALID_BITS(-1); |
| 1514 | return 0; |
| 1515 | } |
| 1516 | f->acc += z << f->valid_bits; |
| 1517 | f->valid_bits += 8; |
| 1518 | } |
| 1519 | } |
| 1520 | if (f->valid_bits < 0) return 0; |
| 1521 | z = f->acc & ((1 << n)-1); |
| 1522 | f->acc >>= n; |
| 1523 | f->valid_bits -= n; |
| 1524 | return z; |
| 1525 | } |
| 1526 | |
| 1527 | static int32 get_bits_signed(vorb *f, int n) |
| 1528 | { |
| 1529 | uint32 z = get_bits(f, n); |
| 1530 | if (z & (1 << (n-1))) |
| 1531 | z += ~((1 << n) - 1); |
| 1532 | return (int32) z; |
| 1533 | } |
| 1534 | |
| 1535 | // @OPTIMIZE: primary accumulator for huffman |
| 1536 | // expand the buffer to as many bits as possible without reading off end of packet |
| 1537 | // it might be nice to allow f->valid_bits and f->acc to be stored in registers, |
| 1538 | // e.g. cache them locally and decode locally |
| 1539 | static __forceinlineinline void prep_huffman(vorb *f) |
| 1540 | { |
| 1541 | if (f->valid_bits <= 24) { |
| 1542 | if (f->valid_bits == 0) f->acc = 0; |
| 1543 | do { |
| 1544 | int z; |
| 1545 | if (f->last_seg && !f->bytes_in_seg) return; |
| 1546 | z = get8_packet_raw(f); |
| 1547 | if (z == EOP(-1)) return; |
| 1548 | f->acc += z << f->valid_bits; |
| 1549 | f->valid_bits += 8; |
| 1550 | } while (f->valid_bits <= 24); |
| 1551 | } |
| 1552 | } |
| 1553 | |
| 1554 | enum |
| 1555 | { |
| 1556 | VORBIS_packet_id = 1, |
| 1557 | VORBIS_packet_comment = 3, |
| 1558 | VORBIS_packet_setup = 5, |
| 1559 | }; |
| 1560 | |
| 1561 | static int codebook_decode_scalar_raw(vorb *f, Codebook *c) |
| 1562 | { |
| 1563 | int i; |
| 1564 | prep_huffman(f); |
| 1565 | |
| 1566 | assert(c->sorted_codewords || c->codewords)((c->sorted_codewords || c->codewords) ? (void)0 : _assert ("c->sorted_codewords || c->codewords", "src/siege/internal/stb/stb_vorbis.c" , 1566)); |
| 1567 | // cases to use binary search: sorted_codewords && !c->codewords |
| 1568 | // sorted_codewords && c->entries > 8 |
| 1569 | if (c->entries > 8 ? c->sorted_codewords!=NULL((void*)0) : !c->codewords) { |
| 1570 | // binary search |
| 1571 | uint32 code = bit_reverse(f->acc); |
| 1572 | int x=0, n=c->sorted_entries, len; |
| 1573 | |
| 1574 | while (n > 1) { |
| 1575 | // invariant: sc[x] <= code < sc[x+n] |
| 1576 | int m = x + (n >> 1); |
| 1577 | if (c->sorted_codewords[m] <= code) { |
| 1578 | x = m; |
| 1579 | n -= (n>>1); |
| 1580 | } else { |
| 1581 | n >>= 1; |
| 1582 | } |
| 1583 | } |
| 1584 | // x is now the sorted index |
| 1585 | if (!c->sparse) x = c->sorted_values[x]; |
| 1586 | // x is now sorted index if sparse, or symbol otherwise |
| 1587 | len = c->codeword_lengths[x]; |
| 1588 | if (f->valid_bits >= len) { |
| 1589 | f->acc >>= len; |
| 1590 | f->valid_bits -= len; |
| 1591 | return x; |
| 1592 | } |
| 1593 | |
| 1594 | f->valid_bits = 0; |
| 1595 | return -1; |
| 1596 | } |
| 1597 | |
| 1598 | // if small, linear search |
| 1599 | assert(!c->sparse)((!c->sparse) ? (void)0 : _assert("!c->sparse", "src/siege/internal/stb/stb_vorbis.c" , 1599)); |
| 1600 | for (i=0; i < c->entries; ++i) { |
| 1601 | if (c->codeword_lengths[i] == NO_CODE255) continue; |
| 1602 | if (c->codewords[i] == (f->acc & ((1 << c->codeword_lengths[i])-1))) { |
| 1603 | if (f->valid_bits >= c->codeword_lengths[i]) { |
| 1604 | f->acc >>= c->codeword_lengths[i]; |
| 1605 | f->valid_bits -= c->codeword_lengths[i]; |
| 1606 | return i; |
| 1607 | } |
| 1608 | f->valid_bits = 0; |
| 1609 | return -1; |
| 1610 | } |
| 1611 | } |
| 1612 | |
| 1613 | error(f, VORBIS_invalid_stream); |
| 1614 | f->valid_bits = 0; |
| 1615 | return -1; |
| 1616 | } |
| 1617 | |
| 1618 | static int codebook_decode_scalar(vorb *f, Codebook *c) |
| 1619 | { |
| 1620 | int i; |
| 1621 | if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH10) |
| 1622 | prep_huffman(f); |
| 1623 | // fast huffman table lookup |
| 1624 | i = f->acc & FAST_HUFFMAN_TABLE_MASK((1 << 10) - 1); |
| 1625 | i = c->fast_huffman[i]; |
| 1626 | if (i >= 0) { |
| 1627 | f->acc >>= c->codeword_lengths[i]; |
| 1628 | f->valid_bits -= c->codeword_lengths[i]; |
| 1629 | if (f->valid_bits < 0) { f->valid_bits = 0; return -1; } |
| 1630 | return i; |
| 1631 | } |
| 1632 | return codebook_decode_scalar_raw(f,c); |
| 1633 | } |
| 1634 | |
| 1635 | #ifndef STB_VORBIS_NO_INLINE_DECODE |
| 1636 | |
| 1637 | #define DECODE_RAW(var, f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } \ |
| 1638 | if (f->valid_bits < STB_VORBIS_FAST_HUFFMAN_LENGTH10) \ |
| 1639 | prep_huffman(f); \ |
| 1640 | var = f->acc & FAST_HUFFMAN_TABLE_MASK((1 << 10) - 1); \ |
| 1641 | var = c->fast_huffman[var]; \ |
| 1642 | if (var >= 0) { \ |
| 1643 | int n = c->codeword_lengths[var]; \ |
| 1644 | f->acc >>= n; \ |
| 1645 | f->valid_bits -= n; \ |
| 1646 | if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } \ |
| 1647 | } else { \ |
| 1648 | var = codebook_decode_scalar_raw(f,c); \ |
| 1649 | } |
| 1650 | |
| 1651 | #else |
| 1652 | |
| 1653 | #define DECODE_RAW(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } var = codebook_decode_scalar(f,c); |
| 1654 | |
| 1655 | #endif |
| 1656 | |
| 1657 | #define DECODE(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } if (c->sparse) var = c->sorted_values[var]; \ |
| 1658 | DECODE_RAW(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } \ |
| 1659 | if (c->sparse) var = c->sorted_values[var]; |
| 1660 | |
| 1661 | #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 1662 | #define DECODE_VQ(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } DECODE_RAW(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } |
| 1663 | #else |
| 1664 | #define DECODE_VQ(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } DECODE(var,f,c)if (f->valid_bits < 10) prep_huffman(f); var = f->acc & ((1 << 10) - 1); var = c->fast_huffman[var]; if (var >= 0) { int n = c->codeword_lengths[var]; f->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; var = -1; } } else { var = codebook_decode_scalar_raw (f,c); } if (c->sparse) var = c->sorted_values[var]; |
| 1665 | #endif |
| 1666 | |
| 1667 | |
| 1668 | |
| 1669 | |
| 1670 | |
| 1671 | |
| 1672 | // CODEBOOK_ELEMENT_FAST is an optimization for the CODEBOOK_FLOATS case |
| 1673 | // where we avoid one addition |
| 1674 | #ifndef STB_VORBIS_CODEBOOK_FLOATS |
| 1675 | #define CODEBOOK_ELEMENT(c,off)(c->multiplicands[off]) (c->multiplicands[off] * c->delta_value + c->minimum_value) |
| 1676 | #define CODEBOOK_ELEMENT_FAST(c,off)(c->multiplicands[off]) (c->multiplicands[off] * c->delta_value) |
| 1677 | #define CODEBOOK_ELEMENT_BASE(c)(0) (c->minimum_value) |
| 1678 | #else |
| 1679 | #define CODEBOOK_ELEMENT(c,off)(c->multiplicands[off]) (c->multiplicands[off]) |
| 1680 | #define CODEBOOK_ELEMENT_FAST(c,off)(c->multiplicands[off]) (c->multiplicands[off]) |
| 1681 | #define CODEBOOK_ELEMENT_BASE(c)(0) (0) |
| 1682 | #endif |
| 1683 | |
| 1684 | static int codebook_decode_start(vorb *f, Codebook *c, int len) |
| 1685 | { |
| 1686 | int z = -1; |
| 1687 | |
| 1688 | // type 0 is only legal in a scalar context |
| 1689 | if (c->lookup_type == 0) |
| 1690 | error(f, VORBIS_invalid_stream); |
| 1691 | else { |
| 1692 | DECODE_VQ(z,f,c)if (f->valid_bits < 10) prep_huffman(f); z = f->acc & ((1 << 10) - 1); z = c->fast_huffman[z]; if (z >= 0) { int n = c->codeword_lengths[z]; f->acc >>= n ; f->valid_bits -= n; if (f->valid_bits < 0) { f-> valid_bits = 0; z = -1; } } else { z = codebook_decode_scalar_raw (f,c); }; |
| 1693 | if (c->sparse) assert(z < c->sorted_entries)((z < c->sorted_entries) ? (void)0 : _assert("z < c->sorted_entries" , "src/siege/internal/stb/stb_vorbis.c", 1693)); |
| 1694 | if (z < 0) { // check for EOP |
| 1695 | if (!f->bytes_in_seg) |
| 1696 | if (f->last_seg) |
| 1697 | return z; |
| 1698 | error(f, VORBIS_invalid_stream); |
| 1699 | } |
| 1700 | } |
| 1701 | return z; |
| 1702 | } |
| 1703 | |
| 1704 | static int codebook_decode(vorb *f, Codebook *c, float *output, int len) |
| 1705 | { |
| 1706 | int i,z = codebook_decode_start(f,c,len); |
| 1707 | if (z < 0) return FALSE0; |
| 1708 | if (len > c->dimensions) len = c->dimensions; |
| 1709 | |
| 1710 | #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 1711 | if (c->lookup_type == 1) { |
| 1712 | float last = CODEBOOK_ELEMENT_BASE(c)(0); |
| 1713 | int div = 1; |
| 1714 | for (i=0; i < len; ++i) { |
| 1715 | int off = (z / div) % c->lookup_values; |
| 1716 | float val = CODEBOOK_ELEMENT_FAST(c,off)(c->multiplicands[off]) + last; |
| 1717 | output[i] += val; |
| 1718 | if (c->sequence_p) last = val + c->minimum_value; |
| 1719 | div *= c->lookup_values; |
| 1720 | } |
| 1721 | return TRUE1; |
| 1722 | } |
| 1723 | #endif |
| 1724 | |
| 1725 | z *= c->dimensions; |
| 1726 | if (c->sequence_p) { |
| 1727 | float last = CODEBOOK_ELEMENT_BASE(c)(0); |
| 1728 | for (i=0; i < len; ++i) { |
| 1729 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1730 | output[i] += val; |
| 1731 | last = val + c->minimum_value; |
| 1732 | } |
| 1733 | } else { |
| 1734 | float last = CODEBOOK_ELEMENT_BASE(c)(0); |
| 1735 | for (i=0; i < len; ++i) { |
| 1736 | output[i] += CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1737 | } |
| 1738 | } |
| 1739 | |
| 1740 | return TRUE1; |
| 1741 | } |
| 1742 | |
| 1743 | static int codebook_decode_step(vorb *f, Codebook *c, float *output, int len, int step) |
| 1744 | { |
| 1745 | int i,z = codebook_decode_start(f,c,len); |
| 1746 | float last = CODEBOOK_ELEMENT_BASE(c)(0); |
| 1747 | if (z < 0) return FALSE0; |
| 1748 | if (len > c->dimensions) len = c->dimensions; |
| 1749 | |
| 1750 | #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 1751 | if (c->lookup_type == 1) { |
| 1752 | int div = 1; |
| 1753 | for (i=0; i < len; ++i) { |
| 1754 | int off = (z / div) % c->lookup_values; |
| 1755 | float val = CODEBOOK_ELEMENT_FAST(c,off)(c->multiplicands[off]) + last; |
| 1756 | output[i*step] += val; |
| 1757 | if (c->sequence_p) last = val; |
| 1758 | div *= c->lookup_values; |
| 1759 | } |
| 1760 | return TRUE1; |
| 1761 | } |
| 1762 | #endif |
| 1763 | |
| 1764 | z *= c->dimensions; |
| 1765 | for (i=0; i < len; ++i) { |
| 1766 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1767 | output[i*step] += val; |
| 1768 | if (c->sequence_p) last = val; |
| 1769 | } |
| 1770 | |
| 1771 | return TRUE1; |
| 1772 | } |
| 1773 | |
| 1774 | static int codebook_decode_deinterleave_repeat(vorb *f, Codebook *c, float **outputs, int ch, int *c_inter_p, int *p_inter_p, int len, int total_decode) |
| 1775 | { |
| 1776 | int c_inter = *c_inter_p; |
| 1777 | int p_inter = *p_inter_p; |
| 1778 | int i,z, effective = c->dimensions; |
| 1779 | |
| 1780 | // type 0 is only legal in a scalar context |
| 1781 | if (c->lookup_type == 0) return error(f, VORBIS_invalid_stream); |
| 1782 | |
| 1783 | while (total_decode > 0) { |
| 1784 | float last = CODEBOOK_ELEMENT_BASE(c)(0); |
| 1785 | DECODE_VQ(z,f,c)if (f->valid_bits < 10) prep_huffman(f); z = f->acc & ((1 << 10) - 1); z = c->fast_huffman[z]; if (z >= 0) { int n = c->codeword_lengths[z]; f->acc >>= n ; f->valid_bits -= n; if (f->valid_bits < 0) { f-> valid_bits = 0; z = -1; } } else { z = codebook_decode_scalar_raw (f,c); }; |
| 1786 | #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 1787 | assert(!c->sparse || z < c->sorted_entries)((!c->sparse || z < c->sorted_entries) ? (void)0 : _assert ("!c->sparse || z < c->sorted_entries", "src/siege/internal/stb/stb_vorbis.c" , 1787)); |
| 1788 | #endif |
| 1789 | if (z < 0) { |
| 1790 | if (!f->bytes_in_seg) |
| 1791 | if (f->last_seg) return FALSE0; |
| 1792 | return error(f, VORBIS_invalid_stream); |
| 1793 | } |
| 1794 | |
| 1795 | // if this will take us off the end of the buffers, stop short! |
| 1796 | // we check by computing the length of the virtual interleaved |
| 1797 | // buffer (len*ch), our current offset within it (p_inter*ch)+(c_inter), |
| 1798 | // and the length we'll be using (effective) |
| 1799 | if (c_inter + p_inter*ch + effective > len * ch) { |
| 1800 | effective = len*ch - (p_inter*ch - c_inter); |
| 1801 | } |
| 1802 | |
| 1803 | #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 1804 | if (c->lookup_type == 1) { |
| 1805 | int div = 1; |
| 1806 | for (i=0; i < effective; ++i) { |
| 1807 | int off = (z / div) % c->lookup_values; |
| 1808 | float val = CODEBOOK_ELEMENT_FAST(c,off)(c->multiplicands[off]) + last; |
| 1809 | outputs[c_inter][p_inter] += val; |
| 1810 | if (++c_inter == ch) { c_inter = 0; ++p_inter; } |
| 1811 | if (c->sequence_p) last = val; |
| 1812 | div *= c->lookup_values; |
| 1813 | } |
| 1814 | } else |
| 1815 | #endif |
| 1816 | { |
| 1817 | z *= c->dimensions; |
| 1818 | if (c->sequence_p) { |
| 1819 | for (i=0; i < effective; ++i) { |
| 1820 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1821 | outputs[c_inter][p_inter] += val; |
| 1822 | if (++c_inter == ch) { c_inter = 0; ++p_inter; } |
| 1823 | last = val; |
| 1824 | } |
| 1825 | } else { |
| 1826 | for (i=0; i < effective; ++i) { |
| 1827 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1828 | outputs[c_inter][p_inter] += val; |
| 1829 | if (++c_inter == ch) { c_inter = 0; ++p_inter; } |
| 1830 | } |
| 1831 | } |
| 1832 | } |
| 1833 | |
| 1834 | total_decode -= effective; |
| 1835 | } |
| 1836 | *c_inter_p = c_inter; |
| 1837 | *p_inter_p = p_inter; |
| 1838 | return TRUE1; |
| 1839 | } |
| 1840 | |
| 1841 | #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 1842 | static int codebook_decode_deinterleave_repeat_2(vorb *f, Codebook *c, float **outputs, int *c_inter_p, int *p_inter_p, int len, int total_decode) |
| 1843 | { |
| 1844 | int c_inter = *c_inter_p; |
| 1845 | int p_inter = *p_inter_p; |
| 1846 | int i,z, effective = c->dimensions; |
| 1847 | |
| 1848 | // type 0 is only legal in a scalar context |
| 1849 | if (c->lookup_type == 0) return error(f, VORBIS_invalid_stream); |
| 1850 | |
| 1851 | while (total_decode > 0) { |
| 1852 | float last = CODEBOOK_ELEMENT_BASE(c)(0); |
| 1853 | DECODE_VQ(z,f,c)if (f->valid_bits < 10) prep_huffman(f); z = f->acc & ((1 << 10) - 1); z = c->fast_huffman[z]; if (z >= 0) { int n = c->codeword_lengths[z]; f->acc >>= n ; f->valid_bits -= n; if (f->valid_bits < 0) { f-> valid_bits = 0; z = -1; } } else { z = codebook_decode_scalar_raw (f,c); }; |
| 1854 | |
| 1855 | if (z < 0) { |
| 1856 | if (!f->bytes_in_seg) |
| 1857 | if (f->last_seg) return FALSE0; |
| 1858 | return error(f, VORBIS_invalid_stream); |
| 1859 | } |
| 1860 | |
| 1861 | // if this will take us off the end of the buffers, stop short! |
| 1862 | // we check by computing the length of the virtual interleaved |
| 1863 | // buffer (len*ch), our current offset within it (p_inter*ch)+(c_inter), |
| 1864 | // and the length we'll be using (effective) |
| 1865 | if (c_inter + p_inter*2 + effective > len * 2) { |
| 1866 | effective = len*2 - (p_inter*2 - c_inter); |
| 1867 | } |
| 1868 | |
| 1869 | { |
| 1870 | z *= c->dimensions; |
| 1871 | stb_prof(11)(void)0; |
| 1872 | if (c->sequence_p) { |
| 1873 | // haven't optimized this case because I don't have any examples |
| 1874 | for (i=0; i < effective; ++i) { |
| 1875 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1876 | outputs[c_inter][p_inter] += val; |
| 1877 | if (++c_inter == 2) { c_inter = 0; ++p_inter; } |
| 1878 | last = val; |
| 1879 | } |
| 1880 | } else { |
| 1881 | i=0; |
| 1882 | if (c_inter == 1) { |
| 1883 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1884 | outputs[c_inter][p_inter] += val; |
| 1885 | c_inter = 0; ++p_inter; |
| 1886 | ++i; |
| 1887 | } |
| 1888 | { |
| 1889 | float *z0 = outputs[0]; |
| 1890 | float *z1 = outputs[1]; |
| 1891 | for (; i+1 < effective;) { |
| 1892 | z0[p_inter] += CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1893 | z1[p_inter] += CODEBOOK_ELEMENT_FAST(c,z+i+1)(c->multiplicands[z+i+1]) + last; |
| 1894 | ++p_inter; |
| 1895 | i += 2; |
| 1896 | } |
| 1897 | } |
| 1898 | if (i < effective) { |
| 1899 | float val = CODEBOOK_ELEMENT_FAST(c,z+i)(c->multiplicands[z+i]) + last; |
| 1900 | outputs[c_inter][p_inter] += val; |
| 1901 | if (++c_inter == 2) { c_inter = 0; ++p_inter; } |
| 1902 | } |
| 1903 | } |
| 1904 | } |
| 1905 | |
| 1906 | total_decode -= effective; |
| 1907 | } |
| 1908 | *c_inter_p = c_inter; |
| 1909 | *p_inter_p = p_inter; |
| 1910 | return TRUE1; |
| 1911 | } |
| 1912 | #endif |
| 1913 | |
| 1914 | static int predict_point(int x, int x0, int x1, int y0, int y1) |
| 1915 | { |
| 1916 | int dy = y1 - y0; |
| 1917 | int adx = x1 - x0; |
| 1918 | // @OPTIMIZE: force int division to round in the right direction... is this necessary on x86? |
| 1919 | int err = abs(dy) * (x - x0); |
| 1920 | int off = err / adx; |
| 1921 | return dy < 0 ? y0 - off : y0 + off; |
| 1922 | } |
| 1923 | |
| 1924 | // the following table is block-copied from the specification |
| 1925 | static float inverse_db_table[256] = |
| 1926 | { |
| 1927 | 1.0649863e-07f, 1.1341951e-07f, 1.2079015e-07f, 1.2863978e-07f, |
| 1928 | 1.3699951e-07f, 1.4590251e-07f, 1.5538408e-07f, 1.6548181e-07f, |
| 1929 | 1.7623575e-07f, 1.8768855e-07f, 1.9988561e-07f, 2.1287530e-07f, |
| 1930 | 2.2670913e-07f, 2.4144197e-07f, 2.5713223e-07f, 2.7384213e-07f, |
| 1931 | 2.9163793e-07f, 3.1059021e-07f, 3.3077411e-07f, 3.5226968e-07f, |
| 1932 | 3.7516214e-07f, 3.9954229e-07f, 4.2550680e-07f, 4.5315863e-07f, |
| 1933 | 4.8260743e-07f, 5.1396998e-07f, 5.4737065e-07f, 5.8294187e-07f, |
| 1934 | 6.2082472e-07f, 6.6116941e-07f, 7.0413592e-07f, 7.4989464e-07f, |
| 1935 | 7.9862701e-07f, 8.5052630e-07f, 9.0579828e-07f, 9.6466216e-07f, |
| 1936 | 1.0273513e-06f, 1.0941144e-06f, 1.1652161e-06f, 1.2409384e-06f, |
| 1937 | 1.3215816e-06f, 1.4074654e-06f, 1.4989305e-06f, 1.5963394e-06f, |
| 1938 | 1.7000785e-06f, 1.8105592e-06f, 1.9282195e-06f, 2.0535261e-06f, |
| 1939 | 2.1869758e-06f, 2.3290978e-06f, 2.4804557e-06f, 2.6416497e-06f, |
| 1940 | 2.8133190e-06f, 2.9961443e-06f, 3.1908506e-06f, 3.3982101e-06f, |
| 1941 | 3.6190449e-06f, 3.8542308e-06f, 4.1047004e-06f, 4.3714470e-06f, |
| 1942 | 4.6555282e-06f, 4.9580707e-06f, 5.2802740e-06f, 5.6234160e-06f, |
| 1943 | 5.9888572e-06f, 6.3780469e-06f, 6.7925283e-06f, 7.2339451e-06f, |
| 1944 | 7.7040476e-06f, 8.2047000e-06f, 8.7378876e-06f, 9.3057248e-06f, |
| 1945 | 9.9104632e-06f, 1.0554501e-05f, 1.1240392e-05f, 1.1970856e-05f, |
| 1946 | 1.2748789e-05f, 1.3577278e-05f, 1.4459606e-05f, 1.5399272e-05f, |
| 1947 | 1.6400004e-05f, 1.7465768e-05f, 1.8600792e-05f, 1.9809576e-05f, |
| 1948 | 2.1096914e-05f, 2.2467911e-05f, 2.3928002e-05f, 2.5482978e-05f, |
| 1949 | 2.7139006e-05f, 2.8902651e-05f, 3.0780908e-05f, 3.2781225e-05f, |
| 1950 | 3.4911534e-05f, 3.7180282e-05f, 3.9596466e-05f, 4.2169667e-05f, |
| 1951 | 4.4910090e-05f, 4.7828601e-05f, 5.0936773e-05f, 5.4246931e-05f, |
| 1952 | 5.7772202e-05f, 6.1526565e-05f, 6.5524908e-05f, 6.9783085e-05f, |
| 1953 | 7.4317983e-05f, 7.9147585e-05f, 8.4291040e-05f, 8.9768747e-05f, |
| 1954 | 9.5602426e-05f, 0.00010181521f, 0.00010843174f, 0.00011547824f, |
| 1955 | 0.00012298267f, 0.00013097477f, 0.00013948625f, 0.00014855085f, |
| 1956 | 0.00015820453f, 0.00016848555f, 0.00017943469f, 0.00019109536f, |
| 1957 | 0.00020351382f, 0.00021673929f, 0.00023082423f, 0.00024582449f, |
| 1958 | 0.00026179955f, 0.00027881276f, 0.00029693158f, 0.00031622787f, |
| 1959 | 0.00033677814f, 0.00035866388f, 0.00038197188f, 0.00040679456f, |
| 1960 | 0.00043323036f, 0.00046138411f, 0.00049136745f, 0.00052329927f, |
| 1961 | 0.00055730621f, 0.00059352311f, 0.00063209358f, 0.00067317058f, |
| 1962 | 0.00071691700f, 0.00076350630f, 0.00081312324f, 0.00086596457f, |
| 1963 | 0.00092223983f, 0.00098217216f, 0.0010459992f, 0.0011139742f, |
| 1964 | 0.0011863665f, 0.0012634633f, 0.0013455702f, 0.0014330129f, |
| 1965 | 0.0015261382f, 0.0016253153f, 0.0017309374f, 0.0018434235f, |
| 1966 | 0.0019632195f, 0.0020908006f, 0.0022266726f, 0.0023713743f, |
| 1967 | 0.0025254795f, 0.0026895994f, 0.0028643847f, 0.0030505286f, |
| 1968 | 0.0032487691f, 0.0034598925f, 0.0036847358f, 0.0039241906f, |
| 1969 | 0.0041792066f, 0.0044507950f, 0.0047400328f, 0.0050480668f, |
| 1970 | 0.0053761186f, 0.0057254891f, 0.0060975636f, 0.0064938176f, |
| 1971 | 0.0069158225f, 0.0073652516f, 0.0078438871f, 0.0083536271f, |
| 1972 | 0.0088964928f, 0.009474637f, 0.010090352f, 0.010746080f, |
| 1973 | 0.011444421f, 0.012188144f, 0.012980198f, 0.013823725f, |
| 1974 | 0.014722068f, 0.015678791f, 0.016697687f, 0.017782797f, |
| 1975 | 0.018938423f, 0.020169149f, 0.021479854f, 0.022875735f, |
| 1976 | 0.024362330f, 0.025945531f, 0.027631618f, 0.029427276f, |
| 1977 | 0.031339626f, 0.033376252f, 0.035545228f, 0.037855157f, |
| 1978 | 0.040315199f, 0.042935108f, 0.045725273f, 0.048696758f, |
| 1979 | 0.051861348f, 0.055231591f, 0.058820850f, 0.062643361f, |
| 1980 | 0.066714279f, 0.071049749f, 0.075666962f, 0.080584227f, |
| 1981 | 0.085821044f, 0.091398179f, 0.097337747f, 0.10366330f, |
| 1982 | 0.11039993f, 0.11757434f, 0.12521498f, 0.13335215f, |
| 1983 | 0.14201813f, 0.15124727f, 0.16107617f, 0.17154380f, |
| 1984 | 0.18269168f, 0.19456402f, 0.20720788f, 0.22067342f, |
| 1985 | 0.23501402f, 0.25028656f, 0.26655159f, 0.28387361f, |
| 1986 | 0.30232132f, 0.32196786f, 0.34289114f, 0.36517414f, |
| 1987 | 0.38890521f, 0.41417847f, 0.44109412f, 0.46975890f, |
| 1988 | 0.50028648f, 0.53279791f, 0.56742212f, 0.60429640f, |
| 1989 | 0.64356699f, 0.68538959f, 0.72993007f, 0.77736504f, |
| 1990 | 0.82788260f, 0.88168307f, 0.9389798f, 1.0f |
| 1991 | }; |
| 1992 | |
| 1993 | |
| 1994 | // @OPTIMIZE: if you want to replace this bresenham line-drawing routine, |
| 1995 | // note that you must produce bit-identical output to decode correctly; |
| 1996 | // this specific sequence of operations is specified in the spec (it's |
| 1997 | // drawing integer-quantized frequency-space lines that the encoder |
| 1998 | // expects to be exactly the same) |
| 1999 | // ... also, isn't the whole point of Bresenham's algorithm to NOT |
| 2000 | // have to divide in the setup? sigh. |
| 2001 | #ifndef STB_VORBIS_NO_DEFER_FLOOR |
| 2002 | #define LINE_OP(a,b)a *= b a *= b |
| 2003 | #else |
| 2004 | #define LINE_OP(a,b)a *= b a = b |
| 2005 | #endif |
| 2006 | |
| 2007 | #ifdef STB_VORBIS_DIVIDE_TABLE |
| 2008 | #define DIVTAB_NUMER 32 |
| 2009 | #define DIVTAB_DENOM 64 |
| 2010 | int8 integer_divide_table[DIVTAB_NUMER][DIVTAB_DENOM]; // 2KB |
| 2011 | #endif |
| 2012 | |
| 2013 | static __forceinlineinline void draw_line(float *output, int x0, int y0, int x1, int y1, int n) |
| 2014 | { |
| 2015 | int dy = y1 - y0; |
| 2016 | int adx = x1 - x0; |
| 2017 | int ady = abs(dy); |
| 2018 | int base; |
| 2019 | int x=x0,y=y0; |
| 2020 | int err = 0; |
| 2021 | int sy; |
| 2022 | |
| 2023 | #ifdef STB_VORBIS_DIVIDE_TABLE |
| 2024 | if (adx < DIVTAB_DENOM && ady < DIVTAB_NUMER) { |
| 2025 | if (dy < 0) { |
| 2026 | base = -integer_divide_table[ady][adx]; |
| 2027 | sy = base-1; |
| 2028 | } else { |
| 2029 | base = integer_divide_table[ady][adx]; |
| 2030 | sy = base+1; |
| 2031 | } |
| 2032 | } else { |
| 2033 | base = dy / adx; |
| 2034 | if (dy < 0) |
| 2035 | sy = base - 1; |
| 2036 | else |
| 2037 | sy = base+1; |
| 2038 | } |
| 2039 | #else |
| 2040 | base = dy / adx; |
| 2041 | if (dy < 0) |
| 2042 | sy = base - 1; |
| 2043 | else |
| 2044 | sy = base+1; |
| 2045 | #endif |
| 2046 | ady -= abs(base) * adx; |
| 2047 | if (x1 > n) x1 = n; |
| 2048 | LINE_OP(output[x], inverse_db_table[y])output[x] *= inverse_db_table[y]; |
| 2049 | for (++x; x < x1; ++x) { |
| 2050 | err += ady; |
| 2051 | if (err >= adx) { |
| 2052 | err -= adx; |
| 2053 | y += sy; |
| 2054 | } else |
| 2055 | y += base; |
| 2056 | LINE_OP(output[x], inverse_db_table[y])output[x] *= inverse_db_table[y]; |
| 2057 | } |
| 2058 | } |
| 2059 | |
| 2060 | static int residue_decode(vorb *f, Codebook *book, float *target, int offset, int n, int rtype) |
| 2061 | { |
| 2062 | int k; |
| 2063 | if (rtype == 0) { |
| 2064 | int step = n / book->dimensions; |
| 2065 | for (k=0; k < step; ++k) |
| 2066 | if (!codebook_decode_step(f, book, target+offset+k, n-offset-k, step)) |
| 2067 | return FALSE0; |
| 2068 | } else { |
| 2069 | for (k=0; k < n; ) { |
| 2070 | if (!codebook_decode(f, book, target+offset, n-k)) |
| 2071 | return FALSE0; |
| 2072 | k += book->dimensions; |
| 2073 | offset += book->dimensions; |
| 2074 | } |
| 2075 | } |
| 2076 | return TRUE1; |
| 2077 | } |
| 2078 | |
| 2079 | static void decode_residue(vorb *f, float *residue_buffers[], int ch, int n, int rn, uint8 *do_not_decode) |
| 2080 | { |
| 2081 | int i,j,pass; |
| 2082 | Residue *r = f->residue_config + rn; |
| 2083 | int rtype = f->residue_types[rn]; |
| 2084 | int c = r->classbook; |
| 2085 | int classwords = f->codebooks[c].dimensions; |
| 2086 | int n_read = r->end - r->begin; |
| 2087 | int part_read = n_read / r->part_size; |
| 2088 | int temp_alloc_point = temp_alloc_save(f)((f)->temp_offset); |
| 2089 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2090 | uint8 ***part_classdata = (uint8 ***) temp_block_array(f,f->channels, part_read * sizeof(**part_classdata))make_block_array((f->alloc.alloc_buffer ? setup_temp_malloc (f,(f->channels*(sizeof(void *)+(part_read * sizeof(**part_classdata ))))) : alloca((f->channels*(sizeof(void *)+(part_read * sizeof (**part_classdata)))))), f->channels, part_read * sizeof(* *part_classdata)); |
| 2091 | #else |
| 2092 | int **classifications = (int **) temp_block_array(f,f->channels, part_read * sizeof(**classifications))make_block_array((f->alloc.alloc_buffer ? setup_temp_malloc (f,(f->channels*(sizeof(void *)+(part_read * sizeof(**classifications ))))) : alloca((f->channels*(sizeof(void *)+(part_read * sizeof (**classifications)))))), f->channels, part_read * sizeof( **classifications)); |
| 2093 | #endif |
| 2094 | |
| 2095 | stb_prof(2)(void)0; |
| 2096 | for (i=0; i < ch; ++i) |
| 2097 | if (!do_not_decode[i]) |
| 2098 | memset(residue_buffers[i], 0, sizeof(float) * n); |
| 2099 | |
| 2100 | if (rtype == 2 && ch != 1) { |
| 2101 | int len = ch * n; |
| 2102 | (void)len; |
| 2103 | for (j=0; j < ch; ++j) |
| 2104 | if (!do_not_decode[j]) |
| 2105 | break; |
| 2106 | if (j == ch) |
| 2107 | goto done; |
| 2108 | |
| 2109 | stb_prof(3)(void)0; |
| 2110 | for (pass=0; pass < 8; ++pass) { |
| 2111 | int pcount = 0, class_set = 0; |
| 2112 | if (ch == 2) { |
| 2113 | stb_prof(13)(void)0; |
| 2114 | while (pcount < part_read) { |
| 2115 | int z = r->begin + pcount*r->part_size; |
| 2116 | int c_inter = (z & 1), p_inter = z>>1; |
| 2117 | if (pass == 0) { |
| 2118 | Codebook *c = f->codebooks+r->classbook; |
| 2119 | int q; |
| 2120 | DECODE(q,f,c)if (f->valid_bits < 10) prep_huffman(f); q = f->acc & ((1 << 10) - 1); q = c->fast_huffman[q]; if (q >= 0) { int n = c->codeword_lengths[q]; f->acc >>= n ; f->valid_bits -= n; if (f->valid_bits < 0) { f-> valid_bits = 0; q = -1; } } else { q = codebook_decode_scalar_raw (f,c); } if (c->sparse) q = c->sorted_values[q];; |
| 2121 | if (q == EOP(-1)) goto done; |
| 2122 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2123 | part_classdata[0][class_set] = r->classdata[q]; |
| 2124 | #else |
| 2125 | for (i=classwords-1; i >= 0; --i) { |
| 2126 | classifications[0][i+pcount] = q % r->classifications; |
| 2127 | q /= r->classifications; |
| 2128 | } |
| 2129 | #endif |
| 2130 | } |
| 2131 | stb_prof(5)(void)0; |
| 2132 | for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { |
| 2133 | int z = r->begin + pcount*r->part_size; |
| 2134 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2135 | int c = part_classdata[0][class_set][i]; |
| 2136 | #else |
| 2137 | int c = classifications[0][pcount]; |
| 2138 | #endif |
| 2139 | int b = r->residue_books[c][pass]; |
| 2140 | if (b >= 0) { |
| 2141 | Codebook *book = f->codebooks + b; |
| 2142 | stb_prof(20)(void)0; // accounts for X time |
| 2143 | #ifdef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 2144 | if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size)) |
| 2145 | goto done; |
| 2146 | #else |
| 2147 | // saves 1% |
| 2148 | if (!codebook_decode_deinterleave_repeat_2(f, book, residue_buffers, &c_inter, &p_inter, n, r->part_size)) |
| 2149 | goto done; |
| 2150 | #endif |
| 2151 | stb_prof(7)(void)0; |
| 2152 | } else { |
| 2153 | z += r->part_size; |
| 2154 | c_inter = z & 1; |
| 2155 | p_inter = z >> 1; |
| 2156 | } |
| 2157 | } |
| 2158 | stb_prof(8)(void)0; |
| 2159 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2160 | ++class_set; |
| 2161 | #endif |
| 2162 | } |
| 2163 | } else if (ch == 1) { |
| 2164 | while (pcount < part_read) { |
| 2165 | int z = r->begin + pcount*r->part_size; |
| 2166 | int c_inter = 0, p_inter = z; |
| 2167 | if (pass == 0) { |
| 2168 | Codebook *c = f->codebooks+r->classbook; |
| 2169 | int q; |
| 2170 | DECODE(q,f,c)if (f->valid_bits < 10) prep_huffman(f); q = f->acc & ((1 << 10) - 1); q = c->fast_huffman[q]; if (q >= 0) { int n = c->codeword_lengths[q]; f->acc >>= n ; f->valid_bits -= n; if (f->valid_bits < 0) { f-> valid_bits = 0; q = -1; } } else { q = codebook_decode_scalar_raw (f,c); } if (c->sparse) q = c->sorted_values[q];; |
| 2171 | if (q == EOP(-1)) goto done; |
| 2172 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2173 | part_classdata[0][class_set] = r->classdata[q]; |
| 2174 | #else |
| 2175 | for (i=classwords-1; i >= 0; --i) { |
| 2176 | classifications[0][i+pcount] = q % r->classifications; |
| 2177 | q /= r->classifications; |
| 2178 | } |
| 2179 | #endif |
| 2180 | } |
| 2181 | for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { |
| 2182 | int z = r->begin + pcount*r->part_size; |
| 2183 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2184 | int c = part_classdata[0][class_set][i]; |
| 2185 | #else |
| 2186 | int c = classifications[0][pcount]; |
| 2187 | #endif |
| 2188 | int b = r->residue_books[c][pass]; |
| 2189 | if (b >= 0) { |
| 2190 | Codebook *book = f->codebooks + b; |
| 2191 | stb_prof(22)(void)0; |
| 2192 | if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size)) |
| 2193 | goto done; |
| 2194 | stb_prof(3)(void)0; |
| 2195 | } else { |
| 2196 | z += r->part_size; |
| 2197 | c_inter = 0; |
| 2198 | p_inter = z; |
| 2199 | } |
| 2200 | } |
| 2201 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2202 | ++class_set; |
| 2203 | #endif |
| 2204 | } |
| 2205 | } else { |
| 2206 | while (pcount < part_read) { |
| 2207 | int z = r->begin + pcount*r->part_size; |
| 2208 | int c_inter = z % ch, p_inter = z/ch; |
| 2209 | if (pass == 0) { |
| 2210 | Codebook *c = f->codebooks+r->classbook; |
| 2211 | int q; |
| 2212 | DECODE(q,f,c)if (f->valid_bits < 10) prep_huffman(f); q = f->acc & ((1 << 10) - 1); q = c->fast_huffman[q]; if (q >= 0) { int n = c->codeword_lengths[q]; f->acc >>= n ; f->valid_bits -= n; if (f->valid_bits < 0) { f-> valid_bits = 0; q = -1; } } else { q = codebook_decode_scalar_raw (f,c); } if (c->sparse) q = c->sorted_values[q];; |
| 2213 | if (q == EOP(-1)) goto done; |
| 2214 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2215 | part_classdata[0][class_set] = r->classdata[q]; |
| 2216 | #else |
| 2217 | for (i=classwords-1; i >= 0; --i) { |
| 2218 | classifications[0][i+pcount] = q % r->classifications; |
| 2219 | q /= r->classifications; |
| 2220 | } |
| 2221 | #endif |
| 2222 | } |
| 2223 | for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { |
| 2224 | int z = r->begin + pcount*r->part_size; |
| 2225 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2226 | int c = part_classdata[0][class_set][i]; |
| 2227 | #else |
| 2228 | int c = classifications[0][pcount]; |
| 2229 | #endif |
| 2230 | int b = r->residue_books[c][pass]; |
| 2231 | if (b >= 0) { |
| 2232 | Codebook *book = f->codebooks + b; |
| 2233 | stb_prof(22)(void)0; |
| 2234 | if (!codebook_decode_deinterleave_repeat(f, book, residue_buffers, ch, &c_inter, &p_inter, n, r->part_size)) |
| 2235 | goto done; |
| 2236 | stb_prof(3)(void)0; |
| 2237 | } else { |
| 2238 | z += r->part_size; |
| 2239 | c_inter = z % ch; |
| 2240 | p_inter = z / ch; |
| 2241 | } |
| 2242 | } |
| 2243 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2244 | ++class_set; |
| 2245 | #endif |
| 2246 | } |
| 2247 | } |
| 2248 | } |
| 2249 | goto done; |
| 2250 | } |
| 2251 | stb_prof(9)(void)0; |
| 2252 | |
| 2253 | for (pass=0; pass < 8; ++pass) { |
| 2254 | int pcount = 0, class_set=0; |
| 2255 | while (pcount < part_read) { |
| 2256 | if (pass == 0) { |
| 2257 | for (j=0; j < ch; ++j) { |
| 2258 | if (!do_not_decode[j]) { |
| 2259 | Codebook *c = f->codebooks+r->classbook; |
| 2260 | int temp; |
| 2261 | DECODE(temp,f,c)if (f->valid_bits < 10) prep_huffman(f); temp = f->acc & ((1 << 10) - 1); temp = c->fast_huffman[temp] ; if (temp >= 0) { int n = c->codeword_lengths[temp]; f ->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; temp = -1; } } else { temp = codebook_decode_scalar_raw(f,c); } if (c->sparse) temp = c ->sorted_values[temp];; |
| 2262 | if (temp == EOP(-1)) goto done; |
| 2263 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2264 | part_classdata[j][class_set] = r->classdata[temp]; |
| 2265 | #else |
| 2266 | for (i=classwords-1; i >= 0; --i) { |
| 2267 | classifications[j][i+pcount] = temp % r->classifications; |
| 2268 | temp /= r->classifications; |
| 2269 | } |
| 2270 | #endif |
| 2271 | } |
| 2272 | } |
| 2273 | } |
| 2274 | for (i=0; i < classwords && pcount < part_read; ++i, ++pcount) { |
| 2275 | for (j=0; j < ch; ++j) { |
| 2276 | if (!do_not_decode[j]) { |
| 2277 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2278 | int c = part_classdata[j][class_set][i]; |
| 2279 | #else |
| 2280 | int c = classifications[j][pcount]; |
| 2281 | #endif |
| 2282 | int b = r->residue_books[c][pass]; |
| 2283 | if (b >= 0) { |
| 2284 | float *target = residue_buffers[j]; |
| 2285 | int offset = r->begin + pcount * r->part_size; |
| 2286 | int n = r->part_size; |
| 2287 | Codebook *book = f->codebooks + b; |
| 2288 | if (!residue_decode(f, book, target, offset, n, rtype)) |
| 2289 | goto done; |
| 2290 | } |
| 2291 | } |
| 2292 | } |
| 2293 | } |
| 2294 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 2295 | ++class_set; |
| 2296 | #endif |
| 2297 | } |
| 2298 | } |
| 2299 | done: |
| 2300 | stb_prof(0)(void)0; |
| 2301 | temp_alloc_restore(f,temp_alloc_point)((f)->temp_offset = (temp_alloc_point)); |
| 2302 | } |
| 2303 | |
| 2304 | |
| 2305 | #if 0 |
| 2306 | // slow way for debugging |
| 2307 | void inverse_mdct_slow(float *buffer, int n) |
| 2308 | { |
| 2309 | int i,j; |
| 2310 | int n2 = n >> 1; |
| 2311 | float *x = (float *) malloc(sizeof(*x) * n2); |
| 2312 | memcpy(x, buffer, sizeof(*x) * n2); |
| 2313 | for (i=0; i < n; ++i) { |
| 2314 | float acc = 0; |
| 2315 | for (j=0; j < n2; ++j) |
| 2316 | // formula from paper: |
| 2317 | //acc += n/4.0f * x[j] * (float) cos(M_PI / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1)); |
| 2318 | // formula from wikipedia |
| 2319 | //acc += 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5)); |
| 2320 | // these are equivalent, except the formula from the paper inverts the multiplier! |
| 2321 | // however, what actually works is NO MULTIPLIER!?! |
| 2322 | //acc += 64 * 2.0f / n2 * x[j] * (float) cos(M_PI/n2 * (i + 0.5 + n2/2)*(j + 0.5)); |
| 2323 | acc += x[j] * (float) cos(M_PI3.14159265358979323846 / 2 / n * (2 * i + 1 + n/2.0)*(2*j+1)); |
| 2324 | buffer[i] = acc; |
| 2325 | } |
| 2326 | free(x); |
| 2327 | } |
| 2328 | #elif 0 |
| 2329 | // same as above, but just barely able to run in real time on modern machines |
| 2330 | void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype) |
| 2331 | { |
| 2332 | float mcos[16384]; |
| 2333 | int i,j; |
| 2334 | int n2 = n >> 1, nmask = (n << 2) -1; |
| 2335 | float *x = (float *) malloc(sizeof(*x) * n2); |
| 2336 | memcpy(x, buffer, sizeof(*x) * n2); |
| 2337 | for (i=0; i < 4*n; ++i) |
| 2338 | mcos[i] = (float) cos(M_PI3.14159265358979323846 / 2 * i / n); |
| 2339 | |
| 2340 | for (i=0; i < n; ++i) { |
| 2341 | float acc = 0; |
| 2342 | for (j=0; j < n2; ++j) |
| 2343 | acc += x[j] * mcos[(2 * i + 1 + n2)*(2*j+1) & nmask]; |
| 2344 | buffer[i] = acc; |
| 2345 | } |
| 2346 | free(x); |
| 2347 | } |
| 2348 | #else |
| 2349 | // transform to use a slow dct-iv; this is STILL basically trivial, |
| 2350 | // but only requires half as many ops |
| 2351 | void dct_iv_slow(float *buffer, int n) |
| 2352 | { |
| 2353 | float mcos[16384]; |
| 2354 | float x[2048]; |
| 2355 | int i,j; |
| 2356 | int n2 = n >> 1, nmask = (n << 3) - 1; |
| 2357 | (void)n2; |
| 2358 | memcpy(x, buffer, sizeof(*x) * n); |
| 2359 | for (i=0; i < 8*n; ++i) |
| 2360 | mcos[i] = (float) cos(M_PI3.14159265358979323846 / 4 * i / n); |
| 2361 | for (i=0; i < n; ++i) { |
| 2362 | float acc = 0; |
| 2363 | for (j=0; j < n; ++j) |
| 2364 | acc += x[j] * mcos[((2 * i + 1)*(2*j+1)) & nmask]; |
| 2365 | //acc += x[j] * cos(M_PI / n * (i + 0.5) * (j + 0.5)); |
| 2366 | buffer[i] = acc; |
| 2367 | } |
| 2368 | } |
| 2369 | |
| 2370 | void inverse_mdct_slow(float *buffer, int n, vorb *f, int blocktype) |
| 2371 | { |
| 2372 | int i, n4 = n >> 2, n2 = n >> 1, n3_4 = n - n4; |
| 2373 | float temp[4096]; |
| 2374 | |
| 2375 | memcpy(temp, buffer, n2 * sizeof(float)); |
| 2376 | dct_iv_slow(temp, n2); // returns -c'-d, a-b' |
| 2377 | |
| 2378 | for (i=0; i < n4 ; ++i) buffer[i] = temp[i+n4]; // a-b' |
| 2379 | for ( ; i < n3_4; ++i) buffer[i] = -temp[n3_4 - i - 1]; // b-a', c+d' |
| 2380 | for ( ; i < n ; ++i) buffer[i] = -temp[i - n3_4]; // c'+d |
| 2381 | } |
| 2382 | #endif |
| 2383 | |
| 2384 | #ifndef LIBVORBIS_MDCT0 |
| 2385 | #define LIBVORBIS_MDCT0 0 |
| 2386 | #endif |
| 2387 | |
| 2388 | #if LIBVORBIS_MDCT0 |
| 2389 | // directly call the vorbis MDCT using an interface documented |
| 2390 | // by Jeff Roberts... useful for performance comparison |
| 2391 | typedef struct |
| 2392 | { |
| 2393 | int n; |
| 2394 | int log2n; |
| 2395 | |
| 2396 | float *trig; |
| 2397 | int *bitrev; |
| 2398 | |
| 2399 | float scale; |
| 2400 | } mdct_lookup; |
| 2401 | |
| 2402 | extern void mdct_init(mdct_lookup *lookup, int n); |
| 2403 | extern void mdct_clear(mdct_lookup *l); |
| 2404 | extern void mdct_backward(mdct_lookup *init, float *in, float *out); |
| 2405 | |
| 2406 | mdct_lookup M1,M2; |
| 2407 | |
| 2408 | void inverse_mdct(float *buffer, int n, vorb *f, int blocktype) |
| 2409 | { |
| 2410 | mdct_lookup *M; |
| 2411 | if (M1.n == n) M = &M1; |
| 2412 | else if (M2.n == n) M = &M2; |
| 2413 | else if (M1.n == 0) { mdct_init(&M1, n); M = &M1; } |
| 2414 | else { |
| 2415 | if (M2.n) __asm int 3; |
| 2416 | mdct_init(&M2, n); |
| 2417 | M = &M2; |
| 2418 | } |
| 2419 | |
| 2420 | mdct_backward(M, buffer, buffer); |
| 2421 | } |
| 2422 | #endif |
| 2423 | |
| 2424 | |
| 2425 | // the following were split out into separate functions while optimizing; |
| 2426 | // they could be pushed back up but eh. __forceinline showed no change; |
| 2427 | // they're probably already being inlined. |
| 2428 | static void imdct_step3_iter0_loop(int n, float *e, int i_off, int k_off, float *A) |
| 2429 | { |
| 2430 | float *ee0 = e + i_off; |
| 2431 | float *ee2 = ee0 + k_off; |
| 2432 | int i; |
| 2433 | |
| 2434 | assert((n & 3) == 0)(((n & 3) == 0) ? (void)0 : _assert("(n & 3) == 0", "src/siege/internal/stb/stb_vorbis.c" , 2434)); |
| 2435 | for (i=(n>>2); i > 0; --i) { |
| 2436 | float k00_20, k01_21; |
| 2437 | k00_20 = ee0[ 0] - ee2[ 0]; |
| 2438 | k01_21 = ee0[-1] - ee2[-1]; |
| 2439 | ee0[ 0] += ee2[ 0];//ee0[ 0] = ee0[ 0] + ee2[ 0]; |
| 2440 | ee0[-1] += ee2[-1];//ee0[-1] = ee0[-1] + ee2[-1]; |
| 2441 | ee2[ 0] = k00_20 * A[0] - k01_21 * A[1]; |
| 2442 | ee2[-1] = k01_21 * A[0] + k00_20 * A[1]; |
| 2443 | A += 8; |
| 2444 | |
| 2445 | k00_20 = ee0[-2] - ee2[-2]; |
| 2446 | k01_21 = ee0[-3] - ee2[-3]; |
| 2447 | ee0[-2] += ee2[-2];//ee0[-2] = ee0[-2] + ee2[-2]; |
| 2448 | ee0[-3] += ee2[-3];//ee0[-3] = ee0[-3] + ee2[-3]; |
| 2449 | ee2[-2] = k00_20 * A[0] - k01_21 * A[1]; |
| 2450 | ee2[-3] = k01_21 * A[0] + k00_20 * A[1]; |
| 2451 | A += 8; |
| 2452 | |
| 2453 | k00_20 = ee0[-4] - ee2[-4]; |
| 2454 | k01_21 = ee0[-5] - ee2[-5]; |
| 2455 | ee0[-4] += ee2[-4];//ee0[-4] = ee0[-4] + ee2[-4]; |
| 2456 | ee0[-5] += ee2[-5];//ee0[-5] = ee0[-5] + ee2[-5]; |
| 2457 | ee2[-4] = k00_20 * A[0] - k01_21 * A[1]; |
| 2458 | ee2[-5] = k01_21 * A[0] + k00_20 * A[1]; |
| 2459 | A += 8; |
| 2460 | |
| 2461 | k00_20 = ee0[-6] - ee2[-6]; |
| 2462 | k01_21 = ee0[-7] - ee2[-7]; |
| 2463 | ee0[-6] += ee2[-6];//ee0[-6] = ee0[-6] + ee2[-6]; |
| 2464 | ee0[-7] += ee2[-7];//ee0[-7] = ee0[-7] + ee2[-7]; |
| 2465 | ee2[-6] = k00_20 * A[0] - k01_21 * A[1]; |
| 2466 | ee2[-7] = k01_21 * A[0] + k00_20 * A[1]; |
| 2467 | A += 8; |
| 2468 | ee0 -= 8; |
| 2469 | ee2 -= 8; |
| 2470 | } |
| 2471 | } |
| 2472 | |
| 2473 | static void imdct_step3_inner_r_loop(int lim, float *e, int d0, int k_off, float *A, int k1) |
| 2474 | { |
| 2475 | int i; |
| 2476 | float k00_20, k01_21; |
| 2477 | |
| 2478 | float *e0 = e + d0; |
| 2479 | float *e2 = e0 + k_off; |
| 2480 | |
| 2481 | for (i=lim >> 2; i > 0; --i) { |
| 2482 | k00_20 = e0[-0] - e2[-0]; |
| 2483 | k01_21 = e0[-1] - e2[-1]; |
| 2484 | e0[-0] += e2[-0];//e0[-0] = e0[-0] + e2[-0]; |
| 2485 | e0[-1] += e2[-1];//e0[-1] = e0[-1] + e2[-1]; |
| 2486 | e2[-0] = (k00_20)*A[0] - (k01_21) * A[1]; |
| 2487 | e2[-1] = (k01_21)*A[0] + (k00_20) * A[1]; |
| 2488 | |
| 2489 | A += k1; |
| 2490 | |
| 2491 | k00_20 = e0[-2] - e2[-2]; |
| 2492 | k01_21 = e0[-3] - e2[-3]; |
| 2493 | e0[-2] += e2[-2];//e0[-2] = e0[-2] + e2[-2]; |
| 2494 | e0[-3] += e2[-3];//e0[-3] = e0[-3] + e2[-3]; |
| 2495 | e2[-2] = (k00_20)*A[0] - (k01_21) * A[1]; |
| 2496 | e2[-3] = (k01_21)*A[0] + (k00_20) * A[1]; |
| 2497 | |
| 2498 | A += k1; |
| 2499 | |
| 2500 | k00_20 = e0[-4] - e2[-4]; |
| 2501 | k01_21 = e0[-5] - e2[-5]; |
| 2502 | e0[-4] += e2[-4];//e0[-4] = e0[-4] + e2[-4]; |
| 2503 | e0[-5] += e2[-5];//e0[-5] = e0[-5] + e2[-5]; |
| 2504 | e2[-4] = (k00_20)*A[0] - (k01_21) * A[1]; |
| 2505 | e2[-5] = (k01_21)*A[0] + (k00_20) * A[1]; |
| 2506 | |
| 2507 | A += k1; |
| 2508 | |
| 2509 | k00_20 = e0[-6] - e2[-6]; |
| 2510 | k01_21 = e0[-7] - e2[-7]; |
| 2511 | e0[-6] += e2[-6];//e0[-6] = e0[-6] + e2[-6]; |
| 2512 | e0[-7] += e2[-7];//e0[-7] = e0[-7] + e2[-7]; |
| 2513 | e2[-6] = (k00_20)*A[0] - (k01_21) * A[1]; |
| 2514 | e2[-7] = (k01_21)*A[0] + (k00_20) * A[1]; |
| 2515 | |
| 2516 | e0 -= 8; |
| 2517 | e2 -= 8; |
| 2518 | |
| 2519 | A += k1; |
| 2520 | } |
| 2521 | } |
| 2522 | |
| 2523 | static void imdct_step3_inner_s_loop(int n, float *e, int i_off, int k_off, float *A, int a_off, int k0) |
| 2524 | { |
| 2525 | int i; |
| 2526 | float A0 = A[0]; |
| 2527 | float A1 = A[0+1]; |
| 2528 | float A2 = A[0+a_off]; |
| 2529 | float A3 = A[0+a_off+1]; |
| 2530 | float A4 = A[0+a_off*2+0]; |
| 2531 | float A5 = A[0+a_off*2+1]; |
| 2532 | float A6 = A[0+a_off*3+0]; |
| 2533 | float A7 = A[0+a_off*3+1]; |
| 2534 | |
| 2535 | float k00,k11; |
| 2536 | |
| 2537 | float *ee0 = e +i_off; |
| 2538 | float *ee2 = ee0+k_off; |
| 2539 | |
| 2540 | for (i=n; i > 0; --i) { |
| 2541 | k00 = ee0[ 0] - ee2[ 0]; |
| 2542 | k11 = ee0[-1] - ee2[-1]; |
| 2543 | ee0[ 0] = ee0[ 0] + ee2[ 0]; |
| 2544 | ee0[-1] = ee0[-1] + ee2[-1]; |
| 2545 | ee2[ 0] = (k00) * A0 - (k11) * A1; |
| 2546 | ee2[-1] = (k11) * A0 + (k00) * A1; |
| 2547 | |
| 2548 | k00 = ee0[-2] - ee2[-2]; |
| 2549 | k11 = ee0[-3] - ee2[-3]; |
| 2550 | ee0[-2] = ee0[-2] + ee2[-2]; |
| 2551 | ee0[-3] = ee0[-3] + ee2[-3]; |
| 2552 | ee2[-2] = (k00) * A2 - (k11) * A3; |
| 2553 | ee2[-3] = (k11) * A2 + (k00) * A3; |
| 2554 | |
| 2555 | k00 = ee0[-4] - ee2[-4]; |
| 2556 | k11 = ee0[-5] - ee2[-5]; |
| 2557 | ee0[-4] = ee0[-4] + ee2[-4]; |
| 2558 | ee0[-5] = ee0[-5] + ee2[-5]; |
| 2559 | ee2[-4] = (k00) * A4 - (k11) * A5; |
| 2560 | ee2[-5] = (k11) * A4 + (k00) * A5; |
| 2561 | |
| 2562 | k00 = ee0[-6] - ee2[-6]; |
| 2563 | k11 = ee0[-7] - ee2[-7]; |
| 2564 | ee0[-6] = ee0[-6] + ee2[-6]; |
| 2565 | ee0[-7] = ee0[-7] + ee2[-7]; |
| 2566 | ee2[-6] = (k00) * A6 - (k11) * A7; |
| 2567 | ee2[-7] = (k11) * A6 + (k00) * A7; |
| 2568 | |
| 2569 | ee0 -= k0; |
| 2570 | ee2 -= k0; |
| 2571 | } |
| 2572 | } |
| 2573 | |
| 2574 | static __forceinlineinline void iter_54(float *z) |
| 2575 | { |
| 2576 | float k00,k11,k22,k33; |
| 2577 | float y0,y1,y2,y3; |
| 2578 | |
| 2579 | k00 = z[ 0] - z[-4]; |
| 2580 | y0 = z[ 0] + z[-4]; |
| 2581 | y2 = z[-2] + z[-6]; |
| 2582 | k22 = z[-2] - z[-6]; |
| 2583 | |
| 2584 | z[-0] = y0 + y2; // z0 + z4 + z2 + z6 |
| 2585 | z[-2] = y0 - y2; // z0 + z4 - z2 - z6 |
| 2586 | |
| 2587 | // done with y0,y2 |
| 2588 | |
| 2589 | k33 = z[-3] - z[-7]; |
| 2590 | |
| 2591 | z[-4] = k00 + k33; // z0 - z4 + z3 - z7 |
| 2592 | z[-6] = k00 - k33; // z0 - z4 - z3 + z7 |
| 2593 | |
| 2594 | // done with k33 |
| 2595 | |
| 2596 | k11 = z[-1] - z[-5]; |
| 2597 | y1 = z[-1] + z[-5]; |
| 2598 | y3 = z[-3] + z[-7]; |
| 2599 | |
| 2600 | z[-1] = y1 + y3; // z1 + z5 + z3 + z7 |
| 2601 | z[-3] = y1 - y3; // z1 + z5 - z3 - z7 |
| 2602 | z[-5] = k11 - k22; // z1 - z5 + z2 - z6 |
| 2603 | z[-7] = k11 + k22; // z1 - z5 - z2 + z6 |
| 2604 | } |
| 2605 | |
| 2606 | static void imdct_step3_inner_s_loop_ld654(int n, float *e, int i_off, float *A, int base_n) |
| 2607 | { |
| 2608 | int k_off = -8; |
| 2609 | (void)k_off; |
| 2610 | int a_off = base_n >> 3; |
| 2611 | float A2 = A[0+a_off]; |
| 2612 | float *z = e + i_off; |
| 2613 | float *base = z - 16 * n; |
| 2614 | |
| 2615 | while (z > base) { |
| 2616 | float k00,k11; |
| 2617 | |
| 2618 | k00 = z[-0] - z[-8]; |
| 2619 | k11 = z[-1] - z[-9]; |
| 2620 | z[-0] = z[-0] + z[-8]; |
| 2621 | z[-1] = z[-1] + z[-9]; |
| 2622 | z[-8] = k00; |
| 2623 | z[-9] = k11 ; |
| 2624 | |
| 2625 | k00 = z[ -2] - z[-10]; |
| 2626 | k11 = z[ -3] - z[-11]; |
| 2627 | z[ -2] = z[ -2] + z[-10]; |
| 2628 | z[ -3] = z[ -3] + z[-11]; |
| 2629 | z[-10] = (k00+k11) * A2; |
| 2630 | z[-11] = (k11-k00) * A2; |
| 2631 | |
| 2632 | k00 = z[-12] - z[ -4]; // reverse to avoid a unary negation |
| 2633 | k11 = z[ -5] - z[-13]; |
| 2634 | z[ -4] = z[ -4] + z[-12]; |
| 2635 | z[ -5] = z[ -5] + z[-13]; |
| 2636 | z[-12] = k11; |
| 2637 | z[-13] = k00; |
| 2638 | |
| 2639 | k00 = z[-14] - z[ -6]; // reverse to avoid a unary negation |
| 2640 | k11 = z[ -7] - z[-15]; |
| 2641 | z[ -6] = z[ -6] + z[-14]; |
| 2642 | z[ -7] = z[ -7] + z[-15]; |
| 2643 | z[-14] = (k00+k11) * A2; |
| 2644 | z[-15] = (k00-k11) * A2; |
| 2645 | |
| 2646 | iter_54(z); |
| 2647 | iter_54(z-8); |
| 2648 | z -= 16; |
| 2649 | } |
| 2650 | } |
| 2651 | |
| 2652 | static void inverse_mdct(float *buffer, int n, vorb *f, int blocktype) |
| 2653 | { |
| 2654 | int n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l; |
| 2655 | int n3_4 = n - n4, ld; |
| 2656 | (void)n3_4; |
| 2657 | // @OPTIMIZE: reduce register pressure by using fewer variables? |
| 2658 | int save_point = temp_alloc_save(f)((f)->temp_offset); |
| 2659 | float *buf2 = (float *) temp_alloc(f, n2 * sizeof(*buf2))(f->alloc.alloc_buffer ? setup_temp_malloc(f,n2 * sizeof(* buf2)) : alloca(n2 * sizeof(*buf2))); |
| 2660 | float *u=NULL((void*)0),*v=NULL((void*)0); |
| 2661 | // twiddle factors |
| 2662 | float *A = f->A[blocktype]; |
| 2663 | |
| 2664 | // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio" |
| 2665 | // See notes about bugs in that paper in less-optimal implementation 'inverse_mdct_old' after this function. |
| 2666 | |
| 2667 | // kernel from paper |
| 2668 | |
| 2669 | |
| 2670 | // merged: |
| 2671 | // copy and reflect spectral data |
| 2672 | // step 0 |
| 2673 | |
| 2674 | // note that it turns out that the items added together during |
| 2675 | // this step are, in fact, being added to themselves (as reflected |
| 2676 | // by step 0). inexplicable inefficiency! this became obvious |
| 2677 | // once I combined the passes. |
| 2678 | |
| 2679 | // so there's a missing 'times 2' here (for adding X to itself). |
| 2680 | // this propogates through linearly to the end, where the numbers |
| 2681 | // are 1/2 too small, and need to be compensated for. |
| 2682 | |
| 2683 | { |
| 2684 | float *d,*e, *AA, *e_stop; |
| 2685 | d = &buf2[n2-2]; |
| 2686 | AA = A; |
| 2687 | e = &buffer[0]; |
| 2688 | e_stop = &buffer[n2]; |
| 2689 | while (e != e_stop) { |
| 2690 | d[1] = (e[0] * AA[0] - e[2]*AA[1]); |
| 2691 | d[0] = (e[0] * AA[1] + e[2]*AA[0]); |
| 2692 | d -= 2; |
| 2693 | AA += 2; |
| 2694 | e += 4; |
| 2695 | } |
| 2696 | |
| 2697 | e = &buffer[n2-3]; |
| 2698 | while (d >= buf2) { |
| 2699 | d[1] = (-e[2] * AA[0] - -e[0]*AA[1]); |
| 2700 | d[0] = (-e[2] * AA[1] + -e[0]*AA[0]); |
| 2701 | d -= 2; |
| 2702 | AA += 2; |
| 2703 | e -= 4; |
| 2704 | } |
| 2705 | } |
| 2706 | |
| 2707 | // now we use symbolic names for these, so that we can |
| 2708 | // possibly swap their meaning as we change which operations |
| 2709 | // are in place |
| 2710 | |
| 2711 | u = buffer; |
| 2712 | v = buf2; |
| 2713 | |
| 2714 | // step 2 (paper output is w, now u) |
| 2715 | // this could be in place, but the data ends up in the wrong |
| 2716 | // place... _somebody_'s got to swap it, so this is nominated |
| 2717 | { |
| 2718 | float *AA = &A[n2-8]; |
| 2719 | float *d0,*d1, *e0, *e1; |
| 2720 | |
| 2721 | e0 = &v[n4]; |
| 2722 | e1 = &v[0]; |
| 2723 | |
| 2724 | d0 = &u[n4]; |
| 2725 | d1 = &u[0]; |
| 2726 | |
| 2727 | while (AA >= A) { |
| 2728 | float v40_20, v41_21; |
| 2729 | |
| 2730 | v41_21 = e0[1] - e1[1]; |
| 2731 | v40_20 = e0[0] - e1[0]; |
| 2732 | d0[1] = e0[1] + e1[1]; |
| 2733 | d0[0] = e0[0] + e1[0]; |
| 2734 | d1[1] = v41_21*AA[4] - v40_20*AA[5]; |
| 2735 | d1[0] = v40_20*AA[4] + v41_21*AA[5]; |
| 2736 | |
| 2737 | v41_21 = e0[3] - e1[3]; |
| 2738 | v40_20 = e0[2] - e1[2]; |
| 2739 | d0[3] = e0[3] + e1[3]; |
| 2740 | d0[2] = e0[2] + e1[2]; |
| 2741 | d1[3] = v41_21*AA[0] - v40_20*AA[1]; |
| 2742 | d1[2] = v40_20*AA[0] + v41_21*AA[1]; |
| 2743 | |
| 2744 | AA -= 8; |
| 2745 | |
| 2746 | d0 += 4; |
| 2747 | d1 += 4; |
| 2748 | e0 += 4; |
| 2749 | e1 += 4; |
| 2750 | } |
| 2751 | } |
| 2752 | |
| 2753 | // step 3 |
| 2754 | ld = ilog(n) - 1; // ilog is off-by-one from normal definitions |
| 2755 | |
| 2756 | // optimized step 3: |
| 2757 | |
| 2758 | // the original step3 loop can be nested r inside s or s inside r; |
| 2759 | // it's written originally as s inside r, but this is dumb when r |
| 2760 | // iterates many times, and s few. So I have two copies of it and |
| 2761 | // switch between them halfway. |
| 2762 | |
| 2763 | // this is iteration 0 of step 3 |
| 2764 | imdct_step3_iter0_loop(n >> 4, u, n2-1-n4*0, -(n >> 3), A); |
| 2765 | imdct_step3_iter0_loop(n >> 4, u, n2-1-n4*1, -(n >> 3), A); |
| 2766 | |
| 2767 | // this is iteration 1 of step 3 |
| 2768 | imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*0, -(n >> 4), A, 16); |
| 2769 | imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*1, -(n >> 4), A, 16); |
| 2770 | imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*2, -(n >> 4), A, 16); |
| 2771 | imdct_step3_inner_r_loop(n >> 5, u, n2-1 - n8*3, -(n >> 4), A, 16); |
| 2772 | |
| 2773 | l=2; |
| 2774 | for (; l < (ld-3)>>1; ++l) { |
| 2775 | int k0 = n >> (l+2), k0_2 = k0>>1; |
| 2776 | int lim = 1 << (l+1); |
| 2777 | int i; |
| 2778 | for (i=0; i < lim; ++i) |
| 2779 | imdct_step3_inner_r_loop(n >> (l+4), u, n2-1 - k0*i, -k0_2, A, 1 << (l+3)); |
| 2780 | } |
| 2781 | |
| 2782 | for (; l < ld-6; ++l) { |
| 2783 | int k0 = n >> (l+2), k1 = 1 << (l+3), k0_2 = k0>>1; |
| 2784 | int rlim = n >> (l+6), r; |
| 2785 | int lim = 1 << (l+1); |
| 2786 | int i_off; |
| 2787 | float *A0 = A; |
| 2788 | i_off = n2-1; |
| 2789 | for (r=rlim; r > 0; --r) { |
| 2790 | imdct_step3_inner_s_loop(lim, u, i_off, -k0_2, A0, k1, k0); |
| 2791 | A0 += k1*4; |
| 2792 | i_off -= 8; |
| 2793 | } |
| 2794 | } |
| 2795 | |
| 2796 | // iterations with count: |
| 2797 | // ld-6,-5,-4 all interleaved together |
| 2798 | // the big win comes from getting rid of needless flops |
| 2799 | // due to the constants on pass 5 & 4 being all 1 and 0; |
| 2800 | // combining them to be simultaneous to improve cache made little difference |
| 2801 | imdct_step3_inner_s_loop_ld654(n >> 5, u, n2-1, A, n); |
| 2802 | |
| 2803 | // output is u |
| 2804 | |
| 2805 | // step 4, 5, and 6 |
| 2806 | // cannot be in-place because of step 5 |
| 2807 | { |
| 2808 | uint16 *bitrev = f->bit_reverse[blocktype]; |
| 2809 | // weirdly, I'd have thought reading sequentially and writing |
| 2810 | // erratically would have been better than vice-versa, but in |
| 2811 | // fact that's not what my testing showed. (That is, with |
| 2812 | // j = bitreverse(i), do you read i and write j, or read j and write i.) |
| 2813 | |
| 2814 | float *d0 = &v[n4-4]; |
| 2815 | float *d1 = &v[n2-4]; |
| 2816 | while (d0 >= v) { |
| 2817 | int k4; |
| 2818 | |
| 2819 | k4 = bitrev[0]; |
| 2820 | d1[3] = u[k4+0]; |
| 2821 | d1[2] = u[k4+1]; |
| 2822 | d0[3] = u[k4+2]; |
| 2823 | d0[2] = u[k4+3]; |
| 2824 | |
| 2825 | k4 = bitrev[1]; |
| 2826 | d1[1] = u[k4+0]; |
| 2827 | d1[0] = u[k4+1]; |
| 2828 | d0[1] = u[k4+2]; |
| 2829 | d0[0] = u[k4+3]; |
| 2830 | |
| 2831 | d0 -= 4; |
| 2832 | d1 -= 4; |
| 2833 | bitrev += 2; |
| 2834 | } |
| 2835 | } |
| 2836 | // (paper output is u, now v) |
| 2837 | |
| 2838 | |
| 2839 | // data must be in buf2 |
| 2840 | assert(v == buf2)((v == buf2) ? (void)0 : _assert("v == buf2", "src/siege/internal/stb/stb_vorbis.c" , 2840)); |
| 2841 | |
| 2842 | // step 7 (paper output is v, now v) |
| 2843 | // this is now in place |
| 2844 | { |
| 2845 | float *C(2 | 4 | 1) = f->C(2 | 4 | 1)[blocktype]; |
| 2846 | float *d, *e; |
| 2847 | |
| 2848 | d = v; |
| 2849 | e = v + n2 - 4; |
| 2850 | |
| 2851 | while (d < e) { |
| 2852 | float a02,a11,b0,b1,b2,b3; |
| 2853 | |
| 2854 | a02 = d[0] - e[2]; |
| 2855 | a11 = d[1] + e[3]; |
| 2856 | |
| 2857 | b0 = C(2 | 4 | 1)[1]*a02 + C(2 | 4 | 1)[0]*a11; |
| 2858 | b1 = C(2 | 4 | 1)[1]*a11 - C(2 | 4 | 1)[0]*a02; |
| 2859 | |
| 2860 | b2 = d[0] + e[ 2]; |
| 2861 | b3 = d[1] - e[ 3]; |
| 2862 | |
| 2863 | d[0] = b2 + b0; |
| 2864 | d[1] = b3 + b1; |
| 2865 | e[2] = b2 - b0; |
| 2866 | e[3] = b1 - b3; |
| 2867 | |
| 2868 | a02 = d[2] - e[0]; |
| 2869 | a11 = d[3] + e[1]; |
| 2870 | |
| 2871 | b0 = C(2 | 4 | 1)[3]*a02 + C(2 | 4 | 1)[2]*a11; |
| 2872 | b1 = C(2 | 4 | 1)[3]*a11 - C(2 | 4 | 1)[2]*a02; |
| 2873 | |
| 2874 | b2 = d[2] + e[ 0]; |
| 2875 | b3 = d[3] - e[ 1]; |
| 2876 | |
| 2877 | d[2] = b2 + b0; |
| 2878 | d[3] = b3 + b1; |
| 2879 | e[0] = b2 - b0; |
| 2880 | e[1] = b1 - b3; |
| 2881 | |
| 2882 | C(2 | 4 | 1) += 4; |
| 2883 | d += 4; |
| 2884 | e -= 4; |
| 2885 | } |
| 2886 | } |
| 2887 | |
| 2888 | // data must be in buf2 |
| 2889 | |
| 2890 | |
| 2891 | // step 8+decode (paper output is X, now buffer) |
| 2892 | // this generates pairs of data a la 8 and pushes them directly through |
| 2893 | // the decode kernel (pushing rather than pulling) to avoid having |
| 2894 | // to make another pass later |
| 2895 | |
| 2896 | // this cannot POSSIBLY be in place, so we refer to the buffers directly |
| 2897 | |
| 2898 | { |
| 2899 | float *d0,*d1,*d2,*d3; |
| 2900 | |
| 2901 | float *B = f->B[blocktype] + n2 - 8; |
| 2902 | float *e = buf2 + n2 - 8; |
| 2903 | d0 = &buffer[0]; |
| 2904 | d1 = &buffer[n2-4]; |
| 2905 | d2 = &buffer[n2]; |
| 2906 | d3 = &buffer[n-4]; |
| 2907 | while (e >= v) { |
| 2908 | float p0,p1,p2,p3; |
| 2909 | |
| 2910 | p3 = e[6]*B[7] - e[7]*B[6]; |
| 2911 | p2 = -e[6]*B[6] - e[7]*B[7]; |
| 2912 | |
| 2913 | d0[0] = p3; |
| 2914 | d1[3] = - p3; |
| 2915 | d2[0] = p2; |
| 2916 | d3[3] = p2; |
| 2917 | |
| 2918 | p1 = e[4]*B[5] - e[5]*B[4]; |
| 2919 | p0 = -e[4]*B[4] - e[5]*B[5]; |
| 2920 | |
| 2921 | d0[1] = p1; |
| 2922 | d1[2] = - p1; |
| 2923 | d2[1] = p0; |
| 2924 | d3[2] = p0; |
| 2925 | |
| 2926 | p3 = e[2]*B[3] - e[3]*B[2]; |
| 2927 | p2 = -e[2]*B[2] - e[3]*B[3]; |
| 2928 | |
| 2929 | d0[2] = p3; |
| 2930 | d1[1] = - p3; |
| 2931 | d2[2] = p2; |
| 2932 | d3[1] = p2; |
| 2933 | |
| 2934 | p1 = e[0]*B[1] - e[1]*B[0]; |
| 2935 | p0 = -e[0]*B[0] - e[1]*B[1]; |
| 2936 | |
| 2937 | d0[3] = p1; |
| 2938 | d1[0] = - p1; |
| 2939 | d2[3] = p0; |
| 2940 | d3[0] = p0; |
| 2941 | |
| 2942 | B -= 8; |
| 2943 | e -= 8; |
| 2944 | d0 += 4; |
| 2945 | d2 += 4; |
| 2946 | d1 -= 4; |
| 2947 | d3 -= 4; |
| 2948 | } |
| 2949 | } |
| 2950 | |
| 2951 | temp_alloc_restore(f,save_point)((f)->temp_offset = (save_point)); |
| 2952 | } |
| 2953 | |
| 2954 | #if 0 |
| 2955 | // this is the original version of the above code, if you want to optimize it from scratch |
| 2956 | void inverse_mdct_naive(float *buffer, int n) |
| 2957 | { |
| 2958 | float s; |
| 2959 | float A[1 << 12], B[1 << 12], C(2 | 4 | 1)[1 << 11]; |
| 2960 | int i,k,k2,k4, n2 = n >> 1, n4 = n >> 2, n8 = n >> 3, l; |
| 2961 | int n3_4 = n - n4, ld; |
| 2962 | // how can they claim this only uses N words?! |
| 2963 | // oh, because they're only used sparsely, whoops |
| 2964 | float u[1 << 13], X[1 << 13], v[1 << 13], w[1 << 13]; |
| 2965 | // set up twiddle factors |
| 2966 | |
| 2967 | for (k=k2=0; k < n4; ++k,k2+=2) { |
| 2968 | A[k2 ] = (float) cos(4*k*M_PI3.14159265358979323846/n); |
| 2969 | A[k2+1] = (float) -sin(4*k*M_PI3.14159265358979323846/n); |
| 2970 | B[k2 ] = (float) cos((k2+1)*M_PI3.14159265358979323846/n/2); |
| 2971 | B[k2+1] = (float) sin((k2+1)*M_PI3.14159265358979323846/n/2); |
| 2972 | } |
| 2973 | for (k=k2=0; k < n8; ++k,k2+=2) { |
| 2974 | C(2 | 4 | 1)[k2 ] = (float) cos(2*(k2+1)*M_PI3.14159265358979323846/n); |
| 2975 | C(2 | 4 | 1)[k2+1] = (float) -sin(2*(k2+1)*M_PI3.14159265358979323846/n); |
| 2976 | } |
| 2977 | |
| 2978 | // IMDCT algorithm from "The use of multirate filter banks for coding of high quality digital audio" |
| 2979 | // Note there are bugs in that pseudocode, presumably due to them attempting |
| 2980 | // to rename the arrays nicely rather than representing the way their actual |
| 2981 | // implementation bounces buffers back and forth. As a result, even in the |
| 2982 | // "some formulars corrected" version, a direct implementation fails. These |
| 2983 | // are noted below as "paper bug". |
| 2984 | |
| 2985 | // copy and reflect spectral data |
| 2986 | for (k=0; k < n2; ++k) u[k] = buffer[k]; |
| 2987 | for ( ; k < n ; ++k) u[k] = -buffer[n - k - 1]; |
| 2988 | // kernel from paper |
| 2989 | // step 1 |
| 2990 | for (k=k2=k4=0; k < n4; k+=1, k2+=2, k4+=4) { |
| 2991 | v[n-k4-1] = (u[k4] - u[n-k4-1]) * A[k2] - (u[k4+2] - u[n-k4-3])*A[k2+1]; |
| 2992 | v[n-k4-3] = (u[k4] - u[n-k4-1]) * A[k2+1] + (u[k4+2] - u[n-k4-3])*A[k2]; |
| 2993 | } |
| 2994 | // step 2 |
| 2995 | for (k=k4=0; k < n8; k+=1, k4+=4) { |
| 2996 | w[n2+3+k4] = v[n2+3+k4] + v[k4+3]; |
| 2997 | w[n2+1+k4] = v[n2+1+k4] + v[k4+1]; |
| 2998 | w[k4+3] = (v[n2+3+k4] - v[k4+3])*A[n2-4-k4] - (v[n2+1+k4]-v[k4+1])*A[n2-3-k4]; |
| 2999 | w[k4+1] = (v[n2+1+k4] - v[k4+1])*A[n2-4-k4] + (v[n2+3+k4]-v[k4+3])*A[n2-3-k4]; |
| 3000 | } |
| 3001 | // step 3 |
| 3002 | ld = ilog(n) - 1; // ilog is off-by-one from normal definitions |
| 3003 | for (l=0; l < ld-3; ++l) { |
| 3004 | int k0 = n >> (l+2), k1 = 1 << (l+3); |
| 3005 | int rlim = n >> (l+4), r4, r; |
| 3006 | int s2lim = 1 << (l+2), s2; |
| 3007 | for (r=r4=0; r < rlim; r4+=4,++r) { |
| 3008 | for (s2=0; s2 < s2lim; s2+=2) { |
| 3009 | u[n-1-k0*s2-r4] = w[n-1-k0*s2-r4] + w[n-1-k0*(s2+1)-r4]; |
| 3010 | u[n-3-k0*s2-r4] = w[n-3-k0*s2-r4] + w[n-3-k0*(s2+1)-r4]; |
| 3011 | u[n-1-k0*(s2+1)-r4] = (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1] |
| 3012 | - (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1+1]; |
| 3013 | u[n-3-k0*(s2+1)-r4] = (w[n-3-k0*s2-r4] - w[n-3-k0*(s2+1)-r4]) * A[r*k1] |
| 3014 | + (w[n-1-k0*s2-r4] - w[n-1-k0*(s2+1)-r4]) * A[r*k1+1]; |
| 3015 | } |
| 3016 | } |
| 3017 | if (l+1 < ld-3) { |
| 3018 | // paper bug: ping-ponging of u&w here is omitted |
| 3019 | memcpy(w, u, sizeof(u)); |
| 3020 | } |
| 3021 | } |
| 3022 | |
| 3023 | // step 4 |
| 3024 | for (i=0; i < n8; ++i) { |
| 3025 | int j = bit_reverse(i) >> (32-ld+3); |
| 3026 | assert(j < n8)((j < n8) ? (void)0 : _assert("j < n8", "src/siege/internal/stb/stb_vorbis.c" , 3026)); |
| 3027 | if (i == j) { |
| 3028 | // paper bug: original code probably swapped in place; if copying, |
| 3029 | // need to directly copy in this case |
| 3030 | int i8 = i << 3; |
| 3031 | v[i8+1] = u[i8+1]; |
| 3032 | v[i8+3] = u[i8+3]; |
| 3033 | v[i8+5] = u[i8+5]; |
| 3034 | v[i8+7] = u[i8+7]; |
| 3035 | } else if (i < j) { |
| 3036 | int i8 = i << 3, j8 = j << 3; |
| 3037 | v[j8+1] = u[i8+1], v[i8+1] = u[j8 + 1]; |
| 3038 | v[j8+3] = u[i8+3], v[i8+3] = u[j8 + 3]; |
| 3039 | v[j8+5] = u[i8+5], v[i8+5] = u[j8 + 5]; |
| 3040 | v[j8+7] = u[i8+7], v[i8+7] = u[j8 + 7]; |
| 3041 | } |
| 3042 | } |
| 3043 | // step 5 |
| 3044 | for (k=0; k < n2; ++k) { |
| 3045 | w[k] = v[k*2+1]; |
| 3046 | } |
| 3047 | // step 6 |
| 3048 | for (k=k2=k4=0; k < n8; ++k, k2 += 2, k4 += 4) { |
| 3049 | u[n-1-k2] = w[k4]; |
| 3050 | u[n-2-k2] = w[k4+1]; |
| 3051 | u[n3_4 - 1 - k2] = w[k4+2]; |
| 3052 | u[n3_4 - 2 - k2] = w[k4+3]; |
| 3053 | } |
| 3054 | // step 7 |
| 3055 | for (k=k2=0; k < n8; ++k, k2 += 2) { |
| 3056 | v[n2 + k2 ] = ( u[n2 + k2] + u[n-2-k2] + C(2 | 4 | 1)[k2+1]*(u[n2+k2]-u[n-2-k2]) + C(2 | 4 | 1)[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2; |
| 3057 | v[n-2 - k2] = ( u[n2 + k2] + u[n-2-k2] - C(2 | 4 | 1)[k2+1]*(u[n2+k2]-u[n-2-k2]) - C(2 | 4 | 1)[k2]*(u[n2+k2+1]+u[n-2-k2+1]))/2; |
| 3058 | v[n2+1+ k2] = ( u[n2+1+k2] - u[n-1-k2] + C(2 | 4 | 1)[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C(2 | 4 | 1)[k2]*(u[n2+k2]-u[n-2-k2]))/2; |
| 3059 | v[n-1 - k2] = (-u[n2+1+k2] + u[n-1-k2] + C(2 | 4 | 1)[k2+1]*(u[n2+1+k2]+u[n-1-k2]) - C(2 | 4 | 1)[k2]*(u[n2+k2]-u[n-2-k2]))/2; |
| 3060 | } |
| 3061 | // step 8 |
| 3062 | for (k=k2=0; k < n4; ++k,k2 += 2) { |
| 3063 | X[k] = v[k2+n2]*B[k2 ] + v[k2+1+n2]*B[k2+1]; |
| 3064 | X[n2-1-k] = v[k2+n2]*B[k2+1] - v[k2+1+n2]*B[k2 ]; |
| 3065 | } |
| 3066 | |
| 3067 | // decode kernel to output |
| 3068 | // determined the following value experimentally |
| 3069 | // (by first figuring out what made inverse_mdct_slow work); then matching that here |
| 3070 | // (probably vorbis encoder premultiplies by n or n/2, to save it on the decoder?) |
| 3071 | s = 0.5; // theoretically would be n4 |
| 3072 | |
| 3073 | // [[[ note! the s value of 0.5 is compensated for by the B[] in the current code, |
| 3074 | // so it needs to use the "old" B values to behave correctly, or else |
| 3075 | // set s to 1.0 ]]] |
| 3076 | for (i=0; i < n4 ; ++i) buffer[i] = s * X[i+n4]; |
| 3077 | for ( ; i < n3_4; ++i) buffer[i] = -s * X[n3_4 - i - 1]; |
| 3078 | for ( ; i < n ; ++i) buffer[i] = -s * X[i - n3_4]; |
| 3079 | } |
| 3080 | #endif |
| 3081 | |
| 3082 | static float *get_window(vorb *f, int len) |
| 3083 | { |
| 3084 | len <<= 1; |
| 3085 | if (len == f->blocksize_0) return f->window[0]; |
| 3086 | if (len == f->blocksize_1) return f->window[1]; |
| 3087 | assert(0)((0) ? (void)0 : _assert("0", "src/siege/internal/stb/stb_vorbis.c" , 3087)); |
| 3088 | return NULL((void*)0); |
| 3089 | } |
| 3090 | |
| 3091 | #ifndef STB_VORBIS_NO_DEFER_FLOOR |
| 3092 | typedef int16 YTYPE; |
| 3093 | #else |
| 3094 | typedef int YTYPE; |
| 3095 | #endif |
| 3096 | static int do_floor(vorb *f, Mapping *map, int i, int n, float *target, YTYPE *finalY, uint8 *step2_flag) |
| 3097 | { |
| 3098 | int n2 = n >> 1; |
| 3099 | int s = map->chan[i].mux, floor; |
| 3100 | floor = map->submap_floor[s]; |
| 3101 | if (f->floor_types[floor] == 0) { |
| 3102 | return error(f, VORBIS_invalid_stream); |
| 3103 | } else { |
| 3104 | Floor1 *g = &f->floor_config[floor].floor1; |
| 3105 | int j,q; |
| 3106 | int lx = 0, ly = finalY[0] * g->floor1_multiplier; |
| 3107 | for (q=1; q < g->values; ++q) { |
| 3108 | j = g->sorted_order[q]; |
| 3109 | #ifndef STB_VORBIS_NO_DEFER_FLOOR |
| 3110 | if (finalY[j] >= 0) |
| 3111 | #else |
| 3112 | if (step2_flag[j]) |
| 3113 | #endif |
| 3114 | { |
| 3115 | int hy = finalY[j] * g->floor1_multiplier; |
| 3116 | int hx = g->Xlist[j]; |
| 3117 | draw_line(target, lx,ly, hx,hy, n2); |
| 3118 | lx = hx, ly = hy; |
| 3119 | } |
| 3120 | } |
| 3121 | if (lx < n2) |
| 3122 | // optimization of: draw_line(target, lx,ly, n,ly, n2); |
| 3123 | for (j=lx; j < n2; ++j) |
| 3124 | LINE_OP(target[j], inverse_db_table[ly])target[j] *= inverse_db_table[ly]; |
| 3125 | } |
| 3126 | return TRUE1; |
| 3127 | } |
| 3128 | |
| 3129 | static int vorbis_decode_initial(vorb *f, int *p_left_start, int *p_left_end, int *p_right_start, int *p_right_end, int *mode) |
| 3130 | { |
| 3131 | Mode *m; |
| 3132 | int i, n, prev, next, window_center; |
| 3133 | f->channel_buffer_start = f->channel_buffer_end = 0; |
| 3134 | |
| 3135 | retry: |
| 3136 | if (f->eof) return FALSE0; |
| 3137 | if (!maybe_start_packet(f)) |
| 3138 | return FALSE0; |
| 3139 | // check packet type |
| 3140 | if (get_bits(f,1) != 0) { |
| 3141 | if (IS_PUSH_MODE(f)((f)->push_mode)) |
| 3142 | return error(f,VORBIS_bad_packet_type); |
| 3143 | while (EOP(-1) != get8_packet(f)); |
| 3144 | goto retry; |
| 3145 | } |
| 3146 | |
| 3147 | if (f->alloc.alloc_buffer) |
| 3148 | assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset)((f->alloc.alloc_buffer_length_in_bytes == f->temp_offset ) ? (void)0 : _assert("f->alloc.alloc_buffer_length_in_bytes == f->temp_offset" , "src/siege/internal/stb/stb_vorbis.c", 3148)); |
| 3149 | |
| 3150 | i = get_bits(f, ilog(f->mode_count-1)); |
| 3151 | if (i == EOP(-1)) return FALSE0; |
| 3152 | if (i >= f->mode_count) return FALSE0; |
| 3153 | *mode = i; |
| 3154 | m = f->mode_config + i; |
| 3155 | if (m->blockflag) { |
| 3156 | n = f->blocksize_1; |
| 3157 | prev = get_bits(f,1); |
| 3158 | next = get_bits(f,1); |
| 3159 | } else { |
| 3160 | prev = next = 0; |
| 3161 | n = f->blocksize_0; |
| 3162 | } |
| 3163 | |
| 3164 | // WINDOWING |
| 3165 | |
| 3166 | window_center = n >> 1; |
| 3167 | if (m->blockflag && !prev) { |
| 3168 | *p_left_start = (n - f->blocksize_0) >> 2; |
| 3169 | *p_left_end = (n + f->blocksize_0) >> 2; |
| 3170 | } else { |
| 3171 | *p_left_start = 0; |
| 3172 | *p_left_end = window_center; |
| 3173 | } |
| 3174 | if (m->blockflag && !next) { |
| 3175 | *p_right_start = (n*3 - f->blocksize_0) >> 2; |
| 3176 | *p_right_end = (n*3 + f->blocksize_0) >> 2; |
| 3177 | } else { |
| 3178 | *p_right_start = window_center; |
| 3179 | *p_right_end = n; |
| 3180 | } |
| 3181 | return TRUE1; |
| 3182 | } |
| 3183 | |
| 3184 | static int vorbis_decode_packet_rest(vorb *f, int *len, Mode *m, int left_start, int left_end, int right_start, int right_end, int *p_left) |
| 3185 | { |
| 3186 | Mapping *map; |
| 3187 | int i,j,k,n,n2; |
| 3188 | int zero_channel[256]; |
| 3189 | int really_zero_channel[256]; |
| 3190 | int window_center; |
| 3191 | |
| 3192 | // WINDOWING |
| 3193 | |
| 3194 | n = f->blocksize[m->blockflag]; |
| 3195 | window_center = n >> 1; |
| 3196 | (void)window_center; |
| 3197 | |
| 3198 | map = &f->mapping[m->mapping]; |
| 3199 | |
| 3200 | // FLOORS |
| 3201 | n2 = n >> 1; |
| 3202 | |
| 3203 | stb_prof(1)(void)0; |
| 3204 | for (i=0; i < f->channels; ++i) { |
| 3205 | int s = map->chan[i].mux, floor; |
| 3206 | zero_channel[i] = FALSE0; |
| 3207 | floor = map->submap_floor[s]; |
| 3208 | if (f->floor_types[floor] == 0) { |
| 3209 | return error(f, VORBIS_invalid_stream); |
| 3210 | } else { |
| 3211 | Floor1 *g = &f->floor_config[floor].floor1; |
| 3212 | if (get_bits(f, 1)) { |
| 3213 | short *finalY; |
| 3214 | uint8 step2_flag[256]; |
| 3215 | static int range_list[4] = { 256, 128, 86, 64 }; |
| 3216 | int range = range_list[g->floor1_multiplier-1]; |
| 3217 | int offset = 2; |
| 3218 | finalY = f->finalY[i]; |
| 3219 | finalY[0] = get_bits(f, ilog(range)-1); |
| 3220 | finalY[1] = get_bits(f, ilog(range)-1); |
| 3221 | for (j=0; j < g->partitions; ++j) { |
| 3222 | int pclass = g->partition_class_list[j]; |
| 3223 | int cdim = g->class_dimensions[pclass]; |
| 3224 | int cbits = g->class_subclasses[pclass]; |
| 3225 | int csub = (1 << cbits)-1; |
| 3226 | int cval = 0; |
| 3227 | if (cbits) { |
| 3228 | Codebook *c = f->codebooks + g->class_masterbooks[pclass]; |
| 3229 | DECODE(cval,f,c)if (f->valid_bits < 10) prep_huffman(f); cval = f->acc & ((1 << 10) - 1); cval = c->fast_huffman[cval] ; if (cval >= 0) { int n = c->codeword_lengths[cval]; f ->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; cval = -1; } } else { cval = codebook_decode_scalar_raw(f,c); } if (c->sparse) cval = c ->sorted_values[cval];; |
| 3230 | } |
| 3231 | for (k=0; k < cdim; ++k) { |
| 3232 | int book = g->subclass_books[pclass][cval & csub]; |
| 3233 | cval = cval >> cbits; |
| 3234 | if (book >= 0) { |
| 3235 | int temp; |
| 3236 | Codebook *c = f->codebooks + book; |
| 3237 | DECODE(temp,f,c)if (f->valid_bits < 10) prep_huffman(f); temp = f->acc & ((1 << 10) - 1); temp = c->fast_huffman[temp] ; if (temp >= 0) { int n = c->codeword_lengths[temp]; f ->acc >>= n; f->valid_bits -= n; if (f->valid_bits < 0) { f->valid_bits = 0; temp = -1; } } else { temp = codebook_decode_scalar_raw(f,c); } if (c->sparse) temp = c ->sorted_values[temp];; |
| 3238 | finalY[offset++] = temp; |
| 3239 | } else |
| 3240 | finalY[offset++] = 0; |
| 3241 | } |
| 3242 | } |
| 3243 | if (f->valid_bits == INVALID_BITS(-1)) goto error; // behavior according to spec |
| 3244 | step2_flag[0] = step2_flag[1] = 1; |
| 3245 | for (j=2; j < g->values; ++j) { |
| 3246 | int low, high, pred, highroom, lowroom, room, val; |
| 3247 | low = g->neighbors[j][0]; |
| 3248 | high = g->neighbors[j][1]; |
| 3249 | //neighbors(g->Xlist, j, &low, &high); |
| 3250 | pred = predict_point(g->Xlist[j], g->Xlist[low], g->Xlist[high], finalY[low], finalY[high]); |
| 3251 | val = finalY[j]; |
| 3252 | highroom = range - pred; |
| 3253 | lowroom = pred; |
| 3254 | if (highroom < lowroom) |
| 3255 | room = highroom * 2; |
| 3256 | else |
| 3257 | room = lowroom * 2; |
| 3258 | if (val) { |
| 3259 | step2_flag[low] = step2_flag[high] = 1; |
| 3260 | step2_flag[j] = 1; |
| 3261 | if (val >= room) |
| 3262 | if (highroom > lowroom) |
| 3263 | finalY[j] = val - lowroom + pred; |
| 3264 | else |
| 3265 | finalY[j] = pred - val + highroom - 1; |
| 3266 | else |
| 3267 | if (val & 1) |
| 3268 | finalY[j] = pred - ((val+1)>>1); |
| 3269 | else |
| 3270 | finalY[j] = pred + (val>>1); |
| 3271 | } else { |
| 3272 | step2_flag[j] = 0; |
| 3273 | finalY[j] = pred; |
| 3274 | } |
| 3275 | } |
| 3276 | |
| 3277 | #ifdef STB_VORBIS_NO_DEFER_FLOOR |
| 3278 | do_floor(f, map, i, n, f->floor_buffers[i], finalY, step2_flag); |
| 3279 | #else |
| 3280 | // defer final floor computation until _after_ residue |
| 3281 | for (j=0; j < g->values; ++j) { |
| 3282 | if (!step2_flag[j]) |
| 3283 | finalY[j] = -1; |
| 3284 | } |
| 3285 | #endif |
| 3286 | } else { |
| 3287 | error: |
| 3288 | zero_channel[i] = TRUE1; |
| 3289 | } |
| 3290 | // So we just defer everything else to later |
| 3291 | |
| 3292 | // at this point we've decoded the floor into buffer |
| 3293 | } |
| 3294 | } |
| 3295 | stb_prof(0)(void)0; |
| 3296 | // at this point we've decoded all floors |
| 3297 | |
| 3298 | if (f->alloc.alloc_buffer) |
| 3299 | assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset)((f->alloc.alloc_buffer_length_in_bytes == f->temp_offset ) ? (void)0 : _assert("f->alloc.alloc_buffer_length_in_bytes == f->temp_offset" , "src/siege/internal/stb/stb_vorbis.c", 3299)); |
| 3300 | |
| 3301 | // re-enable coupled channels if necessary |
| 3302 | memcpy(really_zero_channel, zero_channel, sizeof(really_zero_channel[0]) * f->channels); |
| 3303 | for (i=0; i < map->coupling_steps; ++i) |
| 3304 | if (!zero_channel[map->chan[i].magnitude] || !zero_channel[map->chan[i].angle]) { |
| 3305 | zero_channel[map->chan[i].magnitude] = zero_channel[map->chan[i].angle] = FALSE0; |
| 3306 | } |
| 3307 | |
| 3308 | // RESIDUE DECODE |
| 3309 | for (i=0; i < map->submaps; ++i) { |
| 3310 | float *residue_buffers[STB_VORBIS_MAX_CHANNELS16]; |
| 3311 | int r,t; |
| 3312 | uint8 do_not_decode[256]; |
| 3313 | int ch = 0; |
| 3314 | for (j=0; j < f->channels; ++j) { |
| 3315 | if (map->chan[j].mux == i) { |
| 3316 | if (zero_channel[j]) { |
| 3317 | do_not_decode[ch] = TRUE1; |
| 3318 | residue_buffers[ch] = NULL((void*)0); |
| 3319 | } else { |
| 3320 | do_not_decode[ch] = FALSE0; |
| 3321 | residue_buffers[ch] = f->channel_buffers[j]; |
| 3322 | } |
| 3323 | ++ch; |
| 3324 | } |
| 3325 | } |
| 3326 | r = map->submap_residue[i]; |
| 3327 | t = f->residue_types[r]; |
| 3328 | (void)t; |
| 3329 | decode_residue(f, residue_buffers, ch, n2, r, do_not_decode); |
| 3330 | } |
| 3331 | |
| 3332 | if (f->alloc.alloc_buffer) |
| 3333 | assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset)((f->alloc.alloc_buffer_length_in_bytes == f->temp_offset ) ? (void)0 : _assert("f->alloc.alloc_buffer_length_in_bytes == f->temp_offset" , "src/siege/internal/stb/stb_vorbis.c", 3333)); |
| 3334 | |
| 3335 | // INVERSE COUPLING |
| 3336 | stb_prof(14)(void)0; |
| 3337 | for (i = map->coupling_steps-1; i >= 0; --i) { |
| 3338 | int n2 = n >> 1; |
| 3339 | float *m = f->channel_buffers[map->chan[i].magnitude]; |
| 3340 | float *a = f->channel_buffers[map->chan[i].angle ]; |
| 3341 | for (j=0; j < n2; ++j) { |
| 3342 | float a2,m2; |
| 3343 | if (m[j] > 0) |
| 3344 | if (a[j] > 0) |
| 3345 | m2 = m[j], a2 = m[j] - a[j]; |
| 3346 | else |
| 3347 | a2 = m[j], m2 = m[j] + a[j]; |
| 3348 | else |
| 3349 | if (a[j] > 0) |
| 3350 | m2 = m[j], a2 = m[j] + a[j]; |
| 3351 | else |
| 3352 | a2 = m[j], m2 = m[j] - a[j]; |
| 3353 | m[j] = m2; |
| 3354 | a[j] = a2; |
| 3355 | } |
| 3356 | } |
| 3357 | |
| 3358 | // finish decoding the floors |
| 3359 | #ifndef STB_VORBIS_NO_DEFER_FLOOR |
| 3360 | stb_prof(15)(void)0; |
| 3361 | for (i=0; i < f->channels; ++i) { |
| 3362 | if (really_zero_channel[i]) { |
| 3363 | memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2); |
| 3364 | } else { |
| 3365 | do_floor(f, map, i, n, f->channel_buffers[i], f->finalY[i], NULL((void*)0)); |
| 3366 | } |
| 3367 | } |
| 3368 | #else |
| 3369 | for (i=0; i < f->channels; ++i) { |
| 3370 | if (really_zero_channel[i]) { |
| 3371 | memset(f->channel_buffers[i], 0, sizeof(*f->channel_buffers[i]) * n2); |
| 3372 | } else { |
| 3373 | for (j=0; j < n2; ++j) |
| 3374 | f->channel_buffers[i][j] *= f->floor_buffers[i][j]; |
| 3375 | } |
| 3376 | } |
| 3377 | #endif |
| 3378 | |
| 3379 | // INVERSE MDCT |
| 3380 | stb_prof(16)(void)0; |
| 3381 | for (i=0; i < f->channels; ++i) |
| 3382 | inverse_mdct(f->channel_buffers[i], n, f, m->blockflag); |
| 3383 | stb_prof(0)(void)0; |
| 3384 | |
| 3385 | // this shouldn't be necessary, unless we exited on an error |
| 3386 | // and want to flush to get to the next packet |
| 3387 | flush_packet(f); |
| 3388 | |
| 3389 | if (f->first_decode) { |
| 3390 | // assume we start so first non-discarded sample is sample 0 |
| 3391 | // this isn't to spec, but spec would require us to read ahead |
| 3392 | // and decode the size of all current frames--could be done, |
| 3393 | // but presumably it's not a commonly used feature |
| 3394 | f->current_loc = -n2; // start of first frame is positioned for discard |
| 3395 | // we might have to discard samples "from" the next frame too, |
| 3396 | // if we're lapping a large block then a small at the start? |
| 3397 | f->discard_samples_deferred = n - right_end; |
| 3398 | f->current_loc_valid = TRUE1; |
| 3399 | f->first_decode = FALSE0; |
| 3400 | } else if (f->discard_samples_deferred) { |
| 3401 | left_start += f->discard_samples_deferred; |
| 3402 | *p_left = left_start; |
| 3403 | f->discard_samples_deferred = 0; |
| 3404 | } else if (f->previous_length == 0 && f->current_loc_valid) { |
| 3405 | // we're recovering from a seek... that means we're going to discard |
| 3406 | // the samples from this packet even though we know our position from |
| 3407 | // the last page header, so we need to update the position based on |
| 3408 | // the discarded samples here |
| 3409 | // but wait, the code below is going to add this in itself even |
| 3410 | // on a discard, so we don't need to do it here... |
| 3411 | } |
| 3412 | |
| 3413 | // check if we have ogg information about the sample # for this packet |
| 3414 | if (f->last_seg_which == f->end_seg_with_known_loc) { |
| 3415 | // if we have a valid current loc, and this is final: |
| 3416 | if (f->current_loc_valid && (f->page_flag & PAGEFLAG_last_page4)) { |
| 3417 | uint32 current_end = f->known_loc_for_packet - (n-right_end); |
| 3418 | // then let's infer the size of the (probably) short final frame |
| 3419 | if (current_end < f->current_loc + right_end) { |
| 3420 | if (current_end < f->current_loc) { |
| 3421 | // negative truncation, that's impossible! |
| 3422 | *len = 0; |
| 3423 | } else { |
| 3424 | *len = current_end - f->current_loc; |
| 3425 | } |
| 3426 | *len += left_start; |
| 3427 | f->current_loc += *len; |
| 3428 | return TRUE1; |
| 3429 | } |
| 3430 | } |
| 3431 | // otherwise, just set our sample loc |
| 3432 | // guess that the ogg granule pos refers to the _middle_ of the |
| 3433 | // last frame? |
| 3434 | // set f->current_loc to the position of left_start |
| 3435 | f->current_loc = f->known_loc_for_packet - (n2-left_start); |
| 3436 | f->current_loc_valid = TRUE1; |
| 3437 | } |
| 3438 | if (f->current_loc_valid) |
| 3439 | f->current_loc += (right_start - left_start); |
| 3440 | |
| 3441 | if (f->alloc.alloc_buffer) |
| 3442 | assert(f->alloc.alloc_buffer_length_in_bytes == f->temp_offset)((f->alloc.alloc_buffer_length_in_bytes == f->temp_offset ) ? (void)0 : _assert("f->alloc.alloc_buffer_length_in_bytes == f->temp_offset" , "src/siege/internal/stb/stb_vorbis.c", 3442)); |
| 3443 | *len = right_end; // ignore samples after the window goes to 0 |
| 3444 | return TRUE1; |
| 3445 | } |
| 3446 | |
| 3447 | static int vorbis_decode_packet(vorb *f, int *len, int *p_left, int *p_right) |
| 3448 | { |
| 3449 | int mode, left_end, right_end; |
| 3450 | if (!vorbis_decode_initial(f, p_left, &left_end, p_right, &right_end, &mode)) return 0; |
| 3451 | return vorbis_decode_packet_rest(f, len, f->mode_config + mode, *p_left, left_end, *p_right, right_end, p_left); |
| 3452 | } |
| 3453 | |
| 3454 | static int vorbis_finish_frame(stb_vorbis *f, int len, int left, int right) |
| 3455 | { |
| 3456 | int prev,i,j; |
| 3457 | // we use right&left (the start of the right- and left-window sin()-regions) |
| 3458 | // to determine how much to return, rather than inferring from the rules |
| 3459 | // (same result, clearer code); 'left' indicates where our sin() window |
| 3460 | // starts, therefore where the previous window's right edge starts, and |
| 3461 | // therefore where to start mixing from the previous buffer. 'right' |
| 3462 | // indicates where our sin() ending-window starts, therefore that's where |
| 3463 | // we start saving, and where our returned-data ends. |
| 3464 | |
| 3465 | // mixin from previous window |
| 3466 | if (f->previous_length) { |
| 3467 | int i,j, n = f->previous_length; |
| 3468 | float *w = get_window(f, n); |
| 3469 | for (i=0; i < f->channels; ++i) { |
| 3470 | for (j=0; j < n; ++j) |
| 3471 | f->channel_buffers[i][left+j] = |
| 3472 | f->channel_buffers[i][left+j]*w[ j] + |
| 3473 | f->previous_window[i][ j]*w[n-1-j]; |
| 3474 | } |
| 3475 | } |
| 3476 | |
| 3477 | prev = f->previous_length; |
| 3478 | |
| 3479 | // last half of this data becomes previous window |
| 3480 | f->previous_length = len - right; |
| 3481 | |
| 3482 | // @OPTIMIZE: could avoid this copy by double-buffering the |
| 3483 | // output (flipping previous_window with channel_buffers), but |
| 3484 | // then previous_window would have to be 2x as large, and |
| 3485 | // channel_buffers couldn't be temp mem (although they're NOT |
| 3486 | // currently temp mem, they could be (unless we want to level |
| 3487 | // performance by spreading out the computation)) |
| 3488 | for (i=0; i < f->channels; ++i) |
| 3489 | for (j=0; right+j < len; ++j) |
| 3490 | f->previous_window[i][j] = f->channel_buffers[i][right+j]; |
| 3491 | |
| 3492 | if (!prev) |
| 3493 | // there was no previous packet, so this data isn't valid... |
| 3494 | // this isn't entirely true, only the would-have-overlapped data |
| 3495 | // isn't valid, but this seems to be what the spec requires |
| 3496 | return 0; |
| 3497 | |
| 3498 | // truncate a short frame |
| 3499 | if (len < right) right = len; |
| 3500 | |
| 3501 | f->samples_output += right-left; |
| 3502 | |
| 3503 | return right - left; |
| 3504 | } |
| 3505 | |
| 3506 | static void vorbis_pump_first_frame(stb_vorbis *f) |
| 3507 | { |
| 3508 | int len, right, left; |
| 3509 | if (vorbis_decode_packet(f, &len, &left, &right)) |
| 3510 | vorbis_finish_frame(f, len, left, right); |
| 3511 | } |
| 3512 | |
| 3513 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 3514 | static int is_whole_packet_present(stb_vorbis *f, int end_page) |
| 3515 | { |
| 3516 | // make sure that we have the packet available before continuing... |
| 3517 | // this requires a full ogg parse, but we know we can fetch from f->stream |
| 3518 | |
| 3519 | // instead of coding this out explicitly, we could save the current read state, |
| 3520 | // read the next packet with get8() until end-of-packet, check f->eof, then |
| 3521 | // reset the state? but that would be slower, esp. since we'd have over 256 bytes |
| 3522 | // of state to restore (primarily the page segment table) |
| 3523 | |
| 3524 | int s = f->next_seg, first = TRUE1; |
| 3525 | uint8 *p = f->stream; |
| 3526 | |
| 3527 | if (s != -1) { // if we're not starting the packet with a 'continue on next page' flag |
| 3528 | for (; s < f->segment_count; ++s) { |
| 3529 | p += f->segments[s]; |
| 3530 | if (f->segments[s] < 255) // stop at first short segment |
| 3531 | break; |
| 3532 | } |
| 3533 | // either this continues, or it ends it... |
| 3534 | if (end_page) |
| 3535 | if (s < f->segment_count-1) return error(f, VORBIS_invalid_stream); |
| 3536 | if (s == f->segment_count) |
| 3537 | s = -1; // set 'crosses page' flag |
| 3538 | if (p > f->stream_end) return error(f, VORBIS_need_more_data); |
| 3539 | first = FALSE0; |
| 3540 | } |
| 3541 | for (; s == -1;) { |
| 3542 | uint8 *q; |
| 3543 | int n; |
| 3544 | |
| 3545 | // check that we have the page header ready |
| 3546 | if (p + 26 >= f->stream_end) return error(f, VORBIS_need_more_data); |
| 3547 | // validate the page |
| 3548 | if (memcmp(p, ogg_page_header, 4)) return error(f, VORBIS_invalid_stream); |
| 3549 | if (p[4] != 0) return error(f, VORBIS_invalid_stream); |
| 3550 | if (first) { // the first segment must NOT have 'continued_packet', later ones MUST |
| 3551 | if (f->previous_length) |
| 3552 | if ((p[5] & PAGEFLAG_continued_packet1)) return error(f, VORBIS_invalid_stream); |
| 3553 | // if no previous length, we're resynching, so we can come in on a continued-packet, |
| 3554 | // which we'll just drop |
| 3555 | } else { |
| 3556 | if (!(p[5] & PAGEFLAG_continued_packet1)) return error(f, VORBIS_invalid_stream); |
| 3557 | } |
| 3558 | n = p[26]; // segment counts |
| 3559 | q = p+27; // q points to segment table |
| 3560 | p = q + n; // advance past header |
| 3561 | // make sure we've read the segment table |
| 3562 | if (p > f->stream_end) return error(f, VORBIS_need_more_data); |
| 3563 | for (s=0; s < n; ++s) { |
| 3564 | p += q[s]; |
| 3565 | if (q[s] < 255) |
| 3566 | break; |
| 3567 | } |
| 3568 | if (end_page) |
| 3569 | if (s < n-1) return error(f, VORBIS_invalid_stream); |
| 3570 | if (s == f->segment_count) |
| 3571 | s = -1; // set 'crosses page' flag |
| 3572 | if (p > f->stream_end) return error(f, VORBIS_need_more_data); |
| 3573 | first = FALSE0; |
| 3574 | } |
| 3575 | return TRUE1; |
| 3576 | } |
| 3577 | #endif // !STB_VORBIS_NO_PUSHDATA_API |
| 3578 | |
| 3579 | static int start_decoder(vorb *f) |
| 3580 | { |
| 3581 | uint8 header[6], x,y; |
| 3582 | int len,i,j,k, max_submaps = 0; |
| 3583 | int longest_floorlist=0; |
| 3584 | |
| 3585 | // first page, first packet |
| 3586 | |
| 3587 | if (!start_page(f)) return FALSE0; |
| 3588 | // validate page flag |
| 3589 | if (!(f->page_flag & PAGEFLAG_first_page2)) return error(f, VORBIS_invalid_first_page); |
| 3590 | if (f->page_flag & PAGEFLAG_last_page4) return error(f, VORBIS_invalid_first_page); |
| 3591 | if (f->page_flag & PAGEFLAG_continued_packet1) return error(f, VORBIS_invalid_first_page); |
| 3592 | // check for expected packet length |
| 3593 | if (f->segment_count != 1) return error(f, VORBIS_invalid_first_page); |
| 3594 | if (f->segments[0] != 30) return error(f, VORBIS_invalid_first_page); |
| 3595 | // read packet |
| 3596 | // check packet header |
| 3597 | if (get8(f) != VORBIS_packet_id) return error(f, VORBIS_invalid_first_page); |
| 3598 | if (!getn(f, header, 6)) return error(f, VORBIS_unexpected_eof); |
| 3599 | if (!vorbis_validate(header)) return error(f, VORBIS_invalid_first_page); |
| 3600 | // vorbis_version |
| 3601 | if (get32(f) != 0) return error(f, VORBIS_invalid_first_page); |
| 3602 | f->channels = get8(f); if (!f->channels) return error(f, VORBIS_invalid_first_page); |
| 3603 | if (f->channels > STB_VORBIS_MAX_CHANNELS16) return error(f, VORBIS_too_many_channels); |
| 3604 | f->sample_rate = get32(f); if (!f->sample_rate) return error(f, VORBIS_invalid_first_page); |
| 3605 | get32(f); // bitrate_maximum |
| 3606 | get32(f); // bitrate_nominal |
| 3607 | get32(f); // bitrate_minimum |
| 3608 | x = get8(f); |
| 3609 | { int log0,log1; |
| 3610 | log0 = x & 15; |
| 3611 | log1 = x >> 4; |
| 3612 | f->blocksize_0 = 1 << log0; |
| 3613 | f->blocksize_1 = 1 << log1; |
| 3614 | if (log0 < 6 || log0 > 13) return error(f, VORBIS_invalid_setup); |
| 3615 | if (log1 < 6 || log1 > 13) return error(f, VORBIS_invalid_setup); |
| 3616 | if (log0 > log1) return error(f, VORBIS_invalid_setup); |
| 3617 | } |
| 3618 | |
| 3619 | // framing_flag |
| 3620 | x = get8(f); |
| 3621 | if (!(x & 1)) return error(f, VORBIS_invalid_first_page); |
| 3622 | |
| 3623 | // second packet! |
| 3624 | if (!start_page(f)) return FALSE0; |
| 3625 | |
| 3626 | if (!start_packet(f)) return FALSE0; |
| 3627 | do { |
| 3628 | len = next_segment(f); |
| 3629 | skip(f, len); |
| 3630 | f->bytes_in_seg = 0; |
| 3631 | } while (len); |
| 3632 | |
| 3633 | // third packet! |
| 3634 | if (!start_packet(f)) return FALSE0; |
| 3635 | |
| 3636 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 3637 | if (IS_PUSH_MODE(f)((f)->push_mode)) { |
| 3638 | if (!is_whole_packet_present(f, TRUE1)) { |
| 3639 | // convert error in ogg header to write type |
| 3640 | if (f->error == VORBIS_invalid_stream) |
| 3641 | f->error = VORBIS_invalid_setup; |
| 3642 | return FALSE0; |
| 3643 | } |
| 3644 | } |
| 3645 | #endif |
| 3646 | |
| 3647 | crc32_init(); // always init it, to avoid multithread race conditions |
| 3648 | |
| 3649 | if (get8_packet(f) != VORBIS_packet_setup) return error(f, VORBIS_invalid_setup); |
| 3650 | for (i=0; i < 6; ++i) header[i] = get8_packet(f); |
| 3651 | if (!vorbis_validate(header)) return error(f, VORBIS_invalid_setup); |
| 3652 | |
| 3653 | // codebooks |
| 3654 | |
| 3655 | f->codebook_count = get_bits(f,8) + 1; |
| 3656 | f->codebooks = (Codebook *) setup_malloc(f, sizeof(*f->codebooks) * f->codebook_count); |
| 3657 | if (f->codebooks == NULL((void*)0)) return error(f, VORBIS_outofmem); |
| 3658 | memset(f->codebooks, 0, sizeof(*f->codebooks) * f->codebook_count); |
| 3659 | for (i=0; i < f->codebook_count; ++i) { |
| 3660 | uint32 *values; |
| 3661 | int ordered, sorted_count; |
| 3662 | int total=0; |
| 3663 | uint8 *lengths; |
| 3664 | Codebook *c = f->codebooks+i; |
| 3665 | x = get_bits(f, 8); if (x != 0x42) return error(f, VORBIS_invalid_setup); |
| 3666 | x = get_bits(f, 8); if (x != 0x43) return error(f, VORBIS_invalid_setup); |
| 3667 | x = get_bits(f, 8); if (x != 0x56) return error(f, VORBIS_invalid_setup); |
| 3668 | x = get_bits(f, 8); |
| 3669 | c->dimensions = (get_bits(f, 8)<<8) + x; |
| 3670 | x = get_bits(f, 8); |
| 3671 | y = get_bits(f, 8); |
| 3672 | c->entries = (get_bits(f, 8)<<16) + (y<<8) + x; |
| 3673 | ordered = get_bits(f,1); |
| 3674 | c->sparse = ordered ? 0 : get_bits(f,1); |
| 3675 | |
| 3676 | if (c->sparse) |
| 3677 | lengths = (uint8 *) setup_temp_malloc(f, c->entries); |
| 3678 | else |
| 3679 | lengths = c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries); |
| 3680 | |
| 3681 | if (!lengths) return error(f, VORBIS_outofmem); |
| 3682 | |
| 3683 | if (ordered) { |
| 3684 | int current_entry = 0; |
| 3685 | int current_length = get_bits(f,5) + 1; |
| 3686 | while (current_entry < c->entries) { |
| 3687 | int limit = c->entries - current_entry; |
| 3688 | int n = get_bits(f, ilog(limit)); |
| 3689 | if (current_entry + n > (int) c->entries) { return error(f, VORBIS_invalid_setup); } |
| 3690 | memset(lengths + current_entry, current_length, n); |
| 3691 | current_entry += n; |
| 3692 | ++current_length; |
| 3693 | } |
| 3694 | } else { |
| 3695 | for (j=0; j < c->entries; ++j) { |
| 3696 | int present = c->sparse ? get_bits(f,1) : 1; |
| 3697 | if (present) { |
| 3698 | lengths[j] = get_bits(f, 5) + 1; |
| 3699 | ++total; |
| 3700 | } else { |
| 3701 | lengths[j] = NO_CODE255; |
| 3702 | } |
| 3703 | } |
| 3704 | } |
| 3705 | |
| 3706 | if (c->sparse && total >= c->entries >> 2) { |
| 3707 | // convert sparse items to non-sparse! |
| 3708 | if (c->entries > (int) f->setup_temp_memory_required) |
| 3709 | f->setup_temp_memory_required = c->entries; |
| 3710 | |
| 3711 | c->codeword_lengths = (uint8 *) setup_malloc(f, c->entries); |
| 3712 | memcpy(c->codeword_lengths, lengths, c->entries); |
| 3713 | setup_temp_free(f, lengths, c->entries); // note this is only safe if there have been no intervening temp mallocs! |
| 3714 | lengths = c->codeword_lengths; |
| 3715 | c->sparse = 0; |
| 3716 | } |
| 3717 | |
| 3718 | // compute the size of the sorted tables |
| 3719 | if (c->sparse) { |
| 3720 | sorted_count = total; |
| 3721 | //assert(total != 0); |
| 3722 | } else { |
| 3723 | sorted_count = 0; |
| 3724 | #ifndef STB_VORBIS_NO_HUFFMAN_BINARY_SEARCH |
| 3725 | for (j=0; j < c->entries; ++j) |
| 3726 | if (lengths[j] > STB_VORBIS_FAST_HUFFMAN_LENGTH10 && lengths[j] != NO_CODE255) |
| 3727 | ++sorted_count; |
| 3728 | #endif |
| 3729 | } |
| 3730 | |
| 3731 | c->sorted_entries = sorted_count; |
| 3732 | values = NULL((void*)0); |
| 3733 | |
| 3734 | if (!c->sparse) { |
| 3735 | c->codewords = (uint32 *) setup_malloc(f, sizeof(c->codewords[0]) * c->entries); |
| 3736 | if (!c->codewords) return error(f, VORBIS_outofmem); |
| 3737 | } else { |
| 3738 | unsigned int size; |
| 3739 | if (c->sorted_entries) { |
| 3740 | c->codeword_lengths = (uint8 *) setup_malloc(f, c->sorted_entries); |
| 3741 | if (!c->codeword_lengths) return error(f, VORBIS_outofmem); |
| 3742 | c->codewords = (uint32 *) setup_temp_malloc(f, sizeof(*c->codewords) * c->sorted_entries); |
| 3743 | if (!c->codewords) return error(f, VORBIS_outofmem); |
| 3744 | values = (uint32 *) setup_temp_malloc(f, sizeof(*values) * c->sorted_entries); |
| 3745 | if (!values) return error(f, VORBIS_outofmem); |
| 3746 | } |
| 3747 | size = c->entries + (sizeof(*c->codewords) + sizeof(*values)) * c->sorted_entries; |
| 3748 | if (size > f->setup_temp_memory_required) |
| 3749 | f->setup_temp_memory_required = size; |
| 3750 | } |
| 3751 | |
| 3752 | if (!compute_codewords(c, lengths, c->entries, values)) { |
| 3753 | if (c->sparse) setup_temp_free(f, values, 0); |
| 3754 | return error(f, VORBIS_invalid_setup); |
| 3755 | } |
| 3756 | |
| 3757 | if (c->sorted_entries) { |
| 3758 | // allocate an extra slot for sentinels |
| 3759 | c->sorted_codewords = (uint32 *) setup_malloc(f, sizeof(*c->sorted_codewords) * (c->sorted_entries+1)); |
| 3760 | // allocate an extra slot at the front so that c->sorted_values[-1] is defined |
| 3761 | // so that we can catch that case without an extra if |
| 3762 | c->sorted_values = ( int *) setup_malloc(f, sizeof(*c->sorted_values ) * (c->sorted_entries+1)); |
| 3763 | if (c->sorted_values) { ++c->sorted_values; c->sorted_values[-1] = -1; } |
| 3764 | compute_sorted_huffman(c, lengths, values); |
| 3765 | } |
| 3766 | |
| 3767 | if (c->sparse) { |
| 3768 | setup_temp_free(f, values, sizeof(*values)*c->sorted_entries); |
| 3769 | setup_temp_free(f, c->codewords, sizeof(*c->codewords)*c->sorted_entries); |
| 3770 | setup_temp_free(f, lengths, c->entries); |
| 3771 | c->codewords = NULL((void*)0); |
| 3772 | } |
| 3773 | |
| 3774 | compute_accelerated_huffman(c); |
| 3775 | |
| 3776 | c->lookup_type = get_bits(f, 4); |
| 3777 | if (c->lookup_type > 2) return error(f, VORBIS_invalid_setup); |
| 3778 | if (c->lookup_type > 0) { |
| 3779 | uint16 *mults; |
| 3780 | c->minimum_value = float32_unpack(get_bits(f, 32)); |
| 3781 | c->delta_value = float32_unpack(get_bits(f, 32)); |
| 3782 | c->value_bits = get_bits(f, 4)+1; |
| 3783 | c->sequence_p = get_bits(f,1); |
| 3784 | if (c->lookup_type == 1) { |
| 3785 | c->lookup_values = lookup1_values(c->entries, c->dimensions); |
| 3786 | } else { |
| 3787 | c->lookup_values = c->entries * c->dimensions; |
| 3788 | } |
| 3789 | mults = (uint16 *) setup_temp_malloc(f, sizeof(mults[0]) * c->lookup_values); |
| 3790 | if (mults == NULL((void*)0)) return error(f, VORBIS_outofmem); |
| 3791 | for (j=0; j < (int) c->lookup_values; ++j) { |
| 3792 | int q = get_bits(f, c->value_bits); |
| 3793 | if (q == EOP(-1)) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_invalid_setup); } |
| 3794 | mults[j] = q; |
| 3795 | } |
| 3796 | |
| 3797 | #ifndef STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 3798 | if (c->lookup_type == 1) { |
| 3799 | int len, sparse = c->sparse; |
| 3800 | // pre-expand the lookup1-style multiplicands, to avoid a divide in the inner loop |
| 3801 | if (sparse) { |
| 3802 | if (c->sorted_entries == 0) goto skip; |
| 3803 | c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->sorted_entries * c->dimensions); |
| 3804 | } else |
| 3805 | c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->entries * c->dimensions); |
| 3806 | if (c->multiplicands == NULL((void*)0)) { setup_temp_free(f,mults,sizeof(mults[0])*c->lookup_values); return error(f, VORBIS_outofmem); } |
| 3807 | len = sparse ? c->sorted_entries : c->entries; |
| 3808 | for (j=0; j < len; ++j) { |
| 3809 | int z = sparse ? c->sorted_values[j] : j, div=1; |
| 3810 | for (k=0; k < c->dimensions; ++k) { |
| 3811 | int off = (z / div) % c->lookup_values; |
| 3812 | c->multiplicands[j*c->dimensions + k] = |
| 3813 | #ifndef STB_VORBIS_CODEBOOK_FLOATS |
| 3814 | mults[off]; |
| 3815 | #else |
| 3816 | mults[off]*c->delta_value + c->minimum_value; |
| 3817 | // in this case (and this case only) we could pre-expand c->sequence_p, |
| 3818 | // and throw away the decode logic for it; have to ALSO do |
| 3819 | // it in the case below, but it can only be done if |
| 3820 | // STB_VORBIS_CODEBOOK_FLOATS |
| 3821 | // !STB_VORBIS_DIVIDES_IN_CODEBOOK |
| 3822 | #endif |
| 3823 | div *= c->lookup_values; |
| 3824 | } |
| 3825 | } |
| 3826 | setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values); |
| 3827 | c->lookup_type = 2; |
| 3828 | } |
| 3829 | else |
| 3830 | #endif |
| 3831 | { |
| 3832 | c->multiplicands = (codetype *) setup_malloc(f, sizeof(c->multiplicands[0]) * c->lookup_values); |
| 3833 | #ifndef STB_VORBIS_CODEBOOK_FLOATS |
| 3834 | memcpy(c->multiplicands, mults, sizeof(c->multiplicands[0]) * c->lookup_values); |
| 3835 | #else |
| 3836 | for (j=0; j < (int) c->lookup_values; ++j) |
| 3837 | c->multiplicands[j] = mults[j] * c->delta_value + c->minimum_value; |
| 3838 | setup_temp_free(f, mults,sizeof(mults[0])*c->lookup_values); |
| 3839 | #endif |
| 3840 | } |
| 3841 | skip:; |
| 3842 | |
| 3843 | #ifdef STB_VORBIS_CODEBOOK_FLOATS |
| 3844 | if (c->lookup_type == 2 && c->sequence_p) { |
| 3845 | for (j=1; j < (int) c->lookup_values; ++j) |
| 3846 | c->multiplicands[j] = c->multiplicands[j-1]; |
| 3847 | c->sequence_p = 0; |
| 3848 | } |
| 3849 | #endif |
| 3850 | } |
| 3851 | } |
| 3852 | |
| 3853 | // time domain transfers (notused) |
| 3854 | |
| 3855 | x = get_bits(f, 6) + 1; |
| 3856 | for (i=0; i < x; ++i) { |
| 3857 | uint32 z = get_bits(f, 16); |
| 3858 | if (z != 0) return error(f, VORBIS_invalid_setup); |
| 3859 | } |
| 3860 | |
| 3861 | // Floors |
| 3862 | f->floor_count = get_bits(f, 6)+1; |
| 3863 | f->floor_config = (Floor *) setup_malloc(f, f->floor_count * sizeof(*f->floor_config)); |
| 3864 | for (i=0; i < f->floor_count; ++i) { |
| 3865 | f->floor_types[i] = get_bits(f, 16); |
| 3866 | if (f->floor_types[i] > 1) return error(f, VORBIS_invalid_setup); |
| 3867 | if (f->floor_types[i] == 0) { |
| 3868 | Floor0 *g = &f->floor_config[i].floor0; |
| 3869 | g->order = get_bits(f,8); |
| 3870 | g->rate = get_bits(f,16); |
| 3871 | g->bark_map_size = get_bits(f,16); |
| 3872 | g->amplitude_bits = get_bits(f,6); |
| 3873 | g->amplitude_offset = get_bits(f,8); |
| 3874 | g->number_of_books = get_bits(f,4) + 1; |
| 3875 | for (j=0; j < g->number_of_books; ++j) |
| 3876 | g->book_list[j] = get_bits(f,8); |
| 3877 | return error(f, VORBIS_feature_not_supported); |
| 3878 | } else { |
| 3879 | Point p[31*8+2]; |
| 3880 | Floor1 *g = &f->floor_config[i].floor1; |
| 3881 | int max_class = -1; |
| 3882 | g->partitions = get_bits(f, 5); |
| 3883 | for (j=0; j < g->partitions; ++j) { |
| 3884 | g->partition_class_list[j] = get_bits(f, 4); |
| 3885 | if (g->partition_class_list[j] > max_class) |
| 3886 | max_class = g->partition_class_list[j]; |
| 3887 | } |
| 3888 | for (j=0; j <= max_class; ++j) { |
| 3889 | g->class_dimensions[j] = get_bits(f, 3)+1; |
| 3890 | g->class_subclasses[j] = get_bits(f, 2); |
| 3891 | if (g->class_subclasses[j]) { |
| 3892 | g->class_masterbooks[j] = get_bits(f, 8); |
| 3893 | if (g->class_masterbooks[j] >= f->codebook_count) return error(f, VORBIS_invalid_setup); |
| 3894 | } |
| 3895 | for (k=0; k < 1 << g->class_subclasses[j]; ++k) { |
| 3896 | g->subclass_books[j][k] = get_bits(f,8)-1; |
| 3897 | if (g->subclass_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup); |
| 3898 | } |
| 3899 | } |
| 3900 | g->floor1_multiplier = get_bits(f,2)+1; |
| 3901 | g->rangebits = get_bits(f,4); |
| 3902 | g->Xlist[0] = 0; |
| 3903 | g->Xlist[1] = 1 << g->rangebits; |
| 3904 | g->values = 2; |
| 3905 | for (j=0; j < g->partitions; ++j) { |
| 3906 | int c = g->partition_class_list[j]; |
| 3907 | for (k=0; k < g->class_dimensions[c]; ++k) { |
| 3908 | g->Xlist[g->values] = get_bits(f, g->rangebits); |
| 3909 | ++g->values; |
| 3910 | } |
| 3911 | } |
| 3912 | // precompute the sorting |
| 3913 | for (j=0; j < g->values; ++j) { |
| 3914 | p[j].x = g->Xlist[j]; |
| 3915 | p[j].y = j; |
| 3916 | } |
| 3917 | qsort(p, g->values, sizeof(p[0]), point_compare); |
| 3918 | for (j=0; j < g->values; ++j) |
| 3919 | g->sorted_order[j] = (uint8) p[j].y; |
| 3920 | // precompute the neighbors |
| 3921 | for (j=2; j < g->values; ++j) { |
| 3922 | int low,hi; |
| 3923 | neighbors(g->Xlist, j, &low,&hi); |
| 3924 | g->neighbors[j][0] = low; |
| 3925 | g->neighbors[j][1] = hi; |
| 3926 | } |
| 3927 | |
| 3928 | if (g->values > longest_floorlist) |
| 3929 | longest_floorlist = g->values; |
| 3930 | } |
| 3931 | } |
| 3932 | |
| 3933 | // Residue |
| 3934 | f->residue_count = get_bits(f, 6)+1; |
| 3935 | f->residue_config = (Residue *) setup_malloc(f, f->residue_count * sizeof(*f->residue_config)); |
| 3936 | for (i=0; i < f->residue_count; ++i) { |
| 3937 | uint8 residue_cascade[64]; |
| 3938 | Residue *r = f->residue_config+i; |
| 3939 | f->residue_types[i] = get_bits(f, 16); |
| 3940 | if (f->residue_types[i] > 2) return error(f, VORBIS_invalid_setup); |
| 3941 | r->begin = get_bits(f, 24); |
| 3942 | r->end = get_bits(f, 24); |
| 3943 | r->part_size = get_bits(f,24)+1; |
| 3944 | r->classifications = get_bits(f,6)+1; |
| 3945 | r->classbook = get_bits(f,8); |
| 3946 | for (j=0; j < r->classifications; ++j) { |
| 3947 | uint8 high_bits=0; |
| 3948 | uint8 low_bits=get_bits(f,3); |
| 3949 | if (get_bits(f,1)) |
| 3950 | high_bits = get_bits(f,5); |
| 3951 | residue_cascade[j] = high_bits*8 + low_bits; |
| 3952 | } |
| 3953 | r->residue_books = (short (*)[8]) setup_malloc(f, sizeof(r->residue_books[0]) * r->classifications); |
| 3954 | for (j=0; j < r->classifications; ++j) { |
| 3955 | for (k=0; k < 8; ++k) { |
| 3956 | if (residue_cascade[j] & (1 << k)) { |
| 3957 | r->residue_books[j][k] = get_bits(f, 8); |
| 3958 | if (r->residue_books[j][k] >= f->codebook_count) return error(f, VORBIS_invalid_setup); |
| 3959 | } else { |
| 3960 | r->residue_books[j][k] = -1; |
| 3961 | } |
| 3962 | } |
| 3963 | } |
| 3964 | // precompute the classifications[] array to avoid inner-loop mod/divide |
| 3965 | // call it 'classdata' since we already have r->classifications |
| 3966 | r->classdata = (uint8 **) setup_malloc(f, sizeof(*r->classdata) * f->codebooks[r->classbook].entries); |
| 3967 | if (!r->classdata) return error(f, VORBIS_outofmem); |
| 3968 | memset(r->classdata, 0, sizeof(*r->classdata) * f->codebooks[r->classbook].entries); |
| 3969 | for (j=0; j < f->codebooks[r->classbook].entries; ++j) { |
| 3970 | int classwords = f->codebooks[r->classbook].dimensions; |
| 3971 | int temp = j; |
| 3972 | r->classdata[j] = (uint8 *) setup_malloc(f, sizeof(r->classdata[j][0]) * classwords); |
| 3973 | for (k=classwords-1; k >= 0; --k) { |
| 3974 | r->classdata[j][k] = temp % r->classifications; |
| 3975 | temp /= r->classifications; |
| 3976 | } |
| 3977 | } |
| 3978 | } |
| 3979 | |
| 3980 | f->mapping_count = get_bits(f,6)+1; |
| 3981 | f->mapping = (Mapping *) setup_malloc(f, f->mapping_count * sizeof(*f->mapping)); |
| 3982 | for (i=0; i < f->mapping_count; ++i) { |
| 3983 | Mapping *m = f->mapping + i; |
| 3984 | int mapping_type = get_bits(f,16); |
| 3985 | if (mapping_type != 0) return error(f, VORBIS_invalid_setup); |
| 3986 | m->chan = (MappingChannel *) setup_malloc(f, f->channels * sizeof(*m->chan)); |
| 3987 | if (get_bits(f,1)) |
| 3988 | m->submaps = get_bits(f,4); |
| 3989 | else |
| 3990 | m->submaps = 1; |
| 3991 | if (m->submaps > max_submaps) |
| 3992 | max_submaps = m->submaps; |
| 3993 | if (get_bits(f,1)) { |
| 3994 | m->coupling_steps = get_bits(f,8)+1; |
| 3995 | for (k=0; k < m->coupling_steps; ++k) { |
| 3996 | m->chan[k].magnitude = get_bits(f, ilog(f->channels)-1); |
| 3997 | m->chan[k].angle = get_bits(f, ilog(f->channels)-1); |
| 3998 | if (m->chan[k].magnitude >= f->channels) return error(f, VORBIS_invalid_setup); |
| 3999 | if (m->chan[k].angle >= f->channels) return error(f, VORBIS_invalid_setup); |
| 4000 | if (m->chan[k].magnitude == m->chan[k].angle) return error(f, VORBIS_invalid_setup); |
| 4001 | } |
| 4002 | } else |
| 4003 | m->coupling_steps = 0; |
| 4004 | |
| 4005 | // reserved field |
| 4006 | if (get_bits(f,2)) return error(f, VORBIS_invalid_setup); |
| 4007 | if (m->submaps > 1) { |
| 4008 | for (j=0; j < f->channels; ++j) { |
| 4009 | m->chan[j].mux = get_bits(f, 4); |
| 4010 | if (m->chan[j].mux >= m->submaps) return error(f, VORBIS_invalid_setup); |
| 4011 | } |
| 4012 | } else |
| 4013 | // @SPECIFICATION: this case is missing from the spec |
| 4014 | for (j=0; j < f->channels; ++j) |
| 4015 | m->chan[j].mux = 0; |
| 4016 | |
| 4017 | for (j=0; j < m->submaps; ++j) { |
| 4018 | get_bits(f,8); // discard |
| 4019 | m->submap_floor[j] = get_bits(f,8); |
| 4020 | m->submap_residue[j] = get_bits(f,8); |
| 4021 | if (m->submap_floor[j] >= f->floor_count) return error(f, VORBIS_invalid_setup); |
| 4022 | if (m->submap_residue[j] >= f->residue_count) return error(f, VORBIS_invalid_setup); |
| 4023 | } |
| 4024 | } |
| 4025 | |
| 4026 | // Modes |
| 4027 | f->mode_count = get_bits(f, 6)+1; |
| 4028 | for (i=0; i < f->mode_count; ++i) { |
| 4029 | Mode *m = f->mode_config+i; |
| 4030 | m->blockflag = get_bits(f,1); |
| 4031 | m->windowtype = get_bits(f,16); |
| 4032 | m->transformtype = get_bits(f,16); |
| 4033 | m->mapping = get_bits(f,8); |
| 4034 | if (m->windowtype != 0) return error(f, VORBIS_invalid_setup); |
| 4035 | if (m->transformtype != 0) return error(f, VORBIS_invalid_setup); |
| 4036 | if (m->mapping >= f->mapping_count) return error(f, VORBIS_invalid_setup); |
| 4037 | } |
| 4038 | |
| 4039 | flush_packet(f); |
| 4040 | |
| 4041 | f->previous_length = 0; |
| 4042 | |
| 4043 | for (i=0; i < f->channels; ++i) { |
| 4044 | f->channel_buffers[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1); |
| 4045 | f->previous_window[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2); |
| 4046 | f->finalY[i] = (int16 *) setup_malloc(f, sizeof(int16) * longest_floorlist); |
| 4047 | #ifdef STB_VORBIS_NO_DEFER_FLOOR |
| 4048 | f->floor_buffers[i] = (float *) setup_malloc(f, sizeof(float) * f->blocksize_1/2); |
| 4049 | #endif |
| 4050 | } |
| 4051 | |
| 4052 | if (!init_blocksize(f, 0, f->blocksize_0)) return FALSE0; |
| 4053 | if (!init_blocksize(f, 1, f->blocksize_1)) return FALSE0; |
| 4054 | f->blocksize[0] = f->blocksize_0; |
| 4055 | f->blocksize[1] = f->blocksize_1; |
| 4056 | |
| 4057 | #ifdef STB_VORBIS_DIVIDE_TABLE |
| 4058 | if (integer_divide_table[1][1]==0) |
| 4059 | for (i=0; i < DIVTAB_NUMER; ++i) |
| 4060 | for (j=1; j < DIVTAB_DENOM; ++j) |
| 4061 | integer_divide_table[i][j] = i / j; |
| 4062 | #endif |
| 4063 | |
| 4064 | // compute how much temporary memory is needed |
| 4065 | |
| 4066 | // 1. |
| 4067 | { |
| 4068 | uint32 imdct_mem = (f->blocksize_1 * sizeof(float) >> 1); |
| 4069 | uint32 classify_mem; |
| 4070 | int i,max_part_read=0; |
| 4071 | for (i=0; i < f->residue_count; ++i) { |
| 4072 | Residue *r = f->residue_config + i; |
| 4073 | int n_read = r->end - r->begin; |
| 4074 | int part_read = n_read / r->part_size; |
| 4075 | if (part_read > max_part_read) |
| 4076 | max_part_read = part_read; |
| 4077 | } |
| 4078 | #ifndef STB_VORBIS_DIVIDES_IN_RESIDUE |
| 4079 | classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(uint8 *)); |
| 4080 | #else |
| 4081 | classify_mem = f->channels * (sizeof(void*) + max_part_read * sizeof(int *)); |
| 4082 | #endif |
| 4083 | |
| 4084 | f->temp_memory_required = classify_mem; |
| 4085 | if (imdct_mem > f->temp_memory_required) |
| 4086 | f->temp_memory_required = imdct_mem; |
| 4087 | } |
| 4088 | |
| 4089 | f->first_decode = TRUE1; |
| 4090 | |
| 4091 | if (f->alloc.alloc_buffer) { |
| 4092 | assert(f->temp_offset == f->alloc.alloc_buffer_length_in_bytes)((f->temp_offset == f->alloc.alloc_buffer_length_in_bytes ) ? (void)0 : _assert("f->temp_offset == f->alloc.alloc_buffer_length_in_bytes" , "src/siege/internal/stb/stb_vorbis.c", 4092)); |
| 4093 | // check if there's enough temp memory so we don't error later |
| 4094 | if (f->setup_offset + sizeof(*f) + f->temp_memory_required > (unsigned) f->temp_offset) |
| 4095 | return error(f, VORBIS_outofmem); |
| 4096 | } |
| 4097 | |
| 4098 | f->first_audio_page_offset = stb_vorbis_get_file_offset(f); |
| 4099 | |
| 4100 | return TRUE1; |
| 4101 | } |
| 4102 | |
| 4103 | static void vorbis_deinit(stb_vorbis *p) |
| 4104 | { |
| 4105 | int i,j; |
| 4106 | for (i=0; i < p->residue_count; ++i) { |
| 4107 | Residue *r = p->residue_config+i; |
| 4108 | if (r->classdata) { |
| 4109 | for (j=0; j < p->codebooks[r->classbook].entries; ++j) |
| 4110 | setup_free(p, r->classdata[j]); |
| 4111 | setup_free(p, r->classdata); |
| 4112 | } |
| 4113 | setup_free(p, r->residue_books); |
| 4114 | } |
| 4115 | |
| 4116 | if (p->codebooks) { |
| 4117 | for (i=0; i < p->codebook_count; ++i) { |
| 4118 | Codebook *c = p->codebooks + i; |
| 4119 | setup_free(p, c->codeword_lengths); |
| 4120 | setup_free(p, c->multiplicands); |
| 4121 | setup_free(p, c->codewords); |
| 4122 | setup_free(p, c->sorted_codewords); |
| 4123 | // c->sorted_values[-1] is the first entry in the array |
| 4124 | setup_free(p, c->sorted_values ? c->sorted_values-1 : NULL((void*)0)); |
| 4125 | } |
| 4126 | setup_free(p, p->codebooks); |
| 4127 | } |
| 4128 | setup_free(p, p->floor_config); |
| 4129 | setup_free(p, p->residue_config); |
| 4130 | for (i=0; i < p->mapping_count; ++i) |
| 4131 | setup_free(p, p->mapping[i].chan); |
| 4132 | setup_free(p, p->mapping); |
| 4133 | for (i=0; i < p->channels; ++i) { |
| 4134 | setup_free(p, p->channel_buffers[i]); |
| 4135 | setup_free(p, p->previous_window[i]); |
| 4136 | #ifdef STB_VORBIS_NO_DEFER_FLOOR |
| 4137 | setup_free(p, p->floor_buffers[i]); |
| 4138 | #endif |
| 4139 | setup_free(p, p->finalY[i]); |
| 4140 | } |
| 4141 | for (i=0; i < 2; ++i) { |
| 4142 | setup_free(p, p->A[i]); |
| 4143 | setup_free(p, p->B[i]); |
| 4144 | setup_free(p, p->C(2 | 4 | 1)[i]); |
| 4145 | setup_free(p, p->window[i]); |
| 4146 | } |
| 4147 | #ifndef STB_VORBIS_NO_STDIO |
| 4148 | if (p->close_on_free) fclose(p->f); |
| 4149 | #endif |
| 4150 | } |
| 4151 | |
| 4152 | void stb_vorbis_close(stb_vorbis *p) |
| 4153 | { |
| 4154 | if (p == NULL((void*)0)) return; |
| 4155 | vorbis_deinit(p); |
| 4156 | setup_free(p,p); |
| 4157 | } |
| 4158 | |
| 4159 | static void vorbis_init(stb_vorbis *p, stb_vorbis_alloc *z) |
| 4160 | { |
| 4161 | memset(p, 0, sizeof(*p)); // NULL out all malloc'd pointers to start |
| 4162 | if (z) { |
| 4163 | p->alloc = *z; |
| 4164 | p->alloc.alloc_buffer_length_in_bytes = (p->alloc.alloc_buffer_length_in_bytes+3) & ~3; |
| 4165 | p->temp_offset = p->alloc.alloc_buffer_length_in_bytes; |
| 4166 | } |
| 4167 | p->eof = 0; |
| 4168 | p->error = VORBIS__no_error; |
| 4169 | p->stream = NULL((void*)0); |
| 4170 | p->codebooks = NULL((void*)0); |
| 4171 | p->page_crc_tests = -1; |
| 4172 | #ifndef STB_VORBIS_NO_STDIO |
| 4173 | p->close_on_free = FALSE0; |
| 4174 | p->f = NULL((void*)0); |
| 4175 | #endif |
| 4176 | } |
| 4177 | |
| 4178 | int stb_vorbis_get_sample_offset(stb_vorbis *f) |
| 4179 | { |
| 4180 | if (f->current_loc_valid) |
| 4181 | return f->current_loc; |
| 4182 | else |
| 4183 | return -1; |
| 4184 | } |
| 4185 | |
| 4186 | stb_vorbis_info stb_vorbis_get_info(stb_vorbis *f) |
| 4187 | { |
| 4188 | stb_vorbis_info d; |
| 4189 | d.channels = f->channels; |
| 4190 | d.sample_rate = f->sample_rate; |
| 4191 | d.setup_memory_required = f->setup_memory_required; |
| 4192 | d.setup_temp_memory_required = f->setup_temp_memory_required; |
| 4193 | d.temp_memory_required = f->temp_memory_required; |
| 4194 | d.max_frame_size = f->blocksize_1 >> 1; |
| 4195 | return d; |
| 4196 | } |
| 4197 | |
| 4198 | int stb_vorbis_get_error(stb_vorbis *f) |
| 4199 | { |
| 4200 | int e = f->error; |
| 4201 | f->error = VORBIS__no_error; |
| 4202 | return e; |
| 4203 | } |
| 4204 | |
| 4205 | static stb_vorbis * vorbis_alloc(stb_vorbis *f) |
| 4206 | { |
| 4207 | stb_vorbis *p = (stb_vorbis *) setup_malloc(f, sizeof(*p)); |
| 4208 | return p; |
| 4209 | } |
| 4210 | |
| 4211 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 4212 | |
| 4213 | void stb_vorbis_flush_pushdata(stb_vorbis *f) |
| 4214 | { |
| 4215 | f->previous_length = 0; |
| 4216 | f->page_crc_tests = 0; |
| 4217 | f->discard_samples_deferred = 0; |
| 4218 | f->current_loc_valid = FALSE0; |
| 4219 | f->first_decode = FALSE0; |
| 4220 | f->samples_output = 0; |
| 4221 | f->channel_buffer_start = 0; |
| 4222 | f->channel_buffer_end = 0; |
| 4223 | } |
| 4224 | |
| 4225 | static int vorbis_search_for_page_pushdata(vorb *f, uint8 *data, int data_len) |
| 4226 | { |
| 4227 | int i,n; |
| 4228 | for (i=0; i < f->page_crc_tests; ++i) |
| 4229 | f->scan[i].bytes_done = 0; |
| 4230 | |
| 4231 | // if we have room for more scans, search for them first, because |
| 4232 | // they may cause us to stop early if their header is incomplete |
| 4233 | if (f->page_crc_tests < STB_VORBIS_PUSHDATA_CRC_COUNT4) { |
| 4234 | if (data_len < 4) return 0; |
| 4235 | data_len -= 3; // need to look for 4-byte sequence, so don't miss |
| 4236 | // one that straddles a boundary |
| 4237 | for (i=0; i < data_len; ++i) { |
| 4238 | if (data[i] == 0x4f) { |
| 4239 | if (0==memcmp(data+i, ogg_page_header, 4)) { |
| 4240 | int j,len; |
| 4241 | uint32 crc; |
| 4242 | // make sure we have the whole page header |
| 4243 | if (i+26 >= data_len || i+27+data[i+26] >= data_len) { |
| 4244 | // only read up to this page start, so hopefully we'll |
| 4245 | // have the whole page header start next time |
| 4246 | data_len = i; |
| 4247 | break; |
| 4248 | } |
| 4249 | // ok, we have it all; compute the length of the page |
| 4250 | len = 27 + data[i+26]; |
| 4251 | for (j=0; j < data[i+26]; ++j) |
| 4252 | len += data[i+27+j]; |
| 4253 | // scan everything up to the embedded crc (which we must 0) |
| 4254 | crc = 0; |
| 4255 | for (j=0; j < 22; ++j) |
| 4256 | crc = crc32_update(crc, data[i+j]); |
| 4257 | // now process 4 0-bytes |
| 4258 | for ( ; j < 26; ++j) |
| 4259 | crc = crc32_update(crc, 0); |
| 4260 | // len is the total number of bytes we need to scan |
| 4261 | n = f->page_crc_tests++; |
| 4262 | f->scan[n].bytes_left = len-j; |
| 4263 | f->scan[n].crc_so_far = crc; |
| 4264 | f->scan[n].goal_crc = data[i+22] + (data[i+23] << 8) + (data[i+24]<<16) + (data[i+25]<<24); |
| 4265 | // if the last frame on a page is continued to the next, then |
| 4266 | // we can't recover the sample_loc immediately |
| 4267 | if (data[i+27+data[i+26]-1] == 255) |
| 4268 | f->scan[n].sample_loc = ~0; |
| 4269 | else |
| 4270 | f->scan[n].sample_loc = data[i+6] + (data[i+7] << 8) + (data[i+ 8]<<16) + (data[i+ 9]<<24); |
| 4271 | f->scan[n].bytes_done = i+j; |
| 4272 | if (f->page_crc_tests == STB_VORBIS_PUSHDATA_CRC_COUNT4) |
| 4273 | break; |
| 4274 | // keep going if we still have room for more |
| 4275 | } |
| 4276 | } |
| 4277 | } |
| 4278 | } |
| 4279 | |
| 4280 | for (i=0; i < f->page_crc_tests;) { |
| 4281 | uint32 crc; |
| 4282 | int j; |
| 4283 | int n = f->scan[i].bytes_done; |
| 4284 | int m = f->scan[i].bytes_left; |
| 4285 | if (m > data_len - n) m = data_len - n; |
| 4286 | // m is the bytes to scan in the current chunk |
| 4287 | crc = f->scan[i].crc_so_far; |
| 4288 | for (j=0; j < m; ++j) |
| 4289 | crc = crc32_update(crc, data[n+j]); |
| 4290 | f->scan[i].bytes_left -= m; |
| 4291 | f->scan[i].crc_so_far = crc; |
| 4292 | if (f->scan[i].bytes_left == 0) { |
| 4293 | // does it match? |
| 4294 | if (f->scan[i].crc_so_far == f->scan[i].goal_crc) { |
| 4295 | // Houston, we have page |
| 4296 | data_len = n+m; // consumption amount is wherever that scan ended |
| 4297 | f->page_crc_tests = -1; // drop out of page scan mode |
| 4298 | f->previous_length = 0; // decode-but-don't-output one frame |
| 4299 | f->next_seg = -1; // start a new page |
| 4300 | f->current_loc = f->scan[i].sample_loc; // set the current sample location |
| 4301 | // to the amount we'd have decoded had we decoded this page |
| 4302 | f->current_loc_valid = f->current_loc != ~0; |
| 4303 | return data_len; |
| 4304 | } |
| 4305 | // delete entry |
| 4306 | f->scan[i] = f->scan[--f->page_crc_tests]; |
| 4307 | } else { |
| 4308 | ++i; |
| 4309 | } |
| 4310 | } |
| 4311 | |
| 4312 | return data_len; |
| 4313 | } |
| 4314 | |
| 4315 | // return value: number of bytes we used |
| 4316 | int stb_vorbis_decode_frame_pushdata( |
| 4317 | stb_vorbis *f, // the file we're decoding |
| 4318 | uint8 *data, int data_len, // the memory available for decoding |
| 4319 | int *channels, // place to write number of float * buffers |
| 4320 | float ***output, // place to write float ** array of float * buffers |
| 4321 | int *samples // place to write number of output samples |
| 4322 | ) |
| 4323 | { |
| 4324 | int i; |
| 4325 | int len,right,left; |
| 4326 | |
| 4327 | if (!IS_PUSH_MODE(f)((f)->push_mode)) return error(f, VORBIS_invalid_api_mixing); |
| 4328 | |
| 4329 | if (f->page_crc_tests >= 0) { |
| 4330 | *samples = 0; |
| 4331 | return vorbis_search_for_page_pushdata(f, data, data_len); |
| 4332 | } |
| 4333 | |
| 4334 | f->stream = data; |
| 4335 | f->stream_end = data + data_len; |
| 4336 | f->error = VORBIS__no_error; |
| 4337 | |
| 4338 | // check that we have the entire packet in memory |
| 4339 | if (!is_whole_packet_present(f, FALSE0)) { |
| 4340 | *samples = 0; |
| 4341 | return 0; |
| 4342 | } |
| 4343 | |
| 4344 | if (!vorbis_decode_packet(f, &len, &left, &right)) { |
| 4345 | // save the actual error we encountered |
| 4346 | enum STBVorbisError error = f->error; |
| 4347 | if (error == VORBIS_bad_packet_type) { |
| 4348 | // flush and resynch |
| 4349 | f->error = VORBIS__no_error; |
| 4350 | while (get8_packet(f) != EOP(-1)) |
| 4351 | if (f->eof) break; |
| 4352 | *samples = 0; |
| 4353 | return f->stream - data; |
| 4354 | } |
| 4355 | if (error == VORBIS_continued_packet_flag_invalid) { |
| 4356 | if (f->previous_length == 0) { |
| 4357 | // we may be resynching, in which case it's ok to hit one |
| 4358 | // of these; just discard the packet |
| 4359 | f->error = VORBIS__no_error; |
| 4360 | while (get8_packet(f) != EOP(-1)) |
| 4361 | if (f->eof) break; |
| 4362 | *samples = 0; |
| 4363 | return f->stream - data; |
| 4364 | } |
| 4365 | } |
| 4366 | // if we get an error while parsing, what to do? |
| 4367 | // well, it DEFINITELY won't work to continue from where we are! |
| 4368 | stb_vorbis_flush_pushdata(f); |
| 4369 | // restore the error that actually made us bail |
| 4370 | f->error = error; |
| 4371 | *samples = 0; |
| 4372 | return 1; |
| 4373 | } |
| 4374 | |
| 4375 | // success! |
| 4376 | len = vorbis_finish_frame(f, len, left, right); |
| 4377 | for (i=0; i < f->channels; ++i) |
| 4378 | f->outputs[i] = f->channel_buffers[i] + left; |
| 4379 | |
| 4380 | if (channels) *channels = f->channels; |
| 4381 | *samples = len; |
| 4382 | *output = f->outputs; |
| 4383 | return f->stream - data; |
| 4384 | } |
| 4385 | |
| 4386 | stb_vorbis *stb_vorbis_open_pushdata( |
| 4387 | unsigned char *data, int data_len, // the memory available for decoding |
| 4388 | int *data_used, // only defined if result is not NULL |
| 4389 | int *error, stb_vorbis_alloc *alloc) |
| 4390 | { |
| 4391 | stb_vorbis *f, p; |
| 4392 | vorbis_init(&p, alloc); |
| 4393 | p.stream = data; |
| 4394 | p.stream_end = data + data_len; |
| 4395 | p.push_mode = TRUE1; |
| 4396 | if (!start_decoder(&p)) { |
| 4397 | if (p.eof) |
| 4398 | *error = VORBIS_need_more_data; |
| 4399 | else |
| 4400 | *error = p.error; |
| 4401 | return NULL((void*)0); |
| 4402 | } |
| 4403 | f = vorbis_alloc(&p); |
| 4404 | if (f) { |
| 4405 | *f = p; |
| 4406 | *data_used = f->stream - data; |
| 4407 | *error = 0; |
| 4408 | return f; |
| 4409 | } else { |
| 4410 | vorbis_deinit(&p); |
| 4411 | return NULL((void*)0); |
| 4412 | } |
| 4413 | } |
| 4414 | #endif // STB_VORBIS_NO_PUSHDATA_API |
| 4415 | |
| 4416 | unsigned int stb_vorbis_get_file_offset(stb_vorbis *f) |
| 4417 | { |
| 4418 | #ifndef STB_VORBIS_NO_PUSHDATA_API |
| 4419 | if (f->push_mode) return 0; |
| 4420 | #endif |
| 4421 | if (USE_MEMORY(f)((f)->stream)) return f->stream - f->stream_start; |
| 4422 | #ifndef STB_VORBIS_NO_STDIO |
| 4423 | return ftell(f->f) - f->f_start; |
| 4424 | #endif |
| 4425 | } |
| 4426 | |
| 4427 | #ifndef STB_VORBIS_NO_PULLDATA_API |
| 4428 | // |
| 4429 | // DATA-PULLING API |
| 4430 | // |
| 4431 | |
| 4432 | static uint32 vorbis_find_page(stb_vorbis *f, uint32 *end, uint32 *last) |
| 4433 | { |
| 4434 | for(;;) { |
| 4435 | int n; |
| 4436 | if (f->eof) return 0; |
| 4437 | n = get8(f); |
| 4438 | if (n == 0x4f) { // page header |
| 4439 | unsigned int retry_loc = stb_vorbis_get_file_offset(f); |
| 4440 | int i; |
| 4441 | // check if we're off the end of a file_section stream |
| 4442 | if (retry_loc - 25 > f->stream_len) |
| 4443 | return 0; |
| 4444 | // check the rest of the header |
| 4445 | for (i=1; i < 4; ++i) |
| 4446 | if (get8(f) != ogg_page_header[i]) |
| 4447 | break; |
| 4448 | if (f->eof) return 0; |
| 4449 | if (i == 4) { |
| 4450 | uint8 header[27]; |
| 4451 | uint32 i, crc, goal, len; |
| 4452 | for (i=0; i < 4; ++i) |
| 4453 | header[i] = ogg_page_header[i]; |
| 4454 | for (; i < 27; ++i) |
| 4455 | header[i] = get8(f); |
| 4456 | if (f->eof) return 0; |
| 4457 | if (header[4] != 0) goto invalid; |
| 4458 | goal = header[22] + (header[23] << 8) + (header[24]<<16) + (header[25]<<24); |
| 4459 | for (i=22; i < 26; ++i) |
| 4460 | header[i] = 0; |
| 4461 | crc = 0; |
| 4462 | for (i=0; i < 27; ++i) |
| 4463 | crc = crc32_update(crc, header[i]); |
| 4464 | len = 0; |
| 4465 | for (i=0; i < header[26]; ++i) { |
| 4466 | int s = get8(f); |
| 4467 | crc = crc32_update(crc, s); |
| 4468 | len += s; |
| 4469 | } |
| 4470 | if (len && f->eof) return 0; |
| 4471 | for (i=0; i < len; ++i) |
| 4472 | crc = crc32_update(crc, get8(f)); |
| 4473 | // finished parsing probable page |
| 4474 | if (crc == goal) { |
| 4475 | // we could now check that it's either got the last |
| 4476 | // page flag set, OR it's followed by the capture |
| 4477 | // pattern, but I guess TECHNICALLY you could have |
| 4478 | // a file with garbage between each ogg page and recover |
| 4479 | // from it automatically? So even though that paranoia |
| 4480 | // might decrease the chance of an invalid decode by |
| 4481 | // another 2^32, not worth it since it would hose those |
| 4482 | // invalid-but-useful files? |
| 4483 | if (end) |
| 4484 | *end = stb_vorbis_get_file_offset(f); |
| 4485 | if (last) |
| 4486 | { |
| 4487 | if (header[5] & 0x04) |
| 4488 | *last = 1; |
| 4489 | else |
| 4490 | *last = 0; |
| 4491 | } |
| 4492 | set_file_offset(f, retry_loc-1); |
| 4493 | return 1; |
| 4494 | } |
| 4495 | } |
| 4496 | invalid: |
| 4497 | // not a valid page, so rewind and look for next one |
| 4498 | set_file_offset(f, retry_loc); |
| 4499 | } |
| 4500 | } |
| 4501 | } |
| 4502 | |
| 4503 | // seek is implemented with 'interpolation search'--this is like |
| 4504 | // binary search, but we use the data values to estimate the likely |
| 4505 | // location of the data item (plus a bit of a bias so when the |
| 4506 | // estimation is wrong we don't waste overly much time) |
| 4507 | |
| 4508 | #define SAMPLE_unknown0xffffffff 0xffffffff |
| 4509 | |
| 4510 | |
| 4511 | // ogg vorbis, in its insane infinite wisdom, only provides |
| 4512 | // information about the sample at the END of the page. |
| 4513 | // therefore we COULD have the data we need in the current |
| 4514 | // page, and not know it. we could just use the end location |
| 4515 | // as our only knowledge for bounds, seek back, and eventually |
| 4516 | // the binary search finds it. or we can try to be smart and |
| 4517 | // not waste time trying to locate more pages. we try to be |
| 4518 | // smart, since this data is already in memory anyway, so |
| 4519 | // doing needless I/O would be crazy! |
| 4520 | static int vorbis_analyze_page(stb_vorbis *f, ProbedPage *z) |
| 4521 | { |
| 4522 | uint8 header[27], lacing[255]; |
| 4523 | uint8 packet_type[255]; |
| 4524 | int num_packet, packet_start, previous =0; |
| 4525 | (void)previous; |
| 4526 | int i,len; |
| 4527 | uint32 samples; |
| 4528 | |
| 4529 | // record where the page starts |
| 4530 | z->page_start = stb_vorbis_get_file_offset(f); |
| 4531 | |
| 4532 | // parse the header |
| 4533 | getn(f, header, 27); |
| 4534 | assert(header[0] == 'O' && header[1] == 'g' && header[2] == 'g' && header[3] == 'S')((header[0] == 'O' && header[1] == 'g' && header [2] == 'g' && header[3] == 'S') ? (void)0 : _assert("header[0] == 'O' && header[1] == 'g' && header[2] == 'g' && header[3] == 'S'" , "src/siege/internal/stb/stb_vorbis.c", 4534)); |
| 4535 | getn(f, lacing, header[26]); |
| 4536 | |
| 4537 | // determine the length of the payload |
| 4538 | len = 0; |
| 4539 | for (i=0; i < header[26]; ++i) |
| 4540 | len += lacing[i]; |
| 4541 | |
| 4542 | // this implies where the page ends |
| 4543 | z->page_end = z->page_start + 27 + header[26] + len; |
| 4544 | |
| 4545 | // read the last-decoded sample out of the data |
| 4546 | z->last_decoded_sample = header[6] + (header[7] << 8) + (header[8] << 16) + (header[9] << 16); |
| 4547 | |
| 4548 | if (header[5] & 4) { |
| 4549 | // if this is the last page, it's not possible to work |
| 4550 | // backwards to figure out the first sample! whoops! fuck. |
| 4551 | z->first_decoded_sample = SAMPLE_unknown0xffffffff; |
| 4552 | set_file_offset(f, z->page_start); |
| 4553 | return 1; |
| 4554 | } |
| 4555 | |
| 4556 | // scan through the frames to determine the sample-count of each one... |
| 4557 | // our goal is the sample # of the first fully-decoded sample on the |
| 4558 | // page, which is the first decoded sample of the 2nd page |
| 4559 | |
| 4560 | num_packet=0; |
| 4561 | |
| 4562 | packet_start = ((header[5] & 1) == 0); |
| 4563 | |
| 4564 | for (i=0; i < header[26]; ++i) { |
| 4565 | if (packet_start) { |
| 4566 | uint8 n,b,m; |
| 4567 | if (lacing[i] == 0) goto bail; // trying to read from zero-length packet |
| 4568 | n = get8(f); |
| 4569 | // if bottom bit is non-zero, we've got corruption |
| 4570 | if (n & 1) goto bail; |
| 4571 | n >>= 1; |
| 4572 | b = ilog(f->mode_count-1); |
| 4573 | m = n >> b; |
| 4574 | n &= (1 << b)-1; |
| 4575 | if (n >= f->mode_count) goto bail; |
| 4576 | if (num_packet == 0 && f->mode_config[n].blockflag) |
| 4577 | previous = (m & 1); |
Value stored to 'previous' is never read | |
| 4578 | packet_type[num_packet++] = f->mode_config[n].blockflag; |
| 4579 | skip(f, lacing[i]-1); |
| 4580 | } else |
| 4581 | skip(f, lacing[i]); |
| 4582 | packet_start = (lacing[i] < 255); |
| 4583 | } |
| 4584 | |
| 4585 | // now that we know the sizes of all the pages, we can start determining |
| 4586 | // how much sample data there is. |
| 4587 | |
| 4588 | samples = 0; |
| 4589 | |
| 4590 | // for the last packet, we step by its whole length, because the definition |
| 4591 | // is that we encoded the end sample loc of the 'last packet completed', |
| 4592 | // where 'completed' refers to packets being split, and we are left to guess |
| 4593 | // what 'end sample loc' means. we assume it means ignoring the fact that |
| 4594 | // the last half of the data is useless without windowing against the next |
| 4595 | // packet... (so it's not REALLY complete in that sense) |
| 4596 | if (num_packet > 1) |
| 4597 | samples += f->blocksize[packet_type[num_packet-1]]; |
| 4598 | |
| 4599 | for (i=num_packet-2; i >= 1; --i) { |
| 4600 | // now, for this packet, how many samples do we have that |
| 4601 | // do not overlap the following packet? |
| 4602 | if (packet_type[i] == 1) |
| 4603 | if (packet_type[i+1] == 1) |
| 4604 | samples += f->blocksize_1 >> 1; |
| 4605 | else |
| 4606 | samples += ((f->blocksize_1 - f->blocksize_0) >> 2) + (f->blocksize_0 >> 1); |
| 4607 | else |
| 4608 | samples += f->blocksize_0 >> 1; |
| 4609 | } |
| 4610 | // now, at this point, we've rewound to the very beginning of the |
| 4611 | // _second_ packet. if we entirely discard the first packet after |
| 4612 | // a seek, this will be exactly the right sample number. HOWEVER! |
| 4613 | // we can't as easily compute this number for the LAST page. The |
| 4614 | // only way to get the sample offset of the LAST page is to use |
| 4615 | // the end loc from the previous page. But what that returns us |
| 4616 | // is _exactly_ the place where we get our first non-overlapped |
| 4617 | // sample. (I think. Stupid spec for being ambiguous.) So for |
| 4618 | // consistency it's better to do that here, too. However, that |
| 4619 | // will then require us to NOT discard all of the first frame we |
| 4620 | // decode, in some cases, which means an even weirder frame size |
| 4621 | // and extra code. what a fucking pain. |
| 4622 | |
| 4623 | // we're going to discard the first packet if we |
| 4624 | // start the seek here, so we don't care about it. (we could actually |
| 4625 | // do better; if the first packet is long, and the previous packet |
| 4626 | // is short, there's actually data in the first half of the first |
| 4627 | // packet that doesn't need discarding... but not worth paying the |
| 4628 | // effort of tracking that of that here and in the seeking logic) |
| 4629 | // except crap, if we infer it from the _previous_ packet's end |
| 4630 | // location, we DO need to use that definition... and we HAVE to |
| 4631 | // infer the start loc of the LAST packet from the previous packet's |
| 4632 | // end location. fuck you, ogg vorbis. |
| 4633 | |
| 4634 | z->first_decoded_sample = z->last_decoded_sample - samples; |
| 4635 | |
| 4636 | // restore file state to where we were |
| 4637 | set_file_offset(f, z->page_start); |
| 4638 | return 1; |
| 4639 | |
| 4640 | // restore file state to where we were |
| 4641 | bail: |
| 4642 | set_file_offset(f, z->page_start); |
| 4643 | return 0; |
| 4644 | } |
| 4645 | |
| 4646 | static int vorbis_seek_frame_from_page(stb_vorbis *f, uint32 page_start, uint32 first_sample, uint32 target_sample, int fine) |
| 4647 | { |
| 4648 | int left_start, left_end, right_start, right_end, mode,i; |
| 4649 | int frame=0; |
| 4650 | uint32 frame_start; |
| 4651 | int frames_to_skip, data_to_skip; |
| 4652 | |
| 4653 | // first_sample is the sample # of the first sample that doesn't |
| 4654 | // overlap the previous page... note that this requires us to |
| 4655 | // _partially_ discard the first packet! bleh. |
| 4656 | set_file_offset(f, page_start); |
| 4657 | |
| 4658 | f->next_seg = -1; // force page resync |
| 4659 | |
| 4660 | frame_start = first_sample; |
| 4661 | // frame start is where the previous packet's last decoded sample |
| 4662 | // was, which corresponds to left_end... EXCEPT if the previous |
| 4663 | // packet was long and this packet is short? Probably a bug here. |
| 4664 | |
| 4665 | |
| 4666 | // now, we can start decoding frames... we'll only FAKE decode them, |
| 4667 | // until we find the frame that contains our sample; then we'll rewind, |
| 4668 | // and try again |
| 4669 | for (;;) { |
| 4670 | int start; |
| 4671 | |
| 4672 | if (!vorbis_decode_initial(f, &left_start, &left_end, &right_start, &right_end, &mode)) |
| 4673 | return error(f, VORBIS_seek_failed); |
| 4674 | |
| 4675 | if (frame == 0) |
| 4676 | start = left_end; |
| 4677 | else |
| 4678 | start = left_start; |
| 4679 | |
| 4680 | // the window starts at left_start; the last valid sample we generate |
| 4681 | // before the next frame's window start is right_start-1 |
| 4682 | if (target_sample < frame_start + right_start-start) |
| 4683 | break; |
| 4684 | |
| 4685 | flush_packet(f); |
| 4686 | if (f->eof) |
| 4687 | return error(f, VORBIS_seek_failed); |
| 4688 | |
| 4689 | frame_start += right_start - start; |
| 4690 | |
| 4691 | ++frame; |
| 4692 | } |
| 4693 | |
| 4694 | // ok, at this point, the sample we want is contained in frame #'frame' |
| 4695 | |
| 4696 | // to decode frame #'frame' normally, we have to decode the |
| 4697 | // previous frame first... but if it's the FIRST frame of the page |
| 4698 | // we can't. if it's the first frame, it means it falls in the part |
| 4699 | // of the first frame that doesn't overlap either of the other frames. |
| 4700 | // so, if we have to handle that case for the first frame, we might |
| 4701 | // as well handle it for all of them, so: |
| 4702 | if (target_sample > frame_start + (left_end - left_start)) { |
| 4703 | // so what we want to do is go ahead and just immediately decode |
| 4704 | // this frame, but then make it so the next get_frame_float() uses |
| 4705 | // this already-decoded data? or do we want to go ahead and rewind, |
| 4706 | // and leave a flag saying to skip the first N data? let's do that |
| 4707 | frames_to_skip = frame; // if this is frame #1, skip 1 frame (#0) |
| 4708 | data_to_skip = left_end - left_start; |
| 4709 | } else { |
| 4710 | // otherwise, we want to skip frames 0, 1, 2, ... frame-2 |
| 4711 | // (which means frame-2+1 total frames) then decode frame-1, |
| 4712 | // then leave frame pending |
| 4713 | frames_to_skip = frame - 1; |
| 4714 | assert(frames_to_skip >= 0)((frames_to_skip >= 0) ? (void)0 : _assert("frames_to_skip >= 0" , "src/siege/internal/stb/stb_vorbis.c", 4714)); |
| 4715 | data_to_skip = -1; |
| 4716 | } |
| 4717 | |
| 4718 | set_file_offset(f, page_start); |
| 4719 | f->next_seg = - 1; // force page resync |
| 4720 | |
| 4721 | for (i=0; i < frames_to_skip; ++i) { |
| 4722 | maybe_start_packet(f); |
| 4723 | flush_packet(f); |
| 4724 | } |
| 4725 | |
| 4726 | if (data_to_skip >= 0) { |
| 4727 | int i,j,n = f->blocksize_0 >> 1; |
| 4728 | f->discard_samples_deferred = data_to_skip; |
| 4729 | for (i=0; i < f->channels; ++i) |
| 4730 | for (j=0; j < n; ++j) |
| 4731 | f->previous_window[i][j] = 0; |
| 4732 | f->previous_length = n; |
| 4733 | frame_start += data_to_skip; |
| 4734 | } else { |
| 4735 | f->previous_length = 0; |
| 4736 | vorbis_pump_first_frame(f); |
| 4737 | } |
| 4738 | |
| 4739 | // at this point, the NEXT decoded frame will generate the desired sample |
| 4740 | if (fine) { |
| 4741 | // so if we're doing sample accurate streaming, we want to go ahead and decode it! |
| 4742 | if (target_sample != frame_start) { |
| 4743 | int n; |
| 4744 | stb_vorbis_get_frame_float(f, &n, NULL((void*)0)); |
| 4745 | assert(target_sample > frame_start)((target_sample > frame_start) ? (void)0 : _assert("target_sample > frame_start" , "src/siege/internal/stb/stb_vorbis.c", 4745)); |
| 4746 | assert(f->channel_buffer_start + (int) (target_sample-frame_start) < f->channel_buffer_end)((f->channel_buffer_start + (int) (target_sample-frame_start ) < f->channel_buffer_end) ? (void)0 : _assert("f->channel_buffer_start + (int) (target_sample-frame_start) < f->channel_buffer_end" , "src/siege/internal/stb/stb_vorbis.c", 4746)); |
| 4747 | f->channel_buffer_start += (target_sample - frame_start); |
| 4748 | } |
| 4749 | } |
| 4750 | |
| 4751 | return 0; |
| 4752 | } |
| 4753 | |
| 4754 | static int vorbis_seek_base(stb_vorbis *f, unsigned int sample_number, int fine) |
| 4755 | { |
| 4756 | ProbedPage p[2],q; |
| 4757 | if (IS_PUSH_MODE(f)((f)->push_mode)) return error(f, VORBIS_invalid_api_mixing); |
| 4758 | |
| 4759 | // do we know the location of the last page? |
| 4760 | if (f->p_last.page_start == 0) { |
| 4761 | uint32 z = stb_vorbis_stream_length_in_samples(f); |
| 4762 | if (z == 0) return error(f, VORBIS_cant_find_last_page); |
| 4763 | } |
| 4764 | |
| 4765 | p[0] = f->p_first; |
| 4766 | p[1] = f->p_last; |
| 4767 | |
| 4768 | if (sample_number >= f->p_last.last_decoded_sample) |
| 4769 | sample_number = f->p_last.last_decoded_sample-1; |
| 4770 | |
| 4771 | if (sample_number < f->p_first.last_decoded_sample) { |
| 4772 | vorbis_seek_frame_from_page(f, p[0].page_start, 0, sample_number, fine); |
| 4773 | return 0; |
| 4774 | } else { |
| 4775 | int attempts=0; |
| 4776 | while (p[0].page_end < p[1].page_start) { |
| 4777 | uint32 probe; |
| 4778 | uint32 start_offset, end_offset; |
| 4779 | uint32 start_sample, end_sample; |
| 4780 | |
| 4781 | // copy these into local variables so we can tweak them |
| 4782 | // if any are unknown |
| 4783 | start_offset = p[0].page_end; |
| 4784 | end_offset = p[1].after_previous_page_start; // an address known to seek to page p[1] |
| 4785 | start_sample = p[0].last_decoded_sample; |
| 4786 | end_sample = p[1].last_decoded_sample; |
| 4787 | |
| 4788 | // currently there is no such tweaking logic needed/possible? |
| 4789 | if (start_sample == SAMPLE_unknown0xffffffff || end_sample == SAMPLE_unknown0xffffffff) |
| 4790 | return error(f, VORBIS_seek_failed); |
| 4791 | |
| 4792 | // now we want to lerp between these for the target samples... |
| 4793 | |
| 4794 | // step 1: we need to bias towards the page start... |
| 4795 | if (start_offset + 4000 < end_offset) |
| 4796 | end_offset -= 4000; |
| 4797 | |
| 4798 | // now compute an interpolated search loc |
| 4799 | probe = start_offset + (int) floor((float) (end_offset - start_offset) / (end_sample - start_sample) * (sample_number - start_sample)); |
| 4800 | |
| 4801 | // next we need to bias towards binary search... |
| 4802 | // code is a little wonky to allow for full 32-bit unsigned values |
| 4803 | if (attempts >= 4) { |
| 4804 | uint32 probe2 = start_offset + ((end_offset - start_offset) >> 1); |
| 4805 | if (attempts >= 8) |
| 4806 | probe = probe2; |
| 4807 | else if (probe < probe2) |
| 4808 | probe = probe + ((probe2 - probe) >> 1); |
| 4809 | else |
| 4810 | probe = probe2 + ((probe - probe2) >> 1); |
| 4811 | } |
| 4812 | ++attempts; |
| 4813 | |
| 4814 | set_file_offset(f, probe); |
| 4815 | if (!vorbis_find_page(f, NULL((void*)0), NULL((void*)0))) return error(f, VORBIS_seek_failed); |
| 4816 | if (!vorbis_analyze_page(f, &q)) return error(f, VORBIS_seek_failed); |
| 4817 | q.after_previous_page_start = probe; |
| 4818 | |
| 4819 | // it's possible we've just found the last page again |
| 4820 | if (q.page_start == p[1].page_start) { |
| 4821 | p[1] = q; |
| 4822 | continue; |
| 4823 | } |
| 4824 | |
| 4825 | if (sample_number < q.last_decoded_sample) |
| 4826 | p[1] = q; |
| 4827 | else |
| 4828 | p[0] = q; |
| 4829 | } |
| 4830 | |
| 4831 | if (p[0].last_decoded_sample <= sample_number && sample_number < p[1].last_decoded_sample) { |
| 4832 | vorbis_seek_frame_from_page(f, p[1].page_start, p[0].last_decoded_sample, sample_number, fine); |
| 4833 | return 0; |
| 4834 | } |
| 4835 | return error(f, VORBIS_seek_failed); |
| 4836 | } |
| 4837 | } |
| 4838 | |
| 4839 | int stb_vorbis_seek_frame(stb_vorbis *f, unsigned int sample_number) |
| 4840 | { |
| 4841 | return vorbis_seek_base(f, sample_number, FALSE0); |
| 4842 | } |
| 4843 | |
| 4844 | int stb_vorbis_seek(stb_vorbis *f, unsigned int sample_number) |
| 4845 | { |
| 4846 | return vorbis_seek_base(f, sample_number, TRUE1); |
| 4847 | } |
| 4848 | |
| 4849 | void stb_vorbis_seek_start(stb_vorbis *f) |
| 4850 | { |
| 4851 | if (IS_PUSH_MODE(f)((f)->push_mode)) { error(f, VORBIS_invalid_api_mixing); return; } |
| 4852 | set_file_offset(f, f->first_audio_page_offset); |
| 4853 | f->previous_length = 0; |
| 4854 | f->first_decode = TRUE1; |
| 4855 | f->next_seg = -1; |
| 4856 | vorbis_pump_first_frame(f); |
| 4857 | } |
| 4858 | |
| 4859 | unsigned int stb_vorbis_stream_length_in_samples(stb_vorbis *f) |
| 4860 | { |
| 4861 | unsigned int restore_offset, previous_safe; |
| 4862 | unsigned int end, last_page_loc; |
| 4863 | |
| 4864 | if (IS_PUSH_MODE(f)((f)->push_mode)) return error(f, VORBIS_invalid_api_mixing); |
| 4865 | if (!f->total_samples) { |
| 4866 | int last; |
| 4867 | uint32 lo,hi; |
| 4868 | char header[6]; |
| 4869 | |
| 4870 | // first, store the current decode position so we can restore it |
| 4871 | restore_offset = stb_vorbis_get_file_offset(f); |
| 4872 | |
| 4873 | // now we want to seek back 64K from the end (the last page must |
| 4874 | // be at most a little less than 64K, but let's allow a little slop) |
| 4875 | if (f->stream_len >= 65536 && f->stream_len-65536 >= f->first_audio_page_offset) |
| 4876 | previous_safe = f->stream_len - 65536; |
| 4877 | else |
| 4878 | previous_safe = f->first_audio_page_offset; |
| 4879 | |
| 4880 | set_file_offset(f, previous_safe); |
| 4881 | // previous_safe is now our candidate 'earliest known place that seeking |
| 4882 | // to will lead to the final page' |
| 4883 | |
| 4884 | if (!vorbis_find_page(f, &end, (int unsigned *)&last)) { |
| 4885 | // if we can't find a page, we're hosed! |
| 4886 | f->error = VORBIS_cant_find_last_page; |
| 4887 | f->total_samples = 0xffffffff; |
| 4888 | goto done; |
| 4889 | } |
| 4890 | |
| 4891 | // check if there are more pages |
| 4892 | last_page_loc = stb_vorbis_get_file_offset(f); |
| 4893 | |
| 4894 | // stop when the last_page flag is set, not when we reach eof; |
| 4895 | // this allows us to stop short of a 'file_section' end without |
| 4896 | // explicitly checking the length of the section |
| 4897 | while (!last) { |
| 4898 | set_file_offset(f, end); |
| 4899 | if (!vorbis_find_page(f, &end, (int unsigned *)&last)) { |
| 4900 | // the last page we found didn't have the 'last page' flag |
| 4901 | // set. whoops! |
| 4902 | break; |
| 4903 | } |
| 4904 | previous_safe = last_page_loc+1; |
| 4905 | last_page_loc = stb_vorbis_get_file_offset(f); |
| 4906 | } |
| 4907 | |
| 4908 | set_file_offset(f, last_page_loc); |
| 4909 | |
| 4910 | // parse the header |
| 4911 | getn(f, (unsigned char *)header, 6); |
| 4912 | // extract the absolute granule position |
| 4913 | lo = get32(f); |
| 4914 | hi = get32(f); |
| 4915 | if (lo == 0xffffffff && hi == 0xffffffff) { |
| 4916 | f->error = VORBIS_cant_find_last_page; |
| 4917 | f->total_samples = SAMPLE_unknown0xffffffff; |
| 4918 | goto done; |
| 4919 | } |
| 4920 | if (hi) |
| 4921 | lo = 0xfffffffe; // saturate |
| 4922 | f->total_samples = lo; |
| 4923 | |
| 4924 | f->p_last.page_start = last_page_loc; |
| 4925 | f->p_last.page_end = end; |
| 4926 | f->p_last.last_decoded_sample = lo; |
| 4927 | f->p_last.first_decoded_sample = SAMPLE_unknown0xffffffff; |
| 4928 | f->p_last.after_previous_page_start = previous_safe; |
| 4929 | |
| 4930 | done: |
| 4931 | set_file_offset(f, restore_offset); |
| 4932 | } |
| 4933 | return f->total_samples == SAMPLE_unknown0xffffffff ? 0 : f->total_samples; |
| 4934 | } |
| 4935 | |
| 4936 | float stb_vorbis_stream_length_in_seconds(stb_vorbis *f) |
| 4937 | { |
| 4938 | return stb_vorbis_stream_length_in_samples(f) / (float) f->sample_rate; |
| 4939 | } |
| 4940 | |
| 4941 | |
| 4942 | |
| 4943 | int stb_vorbis_get_frame_float(stb_vorbis *f, int *channels, float ***output) |
| 4944 | { |
| 4945 | int len, right,left,i; |
| 4946 | if (IS_PUSH_MODE(f)((f)->push_mode)) return error(f, VORBIS_invalid_api_mixing); |
| 4947 | |
| 4948 | if (!vorbis_decode_packet(f, &len, &left, &right)) { |
| 4949 | f->channel_buffer_start = f->channel_buffer_end = 0; |
| 4950 | return 0; |
| 4951 | } |
| 4952 | |
| 4953 | len = vorbis_finish_frame(f, len, left, right); |
| 4954 | for (i=0; i < f->channels; ++i) |
| 4955 | f->outputs[i] = f->channel_buffers[i] + left; |
| 4956 | |
| 4957 | f->channel_buffer_start = left; |
| 4958 | f->channel_buffer_end = left+len; |
| 4959 | |
| 4960 | if (channels) *channels = f->channels; |
| 4961 | if (output) *output = f->outputs; |
| 4962 | return len; |
| 4963 | } |
| 4964 | |
| 4965 | #ifndef STB_VORBIS_NO_STDIO |
| 4966 | |
| 4967 | stb_vorbis * stb_vorbis_open_file_section(FILE *file, int close_on_free, int *error, stb_vorbis_alloc *alloc, unsigned int length) |
| 4968 | { |
| 4969 | stb_vorbis *f, p; |
| 4970 | vorbis_init(&p, alloc); |
| 4971 | p.f = file; |
| 4972 | p.f_start = ftell(file); |
| 4973 | p.stream_len = length; |
| 4974 | p.close_on_free = close_on_free; |
| 4975 | if (start_decoder(&p)) { |
| 4976 | f = vorbis_alloc(&p); |
| 4977 | if (f) { |
| 4978 | *f = p; |
| 4979 | vorbis_pump_first_frame(f); |
| 4980 | return f; |
| 4981 | } |
| 4982 | } |
| 4983 | if (error) *error = p.error; |
| 4984 | vorbis_deinit(&p); |
| 4985 | return NULL((void*)0); |
| 4986 | } |
| 4987 | |
| 4988 | stb_vorbis * stb_vorbis_open_file(FILE *file, int close_on_free, int *error, stb_vorbis_alloc *alloc) |
| 4989 | { |
| 4990 | unsigned int len, start; |
| 4991 | start = ftell(file); |
| 4992 | fseek(file, 0, SEEK_END2); |
| 4993 | len = ftell(file) - start; |
| 4994 | fseek(file, start, SEEK_SET0); |
| 4995 | return stb_vorbis_open_file_section(file, close_on_free, error, alloc, len); |
| 4996 | } |
| 4997 | |
| 4998 | stb_vorbis * stb_vorbis_open_filename(char *filename, int *error, stb_vorbis_alloc *alloc) |
| 4999 | { |
| 5000 | FILE *f = fopen(filename, "rb"); |
| 5001 | if (f) |
| 5002 | return stb_vorbis_open_file(f, TRUE1, error, alloc); |
| 5003 | if (error) *error = VORBIS_file_open_failure; |
| 5004 | return NULL((void*)0); |
| 5005 | } |
| 5006 | #endif // STB_VORBIS_NO_STDIO |
| 5007 | |
| 5008 | stb_vorbis * stb_vorbis_open_memory(unsigned char *data, int len, int *error, stb_vorbis_alloc *alloc) |
| 5009 | { |
| 5010 | stb_vorbis *f, p; |
| 5011 | if (data == NULL((void*)0)) return NULL((void*)0); |
| 5012 | vorbis_init(&p, alloc); |
| 5013 | p.stream = data; |
| 5014 | p.stream_end = data + len; |
| 5015 | p.stream_start = p.stream; |
| 5016 | p.stream_len = len; |
| 5017 | p.push_mode = FALSE0; |
| 5018 | if (start_decoder(&p)) { |
| 5019 | f = vorbis_alloc(&p); |
| 5020 | if (f) { |
| 5021 | *f = p; |
| 5022 | vorbis_pump_first_frame(f); |
| 5023 | return f; |
| 5024 | } |
| 5025 | } |
| 5026 | if (error) *error = p.error; |
| 5027 | vorbis_deinit(&p); |
| 5028 | return NULL((void*)0); |
| 5029 | } |
| 5030 | |
| 5031 | #ifndef STB_VORBIS_NO_INTEGER_CONVERSION |
| 5032 | #define PLAYBACK_MONO1 1 |
| 5033 | #define PLAYBACK_LEFT2 2 |
| 5034 | #define PLAYBACK_RIGHT4 4 |
| 5035 | |
| 5036 | #define L(2 | 1) (PLAYBACK_LEFT2 | PLAYBACK_MONO1) |
| 5037 | #define C(2 | 4 | 1) (PLAYBACK_LEFT2 | PLAYBACK_RIGHT4 | PLAYBACK_MONO1) |
| 5038 | #define R(4 | 1) (PLAYBACK_RIGHT4 | PLAYBACK_MONO1) |
| 5039 | |
| 5040 | static int8 channel_position[7][6] = |
| 5041 | { |
| 5042 | { 0 }, |
| 5043 | { C(2 | 4 | 1) }, |
| 5044 | { L(2 | 1), R(4 | 1) }, |
| 5045 | { L(2 | 1), C(2 | 4 | 1), R(4 | 1) }, |
| 5046 | { L(2 | 1), R(4 | 1), L(2 | 1), R(4 | 1) }, |
| 5047 | { L(2 | 1), C(2 | 4 | 1), R(4 | 1), L(2 | 1), R(4 | 1) }, |
| 5048 | { L(2 | 1), C(2 | 4 | 1), R(4 | 1), L(2 | 1), R(4 | 1), C(2 | 4 | 1) }, |
| 5049 | }; |
| 5050 | |
| 5051 | |
| 5052 | #ifndef STB_VORBIS_NO_FAST_SCALED_FLOAT |
| 5053 | typedef union { |
| 5054 | float f; |
| 5055 | int i; |
| 5056 | } float_conv; |
| 5057 | typedef char stb_vorbis_float_size_test[sizeof(float)==4 && sizeof(int) == 4]; |
| 5058 | #define FASTDEF(x)float_conv x float_conv x |
| 5059 | // add (1<<23) to convert to int, then divide by 2^SHIFT, then add 0.5/2^SHIFT to round |
| 5060 | #define MAGIC(SHIFT)(1.5f * (1 << (23-SHIFT)) + 0.5f/(1 << SHIFT)) (1.5f * (1 << (23-SHIFT)) + 0.5f/(1 << SHIFT)) |
| 5061 | #define ADDEND(SHIFT)(((150-SHIFT) << 23) + (1 << 22)) (((150-SHIFT) << 23) + (1 << 22)) |
| 5062 | #define FAST_SCALED_FLOAT_TO_INT(temp,x,s)(temp.f = (x) + (1.5f * (1 << (23-s)) + 0.5f/(1 << s)), temp.i - (((150-s) << 23) + (1 << 22))) (temp.f = (x) + MAGIC(s)(1.5f * (1 << (23-s)) + 0.5f/(1 << s)), temp.i - ADDEND(s)(((150-s) << 23) + (1 << 22))) |
| 5063 | #define check_endianness() |
| 5064 | #else |
| 5065 | #define FAST_SCALED_FLOAT_TO_INT(temp,x,s)(temp.f = (x) + (1.5f * (1 << (23-s)) + 0.5f/(1 << s)), temp.i - (((150-s) << 23) + (1 << 22))) ((int) ((x) * (1 << (s)))) |
| 5066 | #define check_endianness() |
| 5067 | #define FASTDEF(x)float_conv x |
| 5068 | #endif |
| 5069 | |
| 5070 | static void copy_samples(short *dest, float *src, int len) |
| 5071 | { |
| 5072 | int i; |
| 5073 | check_endianness(); |
| 5074 | for (i=0; i < len; ++i) { |
| 5075 | FASTDEF(temp)float_conv temp; |
| 5076 | int v = FAST_SCALED_FLOAT_TO_INT(temp, src[i],15)(temp.f = (src[i]) + (1.5f * (1 << (23-15)) + 0.5f/(1 << 15)), temp.i - (((150-15) << 23) + (1 << 22))); |
| 5077 | if ((unsigned int) (v + 32768) > 65535) |
| 5078 | v = v < 0 ? -32768 : 32767; |
| 5079 | dest[i] = v; |
| 5080 | } |
| 5081 | } |
| 5082 | |
| 5083 | static void compute_samples(int mask, short *output, int num_c, float **data, int d_offset, int len) |
| 5084 | { |
| 5085 | #define BUFFER_SIZE32 32 |
| 5086 | float buffer[BUFFER_SIZE32]; |
| 5087 | int i,j,o,n = BUFFER_SIZE32; |
| 5088 | check_endianness(); |
| 5089 | for (o = 0; o < len; o += BUFFER_SIZE32) { |
| 5090 | memset(buffer, 0, sizeof(buffer)); |
| 5091 | if (o + n > len) n = len - o; |
| 5092 | for (j=0; j < num_c; ++j) { |
| 5093 | if (channel_position[num_c][j] & mask) { |
| 5094 | for (i=0; i < n; ++i) |
| 5095 | buffer[i] += data[j][d_offset+o+i]; |
| 5096 | } |
| 5097 | } |
| 5098 | for (i=0; i < n; ++i) { |
| 5099 | FASTDEF(temp)float_conv temp; |
| 5100 | int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15)(temp.f = (buffer[i]) + (1.5f * (1 << (23-15)) + 0.5f/( 1 << 15)), temp.i - (((150-15) << 23) + (1 << 22))); |
| 5101 | if ((unsigned int) (v + 32768) > 65535) |
| 5102 | v = v < 0 ? -32768 : 32767; |
| 5103 | output[o+i] = v; |
| 5104 | } |
| 5105 | } |
| 5106 | } |
| 5107 | |
| 5108 | static int channel_selector[3][2] = { {0}, {PLAYBACK_MONO1}, {PLAYBACK_LEFT2, PLAYBACK_RIGHT4} }; |
| 5109 | static void compute_stereo_samples(short *output, int num_c, float **data, int d_offset, int len) |
| 5110 | { |
| 5111 | #define BUFFER_SIZE32 32 |
| 5112 | float buffer[BUFFER_SIZE32]; |
| 5113 | int i,j,o,n = BUFFER_SIZE32 >> 1; |
| 5114 | // o is the offset in the source data |
| 5115 | check_endianness(); |
| 5116 | for (o = 0; o < len; o += BUFFER_SIZE32 >> 1) { |
| 5117 | // o2 is the offset in the output data |
| 5118 | int o2 = o << 1; |
| 5119 | memset(buffer, 0, sizeof(buffer)); |
| 5120 | if (o + n > len) n = len - o; |
| 5121 | for (j=0; j < num_c; ++j) { |
| 5122 | int m = channel_position[num_c][j] & (PLAYBACK_LEFT2 | PLAYBACK_RIGHT4); |
| 5123 | if (m == (PLAYBACK_LEFT2 | PLAYBACK_RIGHT4)) { |
| 5124 | for (i=0; i < n; ++i) { |
| 5125 | buffer[i*2+0] += data[j][d_offset+o+i]; |
| 5126 | buffer[i*2+1] += data[j][d_offset+o+i]; |
| 5127 | } |
| 5128 | } else if (m == PLAYBACK_LEFT2) { |
| 5129 | for (i=0; i < n; ++i) { |
| 5130 | buffer[i*2+0] += data[j][d_offset+o+i]; |
| 5131 | } |
| 5132 | } else if (m == PLAYBACK_RIGHT4) { |
| 5133 | for (i=0; i < n; ++i) { |
| 5134 | buffer[i*2+1] += data[j][d_offset+o+i]; |
| 5135 | } |
| 5136 | } |
| 5137 | } |
| 5138 | for (i=0; i < (n<<1); ++i) { |
| 5139 | FASTDEF(temp)float_conv temp; |
| 5140 | int v = FAST_SCALED_FLOAT_TO_INT(temp,buffer[i],15)(temp.f = (buffer[i]) + (1.5f * (1 << (23-15)) + 0.5f/( 1 << 15)), temp.i - (((150-15) << 23) + (1 << 22))); |
| 5141 | if ((unsigned int) (v + 32768) > 65535) |
| 5142 | v = v < 0 ? -32768 : 32767; |
| 5143 | output[o2+i] = v; |
| 5144 | } |
| 5145 | } |
| 5146 | } |
| 5147 | |
| 5148 | static void convert_samples_short(int buf_c, short **buffer, int b_offset, int data_c, float **data, int d_offset, int samples) |
| 5149 | { |
| 5150 | int i; |
| 5151 | if (buf_c != data_c && buf_c <= 2 && data_c <= 6) { |
| 5152 | static int channel_selector[3][2] = { {0}, {PLAYBACK_MONO1}, {PLAYBACK_LEFT2, PLAYBACK_RIGHT4} }; |
| 5153 | for (i=0; i < buf_c; ++i) |
| 5154 | compute_samples(channel_selector[buf_c][i], buffer[i]+b_offset, data_c, data, d_offset, samples); |
| 5155 | } else { |
| 5156 | int limit = buf_c < data_c ? buf_c : data_c; |
| 5157 | for (i=0; i < limit; ++i) |
| 5158 | copy_samples(buffer[i]+b_offset, data[i], samples); |
| 5159 | for ( ; i < buf_c; ++i) |
| 5160 | memset(buffer[i]+b_offset, 0, sizeof(short) * samples); |
| 5161 | } |
| 5162 | } |
| 5163 | |
| 5164 | int stb_vorbis_get_frame_short(stb_vorbis *f, int num_c, short **buffer, int num_samples) |
| 5165 | { |
| 5166 | float **output; |
| 5167 | int len = stb_vorbis_get_frame_float(f, NULL((void*)0), &output); |
| 5168 | if (len > num_samples) len = num_samples; |
| 5169 | if (len) |
| 5170 | convert_samples_short(num_c, buffer, 0, f->channels, output, 0, len); |
| 5171 | return len; |
| 5172 | } |
| 5173 | |
| 5174 | static void convert_channels_short_interleaved(int buf_c, short *buffer, int data_c, float **data, int d_offset, int len) |
| 5175 | { |
| 5176 | int i; |
| 5177 | check_endianness(); |
| 5178 | if (buf_c != data_c && buf_c <= 2 && data_c <= 6) { |
| 5179 | assert(buf_c == 2)((buf_c == 2) ? (void)0 : _assert("buf_c == 2", "src/siege/internal/stb/stb_vorbis.c" , 5179)); |
| 5180 | for (i=0; i < buf_c; ++i) |
| 5181 | compute_stereo_samples(buffer, data_c, data, d_offset, len); |
| 5182 | } else { |
| 5183 | int limit = buf_c < data_c ? buf_c : data_c; |
| 5184 | int j; |
| 5185 | for (j=0; j < len; ++j) { |
| 5186 | for (i=0; i < limit; ++i) { |
| 5187 | FASTDEF(temp)float_conv temp; |
| 5188 | float f = data[i][d_offset+j]; |
| 5189 | int v = FAST_SCALED_FLOAT_TO_INT(temp, f,15)(temp.f = (f) + (1.5f * (1 << (23-15)) + 0.5f/(1 << 15)), temp.i - (((150-15) << 23) + (1 << 22)));//data[i][d_offset+j],15); |
| 5190 | if ((unsigned int) (v + 32768) > 65535) |
| 5191 | v = v < 0 ? -32768 : 32767; |
| 5192 | *buffer++ = v; |
| 5193 | } |
| 5194 | for ( ; i < buf_c; ++i) |
| 5195 | *buffer++ = 0; |
| 5196 | } |
| 5197 | } |
| 5198 | } |
| 5199 | |
| 5200 | int stb_vorbis_get_frame_short_interleaved(stb_vorbis *f, int num_c, short *buffer, int num_shorts) |
| 5201 | { |
| 5202 | float **output; |
| 5203 | int len; |
| 5204 | if (num_c == 1) return stb_vorbis_get_frame_short(f,num_c,&buffer, num_shorts); |
| 5205 | len = stb_vorbis_get_frame_float(f, NULL((void*)0), &output); |
| 5206 | if (len) { |
| 5207 | if (len*num_c > num_shorts) len = num_shorts / num_c; |
| 5208 | convert_channels_short_interleaved(num_c, buffer, f->channels, output, 0, len); |
| 5209 | } |
| 5210 | return len; |
| 5211 | } |
| 5212 | |
| 5213 | int stb_vorbis_get_samples_short_interleaved(stb_vorbis *f, int channels, short *buffer, int num_shorts) |
| 5214 | { |
| 5215 | float **outputs; |
| 5216 | int len = num_shorts / channels; |
| 5217 | int n=0; |
| 5218 | int z = f->channels; |
| 5219 | if (z > channels) z = channels; |
| 5220 | while (n < len) { |
| 5221 | int k = f->channel_buffer_end - f->channel_buffer_start; |
| 5222 | if (n+k >= len) k = len - n; |
| 5223 | if (k) |
| 5224 | convert_channels_short_interleaved(channels, buffer, f->channels, f->channel_buffers, f->channel_buffer_start, k); |
| 5225 | buffer += k*channels; |
| 5226 | n += k; |
| 5227 | f->channel_buffer_start += k; |
| 5228 | if (n == len) break; |
| 5229 | if (!stb_vorbis_get_frame_float(f, NULL((void*)0), &outputs)) break; |
| 5230 | } |
| 5231 | return n; |
| 5232 | } |
| 5233 | |
| 5234 | int stb_vorbis_get_samples_short(stb_vorbis *f, int channels, short **buffer, int len) |
| 5235 | { |
| 5236 | float **outputs; |
| 5237 | int n=0; |
| 5238 | int z = f->channels; |
| 5239 | if (z > channels) z = channels; |
| 5240 | while (n < len) { |
| 5241 | int k = f->channel_buffer_end - f->channel_buffer_start; |
| 5242 | if (n+k >= len) k = len - n; |
| 5243 | if (k) |
| 5244 | convert_samples_short(channels, buffer, n, f->channels, f->channel_buffers, f->channel_buffer_start, k); |
| 5245 | n += k; |
| 5246 | f->channel_buffer_start += k; |
| 5247 | if (n == len) break; |
| 5248 | if (!stb_vorbis_get_frame_float(f, NULL((void*)0), &outputs)) break; |
| 5249 | } |
| 5250 | return n; |
| 5251 | } |
| 5252 | |
| 5253 | #ifndef STB_VORBIS_NO_STDIO |
| 5254 | int stb_vorbis_decode_filename(char *filename, int *channels, short **output) |
| 5255 | { |
| 5256 | int data_len, offset, total, limit, error; |
| 5257 | short *data; |
| 5258 | stb_vorbis *v = stb_vorbis_open_filename(filename, &error, NULL((void*)0)); |
| 5259 | if (v == NULL((void*)0)) return -1; |
| 5260 | limit = v->channels * 4096; |
| 5261 | *channels = v->channels; |
| 5262 | offset = data_len = 0; |
| 5263 | total = limit; |
| 5264 | data = (short *) malloc(total * sizeof(*data)); |
| 5265 | if (data == NULL((void*)0)) { |
| 5266 | stb_vorbis_close(v); |
| 5267 | return -2; |
| 5268 | } |
| 5269 | for (;;) { |
| 5270 | int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset); |
| 5271 | if (n == 0) break; |
| 5272 | data_len += n; |
| 5273 | offset += n * v->channels; |
| 5274 | if (offset + limit > total) { |
| 5275 | short *data2; |
| 5276 | total *= 2; |
| 5277 | data2 = (short *) realloc(data, total * sizeof(*data)); |
| 5278 | if (data2 == NULL((void*)0)) { |
| 5279 | free(data); |
| 5280 | stb_vorbis_close(v); |
| 5281 | return -2; |
| 5282 | } |
| 5283 | data = data2; |
| 5284 | } |
| 5285 | } |
| 5286 | *output = data; |
| 5287 | return data_len; |
| 5288 | } |
| 5289 | #endif // NO_STDIO |
| 5290 | |
| 5291 | int stb_vorbis_decode_memory(uint8 *mem, int len, int *channels, short **output) |
| 5292 | { |
| 5293 | int data_len, offset, total, limit, error; |
| 5294 | short *data; |
| 5295 | stb_vorbis *v = stb_vorbis_open_memory(mem, len, &error, NULL((void*)0)); |
| 5296 | if (v == NULL((void*)0)) return -1; |
| 5297 | limit = v->channels * 4096; |
| 5298 | *channels = v->channels; |
| 5299 | offset = data_len = 0; |
| 5300 | total = limit; |
| 5301 | data = (short *) malloc(total * sizeof(*data)); |
| 5302 | if (data == NULL((void*)0)) { |
| 5303 | stb_vorbis_close(v); |
| 5304 | return -2; |
| 5305 | } |
| 5306 | for (;;) { |
| 5307 | int n = stb_vorbis_get_frame_short_interleaved(v, v->channels, data+offset, total-offset); |
| 5308 | if (n == 0) break; |
| 5309 | data_len += n; |
| 5310 | offset += n * v->channels; |
| 5311 | if (offset + limit > total) { |
| 5312 | short *data2; |
| 5313 | total *= 2; |
| 5314 | data2 = (short *) realloc(data, total * sizeof(*data)); |
| 5315 | if (data2 == NULL((void*)0)) { |
| 5316 | free(data); |
| 5317 | stb_vorbis_close(v); |
| 5318 | return -2; |
| 5319 | } |
| 5320 | data = data2; |
| 5321 | } |
| 5322 | } |
| 5323 | *output = data; |
| 5324 | return data_len; |
| 5325 | } |
| 5326 | #endif |
| 5327 | |
| 5328 | int stb_vorbis_get_samples_float_interleaved(stb_vorbis *f, int channels, float *buffer, int num_floats) |
| 5329 | { |
| 5330 | float **outputs; |
| 5331 | int len = num_floats / channels; |
| 5332 | int n=0; |
| 5333 | int z = f->channels; |
| 5334 | if (z > channels) z = channels; |
| 5335 | while (n < len) { |
| 5336 | int i,j; |
| 5337 | int k = f->channel_buffer_end - f->channel_buffer_start; |
| 5338 | if (n+k >= len) k = len - n; |
| 5339 | for (j=0; j < k; ++j) { |
| 5340 | for (i=0; i < z; ++i) |
| 5341 | *buffer++ = f->channel_buffers[i][f->channel_buffer_start+j]; |
| 5342 | for ( ; i < channels; ++i) |
| 5343 | *buffer++ = 0; |
| 5344 | } |
| 5345 | n += k; |
| 5346 | f->channel_buffer_start += k; |
| 5347 | if (n == len) break; |
| 5348 | if (!stb_vorbis_get_frame_float(f, NULL((void*)0), &outputs)) break; |
| 5349 | } |
| 5350 | return n; |
| 5351 | } |
| 5352 | |
| 5353 | int stb_vorbis_get_samples_float(stb_vorbis *f, int channels, float **buffer, int num_samples) |
| 5354 | { |
| 5355 | float **outputs; |
| 5356 | int n=0; |
| 5357 | int z = f->channels; |
| 5358 | if (z > channels) z = channels; |
| 5359 | while (n < num_samples) { |
| 5360 | int i; |
| 5361 | int k = f->channel_buffer_end - f->channel_buffer_start; |
| 5362 | if (n+k >= num_samples) k = num_samples - n; |
| 5363 | if (k) { |
| 5364 | for (i=0; i < z; ++i) |
| 5365 | memcpy(buffer[i]+n, f->channel_buffers+f->channel_buffer_start, sizeof(float)*k); |
| 5366 | for ( ; i < channels; ++i) |
| 5367 | memset(buffer[i]+n, 0, sizeof(float) * k); |
| 5368 | } |
| 5369 | n += k; |
| 5370 | f->channel_buffer_start += k; |
| 5371 | if (n == num_samples) break; |
| 5372 | if (!stb_vorbis_get_frame_float(f, NULL((void*)0), &outputs)) break; |
| 5373 | } |
| 5374 | return n; |
| 5375 | } |
| 5376 | #endif // STB_VORBIS_NO_PULLDATA_API |
| 5377 | |
| 5378 | void _dummyFunction(void) |
| 5379 | { |
| 5380 | (void)get_bits_signed; |
| 5381 | (void)codebook_decode_scalar; |
| 5382 | (void)channel_selector; |
| 5383 | } |
| 5384 | |
| 5385 | #endif // STB_VORBIS_HEADER_ONLY |