niklase@google.com | 470e71d | 2011-07-07 08:21:25 +0000 | [diff] [blame] | 1 | /* |
| 2 | * SpanDSP - a series of DSP components for telephony |
| 3 | * |
| 4 | * g711.h - In line A-law and u-law conversion routines |
| 5 | * |
| 6 | * Written by Steve Underwood <steveu@coppice.org> |
| 7 | * |
| 8 | * Copyright (C) 2001 Steve Underwood |
| 9 | * |
| 10 | * Despite my general liking of the GPL, I place this code in the |
| 11 | * public domain for the benefit of all mankind - even the slimy |
| 12 | * ones who might try to proprietize my work and use it to my |
| 13 | * detriment. |
| 14 | * |
| 15 | * $Id: g711.h,v 1.1 2006/06/07 15:46:39 steveu Exp $ |
| 16 | * |
| 17 | * Modifications for WebRtc, 2011/04/28, by tlegrand: |
| 18 | * -Changed to use WebRtc types |
| 19 | * -Changed __inline__ to __inline |
| 20 | * -Two changes to make implementation bitexact with ITU-T reference implementation |
| 21 | */ |
| 22 | |
| 23 | /*! \file */ |
| 24 | |
| 25 | /*! \page g711_page A-law and mu-law handling |
| 26 | Lookup tables for A-law and u-law look attractive, until you consider the impact |
| 27 | on the CPU cache. If it causes a substantial area of your processor cache to get |
| 28 | hit too often, cache sloshing will severely slow things down. The main reason |
| 29 | these routines are slow in C, is the lack of direct access to the CPU's "find |
| 30 | the first 1" instruction. A little in-line assembler fixes that, and the |
| 31 | conversion routines can be faster than lookup tables, in most real world usage. |
| 32 | A "find the first 1" instruction is available on most modern CPUs, and is a |
| 33 | much underused feature. |
| 34 | |
| 35 | If an assembly language method of bit searching is not available, these routines |
| 36 | revert to a method that can be a little slow, so the cache thrashing might not |
| 37 | seem so bad :( |
| 38 | |
| 39 | Feel free to submit patches to add fast "find the first 1" support for your own |
| 40 | favourite processor. |
| 41 | |
| 42 | Look up tables are used for transcoding between A-law and u-law, since it is |
| 43 | difficult to achieve the precise transcoding procedure laid down in the G.711 |
| 44 | specification by other means. |
| 45 | */ |
| 46 | |
| 47 | #if !defined(_G711_H_) |
| 48 | #define _G711_H_ |
| 49 | |
| 50 | #ifdef __cplusplus |
| 51 | extern "C" { |
| 52 | #endif |
| 53 | |
| 54 | #include "typedefs.h" |
| 55 | |
| 56 | #if defined(__i386__) |
| 57 | /*! \brief Find the bit position of the highest set bit in a word |
| 58 | \param bits The word to be searched |
| 59 | \return The bit number of the highest set bit, or -1 if the word is zero. */ |
| 60 | static __inline__ int top_bit(unsigned int bits) |
| 61 | { |
| 62 | int res; |
| 63 | |
| 64 | __asm__ __volatile__(" movl $-1,%%edx;\n" |
| 65 | " bsrl %%eax,%%edx;\n" |
| 66 | : "=d" (res) |
| 67 | : "a" (bits)); |
| 68 | return res; |
| 69 | } |
| 70 | /*- End of function --------------------------------------------------------*/ |
| 71 | |
| 72 | /*! \brief Find the bit position of the lowest set bit in a word |
| 73 | \param bits The word to be searched |
| 74 | \return The bit number of the lowest set bit, or -1 if the word is zero. */ |
| 75 | static __inline__ int bottom_bit(unsigned int bits) |
| 76 | { |
| 77 | int res; |
| 78 | |
| 79 | __asm__ __volatile__(" movl $-1,%%edx;\n" |
| 80 | " bsfl %%eax,%%edx;\n" |
| 81 | : "=d" (res) |
| 82 | : "a" (bits)); |
| 83 | return res; |
| 84 | } |
| 85 | /*- End of function --------------------------------------------------------*/ |
| 86 | #elif defined(__x86_64__) |
| 87 | static __inline__ int top_bit(unsigned int bits) |
| 88 | { |
| 89 | int res; |
| 90 | |
| 91 | __asm__ __volatile__(" movq $-1,%%rdx;\n" |
| 92 | " bsrq %%rax,%%rdx;\n" |
| 93 | : "=d" (res) |
| 94 | : "a" (bits)); |
| 95 | return res; |
| 96 | } |
| 97 | /*- End of function --------------------------------------------------------*/ |
| 98 | |
| 99 | static __inline__ int bottom_bit(unsigned int bits) |
| 100 | { |
| 101 | int res; |
| 102 | |
| 103 | __asm__ __volatile__(" movq $-1,%%rdx;\n" |
| 104 | " bsfq %%rax,%%rdx;\n" |
| 105 | : "=d" (res) |
| 106 | : "a" (bits)); |
| 107 | return res; |
| 108 | } |
| 109 | /*- End of function --------------------------------------------------------*/ |
| 110 | #else |
| 111 | static __inline int top_bit(unsigned int bits) |
| 112 | { |
| 113 | int i; |
| 114 | |
| 115 | if (bits == 0) |
| 116 | return -1; |
| 117 | i = 0; |
| 118 | if (bits & 0xFFFF0000) |
| 119 | { |
| 120 | bits &= 0xFFFF0000; |
| 121 | i += 16; |
| 122 | } |
| 123 | if (bits & 0xFF00FF00) |
| 124 | { |
| 125 | bits &= 0xFF00FF00; |
| 126 | i += 8; |
| 127 | } |
| 128 | if (bits & 0xF0F0F0F0) |
| 129 | { |
| 130 | bits &= 0xF0F0F0F0; |
| 131 | i += 4; |
| 132 | } |
| 133 | if (bits & 0xCCCCCCCC) |
| 134 | { |
| 135 | bits &= 0xCCCCCCCC; |
| 136 | i += 2; |
| 137 | } |
| 138 | if (bits & 0xAAAAAAAA) |
| 139 | { |
| 140 | bits &= 0xAAAAAAAA; |
| 141 | i += 1; |
| 142 | } |
| 143 | return i; |
| 144 | } |
| 145 | /*- End of function --------------------------------------------------------*/ |
| 146 | |
| 147 | static __inline int bottom_bit(unsigned int bits) |
| 148 | { |
| 149 | int i; |
| 150 | |
| 151 | if (bits == 0) |
| 152 | return -1; |
| 153 | i = 32; |
| 154 | if (bits & 0x0000FFFF) |
| 155 | { |
| 156 | bits &= 0x0000FFFF; |
| 157 | i -= 16; |
| 158 | } |
| 159 | if (bits & 0x00FF00FF) |
| 160 | { |
| 161 | bits &= 0x00FF00FF; |
| 162 | i -= 8; |
| 163 | } |
| 164 | if (bits & 0x0F0F0F0F) |
| 165 | { |
| 166 | bits &= 0x0F0F0F0F; |
| 167 | i -= 4; |
| 168 | } |
| 169 | if (bits & 0x33333333) |
| 170 | { |
| 171 | bits &= 0x33333333; |
| 172 | i -= 2; |
| 173 | } |
| 174 | if (bits & 0x55555555) |
| 175 | { |
| 176 | bits &= 0x55555555; |
| 177 | i -= 1; |
| 178 | } |
| 179 | return i; |
| 180 | } |
| 181 | /*- End of function --------------------------------------------------------*/ |
| 182 | #endif |
| 183 | |
| 184 | /* N.B. It is tempting to use look-up tables for A-law and u-law conversion. |
| 185 | * However, you should consider the cache footprint. |
| 186 | * |
| 187 | * A 64K byte table for linear to x-law and a 512 byte table for x-law to |
| 188 | * linear sound like peanuts these days, and shouldn't an array lookup be |
| 189 | * real fast? No! When the cache sloshes as badly as this one will, a tight |
| 190 | * calculation may be better. The messiest part is normally finding the |
| 191 | * segment, but a little inline assembly can fix that on an i386, x86_64 and |
| 192 | * many other modern processors. |
| 193 | */ |
| 194 | |
| 195 | /* |
| 196 | * Mu-law is basically as follows: |
| 197 | * |
| 198 | * Biased Linear Input Code Compressed Code |
| 199 | * ------------------------ --------------- |
| 200 | * 00000001wxyza 000wxyz |
| 201 | * 0000001wxyzab 001wxyz |
| 202 | * 000001wxyzabc 010wxyz |
| 203 | * 00001wxyzabcd 011wxyz |
| 204 | * 0001wxyzabcde 100wxyz |
| 205 | * 001wxyzabcdef 101wxyz |
| 206 | * 01wxyzabcdefg 110wxyz |
| 207 | * 1wxyzabcdefgh 111wxyz |
| 208 | * |
| 209 | * Each biased linear code has a leading 1 which identifies the segment |
| 210 | * number. The value of the segment number is equal to 7 minus the number |
| 211 | * of leading 0's. The quantization interval is directly available as the |
| 212 | * four bits wxyz. * The trailing bits (a - h) are ignored. |
| 213 | * |
| 214 | * Ordinarily the complement of the resulting code word is used for |
| 215 | * transmission, and so the code word is complemented before it is returned. |
| 216 | * |
| 217 | * For further information see John C. Bellamy's Digital Telephony, 1982, |
| 218 | * John Wiley & Sons, pps 98-111 and 472-476. |
| 219 | */ |
| 220 | |
| 221 | //#define ULAW_ZEROTRAP /* turn on the trap as per the MIL-STD */ |
| 222 | #define ULAW_BIAS 0x84 /* Bias for linear code. */ |
| 223 | |
| 224 | /*! \brief Encode a linear sample to u-law |
| 225 | \param linear The sample to encode. |
| 226 | \return The u-law value. |
| 227 | */ |
| 228 | static __inline WebRtc_UWord8 linear_to_ulaw(int linear) |
| 229 | { |
| 230 | WebRtc_UWord8 u_val; |
| 231 | int mask; |
| 232 | int seg; |
| 233 | |
| 234 | /* Get the sign and the magnitude of the value. */ |
| 235 | if (linear < 0) |
| 236 | { |
| 237 | /* WebRtc, tlegrand: -1 added to get bitexact to reference implementation */ |
| 238 | linear = ULAW_BIAS - linear - 1; |
| 239 | mask = 0x7F; |
| 240 | } |
| 241 | else |
| 242 | { |
| 243 | linear = ULAW_BIAS + linear; |
| 244 | mask = 0xFF; |
| 245 | } |
| 246 | |
| 247 | seg = top_bit(linear | 0xFF) - 7; |
| 248 | |
| 249 | /* |
| 250 | * Combine the sign, segment, quantization bits, |
| 251 | * and complement the code word. |
| 252 | */ |
| 253 | if (seg >= 8) |
| 254 | u_val = (WebRtc_UWord8) (0x7F ^ mask); |
| 255 | else |
| 256 | u_val = (WebRtc_UWord8) (((seg << 4) | ((linear >> (seg + 3)) & 0xF)) ^ mask); |
| 257 | #ifdef ULAW_ZEROTRAP |
| 258 | /* Optional ITU trap */ |
| 259 | if (u_val == 0) |
| 260 | u_val = 0x02; |
| 261 | #endif |
| 262 | return u_val; |
| 263 | } |
| 264 | /*- End of function --------------------------------------------------------*/ |
| 265 | |
| 266 | /*! \brief Decode an u-law sample to a linear value. |
| 267 | \param ulaw The u-law sample to decode. |
| 268 | \return The linear value. |
| 269 | */ |
| 270 | static __inline WebRtc_Word16 ulaw_to_linear(WebRtc_UWord8 ulaw) |
| 271 | { |
| 272 | int t; |
| 273 | |
| 274 | /* Complement to obtain normal u-law value. */ |
| 275 | ulaw = ~ulaw; |
| 276 | /* |
| 277 | * Extract and bias the quantization bits. Then |
| 278 | * shift up by the segment number and subtract out the bias. |
| 279 | */ |
| 280 | t = (((ulaw & 0x0F) << 3) + ULAW_BIAS) << (((int) ulaw & 0x70) >> 4); |
| 281 | return (WebRtc_Word16) ((ulaw & 0x80) ? (ULAW_BIAS - t) : (t - ULAW_BIAS)); |
| 282 | } |
| 283 | /*- End of function --------------------------------------------------------*/ |
| 284 | |
| 285 | /* |
| 286 | * A-law is basically as follows: |
| 287 | * |
| 288 | * Linear Input Code Compressed Code |
| 289 | * ----------------- --------------- |
| 290 | * 0000000wxyza 000wxyz |
| 291 | * 0000001wxyza 001wxyz |
| 292 | * 000001wxyzab 010wxyz |
| 293 | * 00001wxyzabc 011wxyz |
| 294 | * 0001wxyzabcd 100wxyz |
| 295 | * 001wxyzabcde 101wxyz |
| 296 | * 01wxyzabcdef 110wxyz |
| 297 | * 1wxyzabcdefg 111wxyz |
| 298 | * |
| 299 | * For further information see John C. Bellamy's Digital Telephony, 1982, |
| 300 | * John Wiley & Sons, pps 98-111 and 472-476. |
| 301 | */ |
| 302 | |
| 303 | #define ALAW_AMI_MASK 0x55 |
| 304 | |
| 305 | /*! \brief Encode a linear sample to A-law |
| 306 | \param linear The sample to encode. |
| 307 | \return The A-law value. |
| 308 | */ |
| 309 | static __inline WebRtc_UWord8 linear_to_alaw(int linear) |
| 310 | { |
| 311 | int mask; |
| 312 | int seg; |
| 313 | |
| 314 | if (linear >= 0) |
| 315 | { |
| 316 | /* Sign (bit 7) bit = 1 */ |
| 317 | mask = ALAW_AMI_MASK | 0x80; |
| 318 | } |
| 319 | else |
| 320 | { |
| 321 | /* Sign (bit 7) bit = 0 */ |
| 322 | mask = ALAW_AMI_MASK; |
| 323 | /* WebRtc, tlegrand: Changed from -8 to -1 to get bitexact to reference |
| 324 | * implementation */ |
| 325 | linear = -linear - 1; |
| 326 | } |
| 327 | |
| 328 | /* Convert the scaled magnitude to segment number. */ |
| 329 | seg = top_bit(linear | 0xFF) - 7; |
| 330 | if (seg >= 8) |
| 331 | { |
| 332 | if (linear >= 0) |
| 333 | { |
| 334 | /* Out of range. Return maximum value. */ |
| 335 | return (WebRtc_UWord8) (0x7F ^ mask); |
| 336 | } |
| 337 | /* We must be just a tiny step below zero */ |
| 338 | return (WebRtc_UWord8) (0x00 ^ mask); |
| 339 | } |
| 340 | /* Combine the sign, segment, and quantization bits. */ |
| 341 | return (WebRtc_UWord8) (((seg << 4) | ((linear >> ((seg) ? (seg + 3) : 4)) & 0x0F)) ^ mask); |
| 342 | } |
| 343 | /*- End of function --------------------------------------------------------*/ |
| 344 | |
| 345 | /*! \brief Decode an A-law sample to a linear value. |
| 346 | \param alaw The A-law sample to decode. |
| 347 | \return The linear value. |
| 348 | */ |
| 349 | static __inline WebRtc_Word16 alaw_to_linear(WebRtc_UWord8 alaw) |
| 350 | { |
| 351 | int i; |
| 352 | int seg; |
| 353 | |
| 354 | alaw ^= ALAW_AMI_MASK; |
| 355 | i = ((alaw & 0x0F) << 4); |
| 356 | seg = (((int) alaw & 0x70) >> 4); |
| 357 | if (seg) |
| 358 | i = (i + 0x108) << (seg - 1); |
| 359 | else |
| 360 | i += 8; |
| 361 | return (WebRtc_Word16) ((alaw & 0x80) ? i : -i); |
| 362 | } |
| 363 | /*- End of function --------------------------------------------------------*/ |
| 364 | |
| 365 | /*! \brief Transcode from A-law to u-law, using the procedure defined in G.711. |
| 366 | \param alaw The A-law sample to transcode. |
| 367 | \return The best matching u-law value. |
| 368 | */ |
| 369 | WebRtc_UWord8 alaw_to_ulaw(WebRtc_UWord8 alaw); |
| 370 | |
| 371 | /*! \brief Transcode from u-law to A-law, using the procedure defined in G.711. |
| 372 | \param alaw The u-law sample to transcode. |
| 373 | \return The best matching A-law value. |
| 374 | */ |
| 375 | WebRtc_UWord8 ulaw_to_alaw(WebRtc_UWord8 ulaw); |
| 376 | |
| 377 | #ifdef __cplusplus |
| 378 | } |
| 379 | #endif |
| 380 | |
| 381 | #endif |
| 382 | /*- End of file ------------------------------------------------------------*/ |