Tristan Matthews | 0a329cc | 2013-07-17 13:20:14 -0400 | [diff] [blame] | 1 | /* Copyright (C) 2005 Jean-Marc Valin */ |
| 2 | /** |
| 3 | @file pseudofloat.h |
| 4 | @brief Pseudo-floating point |
| 5 | * This header file provides a lightweight floating point type for |
| 6 | * use on fixed-point platforms when a large dynamic range is |
| 7 | * required. The new type is not compatible with the 32-bit IEEE format, |
| 8 | * it is not even remotely as accurate as 32-bit floats, and is not |
| 9 | * even guaranteed to produce even remotely correct results for code |
| 10 | * other than Speex. It makes all kinds of shortcuts that are acceptable |
| 11 | * for Speex, but may not be acceptable for your application. You're |
| 12 | * quite welcome to reuse this code and improve it, but don't assume |
| 13 | * it works out of the box. Most likely, it doesn't. |
| 14 | */ |
| 15 | /* |
| 16 | Redistribution and use in source and binary forms, with or without |
| 17 | modification, are permitted provided that the following conditions |
| 18 | are met: |
| 19 | |
| 20 | - Redistributions of source code must retain the above copyright |
| 21 | notice, this list of conditions and the following disclaimer. |
| 22 | |
| 23 | - Redistributions in binary form must reproduce the above copyright |
| 24 | notice, this list of conditions and the following disclaimer in the |
| 25 | documentation and/or other materials provided with the distribution. |
| 26 | |
| 27 | - Neither the name of the Xiph.org Foundation nor the names of its |
| 28 | contributors may be used to endorse or promote products derived from |
| 29 | this software without specific prior written permission. |
| 30 | |
| 31 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 32 | ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 33 | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 34 | A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR |
| 35 | CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
| 36 | EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| 37 | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
| 38 | PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF |
| 39 | LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING |
| 40 | NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
| 41 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 42 | */ |
| 43 | |
| 44 | #ifndef PSEUDOFLOAT_H |
| 45 | #define PSEUDOFLOAT_H |
| 46 | |
| 47 | #include "arch.h" |
| 48 | #include "os_support.h" |
| 49 | #include "math_approx.h" |
| 50 | #include <math.h> |
| 51 | |
| 52 | #ifdef FIXED_POINT |
| 53 | |
| 54 | typedef struct { |
| 55 | spx_int16_t m; |
| 56 | spx_int16_t e; |
| 57 | } spx_float_t; |
| 58 | |
| 59 | static const spx_float_t FLOAT_ZERO = {0,0}; |
| 60 | static const spx_float_t FLOAT_ONE = {16384,-14}; |
| 61 | static const spx_float_t FLOAT_HALF = {16384,-15}; |
| 62 | |
| 63 | #define MIN(a,b) ((a)<(b)?(a):(b)) |
| 64 | static inline spx_float_t PSEUDOFLOAT(spx_int32_t x) |
| 65 | { |
| 66 | int e=0; |
| 67 | int sign=0; |
| 68 | if (x<0) |
| 69 | { |
| 70 | sign = 1; |
| 71 | x = -x; |
| 72 | } |
| 73 | if (x==0) |
| 74 | { |
| 75 | spx_float_t r = {0,0}; |
| 76 | return r; |
| 77 | } |
| 78 | e = spx_ilog2(ABS32(x))-14; |
| 79 | x = VSHR32(x, e); |
| 80 | if (sign) |
| 81 | { |
| 82 | spx_float_t r; |
| 83 | r.m = -x; |
| 84 | r.e = e; |
| 85 | return r; |
| 86 | } |
| 87 | else |
| 88 | { |
| 89 | spx_float_t r; |
| 90 | r.m = x; |
| 91 | r.e = e; |
| 92 | return r; |
| 93 | } |
| 94 | } |
| 95 | |
| 96 | |
| 97 | static inline spx_float_t FLOAT_ADD(spx_float_t a, spx_float_t b) |
| 98 | { |
| 99 | spx_float_t r; |
| 100 | if (a.m==0) |
| 101 | return b; |
| 102 | else if (b.m==0) |
| 103 | return a; |
| 104 | if ((a).e > (b).e) |
| 105 | { |
| 106 | r.m = ((a).m>>1) + ((b).m>>MIN(15,(a).e-(b).e+1)); |
| 107 | r.e = (a).e+1; |
| 108 | } |
| 109 | else |
| 110 | { |
| 111 | r.m = ((b).m>>1) + ((a).m>>MIN(15,(b).e-(a).e+1)); |
| 112 | r.e = (b).e+1; |
| 113 | } |
| 114 | if (r.m>0) |
| 115 | { |
| 116 | if (r.m<16384) |
| 117 | { |
| 118 | r.m<<=1; |
| 119 | r.e-=1; |
| 120 | } |
| 121 | } else { |
| 122 | if (r.m>-16384) |
| 123 | { |
| 124 | r.m<<=1; |
| 125 | r.e-=1; |
| 126 | } |
| 127 | } |
| 128 | /*printf ("%f + %f = %f\n", REALFLOAT(a), REALFLOAT(b), REALFLOAT(r));*/ |
| 129 | return r; |
| 130 | } |
| 131 | |
| 132 | static inline spx_float_t FLOAT_SUB(spx_float_t a, spx_float_t b) |
| 133 | { |
| 134 | spx_float_t r; |
| 135 | if (a.m==0) |
| 136 | return b; |
| 137 | else if (b.m==0) |
| 138 | return a; |
| 139 | if ((a).e > (b).e) |
| 140 | { |
| 141 | r.m = ((a).m>>1) - ((b).m>>MIN(15,(a).e-(b).e+1)); |
| 142 | r.e = (a).e+1; |
| 143 | } |
| 144 | else |
| 145 | { |
| 146 | r.m = ((a).m>>MIN(15,(b).e-(a).e+1)) - ((b).m>>1); |
| 147 | r.e = (b).e+1; |
| 148 | } |
| 149 | if (r.m>0) |
| 150 | { |
| 151 | if (r.m<16384) |
| 152 | { |
| 153 | r.m<<=1; |
| 154 | r.e-=1; |
| 155 | } |
| 156 | } else { |
| 157 | if (r.m>-16384) |
| 158 | { |
| 159 | r.m<<=1; |
| 160 | r.e-=1; |
| 161 | } |
| 162 | } |
| 163 | /*printf ("%f + %f = %f\n", REALFLOAT(a), REALFLOAT(b), REALFLOAT(r));*/ |
| 164 | return r; |
| 165 | } |
| 166 | |
| 167 | static inline int FLOAT_LT(spx_float_t a, spx_float_t b) |
| 168 | { |
| 169 | if (a.m==0) |
| 170 | return b.m>0; |
| 171 | else if (b.m==0) |
| 172 | return a.m<0; |
| 173 | if ((a).e > (b).e) |
| 174 | return ((a).m>>1) < ((b).m>>MIN(15,(a).e-(b).e+1)); |
| 175 | else |
| 176 | return ((b).m>>1) > ((a).m>>MIN(15,(b).e-(a).e+1)); |
| 177 | |
| 178 | } |
| 179 | |
| 180 | static inline int FLOAT_GT(spx_float_t a, spx_float_t b) |
| 181 | { |
| 182 | return FLOAT_LT(b,a); |
| 183 | } |
| 184 | |
| 185 | static inline spx_float_t FLOAT_MULT(spx_float_t a, spx_float_t b) |
| 186 | { |
| 187 | spx_float_t r; |
| 188 | r.m = (spx_int16_t)((spx_int32_t)(a).m*(b).m>>15); |
| 189 | r.e = (a).e+(b).e+15; |
| 190 | if (r.m>0) |
| 191 | { |
| 192 | if (r.m<16384) |
| 193 | { |
| 194 | r.m<<=1; |
| 195 | r.e-=1; |
| 196 | } |
| 197 | } else { |
| 198 | if (r.m>-16384) |
| 199 | { |
| 200 | r.m<<=1; |
| 201 | r.e-=1; |
| 202 | } |
| 203 | } |
| 204 | /*printf ("%f * %f = %f\n", REALFLOAT(a), REALFLOAT(b), REALFLOAT(r));*/ |
| 205 | return r; |
| 206 | } |
| 207 | |
| 208 | static inline spx_float_t FLOAT_AMULT(spx_float_t a, spx_float_t b) |
| 209 | { |
| 210 | spx_float_t r; |
| 211 | r.m = (spx_int16_t)((spx_int32_t)(a).m*(b).m>>15); |
| 212 | r.e = (a).e+(b).e+15; |
| 213 | return r; |
| 214 | } |
| 215 | |
| 216 | |
| 217 | static inline spx_float_t FLOAT_SHL(spx_float_t a, int b) |
| 218 | { |
| 219 | spx_float_t r; |
| 220 | r.m = a.m; |
| 221 | r.e = a.e+b; |
| 222 | return r; |
| 223 | } |
| 224 | |
| 225 | static inline spx_int16_t FLOAT_EXTRACT16(spx_float_t a) |
| 226 | { |
| 227 | if (a.e<0) |
| 228 | return EXTRACT16((EXTEND32(a.m)+(EXTEND32(1)<<(-a.e-1)))>>-a.e); |
| 229 | else |
| 230 | return a.m<<a.e; |
| 231 | } |
| 232 | |
| 233 | static inline spx_int32_t FLOAT_EXTRACT32(spx_float_t a) |
| 234 | { |
| 235 | if (a.e<0) |
| 236 | return (EXTEND32(a.m)+(EXTEND32(1)<<(-a.e-1)))>>-a.e; |
| 237 | else |
| 238 | return EXTEND32(a.m)<<a.e; |
| 239 | } |
| 240 | |
| 241 | static inline spx_int32_t FLOAT_MUL32(spx_float_t a, spx_word32_t b) |
| 242 | { |
| 243 | return VSHR32(MULT16_32_Q15(a.m, b),-a.e-15); |
| 244 | } |
| 245 | |
| 246 | static inline spx_float_t FLOAT_MUL32U(spx_word32_t a, spx_word32_t b) |
| 247 | { |
| 248 | int e1, e2; |
| 249 | spx_float_t r; |
| 250 | if (a==0 || b==0) |
| 251 | { |
| 252 | return FLOAT_ZERO; |
| 253 | } |
| 254 | e1 = spx_ilog2(ABS32(a)); |
| 255 | a = VSHR32(a, e1-14); |
| 256 | e2 = spx_ilog2(ABS32(b)); |
| 257 | b = VSHR32(b, e2-14); |
| 258 | r.m = MULT16_16_Q15(a,b); |
| 259 | r.e = e1+e2-13; |
| 260 | return r; |
| 261 | } |
| 262 | |
| 263 | /* Do NOT attempt to divide by a negative number */ |
| 264 | static inline spx_float_t FLOAT_DIV32_FLOAT(spx_word32_t a, spx_float_t b) |
| 265 | { |
| 266 | int e=0; |
| 267 | spx_float_t r; |
| 268 | if (a==0) |
| 269 | { |
| 270 | return FLOAT_ZERO; |
| 271 | } |
| 272 | e = spx_ilog2(ABS32(a))-spx_ilog2(b.m-1)-15; |
| 273 | a = VSHR32(a, e); |
| 274 | if (ABS32(a)>=SHL32(EXTEND32(b.m-1),15)) |
| 275 | { |
| 276 | a >>= 1; |
| 277 | e++; |
| 278 | } |
| 279 | r.m = DIV32_16(a,b.m); |
| 280 | r.e = e-b.e; |
| 281 | return r; |
| 282 | } |
| 283 | |
| 284 | |
| 285 | /* Do NOT attempt to divide by a negative number */ |
| 286 | static inline spx_float_t FLOAT_DIV32(spx_word32_t a, spx_word32_t b) |
| 287 | { |
| 288 | int e0=0,e=0; |
| 289 | spx_float_t r; |
| 290 | if (a==0) |
| 291 | { |
| 292 | return FLOAT_ZERO; |
| 293 | } |
| 294 | if (b>32767) |
| 295 | { |
| 296 | e0 = spx_ilog2(b)-14; |
| 297 | b = VSHR32(b, e0); |
| 298 | e0 = -e0; |
| 299 | } |
| 300 | e = spx_ilog2(ABS32(a))-spx_ilog2(b-1)-15; |
| 301 | a = VSHR32(a, e); |
| 302 | if (ABS32(a)>=SHL32(EXTEND32(b-1),15)) |
| 303 | { |
| 304 | a >>= 1; |
| 305 | e++; |
| 306 | } |
| 307 | e += e0; |
| 308 | r.m = DIV32_16(a,b); |
| 309 | r.e = e; |
| 310 | return r; |
| 311 | } |
| 312 | |
| 313 | /* Do NOT attempt to divide by a negative number */ |
| 314 | static inline spx_float_t FLOAT_DIVU(spx_float_t a, spx_float_t b) |
| 315 | { |
| 316 | int e=0; |
| 317 | spx_int32_t num; |
| 318 | spx_float_t r; |
| 319 | if (b.m<=0) |
| 320 | { |
| 321 | speex_warning_int("Attempted to divide by", b.m); |
| 322 | return FLOAT_ONE; |
| 323 | } |
| 324 | num = a.m; |
| 325 | a.m = ABS16(a.m); |
| 326 | while (a.m >= b.m) |
| 327 | { |
| 328 | e++; |
| 329 | a.m >>= 1; |
| 330 | } |
| 331 | num = num << (15-e); |
| 332 | r.m = DIV32_16(num,b.m); |
| 333 | r.e = a.e-b.e-15+e; |
| 334 | return r; |
| 335 | } |
| 336 | |
| 337 | static inline spx_float_t FLOAT_SQRT(spx_float_t a) |
| 338 | { |
| 339 | spx_float_t r; |
| 340 | spx_int32_t m; |
| 341 | m = SHL32(EXTEND32(a.m), 14); |
| 342 | r.e = a.e - 14; |
| 343 | if (r.e & 1) |
| 344 | { |
| 345 | r.e -= 1; |
| 346 | m <<= 1; |
| 347 | } |
| 348 | r.e >>= 1; |
| 349 | r.m = spx_sqrt(m); |
| 350 | return r; |
| 351 | } |
| 352 | |
| 353 | #else |
| 354 | |
| 355 | #define spx_float_t float |
| 356 | #define FLOAT_ZERO 0.f |
| 357 | #define FLOAT_ONE 1.f |
| 358 | #define FLOAT_HALF 0.5f |
| 359 | #define PSEUDOFLOAT(x) (x) |
| 360 | #define FLOAT_MULT(a,b) ((a)*(b)) |
| 361 | #define FLOAT_AMULT(a,b) ((a)*(b)) |
| 362 | #define FLOAT_MUL32(a,b) ((a)*(b)) |
| 363 | #define FLOAT_DIV32(a,b) ((a)/(b)) |
| 364 | #define FLOAT_EXTRACT16(a) (a) |
| 365 | #define FLOAT_EXTRACT32(a) (a) |
| 366 | #define FLOAT_ADD(a,b) ((a)+(b)) |
| 367 | #define FLOAT_SUB(a,b) ((a)-(b)) |
| 368 | #define REALFLOAT(x) (x) |
| 369 | #define FLOAT_DIV32_FLOAT(a,b) ((a)/(b)) |
| 370 | #define FLOAT_MUL32U(a,b) ((a)*(b)) |
| 371 | #define FLOAT_SHL(a,b) (a) |
| 372 | #define FLOAT_LT(a,b) ((a)<(b)) |
| 373 | #define FLOAT_GT(a,b) ((a)>(b)) |
| 374 | #define FLOAT_DIVU(a,b) ((a)/(b)) |
| 375 | #define FLOAT_SQRT(a) (spx_sqrt(a)) |
| 376 | |
| 377 | #endif |
| 378 | |
| 379 | #endif |