| 1 | #include <math/fixed.h> |
| 2 | #include <math/fixed_extended.h> |
| 3 | |
| 4 | fx32_32_t fx_exp(fx32_32_t x) { |
| 5 | int64_t k = fx_to_int(x: fx_mul(a: x, FX(1.44269504089))); |
| 6 | fx32_32_t k_ln2 = fx_mul(a: fx_from_int(x: k), FX(0.69314718056)); |
| 7 | fx32_32_t r = x - k_ln2; |
| 8 | static const fx32_32_t coeffs[] = {FX(1.0), FX(1.0), FX(0.5), FX(1.0 / 6.0), |
| 9 | FX(1.0 / 24.0)}; |
| 10 | fx32_32_t poly = fx_poly_eval(x: r, c: coeffs, n: 5); |
| 11 | if (k >= 0) |
| 12 | return poly << k; |
| 13 | else |
| 14 | return poly >> -k; |
| 15 | } |
| 16 | |
| 17 | fx32_32_t fx_poly_eval(fx32_32_t x, const fx32_32_t *c, int n) { |
| 18 | fx32_32_t r = c[n - 1]; |
| 19 | for (int i = n - 2; i >= 0; i--) { |
| 20 | r = fx_mul(a: r, b: x) + c[i]; |
| 21 | } |
| 22 | return r; |
| 23 | } |
| 24 | |
| 25 | static const fx32_32_t FX_LN2 = FX(0.69314718056); |
| 26 | |
| 27 | fx32_32_t fx_ln(fx32_32_t x) { |
| 28 | if (x <= 0) |
| 29 | return INT64_MIN; |
| 30 | int64_t k = 0; |
| 31 | /* Normalize x into [1,2) */ |
| 32 | while (x < FX(1.0)) { |
| 33 | x <<= 1; |
| 34 | k--; |
| 35 | } |
| 36 | while (x >= FX(2.0)) { |
| 37 | x >>= 1; |
| 38 | k++; |
| 39 | } |
| 40 | /* z = (x - 1) / (x + 1) */ |
| 41 | fx32_32_t num = x - FX(1.0); |
| 42 | fx32_32_t den = x + FX(1.0); |
| 43 | fx32_32_t z = fx_div(a: num, b: den); |
| 44 | /* Polynomial: z + z^3/3 + z^5/5 */ |
| 45 | fx32_32_t z2 = fx_mul(a: z, b: z); |
| 46 | fx32_32_t z3 = fx_mul(a: z2, b: z); |
| 47 | fx32_32_t z5 = fx_mul(a: z3, b: z2); |
| 48 | fx32_32_t series = z + fx_mul(a: z3, b: fx_div(FX(1.0), FX(3.0))) + |
| 49 | fx_mul(a: z5, b: fx_div(FX(1.0), FX(5.0))); |
| 50 | fx32_32_t ln_m = fx_mul(a: series, FX(2.0)); |
| 51 | return ln_m + fx_mul(a: fx_from_int(x: k), b: FX_LN2); |
| 52 | } |
| 53 | |
| 54 | static const fx32_32_t cordic_atan[] = { |
| 55 | FX(0.7853981633974483), FX(0.4636476090008061), FX(0.2449786631268641), |
| 56 | FX(0.1243549945467614), FX(0.0624188099959574), FX(0.0312398334302683), |
| 57 | FX(0.0156237286204768), FX(0.0078123410601011), FX(0.0039062301319669), |
| 58 | FX(0.0019531225164788), FX(0.0009765621895593), FX(0.0004882812111949), |
| 59 | FX(0.0002441406201494), FX(0.0001220703118937), FX(0.0000610351561742), |
| 60 | FX(0.0000305175781166), FX(0.0000152587890863), FX(0.0000076293945430), |
| 61 | FX(0.0000038146972715), FX(0.0000019073486358), FX(0.0000009536743179), |
| 62 | FX(0.0000004768371590), FX(0.0000002384185795), FX(0.0000001192092898), |
| 63 | FX(0.0000000596046449), FX(0.0000000298023224), FX(0.0000000149011612), |
| 64 | FX(0.0000000074505806), FX(0.0000000037252903), FX(0.0000000018626452), |
| 65 | FX(0.0000000009313226), FX(0.0000000004656613), |
| 66 | }; |
| 67 | |
| 68 | static const fx32_32_t FX_CORDIC_K = FX(0.6072529350088812561694); |
| 69 | |
| 70 | static void fx_cordic(fx32_32_t angle, fx32_32_t *out_sin, fx32_32_t *out_cos) { |
| 71 | fx32_32_t x = FX_CORDIC_K; |
| 72 | fx32_32_t y = 0; |
| 73 | fx32_32_t z = angle; |
| 74 | for (int i = 0; i < 32; i++) { |
| 75 | fx32_32_t x_new, y_new; |
| 76 | if (z >= 0) { |
| 77 | x_new = x - (y >> i); |
| 78 | y_new = y + (x >> i); |
| 79 | z -= cordic_atan[i]; |
| 80 | } else { |
| 81 | x_new = x + (y >> i); |
| 82 | y_new = y - (x >> i); |
| 83 | z += cordic_atan[i]; |
| 84 | } |
| 85 | x = x_new; |
| 86 | y = y_new; |
| 87 | } |
| 88 | *out_sin = y; |
| 89 | *out_cos = x; |
| 90 | } |
| 91 | |
| 92 | fx32_32_t fx_sin(fx32_32_t angle) { |
| 93 | fx32_32_t s, c; |
| 94 | fx_cordic(angle, out_sin: &s, out_cos: &c); |
| 95 | return s; |
| 96 | } |
| 97 | |
| 98 | fx32_32_t fx_cos(fx32_32_t angle) { |
| 99 | fx32_32_t s, c; |
| 100 | fx_cordic(angle, out_sin: &s, out_cos: &c); |
| 101 | return c; |
| 102 | } |
| 103 | |
| 104 | void fx_sincos(fx32_32_t angle, fx32_32_t *sin_out, fx32_32_t *cos_out) { |
| 105 | fx_cordic(angle, out_sin: sin_out, out_cos: cos_out); |
| 106 | } |
| 107 | |
| 108 | static fx32_32_t _fx_reduce_pi(fx32_32_t angle) { |
| 109 | static const fx32_32_t TWO_PI_R = FX(0.15915494309189535); |
| 110 | |
| 111 | int64_t cycles = fx_to_int(x: fx_mul(a: angle, b: TWO_PI_R)); |
| 112 | angle -= fx_mul(a: fx_from_int(x: cycles), FX(2.0 * FP_PI)); |
| 113 | |
| 114 | if (angle > FX_PI) |
| 115 | angle -= FX(2.0 * FP_PI); |
| 116 | if (angle < -FX_PI) |
| 117 | angle += FX(2.0 * FP_PI); |
| 118 | return angle; |
| 119 | } |
| 120 | |
| 121 | fx32_32_t fx_sin_t(fx32_32_t angle) { |
| 122 | |
| 123 | angle = _fx_reduce_pi(angle); |
| 124 | |
| 125 | int negate = 0; |
| 126 | if (angle < 0) { |
| 127 | angle = -angle; |
| 128 | negate ^= 1; |
| 129 | } |
| 130 | if (angle > FX(FP_PI / 2.0)) { |
| 131 | angle = FX_PI - angle; |
| 132 | } |
| 133 | |
| 134 | fx32_32_t x2 = fx_mul(a: angle, b: angle); |
| 135 | fx32_32_t x3 = fx_mul(a: x2, b: angle); |
| 136 | fx32_32_t x5 = fx_mul(a: x3, b: x2); |
| 137 | fx32_32_t x7 = fx_mul(a: x5, b: x2); |
| 138 | fx32_32_t x9 = fx_mul(a: x7, b: x2); |
| 139 | fx32_32_t x11 = fx_mul(a: x9, b: x2); |
| 140 | fx32_32_t x13 = fx_mul(a: x11, b: x2); |
| 141 | |
| 142 | fx32_32_t result = |
| 143 | angle - fx_mul(a: x3, FX(1.0 / 6.0)) + fx_mul(a: x5, FX(1.0 / 120.0)) - |
| 144 | fx_mul(a: x7, FX(1.0 / 5040.0)) + fx_mul(a: x9, FX(1.0 / 362880.0)) - |
| 145 | fx_mul(a: x11, FX(1.0 / 39916800.0)) + fx_mul(a: x13, FX(1.0 / 6227020800.0)); |
| 146 | |
| 147 | return negate ? -result : result; |
| 148 | } |
| 149 | |
| 150 | fx32_32_t fx_cos_t(fx32_32_t angle) { |
| 151 | |
| 152 | angle = _fx_reduce_pi(angle); |
| 153 | |
| 154 | /* fold: cos(-x) = cos(x); cos(pi - x) = -cos(x) */ |
| 155 | int negate = 0; |
| 156 | if (angle < 0) |
| 157 | angle = -angle; |
| 158 | if (angle > FX(FP_PI / 2.0)) { |
| 159 | angle = FX_PI - angle; |
| 160 | negate = 1; |
| 161 | } |
| 162 | |
| 163 | fx32_32_t x2 = fx_mul(a: angle, b: angle); |
| 164 | fx32_32_t x4 = fx_mul(a: x2, b: x2); |
| 165 | fx32_32_t x6 = fx_mul(a: x4, b: x2); |
| 166 | fx32_32_t x8 = fx_mul(a: x6, b: x2); |
| 167 | fx32_32_t x10 = fx_mul(a: x8, b: x2); |
| 168 | fx32_32_t x12 = fx_mul(a: x10, b: x2); |
| 169 | |
| 170 | fx32_32_t result = |
| 171 | FX(1.0) - fx_mul(a: x2, FX(1.0 / 2.0)) + fx_mul(a: x4, FX(1.0 / 24.0)) - |
| 172 | fx_mul(a: x6, FX(1.0 / 720.0)) + fx_mul(a: x8, FX(1.0 / 40320.0)) - |
| 173 | fx_mul(a: x10, FX(1.0 / 3628800.0)) + fx_mul(a: x12, FX(1.0 / 479001600.0)); |
| 174 | |
| 175 | return negate ? -result : result; |
| 176 | } |
| 177 | |
| 178 | void fx_sincos_t(fx32_32_t angle, fx32_32_t *sin_out, fx32_32_t *cos_out) { |
| 179 | angle = _fx_reduce_pi(angle); |
| 180 | |
| 181 | int sin_neg = 0, cos_neg = 0; |
| 182 | if (angle < 0) { |
| 183 | angle = -angle; |
| 184 | sin_neg ^= 1; |
| 185 | } |
| 186 | if (angle > FX(FP_PI / 2.0)) { |
| 187 | angle = FX_PI - angle; |
| 188 | cos_neg = 1; |
| 189 | } |
| 190 | |
| 191 | fx32_32_t x2 = fx_mul(a: angle, b: angle); |
| 192 | fx32_32_t x3 = fx_mul(a: x2, b: angle); |
| 193 | fx32_32_t x4 = fx_mul(a: x3, b: angle); |
| 194 | fx32_32_t x5 = fx_mul(a: x4, b: angle); |
| 195 | fx32_32_t x6 = fx_mul(a: x5, b: angle); |
| 196 | fx32_32_t x7 = fx_mul(a: x6, b: angle); |
| 197 | fx32_32_t x8 = fx_mul(a: x7, b: angle); |
| 198 | fx32_32_t x9 = fx_mul(a: x8, b: angle); |
| 199 | fx32_32_t x10 = fx_mul(a: x9, b: angle); |
| 200 | fx32_32_t x11 = fx_mul(a: x10, b: angle); |
| 201 | fx32_32_t x12 = fx_mul(a: x11, b: angle); |
| 202 | fx32_32_t x13 = fx_mul(a: x12, b: angle); |
| 203 | |
| 204 | fx32_32_t s = |
| 205 | angle - fx_mul(a: x3, FX(1.0 / 6.0)) + fx_mul(a: x5, FX(1.0 / 120.0)) - |
| 206 | fx_mul(a: x7, FX(1.0 / 5040.0)) + fx_mul(a: x9, FX(1.0 / 362880.0)) - |
| 207 | fx_mul(a: x11, FX(1.0 / 39916800.0)) + fx_mul(a: x13, FX(1.0 / 6227020800.0)); |
| 208 | |
| 209 | fx32_32_t c = |
| 210 | FX(1.0) - fx_mul(a: x2, FX(1.0 / 2.0)) + fx_mul(a: x4, FX(1.0 / 24.0)) - |
| 211 | fx_mul(a: x6, FX(1.0 / 720.0)) + fx_mul(a: x8, FX(1.0 / 40320.0)) - |
| 212 | fx_mul(a: x10, FX(1.0 / 3628800.0)) + fx_mul(a: x12, FX(1.0 / 479001600.0)); |
| 213 | |
| 214 | *sin_out = sin_neg ? -s : s; |
| 215 | *cos_out = cos_neg ? -c : c; |
| 216 | } |
| 217 | |
| 218 | fx32_32_t fx_ln_t(fx32_32_t x) { |
| 219 | static const fx32_32_t LN2 = FX(0.69314718055994530941); |
| 220 | |
| 221 | if (x <= 0) |
| 222 | return INT64_MIN; |
| 223 | |
| 224 | int64_t k = 0; |
| 225 | while (x < FX(1.0)) { |
| 226 | x <<= 1; |
| 227 | k--; |
| 228 | } |
| 229 | while (x >= FX(2.0)) { |
| 230 | x >>= 1; |
| 231 | k++; |
| 232 | } |
| 233 | |
| 234 | fx32_32_t z = fx_div(a: x - FX(1.0), b: x + FX(1.0)); |
| 235 | fx32_32_t z2 = fx_mul(a: z, b: z); |
| 236 | |
| 237 | fx32_32_t series = |
| 238 | FX(2.0 / 1.0) + |
| 239 | fx_mul( |
| 240 | a: z2, |
| 241 | FX(2.0 / 3.0) + |
| 242 | fx_mul( |
| 243 | a: z2, |
| 244 | FX(2.0 / 5.0) + |
| 245 | fx_mul( |
| 246 | a: z2, |
| 247 | FX(2.0 / 7.0) + |
| 248 | fx_mul( |
| 249 | a: z2, |
| 250 | FX(2.0 / 9.0) + |
| 251 | fx_mul( |
| 252 | a: z2, |
| 253 | FX(2.0 / 11.0) + |
| 254 | fx_mul( |
| 255 | a: z2, |
| 256 | FX(2.0 / 13.0) + |
| 257 | fx_mul(a: z2, |
| 258 | FX(2.0 / |
| 259 | 15.0)))))))); |
| 260 | |
| 261 | return fx_mul(a: z, b: series) + fx_mul(a: fx_from_int(x: k), b: LN2); |
| 262 | } |
| 263 | |
| 264 | fx32_32_t fx_exp_t(fx32_32_t x) { |
| 265 | static const fx32_32_t LN2 = FX(0.69314718055994530941); |
| 266 | static const fx32_32_t LOG2_E = FX(1.44269504088896340735); |
| 267 | |
| 268 | int64_t k = fx_to_int(x: fx_mul(a: x, b: LOG2_E)); |
| 269 | fx32_32_t r = x - fx_mul(a: fx_from_int(x: k), b: LN2); |
| 270 | |
| 271 | fx32_32_t poly = |
| 272 | FX(1.0) + |
| 273 | fx_mul( |
| 274 | a: r, |
| 275 | FX(1.0) + |
| 276 | fx_mul( |
| 277 | a: r, |
| 278 | FX(1.0 / 2.0) + |
| 279 | fx_mul( |
| 280 | a: r, |
| 281 | FX(1.0 / 6.0) + |
| 282 | fx_mul( |
| 283 | a: r, |
| 284 | FX(1.0 / 24.0) + |
| 285 | fx_mul( |
| 286 | a: r, |
| 287 | FX(1.0 / 120.0) + |
| 288 | fx_mul( |
| 289 | a: r, |
| 290 | FX(1.0 / 720.0) + |
| 291 | fx_mul( |
| 292 | a: r, |
| 293 | FX(1.0 / |
| 294 | 5040.0)))))))); |
| 295 | |
| 296 | if (k >= 0) |
| 297 | return poly << k; |
| 298 | else |
| 299 | return poly >> -k; |
| 300 | } |
| 301 | |
| 302 | fx32_32_t fx_parse(const char *str, char **endptr) { |
| 303 | const char *s = str; |
| 304 | |
| 305 | while (*s == ' ' || *s == '\t') |
| 306 | s++; |
| 307 | |
| 308 | bool neg = false; |
| 309 | if (*s == '+' || *s == '-') { |
| 310 | neg = (*s == '-'); |
| 311 | s++; |
| 312 | } |
| 313 | |
| 314 | bool any_digit = false; |
| 315 | |
| 316 | uint64_t ipart = 0; |
| 317 | while (*s >= '0' && *s <= '9') { |
| 318 | ipart = ipart * 10 + (uint64_t) (*s - '0'); |
| 319 | any_digit = true; |
| 320 | s++; |
| 321 | } |
| 322 | |
| 323 | uint64_t frac_num = 0; /* accumulated fractional digits */ |
| 324 | uint64_t frac_den = 1; /* 10^(fractional digit count), <= 10^18 */ |
| 325 | if (*s == '.') { |
| 326 | s++; |
| 327 | int frac_digits = 0; |
| 328 | while (*s >= '0' && *s <= '9') { |
| 329 | if (frac_digits < 18) { |
| 330 | frac_num = frac_num * 10 + (uint64_t) (*s - '0'); |
| 331 | frac_den *= 10; |
| 332 | frac_digits++; |
| 333 | } |
| 334 | any_digit = true; |
| 335 | s++; |
| 336 | } |
| 337 | } |
| 338 | |
| 339 | if (!any_digit) { |
| 340 | if (endptr) |
| 341 | *endptr = (char *) str; |
| 342 | return 0; |
| 343 | } |
| 344 | |
| 345 | fx32_32_t result = (fx32_32_t) (ipart << 32); |
| 346 | if (frac_den > 1) |
| 347 | result += (fx32_32_t) (((uint128_t) frac_num << 32) / frac_den); |
| 348 | |
| 349 | if (endptr) |
| 350 | *endptr = (char *) s; |
| 351 | |
| 352 | return neg ? -result : result; |
| 353 | } |
| 354 | |