1 /* 2 * Utility compute operations used by translated code. 3 * 4 * Copyright (c) 2003 Fabrice Bellard 5 * Copyright (c) 2007 Aurelien Jarno 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 26 #include "qemu/osdep.h" 27 #include "qemu/host-utils.h" 28 29 #ifndef CONFIG_INT128 30 /* Long integer helpers */ 31 static inline void mul64(uint64_t *plow, uint64_t *phigh, 32 uint64_t a, uint64_t b) 33 { 34 typedef union { 35 uint64_t ll; 36 struct { 37 #if HOST_BIG_ENDIAN 38 uint32_t high, low; 39 #else 40 uint32_t low, high; 41 #endif 42 } l; 43 } LL; 44 LL rl, rm, rn, rh, a0, b0; 45 uint64_t c; 46 47 a0.ll = a; 48 b0.ll = b; 49 50 rl.ll = (uint64_t)a0.l.low * b0.l.low; 51 rm.ll = (uint64_t)a0.l.low * b0.l.high; 52 rn.ll = (uint64_t)a0.l.high * b0.l.low; 53 rh.ll = (uint64_t)a0.l.high * b0.l.high; 54 55 c = (uint64_t)rl.l.high + rm.l.low + rn.l.low; 56 rl.l.high = c; 57 c >>= 32; 58 c = c + rm.l.high + rn.l.high + rh.l.low; 59 rh.l.low = c; 60 rh.l.high += (uint32_t)(c >> 32); 61 62 *plow = rl.ll; 63 *phigh = rh.ll; 64 } 65 66 /* Unsigned 64x64 -> 128 multiplication */ 67 void mulu64 (uint64_t *plow, uint64_t *phigh, uint64_t a, uint64_t b) 68 { 69 mul64(plow, phigh, a, b); 70 } 71 72 /* Signed 64x64 -> 128 multiplication */ 73 void muls64 (uint64_t *plow, uint64_t *phigh, int64_t a, int64_t b) 74 { 75 uint64_t rh; 76 77 mul64(plow, &rh, a, b); 78 79 /* Adjust for signs. */ 80 if (b < 0) { 81 rh -= a; 82 } 83 if (a < 0) { 84 rh -= b; 85 } 86 *phigh = rh; 87 } 88 89 /* 90 * Unsigned 128-by-64 division. 91 * Returns the remainder. 92 * Returns quotient via plow and phigh. 93 * Also returns the remainder via the function return value. 94 */ 95 uint64_t divu128(uint64_t *plow, uint64_t *phigh, uint64_t divisor) 96 { 97 uint64_t dhi = *phigh; 98 uint64_t dlo = *plow; 99 uint64_t rem, dhighest; 100 int sh; 101 102 if (divisor == 0 || dhi == 0) { 103 *plow = dlo / divisor; 104 *phigh = 0; 105 return dlo % divisor; 106 } else { 107 sh = clz64(divisor); 108 109 if (dhi < divisor) { 110 if (sh != 0) { 111 /* normalize the divisor, shifting the dividend accordingly */ 112 divisor <<= sh; 113 dhi = (dhi << sh) | (dlo >> (64 - sh)); 114 dlo <<= sh; 115 } 116 117 *phigh = 0; 118 *plow = udiv_qrnnd(&rem, dhi, dlo, divisor); 119 } else { 120 if (sh != 0) { 121 /* normalize the divisor, shifting the dividend accordingly */ 122 divisor <<= sh; 123 dhighest = dhi >> (64 - sh); 124 dhi = (dhi << sh) | (dlo >> (64 - sh)); 125 dlo <<= sh; 126 127 *phigh = udiv_qrnnd(&dhi, dhighest, dhi, divisor); 128 } else { 129 /** 130 * dhi >= divisor 131 * Since the MSB of divisor is set (sh == 0), 132 * (dhi - divisor) < divisor 133 * 134 * Thus, the high part of the quotient is 1, and we can 135 * calculate the low part with a single call to udiv_qrnnd 136 * after subtracting divisor from dhi 137 */ 138 dhi -= divisor; 139 *phigh = 1; 140 } 141 142 *plow = udiv_qrnnd(&rem, dhi, dlo, divisor); 143 } 144 145 /* 146 * since the dividend/divisor might have been normalized, 147 * the remainder might also have to be shifted back 148 */ 149 return rem >> sh; 150 } 151 } 152 153 /* 154 * Signed 128-by-64 division. 155 * Returns quotient via plow and phigh. 156 * Also returns the remainder via the function return value. 157 */ 158 int64_t divs128(uint64_t *plow, int64_t *phigh, int64_t divisor) 159 { 160 bool neg_quotient = false, neg_remainder = false; 161 uint64_t unsig_hi = *phigh, unsig_lo = *plow; 162 uint64_t rem; 163 164 if (*phigh < 0) { 165 neg_quotient = !neg_quotient; 166 neg_remainder = !neg_remainder; 167 168 if (unsig_lo == 0) { 169 unsig_hi = -unsig_hi; 170 } else { 171 unsig_hi = ~unsig_hi; 172 unsig_lo = -unsig_lo; 173 } 174 } 175 176 if (divisor < 0) { 177 neg_quotient = !neg_quotient; 178 179 divisor = -divisor; 180 } 181 182 rem = divu128(&unsig_lo, &unsig_hi, (uint64_t)divisor); 183 184 if (neg_quotient) { 185 if (unsig_lo == 0) { 186 *phigh = -unsig_hi; 187 *plow = 0; 188 } else { 189 *phigh = ~unsig_hi; 190 *plow = -unsig_lo; 191 } 192 } else { 193 *phigh = unsig_hi; 194 *plow = unsig_lo; 195 } 196 197 if (neg_remainder) { 198 return -rem; 199 } else { 200 return rem; 201 } 202 } 203 #endif 204 205 /** 206 * urshift - 128-bit Unsigned Right Shift. 207 * @plow: in/out - lower 64-bit integer. 208 * @phigh: in/out - higher 64-bit integer. 209 * @shift: in - bytes to shift, between 0 and 127. 210 * 211 * Result is zero-extended and stored in plow/phigh, which are 212 * input/output variables. Shift values outside the range will 213 * be mod to 128. In other words, the caller is responsible to 214 * verify/assert both the shift range and plow/phigh pointers. 215 */ 216 void urshift(uint64_t *plow, uint64_t *phigh, int32_t shift) 217 { 218 shift &= 127; 219 if (shift == 0) { 220 return; 221 } 222 223 uint64_t h = *phigh >> (shift & 63); 224 if (shift >= 64) { 225 *plow = h; 226 *phigh = 0; 227 } else { 228 *plow = (*plow >> (shift & 63)) | (*phigh << (64 - (shift & 63))); 229 *phigh = h; 230 } 231 } 232 233 /** 234 * ulshift - 128-bit Unsigned Left Shift. 235 * @plow: in/out - lower 64-bit integer. 236 * @phigh: in/out - higher 64-bit integer. 237 * @shift: in - bytes to shift, between 0 and 127. 238 * @overflow: out - true if any 1-bit is shifted out. 239 * 240 * Result is zero-extended and stored in plow/phigh, which are 241 * input/output variables. Shift values outside the range will 242 * be mod to 128. In other words, the caller is responsible to 243 * verify/assert both the shift range and plow/phigh pointers. 244 */ 245 void ulshift(uint64_t *plow, uint64_t *phigh, int32_t shift, bool *overflow) 246 { 247 uint64_t low = *plow; 248 uint64_t high = *phigh; 249 250 shift &= 127; 251 if (shift == 0) { 252 return; 253 } 254 255 /* check if any bit will be shifted out */ 256 urshift(&low, &high, 128 - shift); 257 if (low | high) { 258 *overflow = true; 259 } 260 261 if (shift >= 64) { 262 *phigh = *plow << (shift & 63); 263 *plow = 0; 264 } else { 265 *phigh = (*plow >> (64 - (shift & 63))) | (*phigh << (shift & 63)); 266 *plow = *plow << shift; 267 } 268 } 269 270 /* 271 * Unsigned 256-by-128 division. 272 * Returns the remainder via r. 273 * Returns lower 128 bit of quotient. 274 * Needs a normalized divisor (most significant bit set to 1). 275 * 276 * Adapted from include/qemu/host-utils.h udiv_qrnnd, 277 * from the GNU Multi Precision Library - longlong.h __udiv_qrnnd 278 * (https://gmplib.org/repo/gmp/file/tip/longlong.h) 279 * 280 * Licensed under the GPLv2/LGPLv3 281 */ 282 static Int128 udiv256_qrnnd(Int128 *r, Int128 n1, Int128 n0, Int128 d) 283 { 284 Int128 d0, d1, q0, q1, r1, r0, m; 285 uint64_t mp0, mp1; 286 287 d0 = int128_make64(int128_getlo(d)); 288 d1 = int128_make64(int128_gethi(d)); 289 290 r1 = int128_remu(n1, d1); 291 q1 = int128_divu(n1, d1); 292 mp0 = int128_getlo(q1); 293 mp1 = int128_gethi(q1); 294 mulu128(&mp0, &mp1, int128_getlo(d0)); 295 m = int128_make128(mp0, mp1); 296 r1 = int128_make128(int128_gethi(n0), int128_getlo(r1)); 297 if (int128_ult(r1, m)) { 298 q1 = int128_sub(q1, int128_one()); 299 r1 = int128_add(r1, d); 300 if (int128_uge(r1, d)) { 301 if (int128_ult(r1, m)) { 302 q1 = int128_sub(q1, int128_one()); 303 r1 = int128_add(r1, d); 304 } 305 } 306 } 307 r1 = int128_sub(r1, m); 308 309 r0 = int128_remu(r1, d1); 310 q0 = int128_divu(r1, d1); 311 mp0 = int128_getlo(q0); 312 mp1 = int128_gethi(q0); 313 mulu128(&mp0, &mp1, int128_getlo(d0)); 314 m = int128_make128(mp0, mp1); 315 r0 = int128_make128(int128_getlo(n0), int128_getlo(r0)); 316 if (int128_ult(r0, m)) { 317 q0 = int128_sub(q0, int128_one()); 318 r0 = int128_add(r0, d); 319 if (int128_uge(r0, d)) { 320 if (int128_ult(r0, m)) { 321 q0 = int128_sub(q0, int128_one()); 322 r0 = int128_add(r0, d); 323 } 324 } 325 } 326 r0 = int128_sub(r0, m); 327 328 *r = r0; 329 return int128_or(int128_lshift(q1, 64), q0); 330 } 331 332 /* 333 * Unsigned 256-by-128 division. 334 * Returns the remainder. 335 * Returns quotient via plow and phigh. 336 * Also returns the remainder via the function return value. 337 */ 338 Int128 divu256(Int128 *plow, Int128 *phigh, Int128 divisor) 339 { 340 Int128 dhi = *phigh; 341 Int128 dlo = *plow; 342 Int128 rem, dhighest; 343 int sh; 344 345 if (!int128_nz(divisor) || !int128_nz(dhi)) { 346 *plow = int128_divu(dlo, divisor); 347 *phigh = int128_zero(); 348 return int128_remu(dlo, divisor); 349 } else { 350 sh = clz128(divisor); 351 352 if (int128_ult(dhi, divisor)) { 353 if (sh != 0) { 354 /* normalize the divisor, shifting the dividend accordingly */ 355 divisor = int128_lshift(divisor, sh); 356 dhi = int128_or(int128_lshift(dhi, sh), 357 int128_urshift(dlo, (128 - sh))); 358 dlo = int128_lshift(dlo, sh); 359 } 360 361 *phigh = int128_zero(); 362 *plow = udiv256_qrnnd(&rem, dhi, dlo, divisor); 363 } else { 364 if (sh != 0) { 365 /* normalize the divisor, shifting the dividend accordingly */ 366 divisor = int128_lshift(divisor, sh); 367 dhighest = int128_rshift(dhi, (128 - sh)); 368 dhi = int128_or(int128_lshift(dhi, sh), 369 int128_urshift(dlo, (128 - sh))); 370 dlo = int128_lshift(dlo, sh); 371 372 *phigh = udiv256_qrnnd(&dhi, dhighest, dhi, divisor); 373 } else { 374 /* 375 * dhi >= divisor 376 * Since the MSB of divisor is set (sh == 0), 377 * (dhi - divisor) < divisor 378 * 379 * Thus, the high part of the quotient is 1, and we can 380 * calculate the low part with a single call to udiv_qrnnd 381 * after subtracting divisor from dhi 382 */ 383 dhi = int128_sub(dhi, divisor); 384 *phigh = int128_one(); 385 } 386 387 *plow = udiv256_qrnnd(&rem, dhi, dlo, divisor); 388 } 389 390 /* 391 * since the dividend/divisor might have been normalized, 392 * the remainder might also have to be shifted back 393 */ 394 rem = int128_urshift(rem, sh); 395 return rem; 396 } 397 } 398 399 /* 400 * Signed 256-by-128 division. 401 * Returns quotient via plow and phigh. 402 * Also returns the remainder via the function return value. 403 */ 404 Int128 divs256(Int128 *plow, Int128 *phigh, Int128 divisor) 405 { 406 bool neg_quotient = false, neg_remainder = false; 407 Int128 unsig_hi = *phigh, unsig_lo = *plow; 408 Int128 rem; 409 410 if (!int128_nonneg(*phigh)) { 411 neg_quotient = !neg_quotient; 412 neg_remainder = !neg_remainder; 413 414 if (!int128_nz(unsig_lo)) { 415 unsig_hi = int128_neg(unsig_hi); 416 } else { 417 unsig_hi = int128_not(unsig_hi); 418 unsig_lo = int128_neg(unsig_lo); 419 } 420 } 421 422 if (!int128_nonneg(divisor)) { 423 neg_quotient = !neg_quotient; 424 425 divisor = int128_neg(divisor); 426 } 427 428 rem = divu256(&unsig_lo, &unsig_hi, divisor); 429 430 if (neg_quotient) { 431 if (!int128_nz(unsig_lo)) { 432 *phigh = int128_neg(unsig_hi); 433 *plow = int128_zero(); 434 } else { 435 *phigh = int128_not(unsig_hi); 436 *plow = int128_neg(unsig_lo); 437 } 438 } else { 439 *phigh = unsig_hi; 440 *plow = unsig_lo; 441 } 442 443 if (neg_remainder) { 444 return int128_neg(rem); 445 } else { 446 return rem; 447 } 448 } 449