1 /* longlong.h -- definitions for mixed size 32/64 bit arithmetic. 2 Copyright (C) 1991-2014 Free Software Foundation, Inc. 3 4 This file is part of the GNU C Library. 5 6 The GNU C Library is free software; you can redistribute it and/or 7 modify it under the terms of the GNU Lesser General Public 8 License as published by the Free Software Foundation; either 9 version 2.1 of the License, or (at your option) any later version. 10 11 In addition to the permissions in the GNU Lesser General Public 12 License, the Free Software Foundation gives you unlimited 13 permission to link the compiled version of this file into 14 combinations with other programs, and to distribute those 15 combinations without any restriction coming from the use of this 16 file. (The Lesser General Public License restrictions do apply in 17 other respects; for example, they cover modification of the file, 18 and distribution when not linked into a combine executable.) 19 20 The GNU C Library is distributed in the hope that it will be useful, 21 but WITHOUT ANY WARRANTY; without even the implied warranty of 22 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 23 Lesser General Public License for more details. 24 25 You should have received a copy of the GNU Lesser General Public 26 License along with the GNU C Library; if not, see 27 <http://www.gnu.org/licenses/>. */ 28 29 /* You have to define the following before including this file: 30 31 UWtype -- An unsigned type, default type for operations (typically a "word") 32 UHWtype -- An unsigned type, at least half the size of UWtype. 33 UDWtype -- An unsigned type, at least twice as large a UWtype 34 W_TYPE_SIZE -- size in bits of UWtype 35 36 UQItype -- Unsigned 8 bit type. 37 SItype, USItype -- Signed and unsigned 32 bit types. 38 DItype, UDItype -- Signed and unsigned 64 bit types. 39 40 On a 32 bit machine UWtype should typically be USItype; 41 on a 64 bit machine, UWtype should typically be UDItype. */ 42 43 #define __BITS4 (W_TYPE_SIZE / 4) 44 #define __ll_B ((UWtype) 1 << (W_TYPE_SIZE / 2)) 45 #define __ll_lowpart(t) ((UWtype) (t) & (__ll_B - 1)) 46 #define __ll_highpart(t) ((UWtype) (t) >> (W_TYPE_SIZE / 2)) 47 48 #ifndef W_TYPE_SIZE 49 #define W_TYPE_SIZE 32 50 #define UWtype USItype 51 #define UHWtype USItype 52 #define UDWtype UDItype 53 #endif 54 55 /* Used in glibc only. */ 56 #ifndef attribute_hidden 57 #define attribute_hidden 58 #endif 59 60 extern const UQItype __clz_tab[256] attribute_hidden; 61 62 /* Define auxiliary asm macros. 63 64 1) umul_ppmm(high_prod, low_prod, multiplier, multiplicand) multiplies two 65 UWtype integers MULTIPLIER and MULTIPLICAND, and generates a two UWtype 66 word product in HIGH_PROD and LOW_PROD. 67 68 2) __umulsidi3(a,b) multiplies two UWtype integers A and B, and returns a 69 UDWtype product. This is just a variant of umul_ppmm. 70 71 3) udiv_qrnnd(quotient, remainder, high_numerator, low_numerator, 72 denominator) divides a UDWtype, composed by the UWtype integers 73 HIGH_NUMERATOR and LOW_NUMERATOR, by DENOMINATOR and places the quotient 74 in QUOTIENT and the remainder in REMAINDER. HIGH_NUMERATOR must be less 75 than DENOMINATOR for correct operation. If, in addition, the most 76 significant bit of DENOMINATOR must be 1, then the pre-processor symbol 77 UDIV_NEEDS_NORMALIZATION is defined to 1. 78 79 4) sdiv_qrnnd(quotient, remainder, high_numerator, low_numerator, 80 denominator). Like udiv_qrnnd but the numbers are signed. The quotient 81 is rounded towards 0. 82 83 5) count_leading_zeros(count, x) counts the number of zero-bits from the 84 msb to the first nonzero bit in the UWtype X. This is the number of 85 steps X needs to be shifted left to set the msb. Undefined for X == 0, 86 unless the symbol COUNT_LEADING_ZEROS_0 is defined to some value. 87 88 6) count_trailing_zeros(count, x) like count_leading_zeros, but counts 89 from the least significant end. 90 91 7) add_ssaaaa(high_sum, low_sum, high_addend_1, low_addend_1, 92 high_addend_2, low_addend_2) adds two UWtype integers, composed by 93 HIGH_ADDEND_1 and LOW_ADDEND_1, and HIGH_ADDEND_2 and LOW_ADDEND_2 94 respectively. The result is placed in HIGH_SUM and LOW_SUM. Overflow 95 (i.e. carry out) is not stored anywhere, and is lost. 96 97 8) sub_ddmmss(high_difference, low_difference, high_minuend, low_minuend, 98 high_subtrahend, low_subtrahend) subtracts two two-word UWtype integers, 99 composed by HIGH_MINUEND_1 and LOW_MINUEND_1, and HIGH_SUBTRAHEND_2 and 100 LOW_SUBTRAHEND_2 respectively. The result is placed in HIGH_DIFFERENCE 101 and LOW_DIFFERENCE. Overflow (i.e. carry out) is not stored anywhere, 102 and is lost. 103 104 If any of these macros are left undefined for a particular CPU, 105 C macros are used. */ 106 107 /* The CPUs come in alphabetical order below. 108 109 Please add support for more CPUs here, or improve the current support 110 for the CPUs below! 111 (E.g. WE32100, IBM360.) */ 112 113 #if defined (__GNUC__) && !defined (NO_ASM) 114 115 /* We sometimes need to clobber "cc" with gcc2, but that would not be 116 understood by gcc1. Use cpp to avoid major code duplication. */ 117 #if __GNUC__ < 2 118 #define __CLOBBER_CC 119 #define __AND_CLOBBER_CC 120 #else /* __GNUC__ >= 2 */ 121 #define __CLOBBER_CC : "cc" 122 #define __AND_CLOBBER_CC , "cc" 123 #endif /* __GNUC__ < 2 */ 124 125 #if defined (__aarch64__) 126 127 #if W_TYPE_SIZE == 32 128 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X)) 129 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X)) 130 #define COUNT_LEADING_ZEROS_0 32 131 #endif /* W_TYPE_SIZE == 32 */ 132 133 #if W_TYPE_SIZE == 64 134 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clzll (X)) 135 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctzll (X)) 136 #define COUNT_LEADING_ZEROS_0 64 137 #endif /* W_TYPE_SIZE == 64 */ 138 139 #endif /* __aarch64__ */ 140 141 #if defined (__alpha) && W_TYPE_SIZE == 64 142 #define umul_ppmm(ph, pl, m0, m1) \ 143 do { \ 144 UDItype __m0 = (m0), __m1 = (m1); \ 145 (ph) = __builtin_alpha_umulh (__m0, __m1); \ 146 (pl) = __m0 * __m1; \ 147 } while (0) 148 #define UMUL_TIME 46 149 #ifndef LONGLONG_STANDALONE 150 #define udiv_qrnnd(q, r, n1, n0, d) \ 151 do { UDItype __r; \ 152 (q) = __udiv_qrnnd (&__r, (n1), (n0), (d)); \ 153 (r) = __r; \ 154 } while (0) 155 extern UDItype __udiv_qrnnd (UDItype *, UDItype, UDItype, UDItype); 156 #define UDIV_TIME 220 157 #endif /* LONGLONG_STANDALONE */ 158 #ifdef __alpha_cix__ 159 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X)) 160 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X)) 161 #define COUNT_LEADING_ZEROS_0 64 162 #else 163 #define count_leading_zeros(COUNT,X) \ 164 do { \ 165 UDItype __xr = (X), __t, __a; \ 166 __t = __builtin_alpha_cmpbge (0, __xr); \ 167 __a = __clz_tab[__t ^ 0xff] - 1; \ 168 __t = __builtin_alpha_extbl (__xr, __a); \ 169 (COUNT) = 64 - (__clz_tab[__t] + __a*8); \ 170 } while (0) 171 #define count_trailing_zeros(COUNT,X) \ 172 do { \ 173 UDItype __xr = (X), __t, __a; \ 174 __t = __builtin_alpha_cmpbge (0, __xr); \ 175 __t = ~__t & -~__t; \ 176 __a = ((__t & 0xCC) != 0) * 2; \ 177 __a += ((__t & 0xF0) != 0) * 4; \ 178 __a += ((__t & 0xAA) != 0); \ 179 __t = __builtin_alpha_extbl (__xr, __a); \ 180 __a <<= 3; \ 181 __t &= -__t; \ 182 __a += ((__t & 0xCC) != 0) * 2; \ 183 __a += ((__t & 0xF0) != 0) * 4; \ 184 __a += ((__t & 0xAA) != 0); \ 185 (COUNT) = __a; \ 186 } while (0) 187 #endif /* __alpha_cix__ */ 188 #endif /* __alpha */ 189 190 #if defined (__arc__) && W_TYPE_SIZE == 32 191 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 192 __asm__ ("add.f %1, %4, %5\n\tadc %0, %2, %3" \ 193 : "=r" ((USItype) (sh)), \ 194 "=&r" ((USItype) (sl)) \ 195 : "%r" ((USItype) (ah)), \ 196 "rIJ" ((USItype) (bh)), \ 197 "%r" ((USItype) (al)), \ 198 "rIJ" ((USItype) (bl))) 199 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 200 __asm__ ("sub.f %1, %4, %5\n\tsbc %0, %2, %3" \ 201 : "=r" ((USItype) (sh)), \ 202 "=&r" ((USItype) (sl)) \ 203 : "r" ((USItype) (ah)), \ 204 "rIJ" ((USItype) (bh)), \ 205 "r" ((USItype) (al)), \ 206 "rIJ" ((USItype) (bl))) 207 208 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v) 209 #ifdef __ARC_NORM__ 210 #define count_leading_zeros(count, x) \ 211 do \ 212 { \ 213 SItype c_; \ 214 \ 215 __asm__ ("norm.f\t%0,%1\n\tmov.mi\t%0,-1" : "=r" (c_) : "r" (x) : "cc");\ 216 (count) = c_ + 1; \ 217 } \ 218 while (0) 219 #define COUNT_LEADING_ZEROS_0 32 220 #endif 221 #endif 222 223 #if defined (__arm__) && (defined (__thumb2__) || !defined (__thumb__)) \ 224 && W_TYPE_SIZE == 32 225 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 226 __asm__ ("adds %1, %4, %5\n\tadc %0, %2, %3" \ 227 : "=r" ((USItype) (sh)), \ 228 "=&r" ((USItype) (sl)) \ 229 : "%r" ((USItype) (ah)), \ 230 "rI" ((USItype) (bh)), \ 231 "%r" ((USItype) (al)), \ 232 "rI" ((USItype) (bl)) __CLOBBER_CC) 233 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 234 __asm__ ("subs %1, %4, %5\n\tsbc %0, %2, %3" \ 235 : "=r" ((USItype) (sh)), \ 236 "=&r" ((USItype) (sl)) \ 237 : "r" ((USItype) (ah)), \ 238 "rI" ((USItype) (bh)), \ 239 "r" ((USItype) (al)), \ 240 "rI" ((USItype) (bl)) __CLOBBER_CC) 241 # if defined(__ARM_ARCH_2__) || defined(__ARM_ARCH_2A__) \ 242 || defined(__ARM_ARCH_3__) 243 # define umul_ppmm(xh, xl, a, b) \ 244 do { \ 245 register USItype __t0, __t1, __t2; \ 246 __asm__ ("%@ Inlined umul_ppmm\n" \ 247 " mov %2, %5, lsr #16\n" \ 248 " mov %0, %6, lsr #16\n" \ 249 " bic %3, %5, %2, lsl #16\n" \ 250 " bic %4, %6, %0, lsl #16\n" \ 251 " mul %1, %3, %4\n" \ 252 " mul %4, %2, %4\n" \ 253 " mul %3, %0, %3\n" \ 254 " mul %0, %2, %0\n" \ 255 " adds %3, %4, %3\n" \ 256 " addcs %0, %0, #65536\n" \ 257 " adds %1, %1, %3, lsl #16\n" \ 258 " adc %0, %0, %3, lsr #16" \ 259 : "=&r" ((USItype) (xh)), \ 260 "=r" ((USItype) (xl)), \ 261 "=&r" (__t0), "=&r" (__t1), "=r" (__t2) \ 262 : "r" ((USItype) (a)), \ 263 "r" ((USItype) (b)) __CLOBBER_CC ); \ 264 } while (0) 265 # define UMUL_TIME 20 266 # else 267 # define umul_ppmm(xh, xl, a, b) \ 268 do { \ 269 /* Generate umull, under compiler control. */ \ 270 register UDItype __t0 = (UDItype)(USItype)(a) * (USItype)(b); \ 271 (xl) = (USItype)__t0; \ 272 (xh) = (USItype)(__t0 >> 32); \ 273 } while (0) 274 # define UMUL_TIME 3 275 # endif 276 # define UDIV_TIME 100 277 #endif /* __arm__ */ 278 279 #if defined(__arm__) 280 /* Let gcc decide how best to implement count_leading_zeros. */ 281 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X)) 282 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X)) 283 #define COUNT_LEADING_ZEROS_0 32 284 #endif 285 286 #if defined (__AVR__) 287 288 #if W_TYPE_SIZE == 16 289 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X)) 290 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctz (X)) 291 #define COUNT_LEADING_ZEROS_0 16 292 #endif /* W_TYPE_SIZE == 16 */ 293 294 #if W_TYPE_SIZE == 32 295 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzl (X)) 296 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzl (X)) 297 #define COUNT_LEADING_ZEROS_0 32 298 #endif /* W_TYPE_SIZE == 32 */ 299 300 #if W_TYPE_SIZE == 64 301 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clzll (X)) 302 #define count_trailing_zeros(COUNT,X) ((COUNT) = __builtin_ctzll (X)) 303 #define COUNT_LEADING_ZEROS_0 64 304 #endif /* W_TYPE_SIZE == 64 */ 305 306 #endif /* defined (__AVR__) */ 307 308 #if defined (__CRIS__) 309 310 #if __CRIS_arch_version >= 3 311 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X)) 312 #define COUNT_LEADING_ZEROS_0 32 313 #endif /* __CRIS_arch_version >= 3 */ 314 315 #if __CRIS_arch_version >= 8 316 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X)) 317 #endif /* __CRIS_arch_version >= 8 */ 318 319 #if __CRIS_arch_version >= 10 320 #define __umulsidi3(u,v) ((UDItype)(USItype) (u) * (UDItype)(USItype) (v)) 321 #else 322 #define __umulsidi3 __umulsidi3 323 extern UDItype __umulsidi3 (USItype, USItype); 324 #endif /* __CRIS_arch_version >= 10 */ 325 326 #define umul_ppmm(w1, w0, u, v) \ 327 do { \ 328 UDItype __x = __umulsidi3 (u, v); \ 329 (w0) = (USItype) (__x); \ 330 (w1) = (USItype) (__x >> 32); \ 331 } while (0) 332 333 /* FIXME: defining add_ssaaaa and sub_ddmmss should be advantageous for 334 DFmode ("double" intrinsics, avoiding two of the three insns handling 335 carry), but defining them as open-code C composing and doing the 336 operation in DImode (UDImode) shows that the DImode needs work: 337 register pressure from requiring neighboring registers and the 338 traffic to and from them come to dominate, in the 4.7 series. */ 339 340 #endif /* defined (__CRIS__) */ 341 342 #if defined (__hppa) && W_TYPE_SIZE == 32 343 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 344 __asm__ ("add %4,%5,%1\n\taddc %2,%3,%0" \ 345 : "=r" ((USItype) (sh)), \ 346 "=&r" ((USItype) (sl)) \ 347 : "%rM" ((USItype) (ah)), \ 348 "rM" ((USItype) (bh)), \ 349 "%rM" ((USItype) (al)), \ 350 "rM" ((USItype) (bl))) 351 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 352 __asm__ ("sub %4,%5,%1\n\tsubb %2,%3,%0" \ 353 : "=r" ((USItype) (sh)), \ 354 "=&r" ((USItype) (sl)) \ 355 : "rM" ((USItype) (ah)), \ 356 "rM" ((USItype) (bh)), \ 357 "rM" ((USItype) (al)), \ 358 "rM" ((USItype) (bl))) 359 #if defined (_PA_RISC1_1) 360 #define umul_ppmm(w1, w0, u, v) \ 361 do { \ 362 union \ 363 { \ 364 UDItype __f; \ 365 struct {USItype __w1, __w0;} __w1w0; \ 366 } __t; \ 367 __asm__ ("xmpyu %1,%2,%0" \ 368 : "=x" (__t.__f) \ 369 : "x" ((USItype) (u)), \ 370 "x" ((USItype) (v))); \ 371 (w1) = __t.__w1w0.__w1; \ 372 (w0) = __t.__w1w0.__w0; \ 373 } while (0) 374 #define UMUL_TIME 8 375 #else 376 #define UMUL_TIME 30 377 #endif 378 #define UDIV_TIME 40 379 #define count_leading_zeros(count, x) \ 380 do { \ 381 USItype __tmp; \ 382 __asm__ ( \ 383 "ldi 1,%0\n" \ 384 " extru,= %1,15,16,%%r0 ; Bits 31..16 zero?\n" \ 385 " extru,tr %1,15,16,%1 ; No. Shift down, skip add.\n"\ 386 " ldo 16(%0),%0 ; Yes. Perform add.\n" \ 387 " extru,= %1,23,8,%%r0 ; Bits 15..8 zero?\n" \ 388 " extru,tr %1,23,8,%1 ; No. Shift down, skip add.\n"\ 389 " ldo 8(%0),%0 ; Yes. Perform add.\n" \ 390 " extru,= %1,27,4,%%r0 ; Bits 7..4 zero?\n" \ 391 " extru,tr %1,27,4,%1 ; No. Shift down, skip add.\n"\ 392 " ldo 4(%0),%0 ; Yes. Perform add.\n" \ 393 " extru,= %1,29,2,%%r0 ; Bits 3..2 zero?\n" \ 394 " extru,tr %1,29,2,%1 ; No. Shift down, skip add.\n"\ 395 " ldo 2(%0),%0 ; Yes. Perform add.\n" \ 396 " extru %1,30,1,%1 ; Extract bit 1.\n" \ 397 " sub %0,%1,%0 ; Subtract it.\n" \ 398 : "=r" (count), "=r" (__tmp) : "1" (x)); \ 399 } while (0) 400 #endif 401 402 #if (defined (__i370__) || defined (__s390__) || defined (__mvs__)) && W_TYPE_SIZE == 32 403 #if !defined (__zarch__) 404 #define smul_ppmm(xh, xl, m0, m1) \ 405 do { \ 406 union {DItype __ll; \ 407 struct {USItype __h, __l;} __i; \ 408 } __x; \ 409 __asm__ ("lr %N0,%1\n\tmr %0,%2" \ 410 : "=&r" (__x.__ll) \ 411 : "r" (m0), "r" (m1)); \ 412 (xh) = __x.__i.__h; (xl) = __x.__i.__l; \ 413 } while (0) 414 #define sdiv_qrnnd(q, r, n1, n0, d) \ 415 do { \ 416 union {DItype __ll; \ 417 struct {USItype __h, __l;} __i; \ 418 } __x; \ 419 __x.__i.__h = n1; __x.__i.__l = n0; \ 420 __asm__ ("dr %0,%2" \ 421 : "=r" (__x.__ll) \ 422 : "0" (__x.__ll), "r" (d)); \ 423 (q) = __x.__i.__l; (r) = __x.__i.__h; \ 424 } while (0) 425 #else 426 #define smul_ppmm(xh, xl, m0, m1) \ 427 do { \ 428 register SItype __r0 __asm__ ("0"); \ 429 register SItype __r1 __asm__ ("1") = (m0); \ 430 \ 431 __asm__ ("mr\t%%r0,%3" \ 432 : "=r" (__r0), "=r" (__r1) \ 433 : "r" (__r1), "r" (m1)); \ 434 (xh) = __r0; (xl) = __r1; \ 435 } while (0) 436 437 #define sdiv_qrnnd(q, r, n1, n0, d) \ 438 do { \ 439 register SItype __r0 __asm__ ("0") = (n1); \ 440 register SItype __r1 __asm__ ("1") = (n0); \ 441 \ 442 __asm__ ("dr\t%%r0,%4" \ 443 : "=r" (__r0), "=r" (__r1) \ 444 : "r" (__r0), "r" (__r1), "r" (d)); \ 445 (q) = __r1; (r) = __r0; \ 446 } while (0) 447 #endif /* __zarch__ */ 448 #endif 449 450 #if (defined (__i386__) || defined (__i486__)) && W_TYPE_SIZE == 32 451 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 452 __asm__ ("add{l} {%5,%1|%1,%5}\n\tadc{l} {%3,%0|%0,%3}" \ 453 : "=r" ((USItype) (sh)), \ 454 "=&r" ((USItype) (sl)) \ 455 : "%0" ((USItype) (ah)), \ 456 "g" ((USItype) (bh)), \ 457 "%1" ((USItype) (al)), \ 458 "g" ((USItype) (bl))) 459 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 460 __asm__ ("sub{l} {%5,%1|%1,%5}\n\tsbb{l} {%3,%0|%0,%3}" \ 461 : "=r" ((USItype) (sh)), \ 462 "=&r" ((USItype) (sl)) \ 463 : "0" ((USItype) (ah)), \ 464 "g" ((USItype) (bh)), \ 465 "1" ((USItype) (al)), \ 466 "g" ((USItype) (bl))) 467 #define umul_ppmm(w1, w0, u, v) \ 468 __asm__ ("mul{l} %3" \ 469 : "=a" ((USItype) (w0)), \ 470 "=d" ((USItype) (w1)) \ 471 : "%0" ((USItype) (u)), \ 472 "rm" ((USItype) (v))) 473 #define udiv_qrnnd(q, r, n1, n0, dv) \ 474 __asm__ ("div{l} %4" \ 475 : "=a" ((USItype) (q)), \ 476 "=d" ((USItype) (r)) \ 477 : "0" ((USItype) (n0)), \ 478 "1" ((USItype) (n1)), \ 479 "rm" ((USItype) (dv))) 480 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x)) 481 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x)) 482 #define UMUL_TIME 40 483 #define UDIV_TIME 40 484 #endif /* 80x86 */ 485 486 #if defined (__x86_64__) && W_TYPE_SIZE == 64 487 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 488 __asm__ ("add{q} {%5,%1|%1,%5}\n\tadc{q} {%3,%0|%0,%3}" \ 489 : "=r" ((UDItype) (sh)), \ 490 "=&r" ((UDItype) (sl)) \ 491 : "%0" ((UDItype) (ah)), \ 492 "rme" ((UDItype) (bh)), \ 493 "%1" ((UDItype) (al)), \ 494 "rme" ((UDItype) (bl))) 495 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 496 __asm__ ("sub{q} {%5,%1|%1,%5}\n\tsbb{q} {%3,%0|%0,%3}" \ 497 : "=r" ((UDItype) (sh)), \ 498 "=&r" ((UDItype) (sl)) \ 499 : "0" ((UDItype) (ah)), \ 500 "rme" ((UDItype) (bh)), \ 501 "1" ((UDItype) (al)), \ 502 "rme" ((UDItype) (bl))) 503 #define umul_ppmm(w1, w0, u, v) \ 504 __asm__ ("mul{q} %3" \ 505 : "=a" ((UDItype) (w0)), \ 506 "=d" ((UDItype) (w1)) \ 507 : "%0" ((UDItype) (u)), \ 508 "rm" ((UDItype) (v))) 509 #define udiv_qrnnd(q, r, n1, n0, dv) \ 510 __asm__ ("div{q} %4" \ 511 : "=a" ((UDItype) (q)), \ 512 "=d" ((UDItype) (r)) \ 513 : "0" ((UDItype) (n0)), \ 514 "1" ((UDItype) (n1)), \ 515 "rm" ((UDItype) (dv))) 516 #define count_leading_zeros(count, x) ((count) = __builtin_clzll (x)) 517 #define count_trailing_zeros(count, x) ((count) = __builtin_ctzll (x)) 518 #define UMUL_TIME 40 519 #define UDIV_TIME 40 520 #endif /* x86_64 */ 521 522 #if defined (__i960__) && W_TYPE_SIZE == 32 523 #define umul_ppmm(w1, w0, u, v) \ 524 ({union {UDItype __ll; \ 525 struct {USItype __l, __h;} __i; \ 526 } __xx; \ 527 __asm__ ("emul %2,%1,%0" \ 528 : "=d" (__xx.__ll) \ 529 : "%dI" ((USItype) (u)), \ 530 "dI" ((USItype) (v))); \ 531 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;}) 532 #define __umulsidi3(u, v) \ 533 ({UDItype __w; \ 534 __asm__ ("emul %2,%1,%0" \ 535 : "=d" (__w) \ 536 : "%dI" ((USItype) (u)), \ 537 "dI" ((USItype) (v))); \ 538 __w; }) 539 #endif /* __i960__ */ 540 541 #if defined (__ia64) && W_TYPE_SIZE == 64 542 /* This form encourages gcc (pre-release 3.4 at least) to emit predicated 543 "sub r=r,r" and "sub r=r,r,1", giving a 2 cycle latency. The generic 544 code using "al<bl" arithmetically comes out making an actual 0 or 1 in a 545 register, which takes an extra cycle. */ 546 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 547 do { \ 548 UWtype __x; \ 549 __x = (al) - (bl); \ 550 if ((al) < (bl)) \ 551 (sh) = (ah) - (bh) - 1; \ 552 else \ 553 (sh) = (ah) - (bh); \ 554 (sl) = __x; \ 555 } while (0) 556 557 /* Do both product parts in assembly, since that gives better code with 558 all gcc versions. Some callers will just use the upper part, and in 559 that situation we waste an instruction, but not any cycles. */ 560 #define umul_ppmm(ph, pl, m0, m1) \ 561 __asm__ ("xma.hu %0 = %2, %3, f0\n\txma.l %1 = %2, %3, f0" \ 562 : "=&f" (ph), "=f" (pl) \ 563 : "f" (m0), "f" (m1)) 564 #define count_leading_zeros(count, x) \ 565 do { \ 566 UWtype _x = (x), _y, _a, _c; \ 567 __asm__ ("mux1 %0 = %1, @rev" : "=r" (_y) : "r" (_x)); \ 568 __asm__ ("czx1.l %0 = %1" : "=r" (_a) : "r" (-_y | _y)); \ 569 _c = (_a - 1) << 3; \ 570 _x >>= _c; \ 571 if (_x >= 1 << 4) \ 572 _x >>= 4, _c += 4; \ 573 if (_x >= 1 << 2) \ 574 _x >>= 2, _c += 2; \ 575 _c += _x >> 1; \ 576 (count) = W_TYPE_SIZE - 1 - _c; \ 577 } while (0) 578 /* similar to what gcc does for __builtin_ffs, but 0 based rather than 1 579 based, and we don't need a special case for x==0 here */ 580 #define count_trailing_zeros(count, x) \ 581 do { \ 582 UWtype __ctz_x = (x); \ 583 __asm__ ("popcnt %0 = %1" \ 584 : "=r" (count) \ 585 : "r" ((__ctz_x-1) & ~__ctz_x)); \ 586 } while (0) 587 #define UMUL_TIME 14 588 #endif 589 590 #if defined (__M32R__) && W_TYPE_SIZE == 32 591 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 592 /* The cmp clears the condition bit. */ \ 593 __asm__ ("cmp %0,%0\n\taddx %1,%5\n\taddx %0,%3" \ 594 : "=r" ((USItype) (sh)), \ 595 "=&r" ((USItype) (sl)) \ 596 : "0" ((USItype) (ah)), \ 597 "r" ((USItype) (bh)), \ 598 "1" ((USItype) (al)), \ 599 "r" ((USItype) (bl)) \ 600 : "cbit") 601 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 602 /* The cmp clears the condition bit. */ \ 603 __asm__ ("cmp %0,%0\n\tsubx %1,%5\n\tsubx %0,%3" \ 604 : "=r" ((USItype) (sh)), \ 605 "=&r" ((USItype) (sl)) \ 606 : "0" ((USItype) (ah)), \ 607 "r" ((USItype) (bh)), \ 608 "1" ((USItype) (al)), \ 609 "r" ((USItype) (bl)) \ 610 : "cbit") 611 #endif /* __M32R__ */ 612 613 #if defined (__mc68000__) && W_TYPE_SIZE == 32 614 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 615 __asm__ ("add%.l %5,%1\n\taddx%.l %3,%0" \ 616 : "=d" ((USItype) (sh)), \ 617 "=&d" ((USItype) (sl)) \ 618 : "%0" ((USItype) (ah)), \ 619 "d" ((USItype) (bh)), \ 620 "%1" ((USItype) (al)), \ 621 "g" ((USItype) (bl))) 622 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 623 __asm__ ("sub%.l %5,%1\n\tsubx%.l %3,%0" \ 624 : "=d" ((USItype) (sh)), \ 625 "=&d" ((USItype) (sl)) \ 626 : "0" ((USItype) (ah)), \ 627 "d" ((USItype) (bh)), \ 628 "1" ((USItype) (al)), \ 629 "g" ((USItype) (bl))) 630 631 /* The '020, '030, '040, '060 and CPU32 have 32x32->64 and 64/32->32q-32r. */ 632 #if (defined (__mc68020__) && !defined (__mc68060__)) 633 #define umul_ppmm(w1, w0, u, v) \ 634 __asm__ ("mulu%.l %3,%1:%0" \ 635 : "=d" ((USItype) (w0)), \ 636 "=d" ((USItype) (w1)) \ 637 : "%0" ((USItype) (u)), \ 638 "dmi" ((USItype) (v))) 639 #define UMUL_TIME 45 640 #define udiv_qrnnd(q, r, n1, n0, d) \ 641 __asm__ ("divu%.l %4,%1:%0" \ 642 : "=d" ((USItype) (q)), \ 643 "=d" ((USItype) (r)) \ 644 : "0" ((USItype) (n0)), \ 645 "1" ((USItype) (n1)), \ 646 "dmi" ((USItype) (d))) 647 #define UDIV_TIME 90 648 #define sdiv_qrnnd(q, r, n1, n0, d) \ 649 __asm__ ("divs%.l %4,%1:%0" \ 650 : "=d" ((USItype) (q)), \ 651 "=d" ((USItype) (r)) \ 652 : "0" ((USItype) (n0)), \ 653 "1" ((USItype) (n1)), \ 654 "dmi" ((USItype) (d))) 655 656 #elif defined (__mcoldfire__) /* not mc68020 */ 657 658 #define umul_ppmm(xh, xl, a, b) \ 659 __asm__ ("| Inlined umul_ppmm\n" \ 660 " move%.l %2,%/d0\n" \ 661 " move%.l %3,%/d1\n" \ 662 " move%.l %/d0,%/d2\n" \ 663 " swap %/d0\n" \ 664 " move%.l %/d1,%/d3\n" \ 665 " swap %/d1\n" \ 666 " move%.w %/d2,%/d4\n" \ 667 " mulu %/d3,%/d4\n" \ 668 " mulu %/d1,%/d2\n" \ 669 " mulu %/d0,%/d3\n" \ 670 " mulu %/d0,%/d1\n" \ 671 " move%.l %/d4,%/d0\n" \ 672 " clr%.w %/d0\n" \ 673 " swap %/d0\n" \ 674 " add%.l %/d0,%/d2\n" \ 675 " add%.l %/d3,%/d2\n" \ 676 " jcc 1f\n" \ 677 " add%.l %#65536,%/d1\n" \ 678 "1: swap %/d2\n" \ 679 " moveq %#0,%/d0\n" \ 680 " move%.w %/d2,%/d0\n" \ 681 " move%.w %/d4,%/d2\n" \ 682 " move%.l %/d2,%1\n" \ 683 " add%.l %/d1,%/d0\n" \ 684 " move%.l %/d0,%0" \ 685 : "=g" ((USItype) (xh)), \ 686 "=g" ((USItype) (xl)) \ 687 : "g" ((USItype) (a)), \ 688 "g" ((USItype) (b)) \ 689 : "d0", "d1", "d2", "d3", "d4") 690 #define UMUL_TIME 100 691 #define UDIV_TIME 400 692 #else /* not ColdFire */ 693 /* %/ inserts REGISTER_PREFIX, %# inserts IMMEDIATE_PREFIX. */ 694 #define umul_ppmm(xh, xl, a, b) \ 695 __asm__ ("| Inlined umul_ppmm\n" \ 696 " move%.l %2,%/d0\n" \ 697 " move%.l %3,%/d1\n" \ 698 " move%.l %/d0,%/d2\n" \ 699 " swap %/d0\n" \ 700 " move%.l %/d1,%/d3\n" \ 701 " swap %/d1\n" \ 702 " move%.w %/d2,%/d4\n" \ 703 " mulu %/d3,%/d4\n" \ 704 " mulu %/d1,%/d2\n" \ 705 " mulu %/d0,%/d3\n" \ 706 " mulu %/d0,%/d1\n" \ 707 " move%.l %/d4,%/d0\n" \ 708 " eor%.w %/d0,%/d0\n" \ 709 " swap %/d0\n" \ 710 " add%.l %/d0,%/d2\n" \ 711 " add%.l %/d3,%/d2\n" \ 712 " jcc 1f\n" \ 713 " add%.l %#65536,%/d1\n" \ 714 "1: swap %/d2\n" \ 715 " moveq %#0,%/d0\n" \ 716 " move%.w %/d2,%/d0\n" \ 717 " move%.w %/d4,%/d2\n" \ 718 " move%.l %/d2,%1\n" \ 719 " add%.l %/d1,%/d0\n" \ 720 " move%.l %/d0,%0" \ 721 : "=g" ((USItype) (xh)), \ 722 "=g" ((USItype) (xl)) \ 723 : "g" ((USItype) (a)), \ 724 "g" ((USItype) (b)) \ 725 : "d0", "d1", "d2", "d3", "d4") 726 #define UMUL_TIME 100 727 #define UDIV_TIME 400 728 729 #endif /* not mc68020 */ 730 731 /* The '020, '030, '040 and '060 have bitfield insns. 732 cpu32 disguises as a 68020, but lacks them. */ 733 #if defined (__mc68020__) && !defined (__mcpu32__) 734 #define count_leading_zeros(count, x) \ 735 __asm__ ("bfffo %1{%b2:%b2},%0" \ 736 : "=d" ((USItype) (count)) \ 737 : "od" ((USItype) (x)), "n" (0)) 738 /* Some ColdFire architectures have a ff1 instruction supported via 739 __builtin_clz. */ 740 #elif defined (__mcfisaaplus__) || defined (__mcfisac__) 741 #define count_leading_zeros(count,x) ((count) = __builtin_clz (x)) 742 #define COUNT_LEADING_ZEROS_0 32 743 #endif 744 #endif /* mc68000 */ 745 746 #if defined (__m88000__) && W_TYPE_SIZE == 32 747 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 748 __asm__ ("addu.co %1,%r4,%r5\n\taddu.ci %0,%r2,%r3" \ 749 : "=r" ((USItype) (sh)), \ 750 "=&r" ((USItype) (sl)) \ 751 : "%rJ" ((USItype) (ah)), \ 752 "rJ" ((USItype) (bh)), \ 753 "%rJ" ((USItype) (al)), \ 754 "rJ" ((USItype) (bl))) 755 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 756 __asm__ ("subu.co %1,%r4,%r5\n\tsubu.ci %0,%r2,%r3" \ 757 : "=r" ((USItype) (sh)), \ 758 "=&r" ((USItype) (sl)) \ 759 : "rJ" ((USItype) (ah)), \ 760 "rJ" ((USItype) (bh)), \ 761 "rJ" ((USItype) (al)), \ 762 "rJ" ((USItype) (bl))) 763 #define count_leading_zeros(count, x) \ 764 do { \ 765 USItype __cbtmp; \ 766 __asm__ ("ff1 %0,%1" \ 767 : "=r" (__cbtmp) \ 768 : "r" ((USItype) (x))); \ 769 (count) = __cbtmp ^ 31; \ 770 } while (0) 771 #define COUNT_LEADING_ZEROS_0 63 /* sic */ 772 #if defined (__mc88110__) 773 #define umul_ppmm(wh, wl, u, v) \ 774 do { \ 775 union {UDItype __ll; \ 776 struct {USItype __h, __l;} __i; \ 777 } __xx; \ 778 __asm__ ("mulu.d %0,%1,%2" \ 779 : "=r" (__xx.__ll) \ 780 : "r" ((USItype) (u)), \ 781 "r" ((USItype) (v))); \ 782 (wh) = __xx.__i.__h; \ 783 (wl) = __xx.__i.__l; \ 784 } while (0) 785 #define udiv_qrnnd(q, r, n1, n0, d) \ 786 ({union {UDItype __ll; \ 787 struct {USItype __h, __l;} __i; \ 788 } __xx; \ 789 USItype __q; \ 790 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \ 791 __asm__ ("divu.d %0,%1,%2" \ 792 : "=r" (__q) \ 793 : "r" (__xx.__ll), \ 794 "r" ((USItype) (d))); \ 795 (r) = (n0) - __q * (d); (q) = __q; }) 796 #define UMUL_TIME 5 797 #define UDIV_TIME 25 798 #else 799 #define UMUL_TIME 17 800 #define UDIV_TIME 150 801 #endif /* __mc88110__ */ 802 #endif /* __m88000__ */ 803 804 #if defined (__mn10300__) 805 # if defined (__AM33__) 806 # define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X)) 807 # define umul_ppmm(w1, w0, u, v) \ 808 asm("mulu %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v)) 809 # define smul_ppmm(w1, w0, u, v) \ 810 asm("mul %3,%2,%1,%0" : "=r"(w0), "=r"(w1) : "r"(u), "r"(v)) 811 # else 812 # define umul_ppmm(w1, w0, u, v) \ 813 asm("nop; nop; mulu %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v)) 814 # define smul_ppmm(w1, w0, u, v) \ 815 asm("nop; nop; mul %3,%0" : "=d"(w0), "=z"(w1) : "%0"(u), "d"(v)) 816 # endif 817 # define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 818 do { \ 819 DWunion __s, __a, __b; \ 820 __a.s.low = (al); __a.s.high = (ah); \ 821 __b.s.low = (bl); __b.s.high = (bh); \ 822 __s.ll = __a.ll + __b.ll; \ 823 (sl) = __s.s.low; (sh) = __s.s.high; \ 824 } while (0) 825 # define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 826 do { \ 827 DWunion __s, __a, __b; \ 828 __a.s.low = (al); __a.s.high = (ah); \ 829 __b.s.low = (bl); __b.s.high = (bh); \ 830 __s.ll = __a.ll - __b.ll; \ 831 (sl) = __s.s.low; (sh) = __s.s.high; \ 832 } while (0) 833 # define udiv_qrnnd(q, r, nh, nl, d) \ 834 asm("divu %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh)) 835 # define sdiv_qrnnd(q, r, nh, nl, d) \ 836 asm("div %2,%0" : "=D"(q), "=z"(r) : "D"(d), "0"(nl), "1"(nh)) 837 # define UMUL_TIME 3 838 # define UDIV_TIME 38 839 #endif 840 841 #if defined (__mips__) && W_TYPE_SIZE == 32 842 #define umul_ppmm(w1, w0, u, v) \ 843 do { \ 844 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \ 845 (w1) = (USItype) (__x >> 32); \ 846 (w0) = (USItype) (__x); \ 847 } while (0) 848 #define UMUL_TIME 10 849 #define UDIV_TIME 100 850 851 #if (__mips == 32 || __mips == 64) && ! defined (__mips16) 852 #define count_leading_zeros(COUNT,X) ((COUNT) = __builtin_clz (X)) 853 #define COUNT_LEADING_ZEROS_0 32 854 #endif 855 #endif /* __mips__ */ 856 857 #if defined (__ns32000__) && W_TYPE_SIZE == 32 858 #define umul_ppmm(w1, w0, u, v) \ 859 ({union {UDItype __ll; \ 860 struct {USItype __l, __h;} __i; \ 861 } __xx; \ 862 __asm__ ("meid %2,%0" \ 863 : "=g" (__xx.__ll) \ 864 : "%0" ((USItype) (u)), \ 865 "g" ((USItype) (v))); \ 866 (w1) = __xx.__i.__h; (w0) = __xx.__i.__l;}) 867 #define __umulsidi3(u, v) \ 868 ({UDItype __w; \ 869 __asm__ ("meid %2,%0" \ 870 : "=g" (__w) \ 871 : "%0" ((USItype) (u)), \ 872 "g" ((USItype) (v))); \ 873 __w; }) 874 #define udiv_qrnnd(q, r, n1, n0, d) \ 875 ({union {UDItype __ll; \ 876 struct {USItype __l, __h;} __i; \ 877 } __xx; \ 878 __xx.__i.__h = (n1); __xx.__i.__l = (n0); \ 879 __asm__ ("deid %2,%0" \ 880 : "=g" (__xx.__ll) \ 881 : "0" (__xx.__ll), \ 882 "g" ((USItype) (d))); \ 883 (r) = __xx.__i.__l; (q) = __xx.__i.__h; }) 884 #define count_trailing_zeros(count,x) \ 885 do { \ 886 __asm__ ("ffsd %2,%0" \ 887 : "=r" ((USItype) (count)) \ 888 : "0" ((USItype) 0), \ 889 "r" ((USItype) (x))); \ 890 } while (0) 891 #endif /* __ns32000__ */ 892 893 /* FIXME: We should test _IBMR2 here when we add assembly support for the 894 system vendor compilers. 895 FIXME: What's needed for gcc PowerPC VxWorks? __vxworks__ is not good 896 enough, since that hits ARM and m68k too. */ 897 #if (defined (_ARCH_PPC) /* AIX */ \ 898 || defined (__powerpc__) /* gcc */ \ 899 || defined (__POWERPC__) /* BEOS */ \ 900 || defined (__ppc__) /* Darwin */ \ 901 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \ 902 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \ 903 && CPU_FAMILY == PPC) \ 904 ) && W_TYPE_SIZE == 32 905 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 906 do { \ 907 if (__builtin_constant_p (bh) && (bh) == 0) \ 908 __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2" \ 909 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ 910 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \ 911 __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2" \ 912 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ 913 else \ 914 __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3" \ 915 : "=r" (sh), "=&r" (sl) \ 916 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \ 917 } while (0) 918 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 919 do { \ 920 if (__builtin_constant_p (ah) && (ah) == 0) \ 921 __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2" \ 922 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ 923 else if (__builtin_constant_p (ah) && (ah) == ~(USItype) 0) \ 924 __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2" \ 925 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ 926 else if (__builtin_constant_p (bh) && (bh) == 0) \ 927 __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2" \ 928 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ 929 else if (__builtin_constant_p (bh) && (bh) == ~(USItype) 0) \ 930 __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2" \ 931 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ 932 else \ 933 __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2" \ 934 : "=r" (sh), "=&r" (sl) \ 935 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \ 936 } while (0) 937 #define count_leading_zeros(count, x) \ 938 __asm__ ("cntlzw %0,%1" : "=r" (count) : "r" (x)) 939 #define COUNT_LEADING_ZEROS_0 32 940 #if defined (_ARCH_PPC) || defined (__powerpc__) || defined (__POWERPC__) \ 941 || defined (__ppc__) \ 942 || (defined (PPC) && ! defined (CPU_FAMILY)) /* gcc 2.7.x GNU&SysV */ \ 943 || (defined (PPC) && defined (CPU_FAMILY) /* VxWorks */ \ 944 && CPU_FAMILY == PPC) 945 #define umul_ppmm(ph, pl, m0, m1) \ 946 do { \ 947 USItype __m0 = (m0), __m1 = (m1); \ 948 __asm__ ("mulhwu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ 949 (pl) = __m0 * __m1; \ 950 } while (0) 951 #define UMUL_TIME 15 952 #define smul_ppmm(ph, pl, m0, m1) \ 953 do { \ 954 SItype __m0 = (m0), __m1 = (m1); \ 955 __asm__ ("mulhw %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ 956 (pl) = __m0 * __m1; \ 957 } while (0) 958 #define SMUL_TIME 14 959 #define UDIV_TIME 120 960 #endif 961 #endif /* 32-bit POWER architecture variants. */ 962 963 /* We should test _IBMR2 here when we add assembly support for the system 964 vendor compilers. */ 965 #if (defined (_ARCH_PPC64) || defined (__powerpc64__)) && W_TYPE_SIZE == 64 966 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 967 do { \ 968 if (__builtin_constant_p (bh) && (bh) == 0) \ 969 __asm__ ("add%I4c %1,%3,%4\n\taddze %0,%2" \ 970 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ 971 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \ 972 __asm__ ("add%I4c %1,%3,%4\n\taddme %0,%2" \ 973 : "=r" (sh), "=&r" (sl) : "r" (ah), "%r" (al), "rI" (bl));\ 974 else \ 975 __asm__ ("add%I5c %1,%4,%5\n\tadde %0,%2,%3" \ 976 : "=r" (sh), "=&r" (sl) \ 977 : "%r" (ah), "r" (bh), "%r" (al), "rI" (bl)); \ 978 } while (0) 979 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 980 do { \ 981 if (__builtin_constant_p (ah) && (ah) == 0) \ 982 __asm__ ("subf%I3c %1,%4,%3\n\tsubfze %0,%2" \ 983 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ 984 else if (__builtin_constant_p (ah) && (ah) == ~(UDItype) 0) \ 985 __asm__ ("subf%I3c %1,%4,%3\n\tsubfme %0,%2" \ 986 : "=r" (sh), "=&r" (sl) : "r" (bh), "rI" (al), "r" (bl));\ 987 else if (__builtin_constant_p (bh) && (bh) == 0) \ 988 __asm__ ("subf%I3c %1,%4,%3\n\taddme %0,%2" \ 989 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ 990 else if (__builtin_constant_p (bh) && (bh) == ~(UDItype) 0) \ 991 __asm__ ("subf%I3c %1,%4,%3\n\taddze %0,%2" \ 992 : "=r" (sh), "=&r" (sl) : "r" (ah), "rI" (al), "r" (bl));\ 993 else \ 994 __asm__ ("subf%I4c %1,%5,%4\n\tsubfe %0,%3,%2" \ 995 : "=r" (sh), "=&r" (sl) \ 996 : "r" (ah), "r" (bh), "rI" (al), "r" (bl)); \ 997 } while (0) 998 #define count_leading_zeros(count, x) \ 999 __asm__ ("cntlzd %0,%1" : "=r" (count) : "r" (x)) 1000 #define COUNT_LEADING_ZEROS_0 64 1001 #define umul_ppmm(ph, pl, m0, m1) \ 1002 do { \ 1003 UDItype __m0 = (m0), __m1 = (m1); \ 1004 __asm__ ("mulhdu %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ 1005 (pl) = __m0 * __m1; \ 1006 } while (0) 1007 #define UMUL_TIME 15 1008 #define smul_ppmm(ph, pl, m0, m1) \ 1009 do { \ 1010 DItype __m0 = (m0), __m1 = (m1); \ 1011 __asm__ ("mulhd %0,%1,%2" : "=r" (ph) : "%r" (m0), "r" (m1)); \ 1012 (pl) = __m0 * __m1; \ 1013 } while (0) 1014 #define SMUL_TIME 14 /* ??? */ 1015 #define UDIV_TIME 120 /* ??? */ 1016 #endif /* 64-bit PowerPC. */ 1017 1018 #if defined (__ibm032__) /* RT/ROMP */ && W_TYPE_SIZE == 32 1019 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1020 __asm__ ("a %1,%5\n\tae %0,%3" \ 1021 : "=r" ((USItype) (sh)), \ 1022 "=&r" ((USItype) (sl)) \ 1023 : "%0" ((USItype) (ah)), \ 1024 "r" ((USItype) (bh)), \ 1025 "%1" ((USItype) (al)), \ 1026 "r" ((USItype) (bl))) 1027 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1028 __asm__ ("s %1,%5\n\tse %0,%3" \ 1029 : "=r" ((USItype) (sh)), \ 1030 "=&r" ((USItype) (sl)) \ 1031 : "0" ((USItype) (ah)), \ 1032 "r" ((USItype) (bh)), \ 1033 "1" ((USItype) (al)), \ 1034 "r" ((USItype) (bl))) 1035 #define umul_ppmm(ph, pl, m0, m1) \ 1036 do { \ 1037 USItype __m0 = (m0), __m1 = (m1); \ 1038 __asm__ ( \ 1039 "s r2,r2\n" \ 1040 " mts r10,%2\n" \ 1041 " m r2,%3\n" \ 1042 " m r2,%3\n" \ 1043 " m r2,%3\n" \ 1044 " m r2,%3\n" \ 1045 " m r2,%3\n" \ 1046 " m r2,%3\n" \ 1047 " m r2,%3\n" \ 1048 " m r2,%3\n" \ 1049 " m r2,%3\n" \ 1050 " m r2,%3\n" \ 1051 " m r2,%3\n" \ 1052 " m r2,%3\n" \ 1053 " m r2,%3\n" \ 1054 " m r2,%3\n" \ 1055 " m r2,%3\n" \ 1056 " m r2,%3\n" \ 1057 " cas %0,r2,r0\n" \ 1058 " mfs r10,%1" \ 1059 : "=r" ((USItype) (ph)), \ 1060 "=r" ((USItype) (pl)) \ 1061 : "%r" (__m0), \ 1062 "r" (__m1) \ 1063 : "r2"); \ 1064 (ph) += ((((SItype) __m0 >> 31) & __m1) \ 1065 + (((SItype) __m1 >> 31) & __m0)); \ 1066 } while (0) 1067 #define UMUL_TIME 20 1068 #define UDIV_TIME 200 1069 #define count_leading_zeros(count, x) \ 1070 do { \ 1071 if ((x) >= 0x10000) \ 1072 __asm__ ("clz %0,%1" \ 1073 : "=r" ((USItype) (count)) \ 1074 : "r" ((USItype) (x) >> 16)); \ 1075 else \ 1076 { \ 1077 __asm__ ("clz %0,%1" \ 1078 : "=r" ((USItype) (count)) \ 1079 : "r" ((USItype) (x))); \ 1080 (count) += 16; \ 1081 } \ 1082 } while (0) 1083 #endif 1084 1085 #if defined(__sh__) && !__SHMEDIA__ && W_TYPE_SIZE == 32 1086 #ifndef __sh1__ 1087 #define umul_ppmm(w1, w0, u, v) \ 1088 __asm__ ( \ 1089 "dmulu.l %2,%3\n\tsts%M1 macl,%1\n\tsts%M0 mach,%0" \ 1090 : "=r<" ((USItype)(w1)), \ 1091 "=r<" ((USItype)(w0)) \ 1092 : "r" ((USItype)(u)), \ 1093 "r" ((USItype)(v)) \ 1094 : "macl", "mach") 1095 #define UMUL_TIME 5 1096 #endif 1097 1098 /* This is the same algorithm as __udiv_qrnnd_c. */ 1099 #define UDIV_NEEDS_NORMALIZATION 1 1100 1101 #define udiv_qrnnd(q, r, n1, n0, d) \ 1102 do { \ 1103 extern UWtype __udiv_qrnnd_16 (UWtype, UWtype) \ 1104 __attribute__ ((visibility ("hidden"))); \ 1105 /* r0: rn r1: qn */ /* r0: n1 r4: n0 r5: d r6: d1 */ /* r2: __m */ \ 1106 __asm__ ( \ 1107 "mov%M4 %4,r5\n" \ 1108 " swap.w %3,r4\n" \ 1109 " swap.w r5,r6\n" \ 1110 " jsr @%5\n" \ 1111 " shll16 r6\n" \ 1112 " swap.w r4,r4\n" \ 1113 " jsr @%5\n" \ 1114 " swap.w r1,%0\n" \ 1115 " or r1,%0" \ 1116 : "=r" (q), "=&z" (r) \ 1117 : "1" (n1), "r" (n0), "rm" (d), "r" (&__udiv_qrnnd_16) \ 1118 : "r1", "r2", "r4", "r5", "r6", "pr", "t"); \ 1119 } while (0) 1120 1121 #define UDIV_TIME 80 1122 1123 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1124 __asm__ ("clrt;subc %5,%1; subc %4,%0" \ 1125 : "=r" (sh), "=r" (sl) \ 1126 : "0" (ah), "1" (al), "r" (bh), "r" (bl) : "t") 1127 1128 #endif /* __sh__ */ 1129 1130 #if defined (__SH5__) && __SHMEDIA__ && W_TYPE_SIZE == 32 1131 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v) 1132 #define count_leading_zeros(count, x) \ 1133 do \ 1134 { \ 1135 UDItype x_ = (USItype)(x); \ 1136 SItype c_; \ 1137 \ 1138 __asm__ ("nsb %1, %0" : "=r" (c_) : "r" (x_)); \ 1139 (count) = c_ - 31; \ 1140 } \ 1141 while (0) 1142 #define COUNT_LEADING_ZEROS_0 32 1143 #endif 1144 1145 #if defined (__sparc__) && !defined (__arch64__) && !defined (__sparcv9) \ 1146 && W_TYPE_SIZE == 32 1147 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1148 __asm__ ("addcc %r4,%5,%1\n\taddx %r2,%3,%0" \ 1149 : "=r" ((USItype) (sh)), \ 1150 "=&r" ((USItype) (sl)) \ 1151 : "%rJ" ((USItype) (ah)), \ 1152 "rI" ((USItype) (bh)), \ 1153 "%rJ" ((USItype) (al)), \ 1154 "rI" ((USItype) (bl)) \ 1155 __CLOBBER_CC) 1156 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1157 __asm__ ("subcc %r4,%5,%1\n\tsubx %r2,%3,%0" \ 1158 : "=r" ((USItype) (sh)), \ 1159 "=&r" ((USItype) (sl)) \ 1160 : "rJ" ((USItype) (ah)), \ 1161 "rI" ((USItype) (bh)), \ 1162 "rJ" ((USItype) (al)), \ 1163 "rI" ((USItype) (bl)) \ 1164 __CLOBBER_CC) 1165 #if defined (__sparc_v9__) 1166 #define umul_ppmm(w1, w0, u, v) \ 1167 do { \ 1168 register USItype __g1 asm ("g1"); \ 1169 __asm__ ("umul\t%2,%3,%1\n\t" \ 1170 "srlx\t%1, 32, %0" \ 1171 : "=r" ((USItype) (w1)), \ 1172 "=r" (__g1) \ 1173 : "r" ((USItype) (u)), \ 1174 "r" ((USItype) (v))); \ 1175 (w0) = __g1; \ 1176 } while (0) 1177 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \ 1178 __asm__ ("mov\t%2,%%y\n\t" \ 1179 "udiv\t%3,%4,%0\n\t" \ 1180 "umul\t%0,%4,%1\n\t" \ 1181 "sub\t%3,%1,%1" \ 1182 : "=&r" ((USItype) (__q)), \ 1183 "=&r" ((USItype) (__r)) \ 1184 : "r" ((USItype) (__n1)), \ 1185 "r" ((USItype) (__n0)), \ 1186 "r" ((USItype) (__d))) 1187 #else 1188 #if defined (__sparc_v8__) 1189 #define umul_ppmm(w1, w0, u, v) \ 1190 __asm__ ("umul %2,%3,%1;rd %%y,%0" \ 1191 : "=r" ((USItype) (w1)), \ 1192 "=r" ((USItype) (w0)) \ 1193 : "r" ((USItype) (u)), \ 1194 "r" ((USItype) (v))) 1195 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \ 1196 __asm__ ("mov %2,%%y;nop;nop;nop;udiv %3,%4,%0;umul %0,%4,%1;sub %3,%1,%1"\ 1197 : "=&r" ((USItype) (__q)), \ 1198 "=&r" ((USItype) (__r)) \ 1199 : "r" ((USItype) (__n1)), \ 1200 "r" ((USItype) (__n0)), \ 1201 "r" ((USItype) (__d))) 1202 #else 1203 #if defined (__sparclite__) 1204 /* This has hardware multiply but not divide. It also has two additional 1205 instructions scan (ffs from high bit) and divscc. */ 1206 #define umul_ppmm(w1, w0, u, v) \ 1207 __asm__ ("umul %2,%3,%1;rd %%y,%0" \ 1208 : "=r" ((USItype) (w1)), \ 1209 "=r" ((USItype) (w0)) \ 1210 : "r" ((USItype) (u)), \ 1211 "r" ((USItype) (v))) 1212 #define udiv_qrnnd(q, r, n1, n0, d) \ 1213 __asm__ ("! Inlined udiv_qrnnd\n" \ 1214 " wr %%g0,%2,%%y ! Not a delayed write for sparclite\n" \ 1215 " tst %%g0\n" \ 1216 " divscc %3,%4,%%g1\n" \ 1217 " divscc %%g1,%4,%%g1\n" \ 1218 " divscc %%g1,%4,%%g1\n" \ 1219 " divscc %%g1,%4,%%g1\n" \ 1220 " divscc %%g1,%4,%%g1\n" \ 1221 " divscc %%g1,%4,%%g1\n" \ 1222 " divscc %%g1,%4,%%g1\n" \ 1223 " divscc %%g1,%4,%%g1\n" \ 1224 " divscc %%g1,%4,%%g1\n" \ 1225 " divscc %%g1,%4,%%g1\n" \ 1226 " divscc %%g1,%4,%%g1\n" \ 1227 " divscc %%g1,%4,%%g1\n" \ 1228 " divscc %%g1,%4,%%g1\n" \ 1229 " divscc %%g1,%4,%%g1\n" \ 1230 " divscc %%g1,%4,%%g1\n" \ 1231 " divscc %%g1,%4,%%g1\n" \ 1232 " divscc %%g1,%4,%%g1\n" \ 1233 " divscc %%g1,%4,%%g1\n" \ 1234 " divscc %%g1,%4,%%g1\n" \ 1235 " divscc %%g1,%4,%%g1\n" \ 1236 " divscc %%g1,%4,%%g1\n" \ 1237 " divscc %%g1,%4,%%g1\n" \ 1238 " divscc %%g1,%4,%%g1\n" \ 1239 " divscc %%g1,%4,%%g1\n" \ 1240 " divscc %%g1,%4,%%g1\n" \ 1241 " divscc %%g1,%4,%%g1\n" \ 1242 " divscc %%g1,%4,%%g1\n" \ 1243 " divscc %%g1,%4,%%g1\n" \ 1244 " divscc %%g1,%4,%%g1\n" \ 1245 " divscc %%g1,%4,%%g1\n" \ 1246 " divscc %%g1,%4,%%g1\n" \ 1247 " divscc %%g1,%4,%0\n" \ 1248 " rd %%y,%1\n" \ 1249 " bl,a 1f\n" \ 1250 " add %1,%4,%1\n" \ 1251 "1: ! End of inline udiv_qrnnd" \ 1252 : "=r" ((USItype) (q)), \ 1253 "=r" ((USItype) (r)) \ 1254 : "r" ((USItype) (n1)), \ 1255 "r" ((USItype) (n0)), \ 1256 "rI" ((USItype) (d)) \ 1257 : "g1" __AND_CLOBBER_CC) 1258 #define UDIV_TIME 37 1259 #define count_leading_zeros(count, x) \ 1260 do { \ 1261 __asm__ ("scan %1,1,%0" \ 1262 : "=r" ((USItype) (count)) \ 1263 : "r" ((USItype) (x))); \ 1264 } while (0) 1265 /* Early sparclites return 63 for an argument of 0, but they warn that future 1266 implementations might change this. Therefore, leave COUNT_LEADING_ZEROS_0 1267 undefined. */ 1268 #else 1269 /* SPARC without integer multiplication and divide instructions. 1270 (i.e. at least Sun4/20,40,60,65,75,110,260,280,330,360,380,470,490) */ 1271 #define umul_ppmm(w1, w0, u, v) \ 1272 __asm__ ("! Inlined umul_ppmm\n" \ 1273 " wr %%g0,%2,%%y ! SPARC has 0-3 delay insn after a wr\n"\ 1274 " sra %3,31,%%o5 ! Don't move this insn\n" \ 1275 " and %2,%%o5,%%o5 ! Don't move this insn\n" \ 1276 " andcc %%g0,0,%%g1 ! Don't move this insn\n" \ 1277 " mulscc %%g1,%3,%%g1\n" \ 1278 " mulscc %%g1,%3,%%g1\n" \ 1279 " mulscc %%g1,%3,%%g1\n" \ 1280 " mulscc %%g1,%3,%%g1\n" \ 1281 " mulscc %%g1,%3,%%g1\n" \ 1282 " mulscc %%g1,%3,%%g1\n" \ 1283 " mulscc %%g1,%3,%%g1\n" \ 1284 " mulscc %%g1,%3,%%g1\n" \ 1285 " mulscc %%g1,%3,%%g1\n" \ 1286 " mulscc %%g1,%3,%%g1\n" \ 1287 " mulscc %%g1,%3,%%g1\n" \ 1288 " mulscc %%g1,%3,%%g1\n" \ 1289 " mulscc %%g1,%3,%%g1\n" \ 1290 " mulscc %%g1,%3,%%g1\n" \ 1291 " mulscc %%g1,%3,%%g1\n" \ 1292 " mulscc %%g1,%3,%%g1\n" \ 1293 " mulscc %%g1,%3,%%g1\n" \ 1294 " mulscc %%g1,%3,%%g1\n" \ 1295 " mulscc %%g1,%3,%%g1\n" \ 1296 " mulscc %%g1,%3,%%g1\n" \ 1297 " mulscc %%g1,%3,%%g1\n" \ 1298 " mulscc %%g1,%3,%%g1\n" \ 1299 " mulscc %%g1,%3,%%g1\n" \ 1300 " mulscc %%g1,%3,%%g1\n" \ 1301 " mulscc %%g1,%3,%%g1\n" \ 1302 " mulscc %%g1,%3,%%g1\n" \ 1303 " mulscc %%g1,%3,%%g1\n" \ 1304 " mulscc %%g1,%3,%%g1\n" \ 1305 " mulscc %%g1,%3,%%g1\n" \ 1306 " mulscc %%g1,%3,%%g1\n" \ 1307 " mulscc %%g1,%3,%%g1\n" \ 1308 " mulscc %%g1,%3,%%g1\n" \ 1309 " mulscc %%g1,0,%%g1\n" \ 1310 " add %%g1,%%o5,%0\n" \ 1311 " rd %%y,%1" \ 1312 : "=r" ((USItype) (w1)), \ 1313 "=r" ((USItype) (w0)) \ 1314 : "%rI" ((USItype) (u)), \ 1315 "r" ((USItype) (v)) \ 1316 : "g1", "o5" __AND_CLOBBER_CC) 1317 #define UMUL_TIME 39 /* 39 instructions */ 1318 /* It's quite necessary to add this much assembler for the sparc. 1319 The default udiv_qrnnd (in C) is more than 10 times slower! */ 1320 #define udiv_qrnnd(__q, __r, __n1, __n0, __d) \ 1321 __asm__ ("! Inlined udiv_qrnnd\n" \ 1322 " mov 32,%%g1\n" \ 1323 " subcc %1,%2,%%g0\n" \ 1324 "1: bcs 5f\n" \ 1325 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \ 1326 " sub %1,%2,%1 ! this kills msb of n\n" \ 1327 " addx %1,%1,%1 ! so this can't give carry\n" \ 1328 " subcc %%g1,1,%%g1\n" \ 1329 "2: bne 1b\n" \ 1330 " subcc %1,%2,%%g0\n" \ 1331 " bcs 3f\n" \ 1332 " addxcc %0,%0,%0 ! shift n1n0 and a q-bit in lsb\n" \ 1333 " b 3f\n" \ 1334 " sub %1,%2,%1 ! this kills msb of n\n" \ 1335 "4: sub %1,%2,%1\n" \ 1336 "5: addxcc %1,%1,%1\n" \ 1337 " bcc 2b\n" \ 1338 " subcc %%g1,1,%%g1\n" \ 1339 "! Got carry from n. Subtract next step to cancel this carry.\n" \ 1340 " bne 4b\n" \ 1341 " addcc %0,%0,%0 ! shift n1n0 and a 0-bit in lsb\n" \ 1342 " sub %1,%2,%1\n" \ 1343 "3: xnor %0,0,%0\n" \ 1344 " ! End of inline udiv_qrnnd" \ 1345 : "=&r" ((USItype) (__q)), \ 1346 "=&r" ((USItype) (__r)) \ 1347 : "r" ((USItype) (__d)), \ 1348 "1" ((USItype) (__n1)), \ 1349 "0" ((USItype) (__n0)) : "g1" __AND_CLOBBER_CC) 1350 #define UDIV_TIME (3+7*32) /* 7 instructions/iteration. 32 iterations. */ 1351 #endif /* __sparclite__ */ 1352 #endif /* __sparc_v8__ */ 1353 #endif /* __sparc_v9__ */ 1354 #endif /* sparc32 */ 1355 1356 #if ((defined (__sparc__) && defined (__arch64__)) || defined (__sparcv9)) \ 1357 && W_TYPE_SIZE == 64 1358 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1359 do { \ 1360 UDItype __carry = 0; \ 1361 __asm__ ("addcc\t%r5,%6,%1\n\t" \ 1362 "add\t%r3,%4,%0\n\t" \ 1363 "movcs\t%%xcc, 1, %2\n\t" \ 1364 "add\t%0, %2, %0" \ 1365 : "=r" ((UDItype)(sh)), \ 1366 "=&r" ((UDItype)(sl)), \ 1367 "+r" (__carry) \ 1368 : "%rJ" ((UDItype)(ah)), \ 1369 "rI" ((UDItype)(bh)), \ 1370 "%rJ" ((UDItype)(al)), \ 1371 "rI" ((UDItype)(bl)) \ 1372 __CLOBBER_CC); \ 1373 } while (0) 1374 1375 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1376 do { \ 1377 UDItype __carry = 0; \ 1378 __asm__ ("subcc\t%r5,%6,%1\n\t" \ 1379 "sub\t%r3,%4,%0\n\t" \ 1380 "movcs\t%%xcc, 1, %2\n\t" \ 1381 "sub\t%0, %2, %0" \ 1382 : "=r" ((UDItype)(sh)), \ 1383 "=&r" ((UDItype)(sl)), \ 1384 "+r" (__carry) \ 1385 : "%rJ" ((UDItype)(ah)), \ 1386 "rI" ((UDItype)(bh)), \ 1387 "%rJ" ((UDItype)(al)), \ 1388 "rI" ((UDItype)(bl)) \ 1389 __CLOBBER_CC); \ 1390 } while (0) 1391 1392 #define umul_ppmm(wh, wl, u, v) \ 1393 do { \ 1394 UDItype tmp1, tmp2, tmp3, tmp4; \ 1395 __asm__ __volatile__ ( \ 1396 "srl %7,0,%3\n\t" \ 1397 "mulx %3,%6,%1\n\t" \ 1398 "srlx %6,32,%2\n\t" \ 1399 "mulx %2,%3,%4\n\t" \ 1400 "sllx %4,32,%5\n\t" \ 1401 "srl %6,0,%3\n\t" \ 1402 "sub %1,%5,%5\n\t" \ 1403 "srlx %5,32,%5\n\t" \ 1404 "addcc %4,%5,%4\n\t" \ 1405 "srlx %7,32,%5\n\t" \ 1406 "mulx %3,%5,%3\n\t" \ 1407 "mulx %2,%5,%5\n\t" \ 1408 "sethi %%hi(0x80000000),%2\n\t" \ 1409 "addcc %4,%3,%4\n\t" \ 1410 "srlx %4,32,%4\n\t" \ 1411 "add %2,%2,%2\n\t" \ 1412 "movcc %%xcc,%%g0,%2\n\t" \ 1413 "addcc %5,%4,%5\n\t" \ 1414 "sllx %3,32,%3\n\t" \ 1415 "add %1,%3,%1\n\t" \ 1416 "add %5,%2,%0" \ 1417 : "=r" ((UDItype)(wh)), \ 1418 "=&r" ((UDItype)(wl)), \ 1419 "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \ 1420 : "r" ((UDItype)(u)), \ 1421 "r" ((UDItype)(v)) \ 1422 __CLOBBER_CC); \ 1423 } while (0) 1424 #define UMUL_TIME 96 1425 #define UDIV_TIME 230 1426 #endif /* sparc64 */ 1427 1428 #if defined (__vax__) && W_TYPE_SIZE == 32 1429 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1430 __asm__ ("addl2 %5,%1\n\tadwc %3,%0" \ 1431 : "=g" ((USItype) (sh)), \ 1432 "=&g" ((USItype) (sl)) \ 1433 : "%0" ((USItype) (ah)), \ 1434 "g" ((USItype) (bh)), \ 1435 "%1" ((USItype) (al)), \ 1436 "g" ((USItype) (bl))) 1437 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1438 __asm__ ("subl2 %5,%1\n\tsbwc %3,%0" \ 1439 : "=g" ((USItype) (sh)), \ 1440 "=&g" ((USItype) (sl)) \ 1441 : "0" ((USItype) (ah)), \ 1442 "g" ((USItype) (bh)), \ 1443 "1" ((USItype) (al)), \ 1444 "g" ((USItype) (bl))) 1445 #define umul_ppmm(xh, xl, m0, m1) \ 1446 do { \ 1447 union { \ 1448 UDItype __ll; \ 1449 struct {USItype __l, __h;} __i; \ 1450 } __xx; \ 1451 USItype __m0 = (m0), __m1 = (m1); \ 1452 __asm__ ("emul %1,%2,$0,%0" \ 1453 : "=r" (__xx.__ll) \ 1454 : "g" (__m0), \ 1455 "g" (__m1)); \ 1456 (xh) = __xx.__i.__h; \ 1457 (xl) = __xx.__i.__l; \ 1458 (xh) += ((((SItype) __m0 >> 31) & __m1) \ 1459 + (((SItype) __m1 >> 31) & __m0)); \ 1460 } while (0) 1461 #define sdiv_qrnnd(q, r, n1, n0, d) \ 1462 do { \ 1463 union {DItype __ll; \ 1464 struct {SItype __l, __h;} __i; \ 1465 } __xx; \ 1466 __xx.__i.__h = n1; __xx.__i.__l = n0; \ 1467 __asm__ ("ediv %3,%2,%0,%1" \ 1468 : "=g" (q), "=g" (r) \ 1469 : "g" (__xx.__ll), "g" (d)); \ 1470 } while (0) 1471 #endif /* __vax__ */ 1472 1473 #ifdef _TMS320C6X 1474 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1475 do \ 1476 { \ 1477 UDItype __ll; \ 1478 __asm__ ("addu .l1 %1, %2, %0" \ 1479 : "=a" (__ll) : "a" (al), "a" (bl)); \ 1480 (sl) = (USItype)__ll; \ 1481 (sh) = ((USItype)(__ll >> 32)) + (ah) + (bh); \ 1482 } \ 1483 while (0) 1484 1485 #ifdef _TMS320C6400_PLUS 1486 #define __umulsidi3(u,v) ((UDItype)(USItype)u*(USItype)v) 1487 #define umul_ppmm(w1, w0, u, v) \ 1488 do { \ 1489 UDItype __x = (UDItype) (USItype) (u) * (USItype) (v); \ 1490 (w1) = (USItype) (__x >> 32); \ 1491 (w0) = (USItype) (__x); \ 1492 } while (0) 1493 #endif /* _TMS320C6400_PLUS */ 1494 1495 #define count_leading_zeros(count, x) ((count) = __builtin_clz (x)) 1496 #ifdef _TMS320C6400 1497 #define count_trailing_zeros(count, x) ((count) = __builtin_ctz (x)) 1498 #endif 1499 #define UMUL_TIME 4 1500 #define UDIV_TIME 40 1501 #endif /* _TMS320C6X */ 1502 1503 #if defined (__xtensa__) && W_TYPE_SIZE == 32 1504 /* This code is not Xtensa-configuration-specific, so rely on the compiler 1505 to expand builtin functions depending on what configuration features 1506 are available. This avoids library calls when the operation can be 1507 performed in-line. */ 1508 #define umul_ppmm(w1, w0, u, v) \ 1509 do { \ 1510 DWunion __w; \ 1511 __w.ll = __builtin_umulsidi3 (u, v); \ 1512 w1 = __w.s.high; \ 1513 w0 = __w.s.low; \ 1514 } while (0) 1515 #define __umulsidi3(u, v) __builtin_umulsidi3 (u, v) 1516 #define count_leading_zeros(COUNT, X) ((COUNT) = __builtin_clz (X)) 1517 #define count_trailing_zeros(COUNT, X) ((COUNT) = __builtin_ctz (X)) 1518 #endif /* __xtensa__ */ 1519 1520 #if defined xstormy16 1521 extern UHItype __stormy16_count_leading_zeros (UHItype); 1522 #define count_leading_zeros(count, x) \ 1523 do \ 1524 { \ 1525 UHItype size; \ 1526 \ 1527 /* We assume that W_TYPE_SIZE is a multiple of 16... */ \ 1528 for ((count) = 0, size = W_TYPE_SIZE; size; size -= 16) \ 1529 { \ 1530 UHItype c; \ 1531 \ 1532 c = __clzhi2 ((x) >> (size - 16)); \ 1533 (count) += c; \ 1534 if (c != 16) \ 1535 break; \ 1536 } \ 1537 } \ 1538 while (0) 1539 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE 1540 #endif 1541 1542 #if defined (__z8000__) && W_TYPE_SIZE == 16 1543 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1544 __asm__ ("add %H1,%H5\n\tadc %H0,%H3" \ 1545 : "=r" ((unsigned int)(sh)), \ 1546 "=&r" ((unsigned int)(sl)) \ 1547 : "%0" ((unsigned int)(ah)), \ 1548 "r" ((unsigned int)(bh)), \ 1549 "%1" ((unsigned int)(al)), \ 1550 "rQR" ((unsigned int)(bl))) 1551 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1552 __asm__ ("sub %H1,%H5\n\tsbc %H0,%H3" \ 1553 : "=r" ((unsigned int)(sh)), \ 1554 "=&r" ((unsigned int)(sl)) \ 1555 : "0" ((unsigned int)(ah)), \ 1556 "r" ((unsigned int)(bh)), \ 1557 "1" ((unsigned int)(al)), \ 1558 "rQR" ((unsigned int)(bl))) 1559 #define umul_ppmm(xh, xl, m0, m1) \ 1560 do { \ 1561 union {long int __ll; \ 1562 struct {unsigned int __h, __l;} __i; \ 1563 } __xx; \ 1564 unsigned int __m0 = (m0), __m1 = (m1); \ 1565 __asm__ ("mult %S0,%H3" \ 1566 : "=r" (__xx.__i.__h), \ 1567 "=r" (__xx.__i.__l) \ 1568 : "%1" (__m0), \ 1569 "rQR" (__m1)); \ 1570 (xh) = __xx.__i.__h; (xl) = __xx.__i.__l; \ 1571 (xh) += ((((signed int) __m0 >> 15) & __m1) \ 1572 + (((signed int) __m1 >> 15) & __m0)); \ 1573 } while (0) 1574 #endif /* __z8000__ */ 1575 1576 #endif /* __GNUC__ */ 1577 1578 /* If this machine has no inline assembler, use C macros. */ 1579 1580 #if !defined (add_ssaaaa) 1581 #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ 1582 do { \ 1583 UWtype __x; \ 1584 __x = (al) + (bl); \ 1585 (sh) = (ah) + (bh) + (__x < (al)); \ 1586 (sl) = __x; \ 1587 } while (0) 1588 #endif 1589 1590 #if !defined (sub_ddmmss) 1591 #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ 1592 do { \ 1593 UWtype __x; \ 1594 __x = (al) - (bl); \ 1595 (sh) = (ah) - (bh) - (__x > (al)); \ 1596 (sl) = __x; \ 1597 } while (0) 1598 #endif 1599 1600 /* If we lack umul_ppmm but have smul_ppmm, define umul_ppmm in terms of 1601 smul_ppmm. */ 1602 #if !defined (umul_ppmm) && defined (smul_ppmm) 1603 #define umul_ppmm(w1, w0, u, v) \ 1604 do { \ 1605 UWtype __w1; \ 1606 UWtype __xm0 = (u), __xm1 = (v); \ 1607 smul_ppmm (__w1, w0, __xm0, __xm1); \ 1608 (w1) = __w1 + (-(__xm0 >> (W_TYPE_SIZE - 1)) & __xm1) \ 1609 + (-(__xm1 >> (W_TYPE_SIZE - 1)) & __xm0); \ 1610 } while (0) 1611 #endif 1612 1613 /* If we still don't have umul_ppmm, define it using plain C. */ 1614 #if !defined (umul_ppmm) 1615 #define umul_ppmm(w1, w0, u, v) \ 1616 do { \ 1617 UWtype __x0, __x1, __x2, __x3; \ 1618 UHWtype __ul, __vl, __uh, __vh; \ 1619 \ 1620 __ul = __ll_lowpart (u); \ 1621 __uh = __ll_highpart (u); \ 1622 __vl = __ll_lowpart (v); \ 1623 __vh = __ll_highpart (v); \ 1624 \ 1625 __x0 = (UWtype) __ul * __vl; \ 1626 __x1 = (UWtype) __ul * __vh; \ 1627 __x2 = (UWtype) __uh * __vl; \ 1628 __x3 = (UWtype) __uh * __vh; \ 1629 \ 1630 __x1 += __ll_highpart (__x0);/* this can't give carry */ \ 1631 __x1 += __x2; /* but this indeed can */ \ 1632 if (__x1 < __x2) /* did we get it? */ \ 1633 __x3 += __ll_B; /* yes, add it in the proper pos. */ \ 1634 \ 1635 (w1) = __x3 + __ll_highpart (__x1); \ 1636 (w0) = __ll_lowpart (__x1) * __ll_B + __ll_lowpart (__x0); \ 1637 } while (0) 1638 #endif 1639 1640 #if !defined (__umulsidi3) 1641 #define __umulsidi3(u, v) \ 1642 ({DWunion __w; \ 1643 umul_ppmm (__w.s.high, __w.s.low, u, v); \ 1644 __w.ll; }) 1645 #endif 1646 1647 /* Define this unconditionally, so it can be used for debugging. */ 1648 #define __udiv_qrnnd_c(q, r, n1, n0, d) \ 1649 do { \ 1650 UWtype __d1, __d0, __q1, __q0; \ 1651 UWtype __r1, __r0, __m; \ 1652 __d1 = __ll_highpart (d); \ 1653 __d0 = __ll_lowpart (d); \ 1654 \ 1655 __r1 = (n1) % __d1; \ 1656 __q1 = (n1) / __d1; \ 1657 __m = (UWtype) __q1 * __d0; \ 1658 __r1 = __r1 * __ll_B | __ll_highpart (n0); \ 1659 if (__r1 < __m) \ 1660 { \ 1661 __q1--, __r1 += (d); \ 1662 if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */\ 1663 if (__r1 < __m) \ 1664 __q1--, __r1 += (d); \ 1665 } \ 1666 __r1 -= __m; \ 1667 \ 1668 __r0 = __r1 % __d1; \ 1669 __q0 = __r1 / __d1; \ 1670 __m = (UWtype) __q0 * __d0; \ 1671 __r0 = __r0 * __ll_B | __ll_lowpart (n0); \ 1672 if (__r0 < __m) \ 1673 { \ 1674 __q0--, __r0 += (d); \ 1675 if (__r0 >= (d)) \ 1676 if (__r0 < __m) \ 1677 __q0--, __r0 += (d); \ 1678 } \ 1679 __r0 -= __m; \ 1680 \ 1681 (q) = (UWtype) __q1 * __ll_B | __q0; \ 1682 (r) = __r0; \ 1683 } while (0) 1684 1685 /* If the processor has no udiv_qrnnd but sdiv_qrnnd, go through 1686 __udiv_w_sdiv (defined in libgcc or elsewhere). */ 1687 #if !defined (udiv_qrnnd) && defined (sdiv_qrnnd) 1688 #define udiv_qrnnd(q, r, nh, nl, d) \ 1689 do { \ 1690 extern UWtype __udiv_w_sdiv (UWtype *, UWtype, UWtype, UWtype); \ 1691 UWtype __r; \ 1692 (q) = __udiv_w_sdiv (&__r, nh, nl, d); \ 1693 (r) = __r; \ 1694 } while (0) 1695 #endif 1696 1697 /* If udiv_qrnnd was not defined for this processor, use __udiv_qrnnd_c. */ 1698 #if !defined (udiv_qrnnd) 1699 #define UDIV_NEEDS_NORMALIZATION 1 1700 #define udiv_qrnnd __udiv_qrnnd_c 1701 #endif 1702 1703 #if !defined (count_leading_zeros) 1704 #define count_leading_zeros(count, x) \ 1705 do { \ 1706 UWtype __xr = (x); \ 1707 UWtype __a; \ 1708 \ 1709 if (W_TYPE_SIZE <= 32) \ 1710 { \ 1711 __a = __xr < ((UWtype)1<<2*__BITS4) \ 1712 ? (__xr < ((UWtype)1<<__BITS4) ? 0 : __BITS4) \ 1713 : (__xr < ((UWtype)1<<3*__BITS4) ? 2*__BITS4 : 3*__BITS4); \ 1714 } \ 1715 else \ 1716 { \ 1717 for (__a = W_TYPE_SIZE - 8; __a > 0; __a -= 8) \ 1718 if (((__xr >> __a) & 0xff) != 0) \ 1719 break; \ 1720 } \ 1721 \ 1722 (count) = W_TYPE_SIZE - (__clz_tab[__xr >> __a] + __a); \ 1723 } while (0) 1724 #define COUNT_LEADING_ZEROS_0 W_TYPE_SIZE 1725 #endif 1726 1727 #if !defined (count_trailing_zeros) 1728 /* Define count_trailing_zeros using count_leading_zeros. The latter might be 1729 defined in asm, but if it is not, the C version above is good enough. */ 1730 #define count_trailing_zeros(count, x) \ 1731 do { \ 1732 UWtype __ctz_x = (x); \ 1733 UWtype __ctz_c; \ 1734 count_leading_zeros (__ctz_c, __ctz_x & -__ctz_x); \ 1735 (count) = W_TYPE_SIZE - 1 - __ctz_c; \ 1736 } while (0) 1737 #endif 1738 1739 #ifndef UDIV_NEEDS_NORMALIZATION 1740 #define UDIV_NEEDS_NORMALIZATION 0 1741 #endif 1742
