| 1 | /* expr.c -operands, expressions- |
| 2 | Copyright (C) 1987, 1990, 1991, 1992 Free Software Foundation, Inc. |
| 3 | |
| 4 | This file is part of GAS, the GNU Assembler. |
| 5 | |
| 6 | GAS is free software; you can redistribute it and/or modify |
| 7 | it under the terms of the GNU General Public License as published by |
| 8 | the Free Software Foundation; either version 2, or (at your option) |
| 9 | any later version. |
| 10 | |
| 11 | GAS is distributed in the hope that it will be useful, |
| 12 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | GNU General Public License for more details. |
| 15 | |
| 16 | You should have received a copy of the GNU General Public License |
| 17 | along with GAS; see the file COPYING. If not, write to |
| 18 | the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */ |
| 19 | |
| 20 | /* |
| 21 | * This is really a branch office of as-read.c. I split it out to clearly |
| 22 | * distinguish the world of expressions from the world of statements. |
| 23 | * (It also gives smaller files to re-compile.) |
| 24 | * Here, "operand"s are of expressions, not instructions. |
| 25 | */ |
| 26 | |
| 27 | #include <ctype.h> |
| 28 | #include <string.h> |
| 29 | |
| 30 | #include "as.h" |
| 31 | |
| 32 | #include "obstack.h" |
| 33 | |
| 34 | #if __STDC__ == 1 |
| 35 | static void clean_up_expression(expressionS *expressionP); |
| 36 | #else /* __STDC__ */ |
| 37 | static void clean_up_expression(); /* Internal. */ |
| 38 | #endif /* not __STDC__ */ |
| 39 | extern const char EXP_CHARS[]; /* JF hide MD floating pt stuff all the same place */ |
| 40 | extern const char FLT_CHARS[]; |
| 41 | |
| 42 | /* |
| 43 | * Build any floating-point literal here. |
| 44 | * Also build any bignum literal here. |
| 45 | */ |
| 46 | |
| 47 | /* LITTLENUM_TYPE generic_buffer [6]; */ /* JF this is a hack */ |
| 48 | /* Seems atof_machine can backscan through generic_bignum and hit whatever |
| 49 | happens to be loaded before it in memory. And its way too complicated |
| 50 | for me to fix right. Thus a hack. JF: Just make generic_bignum bigger, |
| 51 | and never write into the early words, thus they'll always be zero. |
| 52 | I hate Dean's floating-point code. Bleh. |
| 53 | */ |
| 54 | LITTLENUM_TYPE generic_bignum [SIZE_OF_LARGE_NUMBER+6]; |
| 55 | FLONUM_TYPE generic_floating_point_number = |
| 56 | { |
| 57 | & generic_bignum [6], /* low (JF: Was 0) */ |
| 58 | & generic_bignum [SIZE_OF_LARGE_NUMBER+6 - 1], /* high JF: (added +6) */ |
| 59 | 0, /* leader */ |
| 60 | 0, /* exponent */ |
| 61 | 0 /* sign */ |
| 62 | }; |
| 63 | /* If nonzero, we've been asked to assemble nan, +inf or -inf */ |
| 64 | int generic_floating_point_magic; |
| 65 | \f |
| 66 | floating_constant(expressionP) |
| 67 | expressionS *expressionP; |
| 68 | { |
| 69 | /* input_line_pointer->*/ |
| 70 | /* floating-point constant. */ |
| 71 | int error_code; |
| 72 | |
| 73 | error_code = atof_generic |
| 74 | (& input_line_pointer, ".", EXP_CHARS, |
| 75 | & generic_floating_point_number); |
| 76 | |
| 77 | if (error_code) |
| 78 | { |
| 79 | if (error_code == ERROR_EXPONENT_OVERFLOW) |
| 80 | { |
| 81 | as_bad("bad floating-point constant: exponent overflow, probably assembling junk"); |
| 82 | } |
| 83 | else |
| 84 | { |
| 85 | as_bad("bad floating-point constant: unknown error code=%d.", error_code); |
| 86 | } |
| 87 | } |
| 88 | expressionP->X_seg = SEG_BIG; |
| 89 | /* input_line_pointer->just after constant, */ |
| 90 | /* which may point to whitespace. */ |
| 91 | expressionP->X_add_number =-1; |
| 92 | |
| 93 | } |
| 94 | |
| 95 | |
| 96 | |
| 97 | integer_constant(radix, expressionP) |
| 98 | int radix; |
| 99 | expressionS *expressionP; |
| 100 | { |
| 101 | register char * digit_2; /*->2nd digit of number. */ |
| 102 | char c; |
| 103 | |
| 104 | register valueT number; /* offset or (absolute) value */ |
| 105 | register short int digit; /* value of next digit in current radix */ |
| 106 | register short int maxdig = 0; /* highest permitted digit value. */ |
| 107 | register int too_many_digits = 0; /* if we see >= this number of */ |
| 108 | register char *name; /* points to name of symbol */ |
| 109 | register symbolS * symbolP; /* points to symbol */ |
| 110 | |
| 111 | int small; /* true if fits in 32 bits. */ |
| 112 | extern char hex_value[]; /* in hex_value.c */ |
| 113 | |
| 114 | /* may be bignum, or may fit in 32 bits. */ |
| 115 | /* |
| 116 | * most numbers fit into 32 bits, and we want this case to be fast. |
| 117 | * so we pretend it will fit into 32 bits. if, after making up a 32 |
| 118 | * bit number, we realise that we have scanned more digits than |
| 119 | * comfortably fit into 32 bits, we re-scan the digits coding |
| 120 | * them into a bignum. for decimal and octal numbers we are conservative: some |
| 121 | * numbers may be assumed bignums when in fact they do fit into 32 bits. |
| 122 | * numbers of any radix can have excess leading zeros: we strive |
| 123 | * to recognise this and cast them back into 32 bits. |
| 124 | * we must check that the bignum really is more than 32 |
| 125 | * bits, and change it back to a 32-bit number if it fits. |
| 126 | * the number we are looking for is expected to be positive, but |
| 127 | * if it fits into 32 bits as an unsigned number, we let it be a 32-bit |
| 128 | * number. the cavalier approach is for speed in ordinary cases. |
| 129 | */ |
| 130 | |
| 131 | switch (radix) |
| 132 | { |
| 133 | |
| 134 | case 2: |
| 135 | maxdig = 2; |
| 136 | too_many_digits = 33; |
| 137 | break; |
| 138 | case 8: |
| 139 | maxdig = radix = 8; |
| 140 | too_many_digits = 11; |
| 141 | break; |
| 142 | case 16: |
| 143 | |
| 144 | |
| 145 | maxdig = radix = 16; |
| 146 | too_many_digits = 9; |
| 147 | break; |
| 148 | case 10: |
| 149 | maxdig = radix = 10; |
| 150 | too_many_digits = 11; |
| 151 | } |
| 152 | c = *input_line_pointer; |
| 153 | input_line_pointer++; |
| 154 | digit_2 = input_line_pointer; |
| 155 | for (number=0; (digit=hex_value[c])<maxdig; c = * input_line_pointer ++) |
| 156 | { |
| 157 | number = number * radix + digit; |
| 158 | } |
| 159 | /* c contains character after number. */ |
| 160 | /* input_line_pointer->char after c. */ |
| 161 | small = input_line_pointer - digit_2 < too_many_digits; |
| 162 | if (! small) |
| 163 | { |
| 164 | /* |
| 165 | * we saw a lot of digits. manufacture a bignum the hard way. |
| 166 | */ |
| 167 | LITTLENUM_TYPE * leader; /*->high order littlenum of the bignum. */ |
| 168 | LITTLENUM_TYPE * pointer; /*->littlenum we are frobbing now. */ |
| 169 | long carry; |
| 170 | |
| 171 | leader = generic_bignum; |
| 172 | generic_bignum [0] = 0; |
| 173 | generic_bignum [1] = 0; |
| 174 | /* we could just use digit_2, but lets be mnemonic. */ |
| 175 | input_line_pointer = --digit_2; /*->1st digit. */ |
| 176 | c = *input_line_pointer++; |
| 177 | for (; (carry = hex_value[c]) < maxdig; c = *input_line_pointer++) |
| 178 | { |
| 179 | for (pointer = generic_bignum; |
| 180 | pointer <= leader; |
| 181 | pointer++) |
| 182 | { |
| 183 | long work; |
| 184 | |
| 185 | work = carry + radix * * pointer; |
| 186 | *pointer = work & LITTLENUM_MASK; |
| 187 | carry = work >> LITTLENUM_NUMBER_OF_BITS; |
| 188 | } |
| 189 | if (carry) |
| 190 | { |
| 191 | if (leader < generic_bignum + SIZE_OF_LARGE_NUMBER - 1) |
| 192 | { /* room to grow a longer bignum. */ |
| 193 | *++leader = carry; |
| 194 | } |
| 195 | } |
| 196 | } |
| 197 | /* again, c is char after number, */ |
| 198 | /* input_line_pointer->after c. */ |
| 199 | know(sizeof (int) * 8 == 32); |
| 200 | know(LITTLENUM_NUMBER_OF_BITS == 16); |
| 201 | /* hence the constant "2" in the next line. */ |
| 202 | if (leader < generic_bignum + 2) |
| 203 | { /* will fit into 32 bits. */ |
| 204 | number = |
| 205 | ((generic_bignum [1] & LITTLENUM_MASK) << LITTLENUM_NUMBER_OF_BITS) |
| 206 | | (generic_bignum [0] & LITTLENUM_MASK); |
| 207 | small = 1; |
| 208 | } |
| 209 | else |
| 210 | { |
| 211 | number = leader - generic_bignum + 1; /* number of littlenums in the bignum. */ |
| 212 | } |
| 213 | } |
| 214 | if (small) { |
| 215 | /* |
| 216 | * here with number, in correct radix. c is the next char. |
| 217 | * note that unlike un*x, we allow "011f" "0x9f" to |
| 218 | * both mean the same as the (conventional) "9f". this is simply easier |
| 219 | * than checking for strict canonical form. syntax sux! |
| 220 | */ |
| 221 | |
| 222 | switch (c) { |
| 223 | |
| 224 | #ifdef LOCAL_LABELS_FB |
| 225 | case 'b': { |
| 226 | /* |
| 227 | * backward ref to local label. |
| 228 | * because it is backward, expect it to be defined. |
| 229 | */ |
| 230 | /* |
| 231 | * construct a local label. |
| 232 | */ |
| 233 | name = fb_label_name((int) number, 0); |
| 234 | |
| 235 | /* seen before, or symbol is defined: ok */ |
| 236 | symbolP = symbol_find(name); |
| 237 | if ((symbolP != NULL) && (S_IS_DEFINED(symbolP))) { |
| 238 | |
| 239 | /* local labels are never absolute. don't waste time checking absoluteness. */ |
| 240 | know(SEG_NORMAL(S_GET_SEGMENT(symbolP))); |
| 241 | |
| 242 | expressionP->X_add_symbol = symbolP; |
| 243 | expressionP->X_seg = S_GET_SEGMENT(symbolP); |
| 244 | |
| 245 | } else { /* either not seen or not defined. */ |
| 246 | as_bad("backw. ref to unknown label \"%d:\", 0 assumed.", number); |
| 247 | expressionP->X_seg = SEG_ABSOLUTE; |
| 248 | } |
| 249 | |
| 250 | expressionP->X_add_number = 0; |
| 251 | break; |
| 252 | } /* case 'b' */ |
| 253 | |
| 254 | case 'f': { |
| 255 | /* |
| 256 | * forward reference. expect symbol to be undefined or |
| 257 | * unknown. undefined: seen it before. unknown: never seen |
| 258 | * it before. |
| 259 | * construct a local label name, then an undefined symbol. |
| 260 | * don't create a xseg frag for it: caller may do that. |
| 261 | * just return it as never seen before. |
| 262 | */ |
| 263 | name = fb_label_name((int) number, 1); |
| 264 | symbolP = symbol_find_or_make(name); |
| 265 | /* we have no need to check symbol properties. */ |
| 266 | #ifndef many_segments |
| 267 | /* since "know" puts its arg into a "string", we |
| 268 | can't have newlines in the argument. */ |
| 269 | know(S_GET_SEGMENT(symbolP) == SEG_UNKNOWN || S_GET_SEGMENT(symbolP) == SEG_TEXT || S_GET_SEGMENT(symbolP) == SEG_DATA); |
| 270 | #endif |
| 271 | expressionP->X_add_symbol = symbolP; |
| 272 | expressionP->X_seg = SEG_UNKNOWN; |
| 273 | expressionP->X_subtract_symbol = NULL; |
| 274 | expressionP->X_add_number = 0; |
| 275 | |
| 276 | break; |
| 277 | } /* case 'f' */ |
| 278 | |
| 279 | #endif /* LOCAL_LABELS_FB */ |
| 280 | |
| 281 | #ifdef LOCAL_LABELS_DOLLAR |
| 282 | |
| 283 | case '$': { |
| 284 | |
| 285 | /* if the dollar label is *currently* defined, then this is just another |
| 286 | reference to it. If it is not *currently* defined, then this is a |
| 287 | fresh instantiation of that number, so create it. */ |
| 288 | |
| 289 | if (dollar_label_defined(number)) { |
| 290 | name = dollar_label_name(number, 0); |
| 291 | symbolP = symbol_find(name); |
| 292 | know(symbolP != NULL); |
| 293 | } else { |
| 294 | name = dollar_label_name(number, 1); |
| 295 | symbolP = symbol_find_or_make(name); |
| 296 | } |
| 297 | |
| 298 | expressionP->X_add_symbol = symbolP; |
| 299 | expressionP->X_add_number = 0; |
| 300 | expressionP->X_seg = S_GET_SEGMENT(symbolP); |
| 301 | |
| 302 | break; |
| 303 | } /* case '$' */ |
| 304 | |
| 305 | #endif /* LOCAL_LABELS_DOLLAR */ |
| 306 | |
| 307 | default: { |
| 308 | expressionP->X_add_number = number; |
| 309 | expressionP->X_seg = SEG_ABSOLUTE; |
| 310 | input_line_pointer--; /* restore following character. */ |
| 311 | break; |
| 312 | } /* really just a number */ |
| 313 | |
| 314 | } /* switch on char following the number */ |
| 315 | |
| 316 | |
| 317 | } else { /* not a small number */ |
| 318 | expressionP->X_add_number = number; |
| 319 | expressionP->X_seg = SEG_BIG; |
| 320 | input_line_pointer --; /*->char following number. */ |
| 321 | } /* if (small) */ |
| 322 | } /* integer_constant() */ |
| 323 | |
| 324 | |
| 325 | /* |
| 326 | * Summary of operand(). |
| 327 | * |
| 328 | * in: Input_line_pointer points to 1st char of operand, which may |
| 329 | * be a space. |
| 330 | * |
| 331 | * out: A expressionS. X_seg determines how to understand the rest of the |
| 332 | * expressionS. |
| 333 | * The operand may have been empty: in this case X_seg == SEG_ABSENT. |
| 334 | * Input_line_pointer->(next non-blank) char after operand. |
| 335 | * |
| 336 | */ |
| 337 | \f |
| 338 | |
| 339 | |
| 340 | static segT |
| 341 | operand (expressionP) |
| 342 | register expressionS * expressionP; |
| 343 | { |
| 344 | register char c; |
| 345 | register symbolS * symbolP; /* points to symbol */ |
| 346 | register char *name; /* points to name of symbol */ |
| 347 | /* invented for humans only, hope */ |
| 348 | /* optimising compiler flushes it! */ |
| 349 | register short int radix; /* 2, 8, 10 or 16, 0 when floating */ |
| 350 | /* 0 means we saw start of a floating- */ |
| 351 | /* point constant. */ |
| 352 | |
| 353 | /* digits, assume it is a bignum. */ |
| 354 | |
| 355 | |
| 356 | |
| 357 | |
| 358 | SKIP_WHITESPACE(); /* leading whitespace is part of operand. */ |
| 359 | c = * input_line_pointer ++; /* input_line_pointer->past char in c. */ |
| 360 | |
| 361 | switch (c) |
| 362 | { |
| 363 | #ifdef MRI |
| 364 | case '%': |
| 365 | integer_constant(2, expressionP); |
| 366 | break; |
| 367 | case '@': |
| 368 | integer_constant(8, expressionP); |
| 369 | break; |
| 370 | case '$': |
| 371 | integer_constant(16, expressionP); |
| 372 | break; |
| 373 | #endif |
| 374 | case '1': |
| 375 | case '2': |
| 376 | case '3': |
| 377 | case '4': |
| 378 | case '5': |
| 379 | case '6': |
| 380 | case '7': |
| 381 | case '8': |
| 382 | case '9': |
| 383 | input_line_pointer--; |
| 384 | |
| 385 | integer_constant(10, expressionP); |
| 386 | break; |
| 387 | |
| 388 | case '0': |
| 389 | /* non-decimal radix */ |
| 390 | |
| 391 | |
| 392 | c = *input_line_pointer; |
| 393 | switch (c) |
| 394 | { |
| 395 | |
| 396 | default: |
| 397 | if (c && strchr(FLT_CHARS,c)) |
| 398 | { |
| 399 | input_line_pointer++; |
| 400 | floating_constant(expressionP); |
| 401 | } |
| 402 | else |
| 403 | { |
| 404 | |
| 405 | |
| 406 | /* The string was only zero */ |
| 407 | expressionP->X_add_symbol = 0; |
| 408 | expressionP->X_add_number = 0; |
| 409 | expressionP->X_seg = SEG_ABSOLUTE; |
| 410 | } |
| 411 | |
| 412 | break; |
| 413 | |
| 414 | case 'x': |
| 415 | case 'X': |
| 416 | input_line_pointer++; |
| 417 | integer_constant(16, expressionP); |
| 418 | break; |
| 419 | case 'B': |
| 420 | case 'b': |
| 421 | input_line_pointer++; |
| 422 | integer_constant(2, expressionP); |
| 423 | break; |
| 424 | |
| 425 | case '0': |
| 426 | case '1': |
| 427 | case '2': |
| 428 | case '3': |
| 429 | case '4': |
| 430 | case '5': |
| 431 | case '6': |
| 432 | case '7': |
| 433 | integer_constant(8, expressionP); |
| 434 | break; |
| 435 | |
| 436 | case 'f': |
| 437 | /* if it says '0f' and the line ends or it doesn't look like |
| 438 | a floating point #, its a local label ref. dtrt */ |
| 439 | /* likewise for the b's. xoxorich. */ |
| 440 | if ((c == 'f' || c == 'b' || c == 'b') |
| 441 | && (!*input_line_pointer || |
| 442 | (!strchr("+-.0123456789",*input_line_pointer) && |
| 443 | !strchr(EXP_CHARS,*input_line_pointer)))) |
| 444 | { |
| 445 | input_line_pointer -= 2; |
| 446 | integer_constant(10, expressionP); |
| 447 | break; |
| 448 | } |
| 449 | |
| 450 | case 'd': |
| 451 | case 'D': |
| 452 | case 'F': |
| 453 | case 'r': |
| 454 | case 'e': |
| 455 | case 'E': |
| 456 | case 'g': |
| 457 | case 'G': |
| 458 | |
| 459 | input_line_pointer++; |
| 460 | floating_constant(expressionP); |
| 461 | break; |
| 462 | } |
| 463 | |
| 464 | break; |
| 465 | case '(': |
| 466 | /* didn't begin with digit & not a name */ |
| 467 | { |
| 468 | (void)expression(expressionP); |
| 469 | /* Expression() will pass trailing whitespace */ |
| 470 | if (* input_line_pointer ++ != ')') |
| 471 | { |
| 472 | as_bad("Missing ')' assumed"); |
| 473 | input_line_pointer --; |
| 474 | } |
| 475 | /* here with input_line_pointer->char after "(...)" */ |
| 476 | } |
| 477 | return expressionP->X_seg; |
| 478 | |
| 479 | |
| 480 | case '\'': |
| 481 | /* |
| 482 | * Warning: to conform to other people's assemblers NO ESCAPEMENT is permitted |
| 483 | * for a single quote. The next character, parity errors and all, is taken |
| 484 | * as the value of the operand. VERY KINKY. |
| 485 | */ |
| 486 | expressionP->X_add_number = * input_line_pointer ++; |
| 487 | expressionP->X_seg = SEG_ABSOLUTE; |
| 488 | break; |
| 489 | |
| 490 | case '~': |
| 491 | case '-': |
| 492 | case '+': |
| 493 | |
| 494 | { |
| 495 | /* unary operator: hope for SEG_ABSOLUTE */ |
| 496 | switch(operand (expressionP)) { |
| 497 | case SEG_ABSOLUTE: |
| 498 | /* input_line_pointer -> char after operand */ |
| 499 | if ( c=='-' ) |
| 500 | { |
| 501 | expressionP -> X_add_number = - expressionP -> X_add_number; |
| 502 | /* |
| 503 | * Notice: '-' may overflow: no warning is given. This is compatible |
| 504 | * with other people's assemblers. Sigh. |
| 505 | */ |
| 506 | } |
| 507 | else |
| 508 | { |
| 509 | expressionP -> X_add_number = ~ expressionP -> X_add_number; |
| 510 | } |
| 511 | break; |
| 512 | |
| 513 | case SEG_TEXT: |
| 514 | case SEG_DATA: |
| 515 | case SEG_BSS: |
| 516 | case SEG_PASS1: |
| 517 | case SEG_UNKNOWN: |
| 518 | if(c=='-') { /* JF I hope this hack works */ |
| 519 | expressionP->X_subtract_symbol=expressionP->X_add_symbol; |
| 520 | expressionP->X_add_symbol=0; |
| 521 | expressionP->X_seg=SEG_DIFFERENCE; |
| 522 | break; |
| 523 | } |
| 524 | default: /* unary on non-absolute is unsuported */ |
| 525 | as_warn("Unary operator %c ignored because bad operand follows", c); |
| 526 | break; |
| 527 | /* Expression undisturbed from operand(). */ |
| 528 | } |
| 529 | } |
| 530 | |
| 531 | |
| 532 | |
| 533 | break; |
| 534 | |
| 535 | case '.': |
| 536 | if( !is_part_of_name(*input_line_pointer)) |
| 537 | { |
| 538 | extern struct obstack frags; |
| 539 | |
| 540 | /* |
| 541 | JF: '.' is pseudo symbol with value of current location in current |
| 542 | segment. . . |
| 543 | */ |
| 544 | symbolP = symbol_new("L0\001", |
| 545 | now_seg, |
| 546 | (valueT)(obstack_next_free(&frags)-frag_now->fr_literal), |
| 547 | frag_now); |
| 548 | |
| 549 | expressionP->X_add_number=0; |
| 550 | expressionP->X_add_symbol=symbolP; |
| 551 | expressionP->X_seg = now_seg; |
| 552 | break; |
| 553 | |
| 554 | } |
| 555 | else |
| 556 | { |
| 557 | goto isname; |
| 558 | |
| 559 | |
| 560 | } |
| 561 | case ',': |
| 562 | case '\n': |
| 563 | /* can't imagine any other kind of operand */ |
| 564 | expressionP->X_seg = SEG_ABSENT; |
| 565 | input_line_pointer --; |
| 566 | md_operand (expressionP); |
| 567 | break; |
| 568 | /* Fall through */ |
| 569 | default: |
| 570 | if (is_name_beginner(c)) /* here if did not begin with a digit */ |
| 571 | { |
| 572 | /* |
| 573 | * Identifier begins here. |
| 574 | * This is kludged for speed, so code is repeated. |
| 575 | */ |
| 576 | isname: |
| 577 | name = -- input_line_pointer; |
| 578 | c = get_symbol_end(); |
| 579 | symbolP = symbol_find_or_make(name); |
| 580 | /* |
| 581 | * If we have an absolute symbol or a reg, then we know its value now. |
| 582 | */ |
| 583 | expressionP->X_seg = S_GET_SEGMENT(symbolP); |
| 584 | switch (expressionP->X_seg) |
| 585 | { |
| 586 | case SEG_ABSOLUTE: |
| 587 | case SEG_REGISTER: |
| 588 | expressionP->X_add_number = S_GET_VALUE(symbolP); |
| 589 | break; |
| 590 | |
| 591 | default: |
| 592 | expressionP->X_add_number = 0; |
| 593 | expressionP->X_add_symbol = symbolP; |
| 594 | } |
| 595 | * input_line_pointer = c; |
| 596 | expressionP->X_subtract_symbol = NULL; |
| 597 | } |
| 598 | else |
| 599 | { |
| 600 | as_bad("Bad expression"); |
| 601 | expressionP->X_add_number = 0; |
| 602 | expressionP->X_seg = SEG_ABSOLUTE; |
| 603 | |
| 604 | } |
| 605 | |
| 606 | } |
| 607 | |
| 608 | |
| 609 | |
| 610 | |
| 611 | |
| 612 | |
| 613 | |
| 614 | /* |
| 615 | * It is more 'efficient' to clean up the expressionS when they are created. |
| 616 | * Doing it here saves lines of code. |
| 617 | */ |
| 618 | clean_up_expression (expressionP); |
| 619 | SKIP_WHITESPACE(); /*->1st char after operand. */ |
| 620 | know(* input_line_pointer != ' '); |
| 621 | return (expressionP->X_seg); |
| 622 | } /* operand() */ |
| 623 | |
| 624 | \f |
| 625 | /* Internal. Simplify a struct expression for use by expr() */ |
| 626 | |
| 627 | /* |
| 628 | * In: address of a expressionS. |
| 629 | * The X_seg field of the expressionS may only take certain values. |
| 630 | * Now, we permit SEG_PASS1 to make code smaller & faster. |
| 631 | * Elsewise we waste time special-case testing. Sigh. Ditto SEG_ABSENT. |
| 632 | * Out: expressionS may have been modified: |
| 633 | * 'foo-foo' symbol references cancelled to 0, |
| 634 | * which changes X_seg from SEG_DIFFERENCE to SEG_ABSOLUTE; |
| 635 | * Unused fields zeroed to help expr(). |
| 636 | */ |
| 637 | |
| 638 | static void |
| 639 | clean_up_expression (expressionP) |
| 640 | register expressionS * expressionP; |
| 641 | { |
| 642 | switch (expressionP->X_seg) |
| 643 | { |
| 644 | case SEG_ABSENT: |
| 645 | case SEG_PASS1: |
| 646 | expressionP->X_add_symbol = NULL; |
| 647 | expressionP->X_subtract_symbol = NULL; |
| 648 | expressionP->X_add_number = 0; |
| 649 | break; |
| 650 | |
| 651 | case SEG_BIG: |
| 652 | case SEG_ABSOLUTE: |
| 653 | expressionP->X_subtract_symbol = NULL; |
| 654 | expressionP->X_add_symbol = NULL; |
| 655 | break; |
| 656 | |
| 657 | case SEG_UNKNOWN: |
| 658 | expressionP->X_subtract_symbol = NULL; |
| 659 | break; |
| 660 | |
| 661 | case SEG_DIFFERENCE: |
| 662 | /* |
| 663 | * It does not hurt to 'cancel' NULL==NULL |
| 664 | * when comparing symbols for 'eq'ness. |
| 665 | * It is faster to re-cancel them to NULL |
| 666 | * than to check for this special case. |
| 667 | */ |
| 668 | if (expressionP->X_subtract_symbol == expressionP->X_add_symbol |
| 669 | || (expressionP->X_subtract_symbol |
| 670 | && expressionP->X_add_symbol |
| 671 | && expressionP->X_subtract_symbol->sy_frag==expressionP->X_add_symbol->sy_frag |
| 672 | && S_GET_VALUE(expressionP->X_subtract_symbol) == S_GET_VALUE(expressionP->X_add_symbol))) { |
| 673 | expressionP->X_subtract_symbol = NULL; |
| 674 | expressionP->X_add_symbol = NULL; |
| 675 | expressionP->X_seg = SEG_ABSOLUTE; |
| 676 | } |
| 677 | break; |
| 678 | |
| 679 | case SEG_REGISTER: |
| 680 | expressionP->X_add_symbol = NULL; |
| 681 | expressionP->X_subtract_symbol = NULL; |
| 682 | break; |
| 683 | |
| 684 | default: |
| 685 | if (SEG_NORMAL(expressionP->X_seg)) { |
| 686 | expressionP->X_subtract_symbol = NULL; |
| 687 | } |
| 688 | else { |
| 689 | BAD_CASE (expressionP->X_seg); |
| 690 | } |
| 691 | break; |
| 692 | } |
| 693 | } /* clean_up_expression() */ |
| 694 | \f |
| 695 | /* |
| 696 | * expr_part () |
| 697 | * |
| 698 | * Internal. Made a function because this code is used in 2 places. |
| 699 | * Generate error or correct X_?????_symbol of expressionS. |
| 700 | */ |
| 701 | |
| 702 | /* |
| 703 | * symbol_1 += symbol_2 ... well ... sort of. |
| 704 | */ |
| 705 | |
| 706 | static segT |
| 707 | expr_part (symbol_1_PP, symbol_2_P) |
| 708 | symbolS ** symbol_1_PP; |
| 709 | symbolS * symbol_2_P; |
| 710 | { |
| 711 | segT return_value; |
| 712 | #ifndef MANY_SEGMENTS |
| 713 | know((* symbol_1_PP) == NULL || (S_GET_SEGMENT(*symbol_1_PP) == SEG_TEXT) || (S_GET_SEGMENT(*symbol_1_PP) == SEG_DATA) || (S_GET_SEGMENT(*symbol_1_PP) == SEG_BSS) || (!S_IS_DEFINED(* symbol_1_PP))); |
| 714 | know(symbol_2_P == NULL || (S_GET_SEGMENT(symbol_2_P) == SEG_TEXT) || (S_GET_SEGMENT(symbol_2_P) == SEG_DATA) || (S_GET_SEGMENT(symbol_2_P) == SEG_BSS) || (!S_IS_DEFINED(symbol_2_P))); |
| 715 | #endif |
| 716 | if (* symbol_1_PP) |
| 717 | { |
| 718 | if (!S_IS_DEFINED(* symbol_1_PP)) |
| 719 | { |
| 720 | if (symbol_2_P) |
| 721 | { |
| 722 | return_value = SEG_PASS1; |
| 723 | * symbol_1_PP = NULL; |
| 724 | } |
| 725 | else |
| 726 | { |
| 727 | know(!S_IS_DEFINED(* symbol_1_PP)); |
| 728 | return_value = SEG_UNKNOWN; |
| 729 | } |
| 730 | } |
| 731 | else |
| 732 | { |
| 733 | if (symbol_2_P) |
| 734 | { |
| 735 | if (!S_IS_DEFINED(symbol_2_P)) |
| 736 | { |
| 737 | * symbol_1_PP = NULL; |
| 738 | return_value = SEG_PASS1; |
| 739 | } |
| 740 | else |
| 741 | { |
| 742 | /* {seg1} - {seg2} */ |
| 743 | as_bad("Expression too complex, 2 symbolS forgotten: \"%s\" \"%s\"", |
| 744 | S_GET_NAME(* symbol_1_PP), S_GET_NAME(symbol_2_P)); |
| 745 | * symbol_1_PP = NULL; |
| 746 | return_value = SEG_ABSOLUTE; |
| 747 | } |
| 748 | } |
| 749 | else |
| 750 | { |
| 751 | return_value = S_GET_SEGMENT(* symbol_1_PP); |
| 752 | } |
| 753 | } |
| 754 | } |
| 755 | else |
| 756 | { /* (* symbol_1_PP) == NULL */ |
| 757 | if (symbol_2_P) |
| 758 | { |
| 759 | * symbol_1_PP = symbol_2_P; |
| 760 | return_value = S_GET_SEGMENT(symbol_2_P); |
| 761 | } |
| 762 | else |
| 763 | { |
| 764 | * symbol_1_PP = NULL; |
| 765 | return_value = SEG_ABSOLUTE; |
| 766 | } |
| 767 | } |
| 768 | #ifndef MANY_SEGMENTS |
| 769 | know(return_value == SEG_ABSOLUTE || return_value == SEG_TEXT || return_value == SEG_DATA || return_value == SEG_BSS || return_value == SEG_UNKNOWN || return_value == SEG_PASS1); |
| 770 | #endif |
| 771 | know((*symbol_1_PP) == NULL || (S_GET_SEGMENT(*symbol_1_PP) == return_value)); |
| 772 | return (return_value); |
| 773 | } /* expr_part() */ |
| 774 | \f |
| 775 | /* Expression parser. */ |
| 776 | |
| 777 | /* |
| 778 | * We allow an empty expression, and just assume (absolute,0) silently. |
| 779 | * Unary operators and parenthetical expressions are treated as operands. |
| 780 | * As usual, Q==quantity==operand, O==operator, X==expression mnemonics. |
| 781 | * |
| 782 | * We used to do a aho/ullman shift-reduce parser, but the logic got so |
| 783 | * warped that I flushed it and wrote a recursive-descent parser instead. |
| 784 | * Now things are stable, would anybody like to write a fast parser? |
| 785 | * Most expressions are either register (which does not even reach here) |
| 786 | * or 1 symbol. Then "symbol+constant" and "symbol-symbol" are common. |
| 787 | * So I guess it doesn't really matter how inefficient more complex expressions |
| 788 | * are parsed. |
| 789 | * |
| 790 | * After expr(RANK,resultP) input_line_pointer->operator of rank <= RANK. |
| 791 | * Also, we have consumed any leading or trailing spaces (operand does that) |
| 792 | * and done all intervening operators. |
| 793 | */ |
| 794 | |
| 795 | typedef enum |
| 796 | { |
| 797 | O_illegal, /* (0) what we get for illegal op */ |
| 798 | |
| 799 | O_multiply, /* (1) * */ |
| 800 | O_divide, /* (2) / */ |
| 801 | O_modulus, /* (3) % */ |
| 802 | O_left_shift, /* (4) < */ |
| 803 | O_right_shift, /* (5) > */ |
| 804 | O_bit_inclusive_or, /* (6) | */ |
| 805 | O_bit_or_not, /* (7) ! */ |
| 806 | O_bit_exclusive_or, /* (8) ^ */ |
| 807 | O_bit_and, /* (9) & */ |
| 808 | O_add, /* (10) + */ |
| 809 | O_subtract /* (11) - */ |
| 810 | } |
| 811 | operatorT; |
| 812 | |
| 813 | #define __ O_illegal |
| 814 | |
| 815 | static const operatorT op_encoding [256] = { /* maps ASCII->operators */ |
| 816 | |
| 817 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 818 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 819 | |
| 820 | __, O_bit_or_not, __, __, __, O_modulus, O_bit_and, __, |
| 821 | __, __, O_multiply, O_add, __, O_subtract, __, O_divide, |
| 822 | __, __, __, __, __, __, __, __, |
| 823 | __, __, __, __, O_left_shift, __, O_right_shift, __, |
| 824 | __, __, __, __, __, __, __, __, |
| 825 | __, __, __, __, __, __, __, __, |
| 826 | __, __, __, __, __, __, __, __, |
| 827 | __, __, __, __, __, __, O_bit_exclusive_or, __, |
| 828 | __, __, __, __, __, __, __, __, |
| 829 | __, __, __, __, __, __, __, __, |
| 830 | __, __, __, __, __, __, __, __, |
| 831 | __, __, __, __, O_bit_inclusive_or, __, __, __, |
| 832 | |
| 833 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 834 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 835 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 836 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 837 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 838 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 839 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, |
| 840 | __, __, __, __, __, __, __, __, __, __, __, __, __, __, __, __ |
| 841 | }; |
| 842 | |
| 843 | |
| 844 | /* |
| 845 | * Rank Examples |
| 846 | * 0 operand, (expression) |
| 847 | * 1 + - |
| 848 | * 2 & ^ ! | |
| 849 | * 3 * / % << >> |
| 850 | */ |
| 851 | static const operator_rankT |
| 852 | op_rank [] = { 0, 3, 3, 3, 3, 3, 2, 2, 2, 2, 1, 1 }; |
| 853 | \f |
| 854 | /* Return resultP->X_seg. */ |
| 855 | segT expr(rank, resultP) |
| 856 | register operator_rankT rank; /* Larger # is higher rank. */ |
| 857 | register expressionS *resultP; /* Deliver result here. */ |
| 858 | { |
| 859 | expressionS right; |
| 860 | register operatorT op_left; |
| 861 | register char c_left; /* 1st operator character. */ |
| 862 | register operatorT op_right; |
| 863 | register char c_right; |
| 864 | |
| 865 | know(rank >= 0); |
| 866 | (void)operand (resultP); |
| 867 | know(* input_line_pointer != ' '); /* Operand() gobbles spaces. */ |
| 868 | c_left = * input_line_pointer; /* Potential operator character. */ |
| 869 | op_left = op_encoding [c_left]; |
| 870 | while (op_left != O_illegal && op_rank [(int) op_left] > rank) |
| 871 | { |
| 872 | input_line_pointer ++; /*->after 1st character of operator. */ |
| 873 | /* Operators "<<" and ">>" have 2 characters. */ |
| 874 | if (* input_line_pointer == c_left && (c_left == '<' || c_left == '>')) |
| 875 | { |
| 876 | input_line_pointer ++; |
| 877 | } /*->after operator. */ |
| 878 | if (SEG_ABSENT == expr (op_rank[(int) op_left], &right)) |
| 879 | { |
| 880 | as_warn("Missing operand value assumed absolute 0."); |
| 881 | resultP->X_add_number = 0; |
| 882 | resultP->X_subtract_symbol = NULL; |
| 883 | resultP->X_add_symbol = NULL; |
| 884 | resultP->X_seg = SEG_ABSOLUTE; |
| 885 | } |
| 886 | know(* input_line_pointer != ' '); |
| 887 | c_right = * input_line_pointer; |
| 888 | op_right = op_encoding [c_right]; |
| 889 | if (* input_line_pointer == c_right && (c_right == '<' || c_right == '>')) |
| 890 | { |
| 891 | input_line_pointer ++; |
| 892 | } /*->after operator. */ |
| 893 | know((int) op_right == 0 || op_rank [(int) op_right] <= op_rank[(int) op_left]); |
| 894 | /* input_line_pointer->after right-hand quantity. */ |
| 895 | /* left-hand quantity in resultP */ |
| 896 | /* right-hand quantity in right. */ |
| 897 | /* operator in op_left. */ |
| 898 | if (resultP->X_seg == SEG_PASS1 || right . X_seg == SEG_PASS1) |
| 899 | { |
| 900 | resultP->X_seg = SEG_PASS1; |
| 901 | } |
| 902 | else |
| 903 | { |
| 904 | if (resultP->X_seg == SEG_BIG) |
| 905 | { |
| 906 | as_warn("Left operand of %c is a %s. Integer 0 assumed.", |
| 907 | c_left, resultP->X_add_number > 0 ? "bignum" : "float"); |
| 908 | resultP->X_seg = SEG_ABSOLUTE; |
| 909 | resultP->X_add_symbol = 0; |
| 910 | resultP->X_subtract_symbol = 0; |
| 911 | resultP->X_add_number = 0; |
| 912 | } |
| 913 | if (right . X_seg == SEG_BIG) |
| 914 | { |
| 915 | as_warn("Right operand of %c is a %s. Integer 0 assumed.", |
| 916 | c_left, right . X_add_number > 0 ? "bignum" : "float"); |
| 917 | right . X_seg = SEG_ABSOLUTE; |
| 918 | right . X_add_symbol = 0; |
| 919 | right . X_subtract_symbol = 0; |
| 920 | right . X_add_number = 0; |
| 921 | } |
| 922 | if (op_left == O_subtract) |
| 923 | { |
| 924 | /* |
| 925 | * Convert - into + by exchanging symbolS and negating number. |
| 926 | * I know -infinity can't be negated in 2's complement: |
| 927 | * but then it can't be subtracted either. This trick |
| 928 | * does not cause any further inaccuracy. |
| 929 | */ |
| 930 | |
| 931 | register symbolS * symbolP; |
| 932 | |
| 933 | right . X_add_number = - right . X_add_number; |
| 934 | symbolP = right . X_add_symbol; |
| 935 | right . X_add_symbol = right . X_subtract_symbol; |
| 936 | right . X_subtract_symbol = symbolP; |
| 937 | if (symbolP) |
| 938 | { |
| 939 | right . X_seg = SEG_DIFFERENCE; |
| 940 | } |
| 941 | op_left = O_add; |
| 942 | } |
| 943 | \f |
| 944 | if (op_left == O_add) |
| 945 | { |
| 946 | segT seg1; |
| 947 | segT seg2; |
| 948 | #ifndef MANY_SEGMENTS |
| 949 | know(resultP->X_seg == SEG_DATA || resultP->X_seg == SEG_TEXT || resultP->X_seg == SEG_BSS || resultP->X_seg == |
| 950 | SEG_UNKNOWN || resultP->X_seg == SEG_DIFFERENCE || resultP->X_seg == SEG_ABSOLUTE || resultP->X_seg == SEG_PASS1 |
| 951 | || resultP->X_seg == SEG_REGISTER); |
| 952 | know(right.X_seg == SEG_DATA || right.X_seg == SEG_TEXT || right.X_seg == SEG_BSS || right.X_seg == SEG_UNKNOWN || right.X_seg == SEG_DIFFERENCE || right.X_seg == SEG_ABSOLUTE || right.X_seg == SEG_PASS1); |
| 953 | #endif |
| 954 | clean_up_expression (& right); |
| 955 | clean_up_expression (resultP); |
| 956 | |
| 957 | seg1 = expr_part (& resultP->X_add_symbol, right . X_add_symbol); |
| 958 | seg2 = expr_part (& resultP->X_subtract_symbol, right . X_subtract_symbol); |
| 959 | if (seg1 == SEG_PASS1 || seg2 == SEG_PASS1) { |
| 960 | need_pass_2 = 1; |
| 961 | resultP->X_seg = SEG_PASS1; |
| 962 | } else if (seg2 == SEG_ABSOLUTE) |
| 963 | resultP->X_seg = seg1; |
| 964 | else if (seg1 != SEG_UNKNOWN |
| 965 | && seg1 != SEG_ABSOLUTE |
| 966 | && seg2 != SEG_UNKNOWN |
| 967 | && seg1 != seg2) { |
| 968 | know(seg2 != SEG_ABSOLUTE); |
| 969 | know(resultP->X_subtract_symbol); |
| 970 | #ifndef MANY_SEGMENTS |
| 971 | know(seg1 == SEG_TEXT || seg1 == SEG_DATA || seg1== SEG_BSS); |
| 972 | know(seg2 == SEG_TEXT || seg2 == SEG_DATA || seg2== SEG_BSS); |
| 973 | #endif |
| 974 | know(resultP->X_add_symbol); |
| 975 | know(resultP->X_subtract_symbol); |
| 976 | as_bad("Expression too complex: forgetting %s - %s", |
| 977 | S_GET_NAME(resultP->X_add_symbol), |
| 978 | S_GET_NAME(resultP->X_subtract_symbol)); |
| 979 | resultP->X_seg = SEG_ABSOLUTE; |
| 980 | /* Clean_up_expression() will do the rest. */ |
| 981 | } else |
| 982 | resultP->X_seg = SEG_DIFFERENCE; |
| 983 | |
| 984 | resultP->X_add_number += right . X_add_number; |
| 985 | clean_up_expression (resultP); |
| 986 | } |
| 987 | else |
| 988 | { /* Not +. */ |
| 989 | if (resultP->X_seg == SEG_UNKNOWN || right . X_seg == SEG_UNKNOWN) |
| 990 | { |
| 991 | resultP->X_seg = SEG_PASS1; |
| 992 | need_pass_2 = 1; |
| 993 | } |
| 994 | else |
| 995 | { |
| 996 | resultP->X_subtract_symbol = NULL; |
| 997 | resultP->X_add_symbol = NULL; |
| 998 | /* Will be SEG_ABSOLUTE. */ |
| 999 | if (resultP->X_seg != SEG_ABSOLUTE || right . X_seg != SEG_ABSOLUTE) |
| 1000 | { |
| 1001 | as_bad("Relocation error. Absolute 0 assumed."); |
| 1002 | resultP->X_seg = SEG_ABSOLUTE; |
| 1003 | resultP->X_add_number = 0; |
| 1004 | } |
| 1005 | else |
| 1006 | { |
| 1007 | switch (op_left) |
| 1008 | { |
| 1009 | case O_bit_inclusive_or: |
| 1010 | resultP->X_add_number |= right . X_add_number; |
| 1011 | break; |
| 1012 | |
| 1013 | case O_modulus: |
| 1014 | if (right . X_add_number) |
| 1015 | { |
| 1016 | resultP->X_add_number %= right . X_add_number; |
| 1017 | } |
| 1018 | else |
| 1019 | { |
| 1020 | as_warn("Division by 0. 0 assumed."); |
| 1021 | resultP->X_add_number = 0; |
| 1022 | } |
| 1023 | break; |
| 1024 | |
| 1025 | case O_bit_and: |
| 1026 | resultP->X_add_number &= right . X_add_number; |
| 1027 | break; |
| 1028 | |
| 1029 | case O_multiply: |
| 1030 | resultP->X_add_number *= right . X_add_number; |
| 1031 | break; |
| 1032 | |
| 1033 | case O_divide: |
| 1034 | if (right . X_add_number) |
| 1035 | { |
| 1036 | resultP->X_add_number /= right . X_add_number; |
| 1037 | } |
| 1038 | else |
| 1039 | { |
| 1040 | as_warn("Division by 0. 0 assumed."); |
| 1041 | resultP->X_add_number = 0; |
| 1042 | } |
| 1043 | break; |
| 1044 | |
| 1045 | case O_left_shift: |
| 1046 | resultP->X_add_number <<= right . X_add_number; |
| 1047 | break; |
| 1048 | |
| 1049 | case O_right_shift: |
| 1050 | resultP->X_add_number >>= right . X_add_number; |
| 1051 | break; |
| 1052 | |
| 1053 | case O_bit_exclusive_or: |
| 1054 | resultP->X_add_number ^= right . X_add_number; |
| 1055 | break; |
| 1056 | |
| 1057 | case O_bit_or_not: |
| 1058 | resultP->X_add_number |= ~ right . X_add_number; |
| 1059 | break; |
| 1060 | |
| 1061 | default: |
| 1062 | BAD_CASE(op_left); |
| 1063 | break; |
| 1064 | } /* switch(operator) */ |
| 1065 | } |
| 1066 | } /* If we have to force need_pass_2. */ |
| 1067 | } /* If operator was +. */ |
| 1068 | } /* If we didn't set need_pass_2. */ |
| 1069 | op_left = op_right; |
| 1070 | } /* While next operator is >= this rank. */ |
| 1071 | return (resultP->X_seg); |
| 1072 | } |
| 1073 | \f |
| 1074 | /* |
| 1075 | * get_symbol_end() |
| 1076 | * |
| 1077 | * This lives here because it belongs equally in expr.c & read.c. |
| 1078 | * Expr.c is just a branch office read.c anyway, and putting it |
| 1079 | * here lessens the crowd at read.c. |
| 1080 | * |
| 1081 | * Assume input_line_pointer is at start of symbol name. |
| 1082 | * Advance input_line_pointer past symbol name. |
| 1083 | * Turn that character into a '\0', returning its former value. |
| 1084 | * This allows a string compare (RMS wants symbol names to be strings) |
| 1085 | * of the symbol name. |
| 1086 | * There will always be a char following symbol name, because all good |
| 1087 | * lines end in end-of-line. |
| 1088 | */ |
| 1089 | char |
| 1090 | get_symbol_end() |
| 1091 | { |
| 1092 | register char c; |
| 1093 | |
| 1094 | while (is_part_of_name(c = * input_line_pointer ++)) |
| 1095 | ; |
| 1096 | * -- input_line_pointer = 0; |
| 1097 | return (c); |
| 1098 | } |
| 1099 | |
| 1100 | |
| 1101 | unsigned int get_single_number() |
| 1102 | { |
| 1103 | expressionS exp; |
| 1104 | operand(&exp); |
| 1105 | return exp.X_add_number; |
| 1106 | |
| 1107 | } |
| 1108 | /* |
| 1109 | * Local Variables: |
| 1110 | * comment-column: 0 |
| 1111 | * fill-column: 131 |
| 1112 | * End: |
| 1113 | */ |
| 1114 | |
| 1115 | /* end of expr.c */ |