daily update
[binutils.git] / opcodes / openrisc-asm.c
blob9f6ee3258888a732af790288e5f9caeba8ec1204
1 /* Assembler interface for targets using CGEN. -*- C -*-
2 CGEN: Cpu tools GENerator
4 THIS FILE IS MACHINE GENERATED WITH CGEN.
5 - the resultant file is machine generated, cgen-asm.in isn't
7 Copyright 1996, 1997, 1998, 1999, 2000, 2001 Free Software Foundation, Inc.
9 This file is part of the GNU Binutils and GDB, the GNU debugger.
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2, or (at your option)
14 any later version.
16 This program is distributed in the hope that it will be useful,
17 but WITHOUT ANY WARRANTY; without even the implied warranty of
18 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 GNU General Public License for more details.
21 You should have received a copy of the GNU General Public License
22 along with this program; if not, write to the Free Software Foundation, Inc.,
23 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
25 /* ??? Eventually more and more of this stuff can go to cpu-independent files.
26 Keep that in mind. */
28 #include "sysdep.h"
29 #include <stdio.h>
30 #include "ansidecl.h"
31 #include "bfd.h"
32 #include "symcat.h"
33 #include "openrisc-desc.h"
34 #include "openrisc-opc.h"
35 #include "opintl.h"
36 #include "xregex.h"
37 #include "libiberty.h"
38 #include "safe-ctype.h"
40 #undef min
41 #define min(a,b) ((a) < (b) ? (a) : (b))
42 #undef max
43 #define max(a,b) ((a) > (b) ? (a) : (b))
45 static const char * parse_insn_normal
46 (CGEN_CPU_DESC, const CGEN_INSN *, const char **, CGEN_FIELDS *);
48 /* -- assembler routines inserted here. */
50 /* -- asm.c */
52 #define CGEN_VERBOSE_ASSEMBLER_ERRORS
54 static const char * parse_hi16
55 PARAMS ((CGEN_CPU_DESC, const char **, int, unsigned long *));
56 static const char * parse_lo16
57 PARAMS ((CGEN_CPU_DESC, const char **, int, unsigned long *));
59 long
60 openrisc_sign_extend_16bit (value)
61 long value;
63 return (long) (short) value;
66 /* Handle hi(). */
68 static const char *
69 parse_hi16 (cd, strp, opindex, valuep)
70 CGEN_CPU_DESC cd;
71 const char **strp;
72 int opindex;
73 unsigned long *valuep;
75 const char *errmsg;
76 enum cgen_parse_operand_result result_type;
77 bfd_vma value;
79 if (**strp == '#')
80 ++*strp;
82 if (strncasecmp (*strp, "hi(", 3) == 0)
84 *strp += 3;
86 #if 0
87 errmsg = cgen_parse_signed_integer (cd, strp, opindex, valuep);
88 if (errmsg != NULL)
89 fprintf (stderr, "parse_hi: %s\n", errmsg);
90 if (errmsg != NULL)
91 #endif
92 errmsg = cgen_parse_address (cd, strp, opindex, BFD_RELOC_HI16,
93 &result_type, &value);
94 if (**strp != ')')
95 return "missing `)'";
96 ++*strp;
97 if (errmsg == NULL
98 && result_type == CGEN_PARSE_OPERAND_RESULT_NUMBER)
99 value >>= 16;
100 *valuep = (long) (short) value;
102 return errmsg;
104 else
106 if (**strp == '-')
107 errmsg = cgen_parse_signed_integer (cd, strp, opindex, (long *) &value);
108 else
109 errmsg = cgen_parse_unsigned_integer (cd, strp, opindex, (unsigned long *) &value);
111 *valuep = (long) (short) (value & 0xffff);
112 return errmsg;
115 /* Handle lo(). */
117 static const char *
118 parse_lo16 (cd, strp, opindex, valuep)
119 CGEN_CPU_DESC cd;
120 const char **strp;
121 int opindex;
122 unsigned long *valuep;
124 const char *errmsg;
125 enum cgen_parse_operand_result result_type;
126 bfd_vma value;
128 if (**strp == '#')
129 ++*strp;
131 if (strncasecmp (*strp, "lo(", 3) == 0)
133 *strp += 3;
135 #if 0
136 errmsg = cgen_parse_signed_integer (cd, strp, opindex, valuep);
137 if (errmsg != NULL)
138 fprintf (stderr, "parse_lo: %s\n", errmsg);
140 if (errmsg != NULL)
141 #endif
142 errmsg = cgen_parse_address (cd, strp, opindex, BFD_RELOC_LO16,
143 &result_type, &value);
144 if (**strp != ')')
145 return "missing `)'";
146 ++*strp;
147 if (errmsg == NULL
148 && result_type == CGEN_PARSE_OPERAND_RESULT_NUMBER)
149 value &= 0xffff;
150 *valuep = (long) (short) value;
152 return errmsg;
155 if (**strp == '-')
156 errmsg = cgen_parse_signed_integer (cd, strp, opindex, (long *) &value);
157 else
158 errmsg = cgen_parse_unsigned_integer (cd, strp, opindex, (unsigned long *) &value);
159 *valuep = (long) (short) (value & 0xffff);
160 return errmsg;
163 /* -- */
165 const char * openrisc_cgen_parse_operand
166 PARAMS ((CGEN_CPU_DESC, int, const char **, CGEN_FIELDS *));
168 /* Main entry point for operand parsing.
170 This function is basically just a big switch statement. Earlier versions
171 used tables to look up the function to use, but
172 - if the table contains both assembler and disassembler functions then
173 the disassembler contains much of the assembler and vice-versa,
174 - there's a lot of inlining possibilities as things grow,
175 - using a switch statement avoids the function call overhead.
177 This function could be moved into `parse_insn_normal', but keeping it
178 separate makes clear the interface between `parse_insn_normal' and each of
179 the handlers. */
181 const char *
182 openrisc_cgen_parse_operand (cd, opindex, strp, fields)
183 CGEN_CPU_DESC cd;
184 int opindex;
185 const char ** strp;
186 CGEN_FIELDS * fields;
188 const char * errmsg = NULL;
189 /* Used by scalar operands that still need to be parsed. */
190 long junk ATTRIBUTE_UNUSED;
192 switch (opindex)
194 case OPENRISC_OPERAND_ABS_26 :
196 bfd_vma value;
197 errmsg = cgen_parse_address (cd, strp, OPENRISC_OPERAND_ABS_26, 0, NULL, & value);
198 fields->f_abs26 = value;
200 break;
201 case OPENRISC_OPERAND_DISP_26 :
203 bfd_vma value;
204 errmsg = cgen_parse_address (cd, strp, OPENRISC_OPERAND_DISP_26, 0, NULL, & value);
205 fields->f_disp26 = value;
207 break;
208 case OPENRISC_OPERAND_HI16 :
209 errmsg = parse_hi16 (cd, strp, OPENRISC_OPERAND_HI16, &fields->f_simm16);
210 break;
211 case OPENRISC_OPERAND_LO16 :
212 errmsg = parse_lo16 (cd, strp, OPENRISC_OPERAND_LO16, &fields->f_lo16);
213 break;
214 case OPENRISC_OPERAND_OP_F_23 :
215 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_OP_F_23, &fields->f_op4);
216 break;
217 case OPENRISC_OPERAND_OP_F_3 :
218 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_OP_F_3, &fields->f_op5);
219 break;
220 case OPENRISC_OPERAND_RA :
221 errmsg = cgen_parse_keyword (cd, strp, & openrisc_cgen_opval_h_gr, & fields->f_r2);
222 break;
223 case OPENRISC_OPERAND_RB :
224 errmsg = cgen_parse_keyword (cd, strp, & openrisc_cgen_opval_h_gr, & fields->f_r3);
225 break;
226 case OPENRISC_OPERAND_RD :
227 errmsg = cgen_parse_keyword (cd, strp, & openrisc_cgen_opval_h_gr, & fields->f_r1);
228 break;
229 case OPENRISC_OPERAND_SIMM_16 :
230 errmsg = cgen_parse_signed_integer (cd, strp, OPENRISC_OPERAND_SIMM_16, &fields->f_simm16);
231 break;
232 case OPENRISC_OPERAND_UI16NC :
233 errmsg = parse_lo16 (cd, strp, OPENRISC_OPERAND_UI16NC, &fields->f_i16nc);
234 break;
235 case OPENRISC_OPERAND_UIMM_16 :
236 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_UIMM_16, &fields->f_uimm16);
237 break;
238 case OPENRISC_OPERAND_UIMM_5 :
239 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_UIMM_5, &fields->f_uimm5);
240 break;
242 default :
243 /* xgettext:c-format */
244 fprintf (stderr, _("Unrecognized field %d while parsing.\n"), opindex);
245 abort ();
248 return errmsg;
251 cgen_parse_fn * const openrisc_cgen_parse_handlers[] =
253 parse_insn_normal,
256 void
257 openrisc_cgen_init_asm (cd)
258 CGEN_CPU_DESC cd;
260 openrisc_cgen_init_opcode_table (cd);
261 openrisc_cgen_init_ibld_table (cd);
262 cd->parse_handlers = & openrisc_cgen_parse_handlers[0];
263 cd->parse_operand = openrisc_cgen_parse_operand;
268 /* Regex construction routine.
270 This translates an opcode syntax string into a regex string,
271 by replacing any non-character syntax element (such as an
272 opcode) with the pattern '.*'
274 It then compiles the regex and stores it in the opcode, for
275 later use by openrisc_cgen_assemble_insn
277 Returns NULL for success, an error message for failure. */
279 char *
280 openrisc_cgen_build_insn_regex (CGEN_INSN *insn)
282 CGEN_OPCODE *opc = (CGEN_OPCODE *) CGEN_INSN_OPCODE (insn);
283 const char *mnem = CGEN_INSN_MNEMONIC (insn);
284 char rxbuf[CGEN_MAX_RX_ELEMENTS];
285 char *rx = rxbuf;
286 const CGEN_SYNTAX_CHAR_TYPE *syn;
287 int reg_err;
289 syn = CGEN_SYNTAX_STRING (CGEN_OPCODE_SYNTAX (opc));
291 /* Mnemonics come first in the syntax string. */
292 if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
293 return _("missing mnemonic in syntax string");
294 ++syn;
296 /* Generate a case sensitive regular expression that emulates case
297 insensitive matching in the "C" locale. We cannot generate a case
298 insensitive regular expression because in Turkish locales, 'i' and 'I'
299 are not equal modulo case conversion. */
301 /* Copy the literal mnemonic out of the insn. */
302 for (; *mnem; mnem++)
304 char c = *mnem;
306 if (ISALPHA (c))
308 *rx++ = '[';
309 *rx++ = TOLOWER (c);
310 *rx++ = TOUPPER (c);
311 *rx++ = ']';
313 else
314 *rx++ = c;
317 /* Copy any remaining literals from the syntax string into the rx. */
318 for(; * syn != 0 && rx <= rxbuf + (CGEN_MAX_RX_ELEMENTS - 7 - 4); ++syn)
320 if (CGEN_SYNTAX_CHAR_P (* syn))
322 char c = CGEN_SYNTAX_CHAR (* syn);
324 switch (c)
326 /* Escape any regex metacharacters in the syntax. */
327 case '.': case '[': case '\\':
328 case '*': case '^': case '$':
330 #ifdef CGEN_ESCAPE_EXTENDED_REGEX
331 case '?': case '{': case '}':
332 case '(': case ')': case '*':
333 case '|': case '+': case ']':
334 #endif
335 *rx++ = '\\';
336 *rx++ = c;
337 break;
339 default:
340 if (ISALPHA (c))
342 *rx++ = '[';
343 *rx++ = TOLOWER (c);
344 *rx++ = TOUPPER (c);
345 *rx++ = ']';
347 else
348 *rx++ = c;
349 break;
352 else
354 /* Replace non-syntax fields with globs. */
355 *rx++ = '.';
356 *rx++ = '*';
360 /* Trailing whitespace ok. */
361 * rx++ = '[';
362 * rx++ = ' ';
363 * rx++ = '\t';
364 * rx++ = ']';
365 * rx++ = '*';
367 /* But anchor it after that. */
368 * rx++ = '$';
369 * rx = '\0';
371 CGEN_INSN_RX (insn) = xmalloc (sizeof (regex_t));
372 reg_err = regcomp ((regex_t *) CGEN_INSN_RX (insn), rxbuf, REG_NOSUB);
374 if (reg_err == 0)
375 return NULL;
376 else
378 static char msg[80];
380 regerror (reg_err, (regex_t *) CGEN_INSN_RX (insn), msg, 80);
381 regfree ((regex_t *) CGEN_INSN_RX (insn));
382 free (CGEN_INSN_RX (insn));
383 (CGEN_INSN_RX (insn)) = NULL;
384 return msg;
389 /* Default insn parser.
391 The syntax string is scanned and operands are parsed and stored in FIELDS.
392 Relocs are queued as we go via other callbacks.
394 ??? Note that this is currently an all-or-nothing parser. If we fail to
395 parse the instruction, we return 0 and the caller will start over from
396 the beginning. Backtracking will be necessary in parsing subexpressions,
397 but that can be handled there. Not handling backtracking here may get
398 expensive in the case of the m68k. Deal with later.
400 Returns NULL for success, an error message for failure. */
402 static const char *
403 parse_insn_normal (CGEN_CPU_DESC cd,
404 const CGEN_INSN *insn,
405 const char **strp,
406 CGEN_FIELDS *fields)
408 /* ??? Runtime added insns not handled yet. */
409 const CGEN_SYNTAX *syntax = CGEN_INSN_SYNTAX (insn);
410 const char *str = *strp;
411 const char *errmsg;
412 const char *p;
413 const CGEN_SYNTAX_CHAR_TYPE * syn;
414 #ifdef CGEN_MNEMONIC_OPERANDS
415 /* FIXME: wip */
416 int past_opcode_p;
417 #endif
419 /* For now we assume the mnemonic is first (there are no leading operands).
420 We can parse it without needing to set up operand parsing.
421 GAS's input scrubber will ensure mnemonics are lowercase, but we may
422 not be called from GAS. */
423 p = CGEN_INSN_MNEMONIC (insn);
424 while (*p && TOLOWER (*p) == TOLOWER (*str))
425 ++p, ++str;
427 if (* p)
428 return _("unrecognized instruction");
430 #ifndef CGEN_MNEMONIC_OPERANDS
431 if (* str && ! ISSPACE (* str))
432 return _("unrecognized instruction");
433 #endif
435 CGEN_INIT_PARSE (cd);
436 cgen_init_parse_operand (cd);
437 #ifdef CGEN_MNEMONIC_OPERANDS
438 past_opcode_p = 0;
439 #endif
441 /* We don't check for (*str != '\0') here because we want to parse
442 any trailing fake arguments in the syntax string. */
443 syn = CGEN_SYNTAX_STRING (syntax);
445 /* Mnemonics come first for now, ensure valid string. */
446 if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
447 abort ();
449 ++syn;
451 while (* syn != 0)
453 /* Non operand chars must match exactly. */
454 if (CGEN_SYNTAX_CHAR_P (* syn))
456 /* FIXME: While we allow for non-GAS callers above, we assume the
457 first char after the mnemonic part is a space. */
458 /* FIXME: We also take inappropriate advantage of the fact that
459 GAS's input scrubber will remove extraneous blanks. */
460 if (TOLOWER (*str) == TOLOWER (CGEN_SYNTAX_CHAR (* syn)))
462 #ifdef CGEN_MNEMONIC_OPERANDS
463 if (CGEN_SYNTAX_CHAR(* syn) == ' ')
464 past_opcode_p = 1;
465 #endif
466 ++ syn;
467 ++ str;
469 else if (*str)
471 /* Syntax char didn't match. Can't be this insn. */
472 static char msg [80];
474 /* xgettext:c-format */
475 sprintf (msg, _("syntax error (expected char `%c', found `%c')"),
476 CGEN_SYNTAX_CHAR(*syn), *str);
477 return msg;
479 else
481 /* Ran out of input. */
482 static char msg [80];
484 /* xgettext:c-format */
485 sprintf (msg, _("syntax error (expected char `%c', found end of instruction)"),
486 CGEN_SYNTAX_CHAR(*syn));
487 return msg;
489 continue;
492 /* We have an operand of some sort. */
493 errmsg = cd->parse_operand (cd, CGEN_SYNTAX_FIELD (*syn),
494 &str, fields);
495 if (errmsg)
496 return errmsg;
498 /* Done with this operand, continue with next one. */
499 ++ syn;
502 /* If we're at the end of the syntax string, we're done. */
503 if (* syn == 0)
505 /* FIXME: For the moment we assume a valid `str' can only contain
506 blanks now. IE: We needn't try again with a longer version of
507 the insn and it is assumed that longer versions of insns appear
508 before shorter ones (eg: lsr r2,r3,1 vs lsr r2,r3). */
509 while (ISSPACE (* str))
510 ++ str;
512 if (* str != '\0')
513 return _("junk at end of line"); /* FIXME: would like to include `str' */
515 return NULL;
518 /* We couldn't parse it. */
519 return _("unrecognized instruction");
522 /* Main entry point.
523 This routine is called for each instruction to be assembled.
524 STR points to the insn to be assembled.
525 We assume all necessary tables have been initialized.
526 The assembled instruction, less any fixups, is stored in BUF.
527 Remember that if CGEN_INT_INSN_P then BUF is an int and thus the value
528 still needs to be converted to target byte order, otherwise BUF is an array
529 of bytes in target byte order.
530 The result is a pointer to the insn's entry in the opcode table,
531 or NULL if an error occured (an error message will have already been
532 printed).
534 Note that when processing (non-alias) macro-insns,
535 this function recurses.
537 ??? It's possible to make this cpu-independent.
538 One would have to deal with a few minor things.
539 At this point in time doing so would be more of a curiosity than useful
540 [for example this file isn't _that_ big], but keeping the possibility in
541 mind helps keep the design clean. */
543 const CGEN_INSN *
544 openrisc_cgen_assemble_insn (CGEN_CPU_DESC cd,
545 const char *str,
546 CGEN_FIELDS *fields,
547 CGEN_INSN_BYTES_PTR buf,
548 char **errmsg)
550 const char *start;
551 CGEN_INSN_LIST *ilist;
552 const char *parse_errmsg = NULL;
553 const char *insert_errmsg = NULL;
554 int recognized_mnemonic = 0;
556 /* Skip leading white space. */
557 while (ISSPACE (* str))
558 ++ str;
560 /* The instructions are stored in hashed lists.
561 Get the first in the list. */
562 ilist = CGEN_ASM_LOOKUP_INSN (cd, str);
564 /* Keep looking until we find a match. */
565 start = str;
566 for ( ; ilist != NULL ; ilist = CGEN_ASM_NEXT_INSN (ilist))
568 const CGEN_INSN *insn = ilist->insn;
569 recognized_mnemonic = 1;
571 #ifdef CGEN_VALIDATE_INSN_SUPPORTED
572 /* Not usually needed as unsupported opcodes
573 shouldn't be in the hash lists. */
574 /* Is this insn supported by the selected cpu? */
575 if (! openrisc_cgen_insn_supported (cd, insn))
576 continue;
577 #endif
578 /* If the RELAXED attribute is set, this is an insn that shouldn't be
579 chosen immediately. Instead, it is used during assembler/linker
580 relaxation if possible. */
581 if (CGEN_INSN_ATTR_VALUE (insn, CGEN_INSN_RELAXED) != 0)
582 continue;
584 str = start;
586 /* Skip this insn if str doesn't look right lexically. */
587 if (CGEN_INSN_RX (insn) != NULL &&
588 regexec ((regex_t *) CGEN_INSN_RX (insn), str, 0, NULL, 0) == REG_NOMATCH)
589 continue;
591 /* Allow parse/insert handlers to obtain length of insn. */
592 CGEN_FIELDS_BITSIZE (fields) = CGEN_INSN_BITSIZE (insn);
594 parse_errmsg = CGEN_PARSE_FN (cd, insn) (cd, insn, & str, fields);
595 if (parse_errmsg != NULL)
596 continue;
598 /* ??? 0 is passed for `pc'. */
599 insert_errmsg = CGEN_INSERT_FN (cd, insn) (cd, insn, fields, buf,
600 (bfd_vma) 0);
601 if (insert_errmsg != NULL)
602 continue;
604 /* It is up to the caller to actually output the insn and any
605 queued relocs. */
606 return insn;
610 static char errbuf[150];
611 #ifdef CGEN_VERBOSE_ASSEMBLER_ERRORS
612 const char *tmp_errmsg;
614 /* If requesting verbose error messages, use insert_errmsg.
615 Failing that, use parse_errmsg. */
616 tmp_errmsg = (insert_errmsg ? insert_errmsg :
617 parse_errmsg ? parse_errmsg :
618 recognized_mnemonic ?
619 _("unrecognized form of instruction") :
620 _("unrecognized instruction"));
622 if (strlen (start) > 50)
623 /* xgettext:c-format */
624 sprintf (errbuf, "%s `%.50s...'", tmp_errmsg, start);
625 else
626 /* xgettext:c-format */
627 sprintf (errbuf, "%s `%.50s'", tmp_errmsg, start);
628 #else
629 if (strlen (start) > 50)
630 /* xgettext:c-format */
631 sprintf (errbuf, _("bad instruction `%.50s...'"), start);
632 else
633 /* xgettext:c-format */
634 sprintf (errbuf, _("bad instruction `%.50s'"), start);
635 #endif
637 *errmsg = errbuf;
638 return NULL;
642 #if 0 /* This calls back to GAS which we can't do without care. */
644 /* Record each member of OPVALS in the assembler's symbol table.
645 This lets GAS parse registers for us.
646 ??? Interesting idea but not currently used. */
648 /* Record each member of OPVALS in the assembler's symbol table.
649 FIXME: Not currently used. */
651 void
652 openrisc_cgen_asm_hash_keywords (CGEN_CPU_DESC cd, CGEN_KEYWORD *opvals)
654 CGEN_KEYWORD_SEARCH search = cgen_keyword_search_init (opvals, NULL);
655 const CGEN_KEYWORD_ENTRY * ke;
657 while ((ke = cgen_keyword_search_next (& search)) != NULL)
659 #if 0 /* Unnecessary, should be done in the search routine. */
660 if (! openrisc_cgen_opval_supported (ke))
661 continue;
662 #endif
663 cgen_asm_record_register (cd, ke->name, ke->value);
667 #endif /* 0 */