daily update
[binutils.git] / opcodes / openrisc-asm.c
blobd47659b8c8fdf041a5b58666b09721db34c8555f
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 PARAMS ((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 (insn)
281 CGEN_INSN *insn;
283 CGEN_OPCODE *opc = (CGEN_OPCODE *) CGEN_INSN_OPCODE (insn);
284 const char *mnem = CGEN_INSN_MNEMONIC (insn);
285 char rxbuf[CGEN_MAX_RX_ELEMENTS];
286 char *rx = rxbuf;
287 const CGEN_SYNTAX_CHAR_TYPE *syn;
288 int reg_err;
290 syn = CGEN_SYNTAX_STRING (CGEN_OPCODE_SYNTAX (opc));
292 /* Mnemonics come first in the syntax string. */
293 if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
294 return _("missing mnemonic in syntax string");
295 ++syn;
297 /* Generate a case sensitive regular expression that emulates case
298 insensitive matching in the "C" locale. We cannot generate a case
299 insensitive regular expression because in Turkish locales, 'i' and 'I'
300 are not equal modulo case conversion. */
302 /* Copy the literal mnemonic out of the insn. */
303 for (; *mnem; mnem++)
305 char c = *mnem;
307 if (ISALPHA (c))
309 *rx++ = '[';
310 *rx++ = TOLOWER (c);
311 *rx++ = TOUPPER (c);
312 *rx++ = ']';
314 else
315 *rx++ = c;
318 /* Copy any remaining literals from the syntax string into the rx. */
319 for(; * syn != 0 && rx <= rxbuf + (CGEN_MAX_RX_ELEMENTS - 7 - 4); ++syn)
321 if (CGEN_SYNTAX_CHAR_P (* syn))
323 char c = CGEN_SYNTAX_CHAR (* syn);
325 switch (c)
327 /* Escape any regex metacharacters in the syntax. */
328 case '.': case '[': case '\\':
329 case '*': case '^': case '$':
331 #ifdef CGEN_ESCAPE_EXTENDED_REGEX
332 case '?': case '{': case '}':
333 case '(': case ')': case '*':
334 case '|': case '+': case ']':
335 #endif
336 *rx++ = '\\';
337 *rx++ = c;
338 break;
340 default:
341 if (ISALPHA (c))
343 *rx++ = '[';
344 *rx++ = TOLOWER (c);
345 *rx++ = TOUPPER (c);
346 *rx++ = ']';
348 else
349 *rx++ = c;
350 break;
353 else
355 /* Replace non-syntax fields with globs. */
356 *rx++ = '.';
357 *rx++ = '*';
361 /* Trailing whitespace ok. */
362 * rx++ = '[';
363 * rx++ = ' ';
364 * rx++ = '\t';
365 * rx++ = ']';
366 * rx++ = '*';
368 /* But anchor it after that. */
369 * rx++ = '$';
370 * rx = '\0';
372 CGEN_INSN_RX (insn) = xmalloc (sizeof (regex_t));
373 reg_err = regcomp ((regex_t *) CGEN_INSN_RX (insn), rxbuf, REG_NOSUB);
375 if (reg_err == 0)
376 return NULL;
377 else
379 static char msg[80];
381 regerror (reg_err, (regex_t *) CGEN_INSN_RX (insn), msg, 80);
382 regfree ((regex_t *) CGEN_INSN_RX (insn));
383 free (CGEN_INSN_RX (insn));
384 (CGEN_INSN_RX (insn)) = NULL;
385 return msg;
390 /* Default insn parser.
392 The syntax string is scanned and operands are parsed and stored in FIELDS.
393 Relocs are queued as we go via other callbacks.
395 ??? Note that this is currently an all-or-nothing parser. If we fail to
396 parse the instruction, we return 0 and the caller will start over from
397 the beginning. Backtracking will be necessary in parsing subexpressions,
398 but that can be handled there. Not handling backtracking here may get
399 expensive in the case of the m68k. Deal with later.
401 Returns NULL for success, an error message for failure. */
403 static const char *
404 parse_insn_normal (cd, insn, strp, fields)
405 CGEN_CPU_DESC cd;
406 const CGEN_INSN *insn;
407 const char **strp;
408 CGEN_FIELDS *fields;
410 /* ??? Runtime added insns not handled yet. */
411 const CGEN_SYNTAX *syntax = CGEN_INSN_SYNTAX (insn);
412 const char *str = *strp;
413 const char *errmsg;
414 const char *p;
415 const CGEN_SYNTAX_CHAR_TYPE * syn;
416 #ifdef CGEN_MNEMONIC_OPERANDS
417 /* FIXME: wip */
418 int past_opcode_p;
419 #endif
421 /* For now we assume the mnemonic is first (there are no leading operands).
422 We can parse it without needing to set up operand parsing.
423 GAS's input scrubber will ensure mnemonics are lowercase, but we may
424 not be called from GAS. */
425 p = CGEN_INSN_MNEMONIC (insn);
426 while (*p && TOLOWER (*p) == TOLOWER (*str))
427 ++p, ++str;
429 if (* p)
430 return _("unrecognized instruction");
432 #ifndef CGEN_MNEMONIC_OPERANDS
433 if (* str && ! ISSPACE (* str))
434 return _("unrecognized instruction");
435 #endif
437 CGEN_INIT_PARSE (cd);
438 cgen_init_parse_operand (cd);
439 #ifdef CGEN_MNEMONIC_OPERANDS
440 past_opcode_p = 0;
441 #endif
443 /* We don't check for (*str != '\0') here because we want to parse
444 any trailing fake arguments in the syntax string. */
445 syn = CGEN_SYNTAX_STRING (syntax);
447 /* Mnemonics come first for now, ensure valid string. */
448 if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
449 abort ();
451 ++syn;
453 while (* syn != 0)
455 /* Non operand chars must match exactly. */
456 if (CGEN_SYNTAX_CHAR_P (* syn))
458 /* FIXME: While we allow for non-GAS callers above, we assume the
459 first char after the mnemonic part is a space. */
460 /* FIXME: We also take inappropriate advantage of the fact that
461 GAS's input scrubber will remove extraneous blanks. */
462 if (TOLOWER (*str) == TOLOWER (CGEN_SYNTAX_CHAR (* syn)))
464 #ifdef CGEN_MNEMONIC_OPERANDS
465 if (CGEN_SYNTAX_CHAR(* syn) == ' ')
466 past_opcode_p = 1;
467 #endif
468 ++ syn;
469 ++ str;
471 else if (*str)
473 /* Syntax char didn't match. Can't be this insn. */
474 static char msg [80];
476 /* xgettext:c-format */
477 sprintf (msg, _("syntax error (expected char `%c', found `%c')"),
478 CGEN_SYNTAX_CHAR(*syn), *str);
479 return msg;
481 else
483 /* Ran out of input. */
484 static char msg [80];
486 /* xgettext:c-format */
487 sprintf (msg, _("syntax error (expected char `%c', found end of instruction)"),
488 CGEN_SYNTAX_CHAR(*syn));
489 return msg;
491 continue;
494 /* We have an operand of some sort. */
495 errmsg = cd->parse_operand (cd, CGEN_SYNTAX_FIELD (*syn),
496 &str, fields);
497 if (errmsg)
498 return errmsg;
500 /* Done with this operand, continue with next one. */
501 ++ syn;
504 /* If we're at the end of the syntax string, we're done. */
505 if (* syn == 0)
507 /* FIXME: For the moment we assume a valid `str' can only contain
508 blanks now. IE: We needn't try again with a longer version of
509 the insn and it is assumed that longer versions of insns appear
510 before shorter ones (eg: lsr r2,r3,1 vs lsr r2,r3). */
511 while (ISSPACE (* str))
512 ++ str;
514 if (* str != '\0')
515 return _("junk at end of line"); /* FIXME: would like to include `str' */
517 return NULL;
520 /* We couldn't parse it. */
521 return _("unrecognized instruction");
524 /* Main entry point.
525 This routine is called for each instruction to be assembled.
526 STR points to the insn to be assembled.
527 We assume all necessary tables have been initialized.
528 The assembled instruction, less any fixups, is stored in BUF.
529 Remember that if CGEN_INT_INSN_P then BUF is an int and thus the value
530 still needs to be converted to target byte order, otherwise BUF is an array
531 of bytes in target byte order.
532 The result is a pointer to the insn's entry in the opcode table,
533 or NULL if an error occured (an error message will have already been
534 printed).
536 Note that when processing (non-alias) macro-insns,
537 this function recurses.
539 ??? It's possible to make this cpu-independent.
540 One would have to deal with a few minor things.
541 At this point in time doing so would be more of a curiosity than useful
542 [for example this file isn't _that_ big], but keeping the possibility in
543 mind helps keep the design clean. */
545 const CGEN_INSN *
546 openrisc_cgen_assemble_insn (cd, str, fields, buf, errmsg)
547 CGEN_CPU_DESC cd;
548 const char *str;
549 CGEN_FIELDS *fields;
550 CGEN_INSN_BYTES_PTR buf;
551 char **errmsg;
553 const char *start;
554 CGEN_INSN_LIST *ilist;
555 const char *parse_errmsg = NULL;
556 const char *insert_errmsg = NULL;
557 int recognized_mnemonic = 0;
559 /* Skip leading white space. */
560 while (ISSPACE (* str))
561 ++ str;
563 /* The instructions are stored in hashed lists.
564 Get the first in the list. */
565 ilist = CGEN_ASM_LOOKUP_INSN (cd, str);
567 /* Keep looking until we find a match. */
568 start = str;
569 for ( ; ilist != NULL ; ilist = CGEN_ASM_NEXT_INSN (ilist))
571 const CGEN_INSN *insn = ilist->insn;
572 recognized_mnemonic = 1;
574 #ifdef CGEN_VALIDATE_INSN_SUPPORTED
575 /* Not usually needed as unsupported opcodes
576 shouldn't be in the hash lists. */
577 /* Is this insn supported by the selected cpu? */
578 if (! openrisc_cgen_insn_supported (cd, insn))
579 continue;
580 #endif
581 /* If the RELAX attribute is set, this is an insn that shouldn't be
582 chosen immediately. Instead, it is used during assembler/linker
583 relaxation if possible. */
584 if (CGEN_INSN_ATTR_VALUE (insn, CGEN_INSN_RELAX) != 0)
585 continue;
587 str = start;
589 /* Skip this insn if str doesn't look right lexically. */
590 if (CGEN_INSN_RX (insn) != NULL &&
591 regexec ((regex_t *) CGEN_INSN_RX (insn), str, 0, NULL, 0) == REG_NOMATCH)
592 continue;
594 /* Allow parse/insert handlers to obtain length of insn. */
595 CGEN_FIELDS_BITSIZE (fields) = CGEN_INSN_BITSIZE (insn);
597 parse_errmsg = CGEN_PARSE_FN (cd, insn) (cd, insn, & str, fields);
598 if (parse_errmsg != NULL)
599 continue;
601 /* ??? 0 is passed for `pc'. */
602 insert_errmsg = CGEN_INSERT_FN (cd, insn) (cd, insn, fields, buf,
603 (bfd_vma) 0);
604 if (insert_errmsg != NULL)
605 continue;
607 /* It is up to the caller to actually output the insn and any
608 queued relocs. */
609 return insn;
613 static char errbuf[150];
614 #ifdef CGEN_VERBOSE_ASSEMBLER_ERRORS
615 const char *tmp_errmsg;
617 /* If requesting verbose error messages, use insert_errmsg.
618 Failing that, use parse_errmsg. */
619 tmp_errmsg = (insert_errmsg ? insert_errmsg :
620 parse_errmsg ? parse_errmsg :
621 recognized_mnemonic ?
622 _("unrecognized form of instruction") :
623 _("unrecognized instruction"));
625 if (strlen (start) > 50)
626 /* xgettext:c-format */
627 sprintf (errbuf, "%s `%.50s...'", tmp_errmsg, start);
628 else
629 /* xgettext:c-format */
630 sprintf (errbuf, "%s `%.50s'", tmp_errmsg, start);
631 #else
632 if (strlen (start) > 50)
633 /* xgettext:c-format */
634 sprintf (errbuf, _("bad instruction `%.50s...'"), start);
635 else
636 /* xgettext:c-format */
637 sprintf (errbuf, _("bad instruction `%.50s'"), start);
638 #endif
640 *errmsg = errbuf;
641 return NULL;
645 #if 0 /* This calls back to GAS which we can't do without care. */
647 /* Record each member of OPVALS in the assembler's symbol table.
648 This lets GAS parse registers for us.
649 ??? Interesting idea but not currently used. */
651 /* Record each member of OPVALS in the assembler's symbol table.
652 FIXME: Not currently used. */
654 void
655 openrisc_cgen_asm_hash_keywords (cd, opvals)
656 CGEN_CPU_DESC cd;
657 CGEN_KEYWORD *opvals;
659 CGEN_KEYWORD_SEARCH search = cgen_keyword_search_init (opvals, NULL);
660 const CGEN_KEYWORD_ENTRY * ke;
662 while ((ke = cgen_keyword_search_next (& search)) != NULL)
664 #if 0 /* Unnecessary, should be done in the search routine. */
665 if (! openrisc_cgen_opval_supported (ke))
666 continue;
667 #endif
668 cgen_asm_record_register (cd, ke->name, ke->value);
672 #endif /* 0 */