2004-08-31 Robert Bowdidge <bowdidge@apple.com>
[binutils.git] / opcodes / openrisc-asm.c
blobad977907d9a9535ddbbffc7bcaf14ff0a9ac1d2b
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 ((value & 0xffff) ^ 0x8000) - 0x8000;
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 unsigned long ret;
79 if (**strp == '#')
80 ++*strp;
82 if (strncasecmp (*strp, "hi(", 3) == 0)
84 bfd_vma value;
86 *strp += 3;
87 #if 0
88 errmsg = cgen_parse_signed_integer (cd, strp, opindex, valuep);
89 if (errmsg != NULL)
90 fprintf (stderr, "parse_hi: %s\n", errmsg);
91 if (errmsg != NULL)
92 #endif
93 errmsg = cgen_parse_address (cd, strp, opindex, BFD_RELOC_HI16,
94 &result_type, &value);
95 if (**strp != ')')
96 return _("missing `)'");
98 ++*strp;
99 if (errmsg == NULL
100 && result_type == CGEN_PARSE_OPERAND_RESULT_NUMBER)
101 value >>= 16;
102 ret = value;
104 else
106 if (**strp == '-')
108 long value;
109 errmsg = cgen_parse_signed_integer (cd, strp, opindex, &value);
110 ret = value;
112 else
114 unsigned long value;
115 errmsg = cgen_parse_unsigned_integer (cd, strp, opindex, &value);
116 ret = value;
120 *valuep = ((ret & 0xffff) ^ 0x8000) - 0x8000;
121 return errmsg;
124 /* Handle lo(). */
126 static const char *
127 parse_lo16 (cd, strp, opindex, valuep)
128 CGEN_CPU_DESC cd;
129 const char **strp;
130 int opindex;
131 unsigned long *valuep;
133 const char *errmsg;
134 enum cgen_parse_operand_result result_type;
135 unsigned long ret;
137 if (**strp == '#')
138 ++*strp;
140 if (strncasecmp (*strp, "lo(", 3) == 0)
142 bfd_vma value;
144 *strp += 3;
145 #if 0
146 errmsg = cgen_parse_signed_integer (cd, strp, opindex, valuep);
147 if (errmsg != NULL)
148 fprintf (stderr, "parse_lo: %s\n", errmsg);
150 if (errmsg != NULL)
151 #endif
152 errmsg = cgen_parse_address (cd, strp, opindex, BFD_RELOC_LO16,
153 &result_type, &value);
154 if (**strp != ')')
155 return _("missing `)'");
157 ++*strp;
158 ret = value;
160 else
162 if (**strp == '-')
164 long value;
165 errmsg = cgen_parse_signed_integer (cd, strp, opindex, &value);
166 ret = value;
168 else
170 unsigned long value;
171 errmsg = cgen_parse_unsigned_integer (cd, strp, opindex, &value);
172 ret = value;
176 *valuep = ((ret & 0xffff) ^ 0x8000) - 0x8000;
177 return errmsg;
180 /* -- */
182 const char * openrisc_cgen_parse_operand
183 PARAMS ((CGEN_CPU_DESC, int, const char **, CGEN_FIELDS *));
185 /* Main entry point for operand parsing.
187 This function is basically just a big switch statement. Earlier versions
188 used tables to look up the function to use, but
189 - if the table contains both assembler and disassembler functions then
190 the disassembler contains much of the assembler and vice-versa,
191 - there's a lot of inlining possibilities as things grow,
192 - using a switch statement avoids the function call overhead.
194 This function could be moved into `parse_insn_normal', but keeping it
195 separate makes clear the interface between `parse_insn_normal' and each of
196 the handlers. */
198 const char *
199 openrisc_cgen_parse_operand (cd, opindex, strp, fields)
200 CGEN_CPU_DESC cd;
201 int opindex;
202 const char ** strp;
203 CGEN_FIELDS * fields;
205 const char * errmsg = NULL;
206 /* Used by scalar operands that still need to be parsed. */
207 long junk ATTRIBUTE_UNUSED;
209 switch (opindex)
211 case OPENRISC_OPERAND_ABS_26 :
213 bfd_vma value;
214 errmsg = cgen_parse_address (cd, strp, OPENRISC_OPERAND_ABS_26, 0, NULL, & value);
215 fields->f_abs26 = value;
217 break;
218 case OPENRISC_OPERAND_DISP_26 :
220 bfd_vma value;
221 errmsg = cgen_parse_address (cd, strp, OPENRISC_OPERAND_DISP_26, 0, NULL, & value);
222 fields->f_disp26 = value;
224 break;
225 case OPENRISC_OPERAND_HI16 :
226 errmsg = parse_hi16 (cd, strp, OPENRISC_OPERAND_HI16, &fields->f_simm16);
227 break;
228 case OPENRISC_OPERAND_LO16 :
229 errmsg = parse_lo16 (cd, strp, OPENRISC_OPERAND_LO16, &fields->f_lo16);
230 break;
231 case OPENRISC_OPERAND_OP_F_23 :
232 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_OP_F_23, &fields->f_op4);
233 break;
234 case OPENRISC_OPERAND_OP_F_3 :
235 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_OP_F_3, &fields->f_op5);
236 break;
237 case OPENRISC_OPERAND_RA :
238 errmsg = cgen_parse_keyword (cd, strp, & openrisc_cgen_opval_h_gr, & fields->f_r2);
239 break;
240 case OPENRISC_OPERAND_RB :
241 errmsg = cgen_parse_keyword (cd, strp, & openrisc_cgen_opval_h_gr, & fields->f_r3);
242 break;
243 case OPENRISC_OPERAND_RD :
244 errmsg = cgen_parse_keyword (cd, strp, & openrisc_cgen_opval_h_gr, & fields->f_r1);
245 break;
246 case OPENRISC_OPERAND_SIMM_16 :
247 errmsg = cgen_parse_signed_integer (cd, strp, OPENRISC_OPERAND_SIMM_16, &fields->f_simm16);
248 break;
249 case OPENRISC_OPERAND_UI16NC :
250 errmsg = parse_lo16 (cd, strp, OPENRISC_OPERAND_UI16NC, &fields->f_i16nc);
251 break;
252 case OPENRISC_OPERAND_UIMM_16 :
253 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_UIMM_16, &fields->f_uimm16);
254 break;
255 case OPENRISC_OPERAND_UIMM_5 :
256 errmsg = cgen_parse_unsigned_integer (cd, strp, OPENRISC_OPERAND_UIMM_5, &fields->f_uimm5);
257 break;
259 default :
260 /* xgettext:c-format */
261 fprintf (stderr, _("Unrecognized field %d while parsing.\n"), opindex);
262 abort ();
265 return errmsg;
268 cgen_parse_fn * const openrisc_cgen_parse_handlers[] =
270 parse_insn_normal,
273 void
274 openrisc_cgen_init_asm (cd)
275 CGEN_CPU_DESC cd;
277 openrisc_cgen_init_opcode_table (cd);
278 openrisc_cgen_init_ibld_table (cd);
279 cd->parse_handlers = & openrisc_cgen_parse_handlers[0];
280 cd->parse_operand = openrisc_cgen_parse_operand;
285 /* Regex construction routine.
287 This translates an opcode syntax string into a regex string,
288 by replacing any non-character syntax element (such as an
289 opcode) with the pattern '.*'
291 It then compiles the regex and stores it in the opcode, for
292 later use by openrisc_cgen_assemble_insn
294 Returns NULL for success, an error message for failure. */
296 char *
297 openrisc_cgen_build_insn_regex (CGEN_INSN *insn)
299 CGEN_OPCODE *opc = (CGEN_OPCODE *) CGEN_INSN_OPCODE (insn);
300 const char *mnem = CGEN_INSN_MNEMONIC (insn);
301 char rxbuf[CGEN_MAX_RX_ELEMENTS];
302 char *rx = rxbuf;
303 const CGEN_SYNTAX_CHAR_TYPE *syn;
304 int reg_err;
306 syn = CGEN_SYNTAX_STRING (CGEN_OPCODE_SYNTAX (opc));
308 /* Mnemonics come first in the syntax string. */
309 if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
310 return _("missing mnemonic in syntax string");
311 ++syn;
313 /* Generate a case sensitive regular expression that emulates case
314 insensitive matching in the "C" locale. We cannot generate a case
315 insensitive regular expression because in Turkish locales, 'i' and 'I'
316 are not equal modulo case conversion. */
318 /* Copy the literal mnemonic out of the insn. */
319 for (; *mnem; mnem++)
321 char c = *mnem;
323 if (ISALPHA (c))
325 *rx++ = '[';
326 *rx++ = TOLOWER (c);
327 *rx++ = TOUPPER (c);
328 *rx++ = ']';
330 else
331 *rx++ = c;
334 /* Copy any remaining literals from the syntax string into the rx. */
335 for(; * syn != 0 && rx <= rxbuf + (CGEN_MAX_RX_ELEMENTS - 7 - 4); ++syn)
337 if (CGEN_SYNTAX_CHAR_P (* syn))
339 char c = CGEN_SYNTAX_CHAR (* syn);
341 switch (c)
343 /* Escape any regex metacharacters in the syntax. */
344 case '.': case '[': case '\\':
345 case '*': case '^': case '$':
347 #ifdef CGEN_ESCAPE_EXTENDED_REGEX
348 case '?': case '{': case '}':
349 case '(': case ')': case '*':
350 case '|': case '+': case ']':
351 #endif
352 *rx++ = '\\';
353 *rx++ = c;
354 break;
356 default:
357 if (ISALPHA (c))
359 *rx++ = '[';
360 *rx++ = TOLOWER (c);
361 *rx++ = TOUPPER (c);
362 *rx++ = ']';
364 else
365 *rx++ = c;
366 break;
369 else
371 /* Replace non-syntax fields with globs. */
372 *rx++ = '.';
373 *rx++ = '*';
377 /* Trailing whitespace ok. */
378 * rx++ = '[';
379 * rx++ = ' ';
380 * rx++ = '\t';
381 * rx++ = ']';
382 * rx++ = '*';
384 /* But anchor it after that. */
385 * rx++ = '$';
386 * rx = '\0';
388 CGEN_INSN_RX (insn) = xmalloc (sizeof (regex_t));
389 reg_err = regcomp ((regex_t *) CGEN_INSN_RX (insn), rxbuf, REG_NOSUB);
391 if (reg_err == 0)
392 return NULL;
393 else
395 static char msg[80];
397 regerror (reg_err, (regex_t *) CGEN_INSN_RX (insn), msg, 80);
398 regfree ((regex_t *) CGEN_INSN_RX (insn));
399 free (CGEN_INSN_RX (insn));
400 (CGEN_INSN_RX (insn)) = NULL;
401 return msg;
406 /* Default insn parser.
408 The syntax string is scanned and operands are parsed and stored in FIELDS.
409 Relocs are queued as we go via other callbacks.
411 ??? Note that this is currently an all-or-nothing parser. If we fail to
412 parse the instruction, we return 0 and the caller will start over from
413 the beginning. Backtracking will be necessary in parsing subexpressions,
414 but that can be handled there. Not handling backtracking here may get
415 expensive in the case of the m68k. Deal with later.
417 Returns NULL for success, an error message for failure. */
419 static const char *
420 parse_insn_normal (CGEN_CPU_DESC cd,
421 const CGEN_INSN *insn,
422 const char **strp,
423 CGEN_FIELDS *fields)
425 /* ??? Runtime added insns not handled yet. */
426 const CGEN_SYNTAX *syntax = CGEN_INSN_SYNTAX (insn);
427 const char *str = *strp;
428 const char *errmsg;
429 const char *p;
430 const CGEN_SYNTAX_CHAR_TYPE * syn;
431 #ifdef CGEN_MNEMONIC_OPERANDS
432 /* FIXME: wip */
433 int past_opcode_p;
434 #endif
436 /* For now we assume the mnemonic is first (there are no leading operands).
437 We can parse it without needing to set up operand parsing.
438 GAS's input scrubber will ensure mnemonics are lowercase, but we may
439 not be called from GAS. */
440 p = CGEN_INSN_MNEMONIC (insn);
441 while (*p && TOLOWER (*p) == TOLOWER (*str))
442 ++p, ++str;
444 if (* p)
445 return _("unrecognized instruction");
447 #ifndef CGEN_MNEMONIC_OPERANDS
448 if (* str && ! ISSPACE (* str))
449 return _("unrecognized instruction");
450 #endif
452 CGEN_INIT_PARSE (cd);
453 cgen_init_parse_operand (cd);
454 #ifdef CGEN_MNEMONIC_OPERANDS
455 past_opcode_p = 0;
456 #endif
458 /* We don't check for (*str != '\0') here because we want to parse
459 any trailing fake arguments in the syntax string. */
460 syn = CGEN_SYNTAX_STRING (syntax);
462 /* Mnemonics come first for now, ensure valid string. */
463 if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
464 abort ();
466 ++syn;
468 while (* syn != 0)
470 /* Non operand chars must match exactly. */
471 if (CGEN_SYNTAX_CHAR_P (* syn))
473 /* FIXME: While we allow for non-GAS callers above, we assume the
474 first char after the mnemonic part is a space. */
475 /* FIXME: We also take inappropriate advantage of the fact that
476 GAS's input scrubber will remove extraneous blanks. */
477 if (TOLOWER (*str) == TOLOWER (CGEN_SYNTAX_CHAR (* syn)))
479 #ifdef CGEN_MNEMONIC_OPERANDS
480 if (CGEN_SYNTAX_CHAR(* syn) == ' ')
481 past_opcode_p = 1;
482 #endif
483 ++ syn;
484 ++ str;
486 else if (*str)
488 /* Syntax char didn't match. Can't be this insn. */
489 static char msg [80];
491 /* xgettext:c-format */
492 sprintf (msg, _("syntax error (expected char `%c', found `%c')"),
493 CGEN_SYNTAX_CHAR(*syn), *str);
494 return msg;
496 else
498 /* Ran out of input. */
499 static char msg [80];
501 /* xgettext:c-format */
502 sprintf (msg, _("syntax error (expected char `%c', found end of instruction)"),
503 CGEN_SYNTAX_CHAR(*syn));
504 return msg;
506 continue;
509 /* We have an operand of some sort. */
510 errmsg = cd->parse_operand (cd, CGEN_SYNTAX_FIELD (*syn),
511 &str, fields);
512 if (errmsg)
513 return errmsg;
515 /* Done with this operand, continue with next one. */
516 ++ syn;
519 /* If we're at the end of the syntax string, we're done. */
520 if (* syn == 0)
522 /* FIXME: For the moment we assume a valid `str' can only contain
523 blanks now. IE: We needn't try again with a longer version of
524 the insn and it is assumed that longer versions of insns appear
525 before shorter ones (eg: lsr r2,r3,1 vs lsr r2,r3). */
526 while (ISSPACE (* str))
527 ++ str;
529 if (* str != '\0')
530 return _("junk at end of line"); /* FIXME: would like to include `str' */
532 return NULL;
535 /* We couldn't parse it. */
536 return _("unrecognized instruction");
539 /* Main entry point.
540 This routine is called for each instruction to be assembled.
541 STR points to the insn to be assembled.
542 We assume all necessary tables have been initialized.
543 The assembled instruction, less any fixups, is stored in BUF.
544 Remember that if CGEN_INT_INSN_P then BUF is an int and thus the value
545 still needs to be converted to target byte order, otherwise BUF is an array
546 of bytes in target byte order.
547 The result is a pointer to the insn's entry in the opcode table,
548 or NULL if an error occured (an error message will have already been
549 printed).
551 Note that when processing (non-alias) macro-insns,
552 this function recurses.
554 ??? It's possible to make this cpu-independent.
555 One would have to deal with a few minor things.
556 At this point in time doing so would be more of a curiosity than useful
557 [for example this file isn't _that_ big], but keeping the possibility in
558 mind helps keep the design clean. */
560 const CGEN_INSN *
561 openrisc_cgen_assemble_insn (CGEN_CPU_DESC cd,
562 const char *str,
563 CGEN_FIELDS *fields,
564 CGEN_INSN_BYTES_PTR buf,
565 char **errmsg)
567 const char *start;
568 CGEN_INSN_LIST *ilist;
569 const char *parse_errmsg = NULL;
570 const char *insert_errmsg = NULL;
571 int recognized_mnemonic = 0;
573 /* Skip leading white space. */
574 while (ISSPACE (* str))
575 ++ str;
577 /* The instructions are stored in hashed lists.
578 Get the first in the list. */
579 ilist = CGEN_ASM_LOOKUP_INSN (cd, str);
581 /* Keep looking until we find a match. */
582 start = str;
583 for ( ; ilist != NULL ; ilist = CGEN_ASM_NEXT_INSN (ilist))
585 const CGEN_INSN *insn = ilist->insn;
586 recognized_mnemonic = 1;
588 #ifdef CGEN_VALIDATE_INSN_SUPPORTED
589 /* Not usually needed as unsupported opcodes
590 shouldn't be in the hash lists. */
591 /* Is this insn supported by the selected cpu? */
592 if (! openrisc_cgen_insn_supported (cd, insn))
593 continue;
594 #endif
595 /* If the RELAXED attribute is set, this is an insn that shouldn't be
596 chosen immediately. Instead, it is used during assembler/linker
597 relaxation if possible. */
598 if (CGEN_INSN_ATTR_VALUE (insn, CGEN_INSN_RELAXED) != 0)
599 continue;
601 str = start;
603 /* Skip this insn if str doesn't look right lexically. */
604 if (CGEN_INSN_RX (insn) != NULL &&
605 regexec ((regex_t *) CGEN_INSN_RX (insn), str, 0, NULL, 0) == REG_NOMATCH)
606 continue;
608 /* Allow parse/insert handlers to obtain length of insn. */
609 CGEN_FIELDS_BITSIZE (fields) = CGEN_INSN_BITSIZE (insn);
611 parse_errmsg = CGEN_PARSE_FN (cd, insn) (cd, insn, & str, fields);
612 if (parse_errmsg != NULL)
613 continue;
615 /* ??? 0 is passed for `pc'. */
616 insert_errmsg = CGEN_INSERT_FN (cd, insn) (cd, insn, fields, buf,
617 (bfd_vma) 0);
618 if (insert_errmsg != NULL)
619 continue;
621 /* It is up to the caller to actually output the insn and any
622 queued relocs. */
623 return insn;
627 static char errbuf[150];
628 #ifdef CGEN_VERBOSE_ASSEMBLER_ERRORS
629 const char *tmp_errmsg;
631 /* If requesting verbose error messages, use insert_errmsg.
632 Failing that, use parse_errmsg. */
633 tmp_errmsg = (insert_errmsg ? insert_errmsg :
634 parse_errmsg ? parse_errmsg :
635 recognized_mnemonic ?
636 _("unrecognized form of instruction") :
637 _("unrecognized instruction"));
639 if (strlen (start) > 50)
640 /* xgettext:c-format */
641 sprintf (errbuf, "%s `%.50s...'", tmp_errmsg, start);
642 else
643 /* xgettext:c-format */
644 sprintf (errbuf, "%s `%.50s'", tmp_errmsg, start);
645 #else
646 if (strlen (start) > 50)
647 /* xgettext:c-format */
648 sprintf (errbuf, _("bad instruction `%.50s...'"), start);
649 else
650 /* xgettext:c-format */
651 sprintf (errbuf, _("bad instruction `%.50s'"), start);
652 #endif
654 *errmsg = errbuf;
655 return NULL;
659 #if 0 /* This calls back to GAS which we can't do without care. */
661 /* Record each member of OPVALS in the assembler's symbol table.
662 This lets GAS parse registers for us.
663 ??? Interesting idea but not currently used. */
665 /* Record each member of OPVALS in the assembler's symbol table.
666 FIXME: Not currently used. */
668 void
669 openrisc_cgen_asm_hash_keywords (CGEN_CPU_DESC cd, CGEN_KEYWORD *opvals)
671 CGEN_KEYWORD_SEARCH search = cgen_keyword_search_init (opvals, NULL);
672 const CGEN_KEYWORD_ENTRY * ke;
674 while ((ke = cgen_keyword_search_next (& search)) != NULL)
676 #if 0 /* Unnecessary, should be done in the search routine. */
677 if (! openrisc_cgen_opval_supported (ke))
678 continue;
679 #endif
680 cgen_asm_record_register (cd, ke->name, ke->value);
684 #endif /* 0 */