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)
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.
33 #include "openrisc-desc.h"
34 #include "openrisc-opc.h"
37 #include "libiberty.h"
38 #include "safe-ctype.h"
41 #define min(a,b) ((a) < (b) ? (a) : (b))
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. */
52 #define CGEN_VERBOSE_ASSEMBLER_ERRORS
54 static const char * parse_hi16
55 PARAMS ((CGEN_CPU_DESC
, const char **, int, long *));
56 static const char * parse_lo16
57 PARAMS ((CGEN_CPU_DESC
, const char **, int, long *));
60 openrisc_sign_extend_16bit (value
)
63 return ((value
& 0xffff) ^ 0x8000) - 0x8000;
69 parse_hi16 (cd
, strp
, opindex
, valuep
)
76 enum cgen_parse_operand_result result_type
;
82 if (strncasecmp (*strp
, "hi(", 3) == 0)
87 errmsg
= cgen_parse_address (cd
, strp
, opindex
, BFD_RELOC_HI16
,
88 &result_type
, &value
);
90 return _("missing `)'");
94 && result_type
== CGEN_PARSE_OPERAND_RESULT_NUMBER
)
104 errmsg
= cgen_parse_signed_integer (cd
, strp
, opindex
, &value
);
111 errmsg
= cgen_parse_unsigned_integer (cd
, strp
, opindex
, &value
);
116 *valuep
= ((ret
& 0xffff) ^ 0x8000) - 0x8000;
123 parse_lo16 (cd
, strp
, opindex
, valuep
)
130 enum cgen_parse_operand_result result_type
;
136 if (strncasecmp (*strp
, "lo(", 3) == 0)
141 errmsg
= cgen_parse_address (cd
, strp
, opindex
, BFD_RELOC_LO16
,
142 &result_type
, &value
);
144 return _("missing `)'");
155 errmsg
= cgen_parse_signed_integer (cd
, strp
, opindex
, &value
);
162 errmsg
= cgen_parse_unsigned_integer (cd
, strp
, opindex
, &value
);
167 *valuep
= ((ret
& 0xffff) ^ 0x8000) - 0x8000;
173 const char * openrisc_cgen_parse_operand
174 PARAMS ((CGEN_CPU_DESC
, int, const char **, CGEN_FIELDS
*));
176 /* Main entry point for operand parsing.
178 This function is basically just a big switch statement. Earlier versions
179 used tables to look up the function to use, but
180 - if the table contains both assembler and disassembler functions then
181 the disassembler contains much of the assembler and vice-versa,
182 - there's a lot of inlining possibilities as things grow,
183 - using a switch statement avoids the function call overhead.
185 This function could be moved into `parse_insn_normal', but keeping it
186 separate makes clear the interface between `parse_insn_normal' and each of
190 openrisc_cgen_parse_operand (cd
, opindex
, strp
, fields
)
194 CGEN_FIELDS
* fields
;
196 const char * errmsg
= NULL
;
197 /* Used by scalar operands that still need to be parsed. */
198 long junk ATTRIBUTE_UNUSED
;
202 case OPENRISC_OPERAND_ABS_26
:
205 errmsg
= cgen_parse_address (cd
, strp
, OPENRISC_OPERAND_ABS_26
, 0, NULL
, & value
);
206 fields
->f_abs26
= value
;
209 case OPENRISC_OPERAND_DISP_26
:
212 errmsg
= cgen_parse_address (cd
, strp
, OPENRISC_OPERAND_DISP_26
, 0, NULL
, & value
);
213 fields
->f_disp26
= value
;
216 case OPENRISC_OPERAND_HI16
:
217 errmsg
= parse_hi16 (cd
, strp
, OPENRISC_OPERAND_HI16
, (long *) (& fields
->f_simm16
));
219 case OPENRISC_OPERAND_LO16
:
220 errmsg
= parse_lo16 (cd
, strp
, OPENRISC_OPERAND_LO16
, (long *) (& fields
->f_lo16
));
222 case OPENRISC_OPERAND_OP_F_23
:
223 errmsg
= cgen_parse_unsigned_integer (cd
, strp
, OPENRISC_OPERAND_OP_F_23
, (unsigned long *) (& fields
->f_op4
));
225 case OPENRISC_OPERAND_OP_F_3
:
226 errmsg
= cgen_parse_unsigned_integer (cd
, strp
, OPENRISC_OPERAND_OP_F_3
, (unsigned long *) (& fields
->f_op5
));
228 case OPENRISC_OPERAND_RA
:
229 errmsg
= cgen_parse_keyword (cd
, strp
, & openrisc_cgen_opval_h_gr
, & fields
->f_r2
);
231 case OPENRISC_OPERAND_RB
:
232 errmsg
= cgen_parse_keyword (cd
, strp
, & openrisc_cgen_opval_h_gr
, & fields
->f_r3
);
234 case OPENRISC_OPERAND_RD
:
235 errmsg
= cgen_parse_keyword (cd
, strp
, & openrisc_cgen_opval_h_gr
, & fields
->f_r1
);
237 case OPENRISC_OPERAND_SIMM_16
:
238 errmsg
= cgen_parse_signed_integer (cd
, strp
, OPENRISC_OPERAND_SIMM_16
, (long *) (& fields
->f_simm16
));
240 case OPENRISC_OPERAND_UI16NC
:
241 errmsg
= parse_lo16 (cd
, strp
, OPENRISC_OPERAND_UI16NC
, (long *) (& fields
->f_i16nc
));
243 case OPENRISC_OPERAND_UIMM_16
:
244 errmsg
= cgen_parse_unsigned_integer (cd
, strp
, OPENRISC_OPERAND_UIMM_16
, (unsigned long *) (& fields
->f_uimm16
));
246 case OPENRISC_OPERAND_UIMM_5
:
247 errmsg
= cgen_parse_unsigned_integer (cd
, strp
, OPENRISC_OPERAND_UIMM_5
, (unsigned long *) (& fields
->f_uimm5
));
251 /* xgettext:c-format */
252 fprintf (stderr
, _("Unrecognized field %d while parsing.\n"), opindex
);
259 cgen_parse_fn
* const openrisc_cgen_parse_handlers
[] =
265 openrisc_cgen_init_asm (cd
)
268 openrisc_cgen_init_opcode_table (cd
);
269 openrisc_cgen_init_ibld_table (cd
);
270 cd
->parse_handlers
= & openrisc_cgen_parse_handlers
[0];
271 cd
->parse_operand
= openrisc_cgen_parse_operand
;
276 /* Regex construction routine.
278 This translates an opcode syntax string into a regex string,
279 by replacing any non-character syntax element (such as an
280 opcode) with the pattern '.*'
282 It then compiles the regex and stores it in the opcode, for
283 later use by openrisc_cgen_assemble_insn
285 Returns NULL for success, an error message for failure. */
288 openrisc_cgen_build_insn_regex (CGEN_INSN
*insn
)
290 CGEN_OPCODE
*opc
= (CGEN_OPCODE
*) CGEN_INSN_OPCODE (insn
);
291 const char *mnem
= CGEN_INSN_MNEMONIC (insn
);
292 char rxbuf
[CGEN_MAX_RX_ELEMENTS
];
294 const CGEN_SYNTAX_CHAR_TYPE
*syn
;
297 syn
= CGEN_SYNTAX_STRING (CGEN_OPCODE_SYNTAX (opc
));
299 /* Mnemonics come first in the syntax string. */
300 if (! CGEN_SYNTAX_MNEMONIC_P (* syn
))
301 return _("missing mnemonic in syntax string");
304 /* Generate a case sensitive regular expression that emulates case
305 insensitive matching in the "C" locale. We cannot generate a case
306 insensitive regular expression because in Turkish locales, 'i' and 'I'
307 are not equal modulo case conversion. */
309 /* Copy the literal mnemonic out of the insn. */
310 for (; *mnem
; mnem
++)
325 /* Copy any remaining literals from the syntax string into the rx. */
326 for(; * syn
!= 0 && rx
<= rxbuf
+ (CGEN_MAX_RX_ELEMENTS
- 7 - 4); ++syn
)
328 if (CGEN_SYNTAX_CHAR_P (* syn
))
330 char c
= CGEN_SYNTAX_CHAR (* syn
);
334 /* Escape any regex metacharacters in the syntax. */
335 case '.': case '[': case '\\':
336 case '*': case '^': case '$':
338 #ifdef CGEN_ESCAPE_EXTENDED_REGEX
339 case '?': case '{': case '}':
340 case '(': case ')': case '*':
341 case '|': case '+': case ']':
362 /* Replace non-syntax fields with globs. */
368 /* Trailing whitespace ok. */
375 /* But anchor it after that. */
379 CGEN_INSN_RX (insn
) = xmalloc (sizeof (regex_t
));
380 reg_err
= regcomp ((regex_t
*) CGEN_INSN_RX (insn
), rxbuf
, REG_NOSUB
);
388 regerror (reg_err
, (regex_t
*) CGEN_INSN_RX (insn
), msg
, 80);
389 regfree ((regex_t
*) CGEN_INSN_RX (insn
));
390 free (CGEN_INSN_RX (insn
));
391 (CGEN_INSN_RX (insn
)) = NULL
;
397 /* Default insn parser.
399 The syntax string is scanned and operands are parsed and stored in FIELDS.
400 Relocs are queued as we go via other callbacks.
402 ??? Note that this is currently an all-or-nothing parser. If we fail to
403 parse the instruction, we return 0 and the caller will start over from
404 the beginning. Backtracking will be necessary in parsing subexpressions,
405 but that can be handled there. Not handling backtracking here may get
406 expensive in the case of the m68k. Deal with later.
408 Returns NULL for success, an error message for failure. */
411 parse_insn_normal (CGEN_CPU_DESC cd
,
412 const CGEN_INSN
*insn
,
416 /* ??? Runtime added insns not handled yet. */
417 const CGEN_SYNTAX
*syntax
= CGEN_INSN_SYNTAX (insn
);
418 const char *str
= *strp
;
421 const CGEN_SYNTAX_CHAR_TYPE
* syn
;
422 #ifdef CGEN_MNEMONIC_OPERANDS
427 /* For now we assume the mnemonic is first (there are no leading operands).
428 We can parse it without needing to set up operand parsing.
429 GAS's input scrubber will ensure mnemonics are lowercase, but we may
430 not be called from GAS. */
431 p
= CGEN_INSN_MNEMONIC (insn
);
432 while (*p
&& TOLOWER (*p
) == TOLOWER (*str
))
436 return _("unrecognized instruction");
438 #ifndef CGEN_MNEMONIC_OPERANDS
439 if (* str
&& ! ISSPACE (* str
))
440 return _("unrecognized instruction");
443 CGEN_INIT_PARSE (cd
);
444 cgen_init_parse_operand (cd
);
445 #ifdef CGEN_MNEMONIC_OPERANDS
449 /* We don't check for (*str != '\0') here because we want to parse
450 any trailing fake arguments in the syntax string. */
451 syn
= CGEN_SYNTAX_STRING (syntax
);
453 /* Mnemonics come first for now, ensure valid string. */
454 if (! CGEN_SYNTAX_MNEMONIC_P (* syn
))
461 /* Non operand chars must match exactly. */
462 if (CGEN_SYNTAX_CHAR_P (* syn
))
464 /* FIXME: While we allow for non-GAS callers above, we assume the
465 first char after the mnemonic part is a space. */
466 /* FIXME: We also take inappropriate advantage of the fact that
467 GAS's input scrubber will remove extraneous blanks. */
468 if (TOLOWER (*str
) == TOLOWER (CGEN_SYNTAX_CHAR (* syn
)))
470 #ifdef CGEN_MNEMONIC_OPERANDS
471 if (CGEN_SYNTAX_CHAR(* syn
) == ' ')
479 /* Syntax char didn't match. Can't be this insn. */
480 static char msg
[80];
482 /* xgettext:c-format */
483 sprintf (msg
, _("syntax error (expected char `%c', found `%c')"),
484 CGEN_SYNTAX_CHAR(*syn
), *str
);
489 /* Ran out of input. */
490 static char msg
[80];
492 /* xgettext:c-format */
493 sprintf (msg
, _("syntax error (expected char `%c', found end of instruction)"),
494 CGEN_SYNTAX_CHAR(*syn
));
500 /* We have an operand of some sort. */
501 errmsg
= cd
->parse_operand (cd
, CGEN_SYNTAX_FIELD (*syn
),
506 /* Done with this operand, continue with next one. */
510 /* If we're at the end of the syntax string, we're done. */
513 /* FIXME: For the moment we assume a valid `str' can only contain
514 blanks now. IE: We needn't try again with a longer version of
515 the insn and it is assumed that longer versions of insns appear
516 before shorter ones (eg: lsr r2,r3,1 vs lsr r2,r3). */
517 while (ISSPACE (* str
))
521 return _("junk at end of line"); /* FIXME: would like to include `str' */
526 /* We couldn't parse it. */
527 return _("unrecognized instruction");
531 This routine is called for each instruction to be assembled.
532 STR points to the insn to be assembled.
533 We assume all necessary tables have been initialized.
534 The assembled instruction, less any fixups, is stored in BUF.
535 Remember that if CGEN_INT_INSN_P then BUF is an int and thus the value
536 still needs to be converted to target byte order, otherwise BUF is an array
537 of bytes in target byte order.
538 The result is a pointer to the insn's entry in the opcode table,
539 or NULL if an error occured (an error message will have already been
542 Note that when processing (non-alias) macro-insns,
543 this function recurses.
545 ??? It's possible to make this cpu-independent.
546 One would have to deal with a few minor things.
547 At this point in time doing so would be more of a curiosity than useful
548 [for example this file isn't _that_ big], but keeping the possibility in
549 mind helps keep the design clean. */
552 openrisc_cgen_assemble_insn (CGEN_CPU_DESC cd
,
555 CGEN_INSN_BYTES_PTR buf
,
559 CGEN_INSN_LIST
*ilist
;
560 const char *parse_errmsg
= NULL
;
561 const char *insert_errmsg
= NULL
;
562 int recognized_mnemonic
= 0;
564 /* Skip leading white space. */
565 while (ISSPACE (* str
))
568 /* The instructions are stored in hashed lists.
569 Get the first in the list. */
570 ilist
= CGEN_ASM_LOOKUP_INSN (cd
, str
);
572 /* Keep looking until we find a match. */
574 for ( ; ilist
!= NULL
; ilist
= CGEN_ASM_NEXT_INSN (ilist
))
576 const CGEN_INSN
*insn
= ilist
->insn
;
577 recognized_mnemonic
= 1;
579 #ifdef CGEN_VALIDATE_INSN_SUPPORTED
580 /* Not usually needed as unsupported opcodes
581 shouldn't be in the hash lists. */
582 /* Is this insn supported by the selected cpu? */
583 if (! openrisc_cgen_insn_supported (cd
, insn
))
586 /* If the RELAXED attribute is set, this is an insn that shouldn't be
587 chosen immediately. Instead, it is used during assembler/linker
588 relaxation if possible. */
589 if (CGEN_INSN_ATTR_VALUE (insn
, CGEN_INSN_RELAXED
) != 0)
594 /* Skip this insn if str doesn't look right lexically. */
595 if (CGEN_INSN_RX (insn
) != NULL
&&
596 regexec ((regex_t
*) CGEN_INSN_RX (insn
), str
, 0, NULL
, 0) == REG_NOMATCH
)
599 /* Allow parse/insert handlers to obtain length of insn. */
600 CGEN_FIELDS_BITSIZE (fields
) = CGEN_INSN_BITSIZE (insn
);
602 parse_errmsg
= CGEN_PARSE_FN (cd
, insn
) (cd
, insn
, & str
, fields
);
603 if (parse_errmsg
!= NULL
)
606 /* ??? 0 is passed for `pc'. */
607 insert_errmsg
= CGEN_INSERT_FN (cd
, insn
) (cd
, insn
, fields
, buf
,
609 if (insert_errmsg
!= NULL
)
612 /* It is up to the caller to actually output the insn and any
618 static char errbuf
[150];
619 #ifdef CGEN_VERBOSE_ASSEMBLER_ERRORS
620 const char *tmp_errmsg
;
622 /* If requesting verbose error messages, use insert_errmsg.
623 Failing that, use parse_errmsg. */
624 tmp_errmsg
= (insert_errmsg
? insert_errmsg
:
625 parse_errmsg
? parse_errmsg
:
626 recognized_mnemonic
?
627 _("unrecognized form of instruction") :
628 _("unrecognized instruction"));
630 if (strlen (start
) > 50)
631 /* xgettext:c-format */
632 sprintf (errbuf
, "%s `%.50s...'", tmp_errmsg
, start
);
634 /* xgettext:c-format */
635 sprintf (errbuf
, "%s `%.50s'", tmp_errmsg
, start
);
637 if (strlen (start
) > 50)
638 /* xgettext:c-format */
639 sprintf (errbuf
, _("bad instruction `%.50s...'"), start
);
641 /* xgettext:c-format */
642 sprintf (errbuf
, _("bad instruction `%.50s'"), start
);
650 #if 0 /* This calls back to GAS which we can't do without care. */
652 /* Record each member of OPVALS in the assembler's symbol table.
653 This lets GAS parse registers for us.
654 ??? Interesting idea but not currently used. */
656 /* Record each member of OPVALS in the assembler's symbol table.
657 FIXME: Not currently used. */
660 openrisc_cgen_asm_hash_keywords (CGEN_CPU_DESC cd
, CGEN_KEYWORD
*opvals
)
662 CGEN_KEYWORD_SEARCH search
= cgen_keyword_search_init (opvals
, NULL
);
663 const CGEN_KEYWORD_ENTRY
* ke
;
665 while ((ke
= cgen_keyword_search_next (& search
)) != NULL
)
667 #if 0 /* Unnecessary, should be done in the search routine. */
668 if (! openrisc_cgen_opval_supported (ke
))
671 cgen_asm_record_register (cd
, ke
->name
, ke
->value
);