/* Assembler interface for targets using CGEN. -*- C -*-
CGEN: Cpu tools GENerator
-This file is used to generate @arch@-asm.c.
+ THIS FILE IS MACHINE GENERATED WITH CGEN.
+ - the resultant file is machine generated, cgen-asm.in isn't
-Copyright (C) 1996, 1997 Free Software Foundation, Inc.
+ Copyright (C) 1996-2015 Free Software Foundation, Inc.
-This file is part of the GNU Binutils and GDB, the GNU debugger.
+ This file is part of libopcodes.
-This program is free software; you can redistribute it and/or modify
-it under the terms of the GNU General Public License as published by
-the Free Software Foundation; either version 2, or (at your option)
-any later version.
+ This library is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3, or (at your option)
+ any later version.
-This program is distributed in the hope that it will be useful,
-but WITHOUT ANY WARRANTY; without even the implied warranty of
-MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-GNU General Public License for more details.
+ It is distributed in the hope that it will be useful, but WITHOUT
+ ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public
+ License for more details.
-You should have received a copy of the GNU General Public License
-along with this program; if not, write to the Free Software
-Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software Foundation, Inc.,
+ 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA. */
+
+
+/* ??? Eventually more and more of this stuff can go to cpu-independent files.
+ Keep that in mind. */
#include "sysdep.h"
-#include <ctype.h>
#include <stdio.h>
#include "ansidecl.h"
#include "bfd.h"
-#include "@arch@-opc.h"
-
-/* ??? The layout of this stuff is still work in progress.
- For speed in assembly/disassembly, we use inline functions. That of course
- will only work for GCC. When this stuff is finished, we can decide whether
- to keep the inline functions (and only get the performance increase when
- compiled with GCC), or switch to macros, or use something else.
-*/
-
-static const char *parse_insn_normal
- PARAMS ((const CGEN_INSN *, const char **, CGEN_FIELDS *));
-static void insert_insn_normal
- PARAMS ((const CGEN_INSN *, CGEN_FIELDS *, cgen_insn_t *));
+#include "symcat.h"
+#include "@prefix@-desc.h"
+#include "@prefix@-opc.h"
+#include "opintl.h"
+#include "xregex.h"
+#include "libiberty.h"
+#include "safe-ctype.h"
+
+#undef min
+#define min(a,b) ((a) < (b) ? (a) : (b))
+#undef max
+#define max(a,b) ((a) > (b) ? (a) : (b))
+
+static const char * parse_insn_normal
+ (CGEN_CPU_DESC, const CGEN_INSN *, const char **, CGEN_FIELDS *);
+\f
+/* -- assembler routines inserted here. */
\f
-/* Default insertion routine.
-
- SHIFT is negative for left shifts, positive for right shifts.
- All bits of VALUE to be inserted must be valid as we don't handle
- signed vs unsigned shifts.
-
- ATTRS is a mask of the boolean attributes. We don't need any at the
- moment, but for consistency with extract_normal we have them. */
-
-/* FIXME: This duplicates functionality with bfd's howto table and
- bfd_install_relocation. */
-/* FIXME: For architectures where insns can be representable as ints,
- store insn in `field' struct and add registers, etc. while parsing. */
-
-static CGEN_INLINE void
-insert_normal (value, attrs, start, length, shift, total_length, buffer)
- long value;
- unsigned int attrs;
- int start, length, shift, total_length;
- char *buffer;
-{
- bfd_vma x;
-#if 0 /*def CGEN_INT_INSN*/
- *buffer |= ((value & ((1 << length) - 1))
- << (total_length - (start + length)));
-#else
- switch (total_length)
+/* Regex construction routine.
+
+ This translates an opcode syntax string into a regex string,
+ by replacing any non-character syntax element (such as an
+ opcode) with the pattern '.*'
+
+ It then compiles the regex and stores it in the opcode, for
+ later use by @arch@_cgen_assemble_insn
+
+ Returns NULL for success, an error message for failure. */
+
+char *
+@arch@_cgen_build_insn_regex (CGEN_INSN *insn)
+{
+ CGEN_OPCODE *opc = (CGEN_OPCODE *) CGEN_INSN_OPCODE (insn);
+ const char *mnem = CGEN_INSN_MNEMONIC (insn);
+ char rxbuf[CGEN_MAX_RX_ELEMENTS];
+ char *rx = rxbuf;
+ const CGEN_SYNTAX_CHAR_TYPE *syn;
+ int reg_err;
+
+ syn = CGEN_SYNTAX_STRING (CGEN_OPCODE_SYNTAX (opc));
+
+ /* Mnemonics come first in the syntax string. */
+ if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
+ return _("missing mnemonic in syntax string");
+ ++syn;
+
+ /* Generate a case sensitive regular expression that emulates case
+ insensitive matching in the "C" locale. We cannot generate a case
+ insensitive regular expression because in Turkish locales, 'i' and 'I'
+ are not equal modulo case conversion. */
+
+ /* Copy the literal mnemonic out of the insn. */
+ for (; *mnem; mnem++)
{
- case 8:
- x = *(unsigned char *) buffer;
- break;
- case 16:
- if (CGEN_CURRENT_ENDIAN == CGEN_ENDIAN_BIG)
- x = bfd_getb16 (buffer);
+ char c = *mnem;
+
+ if (ISALPHA (c))
+ {
+ *rx++ = '[';
+ *rx++ = TOLOWER (c);
+ *rx++ = TOUPPER (c);
+ *rx++ = ']';
+ }
else
- x = bfd_getl16 (buffer);
- break;
- case 32:
- if (CGEN_CURRENT_ENDIAN == CGEN_ENDIAN_BIG)
- x = bfd_getb32 (buffer);
+ *rx++ = c;
+ }
+
+ /* Copy any remaining literals from the syntax string into the rx. */
+ for(; * syn != 0 && rx <= rxbuf + (CGEN_MAX_RX_ELEMENTS - 7 - 4); ++syn)
+ {
+ if (CGEN_SYNTAX_CHAR_P (* syn))
+ {
+ char c = CGEN_SYNTAX_CHAR (* syn);
+
+ switch (c)
+ {
+ /* Escape any regex metacharacters in the syntax. */
+ case '.': case '[': case '\\':
+ case '*': case '^': case '$':
+
+#ifdef CGEN_ESCAPE_EXTENDED_REGEX
+ case '?': case '{': case '}':
+ case '(': case ')': case '*':
+ case '|': case '+': case ']':
+#endif
+ *rx++ = '\\';
+ *rx++ = c;
+ break;
+
+ default:
+ if (ISALPHA (c))
+ {
+ *rx++ = '[';
+ *rx++ = TOLOWER (c);
+ *rx++ = TOUPPER (c);
+ *rx++ = ']';
+ }
+ else
+ *rx++ = c;
+ break;
+ }
+ }
else
- x = bfd_getl32 (buffer);
- break;
- default :
- abort ();
+ {
+ /* Replace non-syntax fields with globs. */
+ *rx++ = '.';
+ *rx++ = '*';
+ }
}
- if (shift < 0)
- value <<= -shift;
- else
- value >>= shift;
+ /* Trailing whitespace ok. */
+ * rx++ = '[';
+ * rx++ = ' ';
+ * rx++ = '\t';
+ * rx++ = ']';
+ * rx++ = '*';
+
+ /* But anchor it after that. */
+ * rx++ = '$';
+ * rx = '\0';
- x |= ((value & ((1 << length) - 1))
- << (total_length - (start + length)));
+ CGEN_INSN_RX (insn) = xmalloc (sizeof (regex_t));
+ reg_err = regcomp ((regex_t *) CGEN_INSN_RX (insn), rxbuf, REG_NOSUB);
- switch (total_length)
+ if (reg_err == 0)
+ return NULL;
+ else
{
- case 8:
- *buffer = value;
- break;
- case 16:
- if (CGEN_CURRENT_ENDIAN == CGEN_ENDIAN_BIG)
- bfd_putb16 (x, buffer);
- else
- bfd_putl16 (x, buffer);
- break;
- case 32:
- if (CGEN_CURRENT_ENDIAN == CGEN_ENDIAN_BIG)
- bfd_putb32 (x, buffer);
- else
- bfd_putl32 (x, buffer);
- break;
- default :
- abort ();
+ static char msg[80];
+
+ regerror (reg_err, (regex_t *) CGEN_INSN_RX (insn), msg, 80);
+ regfree ((regex_t *) CGEN_INSN_RX (insn));
+ free (CGEN_INSN_RX (insn));
+ (CGEN_INSN_RX (insn)) = NULL;
+ return msg;
}
-#endif
}
-\f
-/* -- assembler routines inserted here */
+
\f
/* Default insn parser.
but that can be handled there. Not handling backtracking here may get
expensive in the case of the m68k. Deal with later.
- Returns NULL for success, an error message for failure.
-*/
+ Returns NULL for success, an error message for failure. */
static const char *
-parse_insn_normal (insn, strp, fields)
- const CGEN_INSN *insn;
- const char **strp;
- CGEN_FIELDS *fields;
+parse_insn_normal (CGEN_CPU_DESC cd,
+ const CGEN_INSN *insn,
+ const char **strp,
+ CGEN_FIELDS *fields)
{
+ /* ??? Runtime added insns not handled yet. */
const CGEN_SYNTAX *syntax = CGEN_INSN_SYNTAX (insn);
const char *str = *strp;
const char *errmsg;
const char *p;
- const unsigned char *syn;
+ const CGEN_SYNTAX_CHAR_TYPE * syn;
#ifdef CGEN_MNEMONIC_OPERANDS
+ /* FIXME: wip */
int past_opcode_p;
#endif
/* For now we assume the mnemonic is first (there are no leading operands).
- We can parse it without needing to set up operand parsing. */
+ We can parse it without needing to set up operand parsing.
+ GAS's input scrubber will ensure mnemonics are lowercase, but we may
+ not be called from GAS. */
p = CGEN_INSN_MNEMONIC (insn);
- while (*p && *p == *str)
+ while (*p && TOLOWER (*p) == TOLOWER (*str))
++p, ++str;
- if (*p || (*str && !isspace (*str)))
- return "unrecognized instruction";
- CGEN_INIT_PARSE ();
- cgen_init_parse_operand ();
+ if (* p)
+ return _("unrecognized instruction");
+
+#ifndef CGEN_MNEMONIC_OPERANDS
+ if (* str && ! ISSPACE (* str))
+ return _("unrecognized instruction");
+#endif
+
+ CGEN_INIT_PARSE (cd);
+ cgen_init_parse_operand (cd);
#ifdef CGEN_MNEMONIC_OPERANDS
past_opcode_p = 0;
#endif
/* We don't check for (*str != '\0') here because we want to parse
any trailing fake arguments in the syntax string. */
- syn = CGEN_SYNTAX_STRING (CGEN_INSN_SYNTAX (insn));
+ syn = CGEN_SYNTAX_STRING (syntax);
+
/* Mnemonics come first for now, ensure valid string. */
- if (! CGEN_SYNTAX_MNEMONIC_P (*syn))
+ if (! CGEN_SYNTAX_MNEMONIC_P (* syn))
abort ();
+
++syn;
- while (*syn != 0)
+
+ while (* syn != 0)
{
/* Non operand chars must match exactly. */
- /* FIXME: Need to better handle whitespace. */
- if (CGEN_SYNTAX_CHAR_P (*syn))
+ if (CGEN_SYNTAX_CHAR_P (* syn))
{
- if (*str == CGEN_SYNTAX_CHAR (*syn))
+ /* FIXME: While we allow for non-GAS callers above, we assume the
+ first char after the mnemonic part is a space. */
+ /* FIXME: We also take inappropriate advantage of the fact that
+ GAS's input scrubber will remove extraneous blanks. */
+ if (TOLOWER (*str) == TOLOWER (CGEN_SYNTAX_CHAR (* syn)))
{
#ifdef CGEN_MNEMONIC_OPERANDS
- if (*syn == ' ')
+ if (CGEN_SYNTAX_CHAR(* syn) == ' ')
past_opcode_p = 1;
#endif
- ++syn;
- ++str;
+ ++ syn;
+ ++ str;
}
- else
+ else if (*str)
{
/* Syntax char didn't match. Can't be this insn. */
- /* FIXME: would like to return "expected char `c'" */
- return "syntax error";
+ static char msg [80];
+
+ /* xgettext:c-format */
+ sprintf (msg, _("syntax error (expected char `%c', found `%c')"),
+ CGEN_SYNTAX_CHAR(*syn), *str);
+ return msg;
+ }
+ else
+ {
+ /* Ran out of input. */
+ static char msg [80];
+
+ /* xgettext:c-format */
+ sprintf (msg, _("syntax error (expected char `%c', found end of instruction)"),
+ CGEN_SYNTAX_CHAR(*syn));
+ return msg;
}
continue;
}
+#ifdef CGEN_MNEMONIC_OPERANDS
+ (void) past_opcode_p;
+#endif
/* We have an operand of some sort. */
- errmsg = @arch@_cgen_parse_operand (CGEN_SYNTAX_FIELD (*syn),
- &str, fields);
+ errmsg = cd->parse_operand (cd, CGEN_SYNTAX_FIELD (*syn), &str, fields);
if (errmsg)
return errmsg;
/* Done with this operand, continue with next one. */
- ++syn;
+ ++ syn;
}
/* If we're at the end of the syntax string, we're done. */
- if (*syn == '\0')
+ if (* syn == 0)
{
/* FIXME: For the moment we assume a valid `str' can only contain
blanks now. IE: We needn't try again with a longer version of
the insn and it is assumed that longer versions of insns appear
before shorter ones (eg: lsr r2,r3,1 vs lsr r2,r3). */
- while (isspace (*str))
- ++str;
+ while (ISSPACE (* str))
+ ++ str;
- if (*str != '\0')
- return "junk at end of line"; /* FIXME: would like to include `str' */
+ if (* str != '\0')
+ return _("junk at end of line"); /* FIXME: would like to include `str' */
return NULL;
}
/* We couldn't parse it. */
- return "unrecognized instruction";
-}
-
-/* Default insn builder (insert handler).
- The instruction is recorded in target byte order. */
-
-static void
-insert_insn_normal (insn, fields, buffer)
- const CGEN_INSN *insn;
- CGEN_FIELDS *fields;
- cgen_insn_t *buffer;
-{
- const CGEN_SYNTAX *syntax = CGEN_INSN_SYNTAX (insn);
- bfd_vma value;
- const unsigned char *syn;
-
- CGEN_INIT_INSERT ();
- value = CGEN_INSN_VALUE (insn);
-
- /* If we're recording insns as numbers (rather than a string of bytes),
- target byte order handling is deferred until later. */
-#undef min
-#define min(a,b) ((a) < (b) ? (a) : (b))
-#if 0 /*def CGEN_INT_INSN*/
- *buffer = value;
-#else
- switch (min (CGEN_BASE_INSN_BITSIZE, CGEN_FIELDS_BITSIZE (fields)))
- {
- case 8:
- *buffer = value;
- break;
- case 16:
- if (CGEN_CURRENT_ENDIAN == CGEN_ENDIAN_BIG)
- bfd_putb16 (value, (char *) buffer);
- else
- bfd_putl16 (value, (char *) buffer);
- break;
- case 32:
- if (CGEN_CURRENT_ENDIAN == CGEN_ENDIAN_BIG)
- bfd_putb32 (value, (char *) buffer);
- else
- bfd_putl32 (value, (char *) buffer);
- break;
- default:
- abort ();
- }
-#endif
-
- /* ??? Rather than scanning the syntax string again, we could store
- in `fields' a null terminated list of the fields that are present. */
-
- for (syn = CGEN_SYNTAX_STRING (syntax); *syn != '\0'; ++syn)
- {
- if (CGEN_SYNTAX_CHAR_P (*syn))
- continue;
-
- @arch@_cgen_insert_operand (CGEN_SYNTAX_FIELD (*syn), fields, buffer);
- }
+ return _("unrecognized instruction");
}
\f
/* Main entry point.
This routine is called for each instruction to be assembled.
STR points to the insn to be assembled.
We assume all necessary tables have been initialized.
+ The assembled instruction, less any fixups, is stored in BUF.
+ Remember that if CGEN_INT_INSN_P then BUF is an int and thus the value
+ still needs to be converted to target byte order, otherwise BUF is an array
+ of bytes in target byte order.
The result is a pointer to the insn's entry in the opcode table,
or NULL if an error occured (an error message will have already been
- printed). */
+ printed).
+
+ Note that when processing (non-alias) macro-insns,
+ this function recurses.
+
+ ??? It's possible to make this cpu-independent.
+ One would have to deal with a few minor things.
+ At this point in time doing so would be more of a curiosity than useful
+ [for example this file isn't _that_ big], but keeping the possibility in
+ mind helps keep the design clean. */
const CGEN_INSN *
-@arch@_cgen_assemble_insn (str, fields, buf, errmsg)
- const char *str;
- CGEN_FIELDS *fields;
- cgen_insn_t *buf;
- char **errmsg;
+@arch@_cgen_assemble_insn (CGEN_CPU_DESC cd,
+ const char *str,
+ CGEN_FIELDS *fields,
+ CGEN_INSN_BYTES_PTR buf,
+ char **errmsg)
{
const char *start;
CGEN_INSN_LIST *ilist;
+ const char *parse_errmsg = NULL;
+ const char *insert_errmsg = NULL;
+ int recognized_mnemonic = 0;
/* Skip leading white space. */
- while (isspace (*str))
- ++str;
+ while (ISSPACE (* str))
+ ++ str;
/* The instructions are stored in hashed lists.
Get the first in the list. */
- ilist = CGEN_ASM_LOOKUP_INSN (str);
+ ilist = CGEN_ASM_LOOKUP_INSN (cd, str);
/* Keep looking until we find a match. */
-
start = str;
for ( ; ilist != NULL ; ilist = CGEN_ASM_NEXT_INSN (ilist))
{
const CGEN_INSN *insn = ilist->insn;
+ recognized_mnemonic = 1;
-#if 0 /* not needed as unsupported opcodes shouldn't be in the hash lists */
+#ifdef CGEN_VALIDATE_INSN_SUPPORTED
+ /* Not usually needed as unsupported opcodes
+ shouldn't be in the hash lists. */
/* Is this insn supported by the selected cpu? */
- if (! @arch@_cgen_insn_supported (insn))
+ if (! @arch@_cgen_insn_supported (cd, insn))
continue;
#endif
-
-#if 1 /* FIXME: wip */
- /* If the RELAX attribute is set, this is an insn that shouldn't be
+ /* If the RELAXED attribute is set, this is an insn that shouldn't be
chosen immediately. Instead, it is used during assembler/linker
relaxation if possible. */
- if (CGEN_INSN_ATTR (insn, CGEN_INSN_RELAX) != 0)
+ if (CGEN_INSN_ATTR_VALUE (insn, CGEN_INSN_RELAXED) != 0)
continue;
-#endif
str = start;
- /* Record a default length for the insn. This will get set to the
- correct value while parsing. */
- /* FIXME: wip */
+ /* Skip this insn if str doesn't look right lexically. */
+ if (CGEN_INSN_RX (insn) != NULL &&
+ regexec ((regex_t *) CGEN_INSN_RX (insn), str, 0, NULL, 0) == REG_NOMATCH)
+ continue;
+
+ /* Allow parse/insert handlers to obtain length of insn. */
CGEN_FIELDS_BITSIZE (fields) = CGEN_INSN_BITSIZE (insn);
- /* ??? The extent to which moving the parse and insert handlers into
- this function (thus removing the function call) will speed things up
- is unclear. The simplicity and flexibility of the current scheme is
- appropriate for now. One could have the best of both worlds with
- inline functions but of course that would only work for gcc. Since
- we're machine generating some code we could do that here too. Maybe
- later. */
- if (! (*CGEN_PARSE_FN (insn)) (insn, &str, fields))
- {
- (*CGEN_INSERT_FN (insn)) (insn, fields, buf);
- /* It is up to the caller to actually output the insn and any
- queued relocs. */
- return insn;
- }
+ parse_errmsg = CGEN_PARSE_FN (cd, insn) (cd, insn, & str, fields);
+ if (parse_errmsg != NULL)
+ continue;
+
+ /* ??? 0 is passed for `pc'. */
+ insert_errmsg = CGEN_INSERT_FN (cd, insn) (cd, insn, fields, buf,
+ (bfd_vma) 0);
+ if (insert_errmsg != NULL)
+ continue;
- /* Try the next entry. */
+ /* It is up to the caller to actually output the insn and any
+ queued relocs. */
+ return insn;
}
- /* FIXME: We can return a better error message than this.
- Need to track why it failed and pick the right one. */
{
- static char errbuf[100];
- sprintf (errbuf, "bad instruction `%.50s%s'",
- start, strlen (start) > 50 ? "..." : "");
+ static char errbuf[150];
+ const char *tmp_errmsg;
+#ifdef CGEN_VERBOSE_ASSEMBLER_ERRORS
+#define be_verbose 1
+#else
+#define be_verbose 0
+#endif
+
+ if (be_verbose)
+ {
+ /* If requesting verbose error messages, use insert_errmsg.
+ Failing that, use parse_errmsg. */
+ tmp_errmsg = (insert_errmsg ? insert_errmsg :
+ parse_errmsg ? parse_errmsg :
+ recognized_mnemonic ?
+ _("unrecognized form of instruction") :
+ _("unrecognized instruction"));
+
+ if (strlen (start) > 50)
+ /* xgettext:c-format */
+ sprintf (errbuf, "%s `%.50s...'", tmp_errmsg, start);
+ else
+ /* xgettext:c-format */
+ sprintf (errbuf, "%s `%.50s'", tmp_errmsg, start);
+ }
+ else
+ {
+ if (strlen (start) > 50)
+ /* xgettext:c-format */
+ sprintf (errbuf, _("bad instruction `%.50s...'"), start);
+ else
+ /* xgettext:c-format */
+ sprintf (errbuf, _("bad instruction `%.50s'"), start);
+ }
+
*errmsg = errbuf;
return NULL;
}
}
-\f
-#if 0 /* This calls back to GAS which we can't do without care. */
-
-/* Record each member of OPVALS in the assembler's symbol table.
- This lets GAS parse registers for us.
- ??? Interesting idea but not currently used. */
-
-/* Record each member of OPVALS in the assembler's symbol table.
- FIXME: Not currently used. */
-
-void
-@arch@_cgen_asm_hash_keywords (opvals)
- CGEN_KEYWORD *opvals;
-{
- CGEN_KEYWORD_SEARCH search = cgen_keyword_search_init (opvals, NULL);
- const CGEN_KEYWORD_ENTRY *ke;
-
- while ((ke = cgen_keyword_search_next (&search)) != NULL)
- {
-#if 0 /* Unnecessary, should be done in the search routine. */
- if (! @arch@_cgen_opval_supported (ke))
- continue;
-#endif
- cgen_asm_record_register (ke->name, ke->value);
- }
-}
-
-#endif /* 0 */