parser: simplify code by keeping a pointer to the current operand
[nasm/avx512.git] / parser.c
blob3450e7768a4d8060d1e64804b41aa0b32e3183e4
1 /* ----------------------------------------------------------------------- *
3 * Copyright 1996-2013 The NASM Authors - All Rights Reserved
4 * See the file AUTHORS included with the NASM distribution for
5 * the specific copyright holders.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following
9 * conditions are met:
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above
14 * copyright notice, this list of conditions and the following
15 * disclaimer in the documentation and/or other materials provided
16 * with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND
19 * CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES,
20 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
21 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
22 * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
23 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
24 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
25 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
29 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
30 * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 * ----------------------------------------------------------------------- */
35 * parser.c source line parser for the Netwide Assembler
38 #include "compiler.h"
40 #include <stdio.h>
41 #include <stdlib.h>
42 #include <stddef.h>
43 #include <string.h>
44 #include <ctype.h>
45 #include <inttypes.h>
47 #include "nasm.h"
48 #include "insns.h"
49 #include "nasmlib.h"
50 #include "stdscan.h"
51 #include "eval.h"
52 #include "parser.h"
53 #include "float.h"
54 #include "tables.h"
56 extern int in_abs_seg; /* ABSOLUTE segment flag */
57 extern int32_t abs_seg; /* ABSOLUTE segment */
58 extern int32_t abs_offset; /* ABSOLUTE segment offset */
60 static int is_comma_next(void);
62 static int i;
63 static struct tokenval tokval;
64 static struct location *location; /* Pointer to current line's segment,offset */
66 void parser_global_info(struct location * locp)
68 location = locp;
71 static int prefix_slot(int prefix)
73 switch (prefix) {
74 case P_WAIT:
75 return PPS_WAIT;
76 case R_CS:
77 case R_DS:
78 case R_SS:
79 case R_ES:
80 case R_FS:
81 case R_GS:
82 return PPS_SEG;
83 case P_LOCK:
84 return PPS_LOCK;
85 case P_REP:
86 case P_REPE:
87 case P_REPZ:
88 case P_REPNE:
89 case P_REPNZ:
90 case P_XACQUIRE:
91 case P_XRELEASE:
92 return PPS_REP;
93 case P_O16:
94 case P_O32:
95 case P_O64:
96 case P_OSP:
97 return PPS_OSIZE;
98 case P_A16:
99 case P_A32:
100 case P_A64:
101 case P_ASP:
102 return PPS_ASIZE;
103 default:
104 nasm_error(ERR_PANIC, "Invalid value %d passed to prefix_slot()", prefix);
105 return -1;
109 static void process_size_override(insn *result, operand *op)
111 if (tasm_compatible_mode) {
112 switch ((int)tokval.t_integer) {
113 /* For TASM compatibility a size override inside the
114 * brackets changes the size of the operand, not the
115 * address type of the operand as it does in standard
116 * NASM syntax. Hence:
118 * mov eax,[DWORD val]
120 * is valid syntax in TASM compatibility mode. Note that
121 * you lose the ability to override the default address
122 * type for the instruction, but we never use anything
123 * but 32-bit flat model addressing in our code.
125 case S_BYTE:
126 op->type |= BITS8;
127 break;
128 case S_WORD:
129 op->type |= BITS16;
130 break;
131 case S_DWORD:
132 case S_LONG:
133 op->type |= BITS32;
134 break;
135 case S_QWORD:
136 op->type |= BITS64;
137 break;
138 case S_TWORD:
139 op->type |= BITS80;
140 break;
141 case S_OWORD:
142 op->type |= BITS128;
143 break;
144 default:
145 nasm_error(ERR_NONFATAL,
146 "invalid operand size specification");
147 break;
149 } else {
150 /* Standard NASM compatible syntax */
151 switch ((int)tokval.t_integer) {
152 case S_NOSPLIT:
153 op->eaflags |= EAF_TIMESTWO;
154 break;
155 case S_REL:
156 op->eaflags |= EAF_REL;
157 break;
158 case S_ABS:
159 op->eaflags |= EAF_ABS;
160 break;
161 case S_BYTE:
162 op->disp_size = 8;
163 op->eaflags |= EAF_BYTEOFFS;
164 break;
165 case P_A16:
166 case P_A32:
167 case P_A64:
168 if (result->prefixes[PPS_ASIZE] &&
169 result->prefixes[PPS_ASIZE] != tokval.t_integer)
170 nasm_error(ERR_NONFATAL,
171 "conflicting address size specifications");
172 else
173 result->prefixes[PPS_ASIZE] = tokval.t_integer;
174 break;
175 case S_WORD:
176 op->disp_size = 16;
177 op->eaflags |= EAF_WORDOFFS;
178 break;
179 case S_DWORD:
180 case S_LONG:
181 op->disp_size = 32;
182 op->eaflags |= EAF_WORDOFFS;
183 break;
184 case S_QWORD:
185 op->disp_size = 64;
186 op->eaflags |= EAF_WORDOFFS;
187 break;
188 default:
189 nasm_error(ERR_NONFATAL, "invalid size specification in"
190 " effective address");
191 break;
197 * when two or more decorators follow a register operand,
198 * consecutive decorators are parsed here.
199 * opmask and zeroing decorators can be placed in any order.
200 * e.g. zmm1 {k2}{z} or zmm2 {z,k3}
201 * decorator(s) are placed at the end of an operand.
203 static bool parse_braces(decoflags_t *decoflags)
205 int i;
206 bool recover = false;
208 i = tokval.t_type;
209 do {
210 if (i == TOKEN_OPMASK) {
211 if (*decoflags & OPMASK_MASK) {
212 nasm_error(ERR_NONFATAL, "opmask k%lu is already set",
213 *decoflags & OPMASK_MASK);
214 *decoflags &= ~OPMASK_MASK;
216 *decoflags |= VAL_OPMASK(nasm_regvals[tokval.t_integer]);
217 } else if (i == TOKEN_DECORATOR) {
218 switch (tokval.t_integer) {
219 case BRC_Z:
221 * according to AVX512 spec, only zeroing/merging decorator
222 * is supported with opmask
224 *decoflags |= GEN_Z(0);
225 break;
226 default:
227 nasm_error(ERR_NONFATAL, "{%s} is not an expected decorator",
228 tokval.t_charptr);
229 break;
231 } else if (i == ',' || i == TOKEN_EOS){
232 break;
233 } else {
234 nasm_error(ERR_NONFATAL, "only a series of valid decorators"
235 " expected");
236 recover = true;
237 break;
239 i = stdscan(NULL, &tokval);
240 } while(1);
242 return recover;
245 insn *parse_line(int pass, char *buffer, insn *result, ldfunc ldef)
247 bool insn_is_label = false;
248 struct eval_hints hints;
249 int opnum;
250 int critical;
251 bool first;
252 bool recover;
254 restart_parse:
255 first = true;
256 result->forw_ref = false;
258 stdscan_reset();
259 stdscan_set(buffer);
260 i = stdscan(NULL, &tokval);
262 result->label = NULL; /* Assume no label */
263 result->eops = NULL; /* must do this, whatever happens */
264 result->operands = 0; /* must initialize this */
265 result->evex_rm = 0; /* Ensure EVEX rounding mode is reset */
266 result->evex_brerop = -1; /* Reset EVEX broadcasting/ER op position */
268 /* Ignore blank lines */
269 if (i == TOKEN_EOS)
270 goto fail;
272 if (i != TOKEN_ID &&
273 i != TOKEN_INSN &&
274 i != TOKEN_PREFIX &&
275 (i != TOKEN_REG || !IS_SREG(tokval.t_integer))) {
276 nasm_error(ERR_NONFATAL,
277 "label or instruction expected at start of line");
278 goto fail;
281 if (i == TOKEN_ID || (insn_is_label && i == TOKEN_INSN)) {
282 /* there's a label here */
283 first = false;
284 result->label = tokval.t_charptr;
285 i = stdscan(NULL, &tokval);
286 if (i == ':') { /* skip over the optional colon */
287 i = stdscan(NULL, &tokval);
288 } else if (i == 0) {
289 nasm_error(ERR_WARNING | ERR_WARN_OL | ERR_PASS1,
290 "label alone on a line without a colon might be in error");
292 if (i != TOKEN_INSN || tokval.t_integer != I_EQU) {
294 * FIXME: location->segment could be NO_SEG, in which case
295 * it is possible we should be passing 'abs_seg'. Look into this.
296 * Work out whether that is *really* what we should be doing.
297 * Generally fix things. I think this is right as it is, but
298 * am still not certain.
300 ldef(result->label, in_abs_seg ? abs_seg : location->segment,
301 location->offset, NULL, true, false);
305 /* Just a label here */
306 if (i == TOKEN_EOS)
307 goto fail;
309 nasm_build_assert(P_none != 0);
310 memset(result->prefixes, P_none, sizeof(result->prefixes));
311 result->times = 1L;
313 while (i == TOKEN_PREFIX ||
314 (i == TOKEN_REG && IS_SREG(tokval.t_integer))) {
315 first = false;
318 * Handle special case: the TIMES prefix.
320 if (i == TOKEN_PREFIX && tokval.t_integer == P_TIMES) {
321 expr *value;
323 i = stdscan(NULL, &tokval);
324 value = evaluate(stdscan, NULL, &tokval, NULL, pass0, nasm_error, NULL);
325 i = tokval.t_type;
326 if (!value) /* Error in evaluator */
327 goto fail;
328 if (!is_simple(value)) {
329 nasm_error(ERR_NONFATAL,
330 "non-constant argument supplied to TIMES");
331 result->times = 1L;
332 } else {
333 result->times = value->value;
334 if (value->value < 0 && pass0 == 2) {
335 nasm_error(ERR_NONFATAL, "TIMES value %"PRId64" is negative",
336 value->value);
337 result->times = 0;
340 } else {
341 int slot = prefix_slot(tokval.t_integer);
342 if (result->prefixes[slot]) {
343 if (result->prefixes[slot] == tokval.t_integer)
344 nasm_error(ERR_WARNING | ERR_PASS1,
345 "instruction has redundant prefixes");
346 else
347 nasm_error(ERR_NONFATAL,
348 "instruction has conflicting prefixes");
350 result->prefixes[slot] = tokval.t_integer;
351 i = stdscan(NULL, &tokval);
355 if (i != TOKEN_INSN) {
356 int j;
357 enum prefixes pfx;
359 for (j = 0; j < MAXPREFIX; j++) {
360 if ((pfx = result->prefixes[j]) != P_none)
361 break;
364 if (i == 0 && pfx != P_none) {
366 * Instruction prefixes are present, but no actual
367 * instruction. This is allowed: at this point we
368 * invent a notional instruction of RESB 0.
370 result->opcode = I_RESB;
371 result->operands = 1;
372 result->oprs[0].type = IMMEDIATE;
373 result->oprs[0].offset = 0L;
374 result->oprs[0].segment = result->oprs[0].wrt = NO_SEG;
375 return result;
376 } else {
377 nasm_error(ERR_NONFATAL, "parser: instruction expected");
378 goto fail;
382 result->opcode = tokval.t_integer;
383 result->condition = tokval.t_inttwo;
386 * INCBIN cannot be satisfied with incorrectly
387 * evaluated operands, since the correct values _must_ be known
388 * on the first pass. Hence, even in pass one, we set the
389 * `critical' flag on calling evaluate(), so that it will bomb
390 * out on undefined symbols.
392 if (result->opcode == I_INCBIN) {
393 critical = (pass0 < 2 ? 1 : 2);
395 } else
396 critical = (pass == 2 ? 2 : 0);
398 if (result->opcode == I_DB || result->opcode == I_DW ||
399 result->opcode == I_DD || result->opcode == I_DQ ||
400 result->opcode == I_DT || result->opcode == I_DO ||
401 result->opcode == I_DY || result->opcode == I_INCBIN) {
402 extop *eop, **tail = &result->eops, **fixptr;
403 int oper_num = 0;
404 int32_t sign;
406 result->eops_float = false;
409 * Begin to read the DB/DW/DD/DQ/DT/DO/INCBIN operands.
411 while (1) {
412 i = stdscan(NULL, &tokval);
413 if (i == TOKEN_EOS)
414 break;
415 else if (first && i == ':') {
416 insn_is_label = true;
417 goto restart_parse;
419 first = false;
420 fixptr = tail;
421 eop = *tail = nasm_malloc(sizeof(extop));
422 tail = &eop->next;
423 eop->next = NULL;
424 eop->type = EOT_NOTHING;
425 oper_num++;
426 sign = +1;
429 * is_comma_next() here is to distinguish this from
430 * a string used as part of an expression...
432 if (i == TOKEN_STR && is_comma_next()) {
433 eop->type = EOT_DB_STRING;
434 eop->stringval = tokval.t_charptr;
435 eop->stringlen = tokval.t_inttwo;
436 i = stdscan(NULL, &tokval); /* eat the comma */
437 } else if (i == TOKEN_STRFUNC) {
438 bool parens = false;
439 const char *funcname = tokval.t_charptr;
440 enum strfunc func = tokval.t_integer;
441 i = stdscan(NULL, &tokval);
442 if (i == '(') {
443 parens = true;
444 i = stdscan(NULL, &tokval);
446 if (i != TOKEN_STR) {
447 nasm_error(ERR_NONFATAL,
448 "%s must be followed by a string constant",
449 funcname);
450 eop->type = EOT_NOTHING;
451 } else {
452 eop->type = EOT_DB_STRING_FREE;
453 eop->stringlen =
454 string_transform(tokval.t_charptr, tokval.t_inttwo,
455 &eop->stringval, func);
456 if (eop->stringlen == (size_t)-1) {
457 nasm_error(ERR_NONFATAL, "invalid string for transform");
458 eop->type = EOT_NOTHING;
461 if (parens && i && i != ')') {
462 i = stdscan(NULL, &tokval);
463 if (i != ')') {
464 nasm_error(ERR_NONFATAL, "unterminated %s function",
465 funcname);
468 if (i && i != ',')
469 i = stdscan(NULL, &tokval);
470 } else if (i == '-' || i == '+') {
471 char *save = stdscan_get();
472 int token = i;
473 sign = (i == '-') ? -1 : 1;
474 i = stdscan(NULL, &tokval);
475 if (i != TOKEN_FLOAT) {
476 stdscan_set(save);
477 i = tokval.t_type = token;
478 goto is_expression;
479 } else {
480 goto is_float;
482 } else if (i == TOKEN_FLOAT) {
483 is_float:
484 eop->type = EOT_DB_STRING;
485 result->eops_float = true;
487 eop->stringlen = idata_bytes(result->opcode);
488 if (eop->stringlen > 16) {
489 nasm_error(ERR_NONFATAL, "floating-point constant"
490 " encountered in DY instruction");
491 eop->stringlen = 0;
492 } else if (eop->stringlen < 1) {
493 nasm_error(ERR_NONFATAL, "floating-point constant"
494 " encountered in unknown instruction");
496 * fix suggested by Pedro Gimeno... original line was:
497 * eop->type = EOT_NOTHING;
499 eop->stringlen = 0;
502 eop = nasm_realloc(eop, sizeof(extop) + eop->stringlen);
503 tail = &eop->next;
504 *fixptr = eop;
505 eop->stringval = (char *)eop + sizeof(extop);
506 if (!eop->stringlen ||
507 !float_const(tokval.t_charptr, sign,
508 (uint8_t *)eop->stringval,
509 eop->stringlen, nasm_error))
510 eop->type = EOT_NOTHING;
511 i = stdscan(NULL, &tokval); /* eat the comma */
512 } else {
513 /* anything else, assume it is an expression */
514 expr *value;
516 is_expression:
517 value = evaluate(stdscan, NULL, &tokval, NULL,
518 critical, nasm_error, NULL);
519 i = tokval.t_type;
520 if (!value) /* Error in evaluator */
521 goto fail;
522 if (is_unknown(value)) {
523 eop->type = EOT_DB_NUMBER;
524 eop->offset = 0; /* doesn't matter what we put */
525 eop->segment = eop->wrt = NO_SEG; /* likewise */
526 } else if (is_reloc(value)) {
527 eop->type = EOT_DB_NUMBER;
528 eop->offset = reloc_value(value);
529 eop->segment = reloc_seg(value);
530 eop->wrt = reloc_wrt(value);
531 } else {
532 nasm_error(ERR_NONFATAL,
533 "operand %d: expression is not simple"
534 " or relocatable", oper_num);
539 * We're about to call stdscan(), which will eat the
540 * comma that we're currently sitting on between
541 * arguments. However, we'd better check first that it
542 * _is_ a comma.
544 if (i == TOKEN_EOS) /* also could be EOL */
545 break;
546 if (i != ',') {
547 nasm_error(ERR_NONFATAL, "comma expected after operand %d",
548 oper_num);
549 goto fail;
553 if (result->opcode == I_INCBIN) {
555 * Correct syntax for INCBIN is that there should be
556 * one string operand, followed by one or two numeric
557 * operands.
559 if (!result->eops || result->eops->type != EOT_DB_STRING)
560 nasm_error(ERR_NONFATAL, "`incbin' expects a file name");
561 else if (result->eops->next &&
562 result->eops->next->type != EOT_DB_NUMBER)
563 nasm_error(ERR_NONFATAL, "`incbin': second parameter is"
564 " non-numeric");
565 else if (result->eops->next && result->eops->next->next &&
566 result->eops->next->next->type != EOT_DB_NUMBER)
567 nasm_error(ERR_NONFATAL, "`incbin': third parameter is"
568 " non-numeric");
569 else if (result->eops->next && result->eops->next->next &&
570 result->eops->next->next->next)
571 nasm_error(ERR_NONFATAL,
572 "`incbin': more than three parameters");
573 else
574 return result;
576 * If we reach here, one of the above errors happened.
577 * Throw the instruction away.
579 goto fail;
580 } else /* DB ... */ if (oper_num == 0)
581 nasm_error(ERR_WARNING | ERR_PASS1,
582 "no operand for data declaration");
583 else
584 result->operands = oper_num;
586 return result;
590 * Now we begin to parse the operands. There may be up to four
591 * of these, separated by commas, and terminated by a zero token.
594 for (opnum = 0; opnum < MAX_OPERANDS; opnum++) {
595 operand *op = &result->oprs[opnum];
596 expr *value; /* used most of the time */
597 int mref; /* is this going to be a memory ref? */
598 int bracket; /* is it a [] mref, or a & mref? */
599 int setsize = 0;
600 decoflags_t brace_flags = 0; /* flags for decorators in braces */
602 op->disp_size = 0; /* have to zero this whatever */
603 op->eaflags = 0; /* and this */
604 op->opflags = 0;
605 op->decoflags = 0;
607 i = stdscan(NULL, &tokval);
608 if (i == TOKEN_EOS)
609 break; /* end of operands: get out of here */
610 else if (first && i == ':') {
611 insn_is_label = true;
612 goto restart_parse;
614 first = false;
615 op->type = 0; /* so far, no override */
616 while (i == TOKEN_SPECIAL) { /* size specifiers */
617 switch ((int)tokval.t_integer) {
618 case S_BYTE:
619 if (!setsize) /* we want to use only the first */
620 op->type |= BITS8;
621 setsize = 1;
622 break;
623 case S_WORD:
624 if (!setsize)
625 op->type |= BITS16;
626 setsize = 1;
627 break;
628 case S_DWORD:
629 case S_LONG:
630 if (!setsize)
631 op->type |= BITS32;
632 setsize = 1;
633 break;
634 case S_QWORD:
635 if (!setsize)
636 op->type |= BITS64;
637 setsize = 1;
638 break;
639 case S_TWORD:
640 if (!setsize)
641 op->type |= BITS80;
642 setsize = 1;
643 break;
644 case S_OWORD:
645 if (!setsize)
646 op->type |= BITS128;
647 setsize = 1;
648 break;
649 case S_YWORD:
650 if (!setsize)
651 op->type |= BITS256;
652 setsize = 1;
653 break;
654 case S_ZWORD:
655 if (!setsize)
656 op->type |= BITS512;
657 setsize = 1;
658 break;
659 case S_TO:
660 op->type |= TO;
661 break;
662 case S_STRICT:
663 op->type |= STRICT;
664 break;
665 case S_FAR:
666 op->type |= FAR;
667 break;
668 case S_NEAR:
669 op->type |= NEAR;
670 break;
671 case S_SHORT:
672 op->type |= SHORT;
673 break;
674 default:
675 nasm_error(ERR_NONFATAL, "invalid operand size specification");
677 i = stdscan(NULL, &tokval);
680 if (i == '[' || i == '&') { /* memory reference */
681 mref = true;
682 bracket = (i == '[');
683 i = stdscan(NULL, &tokval); /* then skip the colon */
684 while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
685 process_size_override(result, op);
686 i = stdscan(NULL, &tokval);
688 } else { /* immediate operand, or register */
689 mref = false;
690 bracket = false; /* placate optimisers */
693 if ((op->type & FAR) && !mref &&
694 result->opcode != I_JMP && result->opcode != I_CALL) {
695 nasm_error(ERR_NONFATAL, "invalid use of FAR operand specifier");
698 value = evaluate(stdscan, NULL, &tokval,
699 &op->opflags,
700 critical, nasm_error, &hints);
701 i = tokval.t_type;
702 if (op->opflags & OPFLAG_FORWARD) {
703 result->forw_ref = true;
705 if (!value) /* Error in evaluator */
706 goto fail;
707 if (i == ':' && mref) { /* it was seg:offset */
709 * Process the segment override.
711 if (value[1].type != 0 ||
712 value->value != 1 ||
713 !IS_SREG(value->type))
714 nasm_error(ERR_NONFATAL, "invalid segment override");
715 else if (result->prefixes[PPS_SEG])
716 nasm_error(ERR_NONFATAL,
717 "instruction has conflicting segment overrides");
718 else {
719 result->prefixes[PPS_SEG] = value->type;
720 if (IS_FSGS(value->type))
721 op->eaflags |= EAF_FSGS;
724 i = stdscan(NULL, &tokval); /* then skip the colon */
725 while (i == TOKEN_SPECIAL || i == TOKEN_PREFIX) {
726 process_size_override(result, op);
727 i = stdscan(NULL, &tokval);
729 value = evaluate(stdscan, NULL, &tokval,
730 &op->opflags,
731 critical, nasm_error, &hints);
732 i = tokval.t_type;
733 if (op->opflags & OPFLAG_FORWARD) {
734 result->forw_ref = true;
736 /* and get the offset */
737 if (!value) /* Error in evaluator */
738 goto fail;
741 recover = false;
742 if (mref && bracket) { /* find ] at the end */
743 if (i != ']') {
744 nasm_error(ERR_NONFATAL, "parser: expecting ]");
745 recover = true;
746 } else { /* we got the required ] */
747 i = stdscan(NULL, &tokval);
748 if ((i == TOKEN_DECORATOR) || (i == TOKEN_OPMASK)) {
750 * according to AVX512 spec, broacast or opmask decorator
751 * is expected for memory reference operands
753 if (tokval.t_flag & TFLAG_BRDCAST) {
754 brace_flags |= GEN_BRDCAST(0);
755 i = stdscan(NULL, &tokval);
756 } else if (i == TOKEN_OPMASK) {
757 brace_flags |= VAL_OPMASK(nasm_regvals[tokval.t_integer]);
758 i = stdscan(NULL, &tokval);
759 } else {
760 nasm_error(ERR_NONFATAL, "broadcast or opmask "
761 "decorator expected inside braces");
762 recover = true;
766 if (i != 0 && i != ',') {
767 nasm_error(ERR_NONFATAL, "comma or end of line expected");
768 recover = true;
771 } else { /* immediate operand */
772 if (i != 0 && i != ',' && i != ':' &&
773 i != TOKEN_DECORATOR && i != TOKEN_OPMASK) {
774 nasm_error(ERR_NONFATAL, "comma, colon, decorator or end of "
775 "line expected after operand");
776 recover = true;
777 } else if (i == ':') {
778 op->type |= COLON;
779 } else if (i == TOKEN_DECORATOR || i == TOKEN_OPMASK) {
780 /* parse opmask (and zeroing) after an operand */
781 recover = parse_braces(&brace_flags);
784 if (recover) {
785 do { /* error recovery */
786 i = stdscan(NULL, &tokval);
787 } while (i != 0 && i != ',');
791 * now convert the exprs returned from evaluate()
792 * into operand descriptions...
795 if (mref) { /* it's a memory reference */
796 expr *e = value;
797 int b, i, s; /* basereg, indexreg, scale */
798 int64_t o; /* offset */
800 b = i = -1, o = s = 0;
801 op->hintbase = hints.base;
802 op->hinttype = hints.type;
804 if (e->type && e->type <= EXPR_REG_END) { /* this bit's a register */
805 bool is_gpr = is_class(REG_GPR,nasm_reg_flags[e->type]);
807 if (is_gpr && e->value == 1)
808 b = e->type; /* It can be basereg */
809 else /* No, it has to be indexreg */
810 i = e->type, s = e->value;
811 e++;
813 if (e->type && e->type <= EXPR_REG_END) { /* it's a 2nd register */
814 bool is_gpr = is_class(REG_GPR,nasm_reg_flags[e->type]);
816 if (b != -1) /* If the first was the base, ... */
817 i = e->type, s = e->value; /* second has to be indexreg */
819 else if (!is_gpr || e->value != 1) {
820 /* If both want to be index */
821 nasm_error(ERR_NONFATAL,
822 "invalid effective address: two index registers");
823 goto fail;
824 } else
825 b = e->type;
826 e++;
828 if (e->type != 0) { /* is there an offset? */
829 if (e->type <= EXPR_REG_END) { /* in fact, is there an error? */
830 nasm_error(ERR_NONFATAL,
831 "beroset-p-603-invalid effective address");
832 goto fail;
833 } else {
834 if (e->type == EXPR_UNKNOWN) {
835 op->opflags |= OPFLAG_UNKNOWN;
836 o = 0; /* doesn't matter what */
837 op->wrt = NO_SEG; /* nor this */
838 op->segment = NO_SEG; /* or this */
839 while (e->type)
840 e++; /* go to the end of the line */
841 } else {
842 if (e->type == EXPR_SIMPLE) {
843 o = e->value;
844 e++;
846 if (e->type == EXPR_WRT) {
847 op->wrt = e->value;
848 e++;
849 } else
850 op->wrt = NO_SEG;
852 * Look for a segment base type.
854 if (e->type && e->type < EXPR_SEGBASE) {
855 nasm_error(ERR_NONFATAL,
856 "beroset-p-630-invalid effective address");
857 goto fail;
859 while (e->type && e->value == 0)
860 e++;
861 if (e->type && e->value != 1) {
862 nasm_error(ERR_NONFATAL,
863 "beroset-p-637-invalid effective address");
864 goto fail;
866 if (e->type) {
867 op->segment =
868 e->type - EXPR_SEGBASE;
869 e++;
870 } else
871 op->segment = NO_SEG;
872 while (e->type && e->value == 0)
873 e++;
874 if (e->type) {
875 nasm_error(ERR_NONFATAL,
876 "beroset-p-650-invalid effective address");
877 goto fail;
881 } else {
882 o = 0;
883 op->wrt = NO_SEG;
884 op->segment = NO_SEG;
887 if (e->type != 0) { /* there'd better be nothing left! */
888 nasm_error(ERR_NONFATAL,
889 "beroset-p-663-invalid effective address");
890 goto fail;
893 /* It is memory, but it can match any r/m operand */
894 op->type |= MEMORY_ANY;
896 if (b == -1 && (i == -1 || s == 0)) {
897 int is_rel = globalbits == 64 &&
898 !(op->eaflags & EAF_ABS) &&
899 ((globalrel &&
900 !(op->eaflags & EAF_FSGS)) ||
901 (op->eaflags & EAF_REL));
903 op->type |= is_rel ? IP_REL : MEM_OFFS;
906 if (i != -1) {
907 opflags_t iclass = nasm_reg_flags[i];
909 if (is_class(XMMREG,iclass))
910 op->type |= XMEM;
911 else if (is_class(YMMREG,iclass))
912 op->type |= YMEM;
913 else if (is_class(ZMMREG,iclass))
914 op->type |= ZMEM;
917 op->basereg = b;
918 op->indexreg = i;
919 op->scale = s;
920 op->offset = o;
921 op->decoflags |= brace_flags;
922 } else { /* it's not a memory reference */
923 if (is_just_unknown(value)) { /* it's immediate but unknown */
924 op->type |= IMMEDIATE;
925 op->opflags |= OPFLAG_UNKNOWN;
926 op->offset = 0; /* don't care */
927 op->segment = NO_SEG; /* don't care again */
928 op->wrt = NO_SEG; /* still don't care */
930 if(optimizing >= 0 && !(op->type & STRICT)) {
931 /* Be optimistic */
932 op->type |=
933 UNITY | SBYTEWORD | SBYTEDWORD | UDWORD | SDWORD;
935 } else if (is_reloc(value)) { /* it's immediate */
936 op->type |= IMMEDIATE;
937 op->offset = reloc_value(value);
938 op->segment = reloc_seg(value);
939 op->wrt = reloc_wrt(value);
941 if (is_simple(value)) {
942 uint64_t n = reloc_value(value);
943 if (n == 1)
944 op->type |= UNITY;
945 if (optimizing >= 0 &&
946 !(op->type & STRICT)) {
947 if ((uint32_t) (n + 128) <= 255)
948 op->type |= SBYTEDWORD;
949 if ((uint16_t) (n + 128) <= 255)
950 op->type |= SBYTEWORD;
951 if (n <= 0xFFFFFFFF)
952 op->type |= UDWORD;
953 if (n + 0x80000000 <= 0xFFFFFFFF)
954 op->type |= SDWORD;
957 } else if(value->type == EXPR_RDSAE) {
959 * it's not an operand but a rounding or SAE decorator.
960 * put the decorator information in the (opflag_t) type field
961 * of previous operand.
963 opnum--; op--;
964 switch (value->value) {
965 case BRC_RN:
966 case BRC_RU:
967 case BRC_RD:
968 case BRC_RZ:
969 case BRC_SAE:
970 op->decoflags |= (value->value == BRC_SAE ? SAE : ER);
971 result->evex_rm = value->value;
972 break;
973 default:
974 nasm_error(ERR_NONFATAL, "invalid decorator");
975 break;
977 } else { /* it's a register */
978 opflags_t rs;
980 if (value->type >= EXPR_SIMPLE || value->value != 1) {
981 nasm_error(ERR_NONFATAL, "invalid operand type");
982 goto fail;
986 * check that its only 1 register, not an expression...
988 for (i = 1; value[i].type; i++)
989 if (value[i].value) {
990 nasm_error(ERR_NONFATAL, "invalid operand type");
991 goto fail;
994 /* clear overrides, except TO which applies to FPU regs */
995 if (op->type & ~TO) {
997 * we want to produce a warning iff the specified size
998 * is different from the register size
1000 rs = op->type & SIZE_MASK;
1001 } else
1002 rs = 0;
1004 op->type &= TO;
1005 op->type |= REGISTER;
1006 op->type |= nasm_reg_flags[value->type];
1007 op->decoflags |= brace_flags;
1008 op->basereg = value->type;
1010 if (rs && (op->type & SIZE_MASK) != rs)
1011 nasm_error(ERR_WARNING | ERR_PASS1,
1012 "register size specification ignored");
1016 /* remember the position of operand having broadcasting/ER mode */
1017 if (op->decoflags & (BRDCAST_MASK | ER | SAE))
1018 result->evex_brerop = opnum;
1021 result->operands = opnum; /* set operand count */
1023 /* clear remaining operands */
1024 while (opnum < MAX_OPERANDS)
1025 result->oprs[opnum++].type = 0;
1028 * Transform RESW, RESD, RESQ, REST, RESO, RESY into RESB.
1030 switch (result->opcode) {
1031 case I_RESW:
1032 result->opcode = I_RESB;
1033 result->oprs[0].offset *= 2;
1034 break;
1035 case I_RESD:
1036 result->opcode = I_RESB;
1037 result->oprs[0].offset *= 4;
1038 break;
1039 case I_RESQ:
1040 result->opcode = I_RESB;
1041 result->oprs[0].offset *= 8;
1042 break;
1043 case I_REST:
1044 result->opcode = I_RESB;
1045 result->oprs[0].offset *= 10;
1046 break;
1047 case I_RESO:
1048 result->opcode = I_RESB;
1049 result->oprs[0].offset *= 16;
1050 break;
1051 case I_RESY:
1052 result->opcode = I_RESB;
1053 result->oprs[0].offset *= 32;
1054 break;
1055 default:
1056 break;
1059 return result;
1061 fail:
1062 result->opcode = I_none;
1063 return result;
1066 static int is_comma_next(void)
1068 struct tokenval tv;
1069 char *p;
1070 int i;
1072 p = stdscan_get();
1073 i = stdscan(NULL, &tv);
1074 stdscan_set(p);
1076 return (i == ',' || i == ';' || !i);
1079 void cleanup_insn(insn * i)
1081 extop *e;
1083 while ((e = i->eops)) {
1084 i->eops = e->next;
1085 if (e->type == EOT_DB_STRING_FREE)
1086 nasm_free(e->stringval);
1087 nasm_free(e);