Suppress noisy pointer signed-ness warnings on x86-64.
[tinycc.git] / i386-gen.c
bloba2564f3db6ab5ed72cc19d0fdb46fd4542546de5
1 /*
2 * X86 code generator for TCC
3 *
4 * Copyright (c) 2001-2004 Fabrice Bellard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 /* number of available registers */
22 #define NB_REGS 4
24 /* a register can belong to several classes. The classes must be
25 sorted from more general to more precise (see gv2() code which does
26 assumptions on it). */
27 #define RC_INT 0x0001 /* generic integer register */
28 #define RC_FLOAT 0x0002 /* generic float register */
29 #define RC_EAX 0x0004
30 #define RC_ST0 0x0008
31 #define RC_ECX 0x0010
32 #define RC_EDX 0x0020
33 #define RC_IRET RC_EAX /* function return: integer register */
34 #define RC_LRET RC_EDX /* function return: second integer register */
35 #define RC_FRET RC_ST0 /* function return: float register */
37 /* pretty names for the registers */
38 enum {
39 TREG_EAX = 0,
40 TREG_ECX,
41 TREG_EDX,
42 TREG_ST0,
45 int reg_classes[NB_REGS] = {
46 /* eax */ RC_INT | RC_EAX,
47 /* ecx */ RC_INT | RC_ECX,
48 /* edx */ RC_INT | RC_EDX,
49 /* st0 */ RC_FLOAT | RC_ST0,
52 /* return registers for function */
53 #define REG_IRET TREG_EAX /* single word int return register */
54 #define REG_LRET TREG_EDX /* second word return register (for long long) */
55 #define REG_FRET TREG_ST0 /* float return register */
57 /* defined if function parameters must be evaluated in reverse order */
58 #define INVERT_FUNC_PARAMS
60 /* defined if structures are passed as pointers. Otherwise structures
61 are directly pushed on stack. */
62 //#define FUNC_STRUCT_PARAM_AS_PTR
64 /* pointer size, in bytes */
65 #define PTR_SIZE 4
67 /* long double size and alignment, in bytes */
68 #define LDOUBLE_SIZE 12
69 #define LDOUBLE_ALIGN 4
70 /* maximum alignment (for aligned attribute support) */
71 #define MAX_ALIGN 8
73 /******************************************************/
74 /* ELF defines */
76 #define EM_TCC_TARGET EM_386
78 /* relocation type for 32 bit data relocation */
79 #define R_DATA_32 R_386_32
80 #define R_JMP_SLOT R_386_JMP_SLOT
81 #define R_COPY R_386_COPY
83 #define ELF_START_ADDR 0x08048000
84 #define ELF_PAGE_SIZE 0x1000
86 /******************************************************/
88 static unsigned long func_sub_sp_offset;
89 static unsigned long func_bound_offset;
90 static int func_ret_sub;
92 /* XXX: make it faster ? */
93 void g(int c)
95 int ind1;
96 ind1 = ind + 1;
97 if (ind1 > cur_text_section->data_allocated)
98 section_realloc(cur_text_section, ind1);
99 cur_text_section->data[ind] = c;
100 ind = ind1;
103 void o(unsigned int c)
105 while (c) {
106 g(c);
107 c = c >> 8;
111 void gen_le32(int c)
113 g(c);
114 g(c >> 8);
115 g(c >> 16);
116 g(c >> 24);
119 /* output a symbol and patch all calls to it */
120 void gsym_addr(int t, int a)
122 int n, *ptr;
123 while (t) {
124 ptr = (int *)(cur_text_section->data + t);
125 n = *ptr; /* next value */
126 *ptr = a - t - 4;
127 t = n;
131 void gsym(int t)
133 gsym_addr(t, ind);
136 /* psym is used to put an instruction with a data field which is a
137 reference to a symbol. It is in fact the same as oad ! */
138 #define psym oad
140 /* instruction + 4 bytes data. Return the address of the data */
141 static int oad(int c, int s)
143 int ind1;
145 o(c);
146 ind1 = ind + 4;
147 if (ind1 > cur_text_section->data_allocated)
148 section_realloc(cur_text_section, ind1);
149 *(int *)(cur_text_section->data + ind) = s;
150 s = ind;
151 ind = ind1;
152 return s;
155 /* output constant with relocation if 'r & VT_SYM' is true */
156 static void gen_addr32(int r, Sym *sym, int c)
158 if (r & VT_SYM)
159 greloc(cur_text_section, sym, ind, R_386_32);
160 gen_le32(c);
163 /* generate a modrm reference. 'op_reg' contains the addtionnal 3
164 opcode bits */
165 static void gen_modrm(int op_reg, int r, Sym *sym, int c)
167 op_reg = op_reg << 3;
168 if ((r & VT_VALMASK) == VT_CONST) {
169 /* constant memory reference */
170 o(0x05 | op_reg);
171 gen_addr32(r, sym, c);
172 } else if ((r & VT_VALMASK) == VT_LOCAL) {
173 /* currently, we use only ebp as base */
174 if (c == (char)c) {
175 /* short reference */
176 o(0x45 | op_reg);
177 g(c);
178 } else {
179 oad(0x85 | op_reg, c);
181 } else {
182 g(0x00 | op_reg | (r & VT_VALMASK));
187 /* load 'r' from value 'sv' */
188 void load(int r, SValue *sv)
190 int v, t, ft, fc, fr;
191 SValue v1;
193 fr = sv->r;
194 ft = sv->type.t;
195 fc = sv->c.ul;
197 v = fr & VT_VALMASK;
198 if (fr & VT_LVAL) {
199 if (v == VT_LLOCAL) {
200 v1.type.t = VT_INT;
201 v1.r = VT_LOCAL | VT_LVAL;
202 v1.c.ul = fc;
203 load(r, &v1);
204 fr = r;
206 if ((ft & VT_BTYPE) == VT_FLOAT) {
207 o(0xd9); /* flds */
208 r = 0;
209 } else if ((ft & VT_BTYPE) == VT_DOUBLE) {
210 o(0xdd); /* fldl */
211 r = 0;
212 } else if ((ft & VT_BTYPE) == VT_LDOUBLE) {
213 o(0xdb); /* fldt */
214 r = 5;
215 } else if ((ft & VT_TYPE) == VT_BYTE) {
216 o(0xbe0f); /* movsbl */
217 } else if ((ft & VT_TYPE) == (VT_BYTE | VT_UNSIGNED)) {
218 o(0xb60f); /* movzbl */
219 } else if ((ft & VT_TYPE) == VT_SHORT) {
220 o(0xbf0f); /* movswl */
221 } else if ((ft & VT_TYPE) == (VT_SHORT | VT_UNSIGNED)) {
222 o(0xb70f); /* movzwl */
223 } else {
224 o(0x8b); /* movl */
226 gen_modrm(r, fr, sv->sym, fc);
227 } else {
228 if (v == VT_CONST) {
229 o(0xb8 + r); /* mov $xx, r */
230 gen_addr32(fr, sv->sym, fc);
231 } else if (v == VT_LOCAL) {
232 o(0x8d); /* lea xxx(%ebp), r */
233 gen_modrm(r, VT_LOCAL, sv->sym, fc);
234 } else if (v == VT_CMP) {
235 oad(0xb8 + r, 0); /* mov $0, r */
236 o(0x0f); /* setxx %br */
237 o(fc);
238 o(0xc0 + r);
239 } else if (v == VT_JMP || v == VT_JMPI) {
240 t = v & 1;
241 oad(0xb8 + r, t); /* mov $1, r */
242 o(0x05eb); /* jmp after */
243 gsym(fc);
244 oad(0xb8 + r, t ^ 1); /* mov $0, r */
245 } else if (v != r) {
246 o(0x89);
247 o(0xc0 + r + v * 8); /* mov v, r */
252 /* store register 'r' in lvalue 'v' */
253 void store(int r, SValue *v)
255 int fr, bt, ft, fc;
257 ft = v->type.t;
258 fc = v->c.ul;
259 fr = v->r & VT_VALMASK;
260 bt = ft & VT_BTYPE;
261 /* XXX: incorrect if float reg to reg */
262 if (bt == VT_FLOAT) {
263 o(0xd9); /* fsts */
264 r = 2;
265 } else if (bt == VT_DOUBLE) {
266 o(0xdd); /* fstpl */
267 r = 2;
268 } else if (bt == VT_LDOUBLE) {
269 o(0xc0d9); /* fld %st(0) */
270 o(0xdb); /* fstpt */
271 r = 7;
272 } else {
273 if (bt == VT_SHORT)
274 o(0x66);
275 if (bt == VT_BYTE || bt == VT_BOOL)
276 o(0x88);
277 else
278 o(0x89);
280 if (fr == VT_CONST ||
281 fr == VT_LOCAL ||
282 (v->r & VT_LVAL)) {
283 gen_modrm(r, v->r, v->sym, fc);
284 } else if (fr != r) {
285 o(0xc0 + fr + r * 8); /* mov r, fr */
289 static void gadd_sp(int val)
291 if (val == (char)val) {
292 o(0xc483);
293 g(val);
294 } else {
295 oad(0xc481, val); /* add $xxx, %esp */
299 /* 'is_jmp' is '1' if it is a jump */
300 static void gcall_or_jmp(int is_jmp)
302 int r;
303 if ((vtop->r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
304 /* constant case */
305 if (vtop->r & VT_SYM) {
306 /* relocation case */
307 greloc(cur_text_section, vtop->sym,
308 ind + 1, R_386_PC32);
309 } else {
310 /* put an empty PC32 relocation */
311 put_elf_reloc(symtab_section, cur_text_section,
312 ind + 1, R_386_PC32, 0);
314 oad(0xe8 + is_jmp, vtop->c.ul - 4); /* call/jmp im */
315 } else {
316 /* otherwise, indirect call */
317 r = gv(RC_INT);
318 o(0xff); /* call/jmp *r */
319 o(0xd0 + r + (is_jmp << 4));
323 static uint8_t fastcall_regs[3] = { TREG_EAX, TREG_EDX, TREG_ECX };
324 static uint8_t fastcallw_regs[2] = { TREG_ECX, TREG_EDX };
326 /* Generate function call. The function address is pushed first, then
327 all the parameters in call order. This functions pops all the
328 parameters and the function address. */
329 void gfunc_call(int nb_args)
331 int size, align, r, args_size, i, func_call;
332 Sym *func_sym;
334 args_size = 0;
335 for(i = 0;i < nb_args; i++) {
336 if ((vtop->type.t & VT_BTYPE) == VT_STRUCT) {
337 size = type_size(&vtop->type, &align);
338 /* align to stack align size */
339 size = (size + 3) & ~3;
340 /* allocate the necessary size on stack */
341 oad(0xec81, size); /* sub $xxx, %esp */
342 /* generate structure store */
343 r = get_reg(RC_INT);
344 o(0x89); /* mov %esp, r */
345 o(0xe0 + r);
346 vset(&vtop->type, r | VT_LVAL, 0);
347 vswap();
348 vstore();
349 args_size += size;
350 } else if (is_float(vtop->type.t)) {
351 gv(RC_FLOAT); /* only one float register */
352 if ((vtop->type.t & VT_BTYPE) == VT_FLOAT)
353 size = 4;
354 else if ((vtop->type.t & VT_BTYPE) == VT_DOUBLE)
355 size = 8;
356 else
357 size = 12;
358 oad(0xec81, size); /* sub $xxx, %esp */
359 if (size == 12)
360 o(0x7cdb);
361 else
362 o(0x5cd9 + size - 4); /* fstp[s|l] 0(%esp) */
363 g(0x24);
364 g(0x00);
365 args_size += size;
366 } else {
367 /* simple type (currently always same size) */
368 /* XXX: implicit cast ? */
369 r = gv(RC_INT);
370 if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
371 size = 8;
372 o(0x50 + vtop->r2); /* push r */
373 } else {
374 size = 4;
376 o(0x50 + r); /* push r */
377 args_size += size;
379 vtop--;
381 save_regs(0); /* save used temporary registers */
382 func_sym = vtop->type.ref;
383 func_call = FUNC_CALL(func_sym->r);
384 /* fast call case */
385 if ((func_call >= FUNC_FASTCALL1 && func_call <= FUNC_FASTCALL3) ||
386 func_call == FUNC_FASTCALLW) {
387 int fastcall_nb_regs;
388 uint8_t *fastcall_regs_ptr;
389 if (func_call == FUNC_FASTCALLW) {
390 fastcall_regs_ptr = fastcallw_regs;
391 fastcall_nb_regs = 2;
392 } else {
393 fastcall_regs_ptr = fastcall_regs;
394 fastcall_nb_regs = func_call - FUNC_FASTCALL1 + 1;
396 for(i = 0;i < fastcall_nb_regs; i++) {
397 if (args_size <= 0)
398 break;
399 o(0x58 + fastcall_regs_ptr[i]); /* pop r */
400 /* XXX: incorrect for struct/floats */
401 args_size -= 4;
404 gcall_or_jmp(0);
405 if (args_size && func_call != FUNC_STDCALL)
406 gadd_sp(args_size);
407 vtop--;
410 #ifdef TCC_TARGET_PE
411 #define FUNC_PROLOG_SIZE 10
412 #else
413 #define FUNC_PROLOG_SIZE 9
414 #endif
416 /* generate function prolog of type 't' */
417 void gfunc_prolog(CType *func_type)
419 int addr, align, size, func_call, fastcall_nb_regs;
420 int param_index, param_addr;
421 uint8_t *fastcall_regs_ptr;
422 Sym *sym;
423 CType *type;
425 sym = func_type->ref;
426 func_call = FUNC_CALL(sym->r);
427 addr = 8;
428 loc = 0;
429 if (func_call >= FUNC_FASTCALL1 && func_call <= FUNC_FASTCALL3) {
430 fastcall_nb_regs = func_call - FUNC_FASTCALL1 + 1;
431 fastcall_regs_ptr = fastcall_regs;
432 } else if (func_call == FUNC_FASTCALLW) {
433 fastcall_nb_regs = 2;
434 fastcall_regs_ptr = fastcallw_regs;
435 } else {
436 fastcall_nb_regs = 0;
437 fastcall_regs_ptr = NULL;
439 param_index = 0;
441 ind += FUNC_PROLOG_SIZE;
442 func_sub_sp_offset = ind;
443 /* if the function returns a structure, then add an
444 implicit pointer parameter */
445 func_vt = sym->type;
446 if ((func_vt.t & VT_BTYPE) == VT_STRUCT) {
447 /* XXX: fastcall case ? */
448 func_vc = addr;
449 addr += 4;
450 param_index++;
452 /* define parameters */
453 while ((sym = sym->next) != NULL) {
454 type = &sym->type;
455 size = type_size(type, &align);
456 size = (size + 3) & ~3;
457 #ifdef FUNC_STRUCT_PARAM_AS_PTR
458 /* structs are passed as pointer */
459 if ((type->t & VT_BTYPE) == VT_STRUCT) {
460 size = 4;
462 #endif
463 if (param_index < fastcall_nb_regs) {
464 /* save FASTCALL register */
465 loc -= 4;
466 o(0x89); /* movl */
467 gen_modrm(fastcall_regs_ptr[param_index], VT_LOCAL, NULL, loc);
468 param_addr = loc;
469 } else {
470 param_addr = addr;
471 addr += size;
473 sym_push(sym->v & ~SYM_FIELD, type,
474 VT_LOCAL | lvalue_type(type->t), param_addr);
475 param_index++;
477 func_ret_sub = 0;
478 /* pascal type call ? */
479 if (func_call == FUNC_STDCALL)
480 func_ret_sub = addr - 8;
482 /* leave some room for bound checking code */
483 if (do_bounds_check) {
484 oad(0xb8, 0); /* lbound section pointer */
485 oad(0xb8, 0); /* call to function */
486 func_bound_offset = lbounds_section->data_offset;
490 /* generate function epilog */
491 void gfunc_epilog(void)
493 int v, saved_ind;
495 #ifdef CONFIG_TCC_BCHECK
496 if (do_bounds_check && func_bound_offset != lbounds_section->data_offset) {
497 int saved_ind;
498 int *bounds_ptr;
499 Sym *sym, *sym_data;
500 /* add end of table info */
501 bounds_ptr = section_ptr_add(lbounds_section, sizeof(int));
502 *bounds_ptr = 0;
503 /* generate bound local allocation */
504 saved_ind = ind;
505 ind = func_sub_sp_offset;
506 sym_data = get_sym_ref(&char_pointer_type, lbounds_section,
507 func_bound_offset, lbounds_section->data_offset);
508 greloc(cur_text_section, sym_data,
509 ind + 1, R_386_32);
510 oad(0xb8, 0); /* mov %eax, xxx */
511 sym = external_global_sym(TOK___bound_local_new, &func_old_type, 0);
512 greloc(cur_text_section, sym,
513 ind + 1, R_386_PC32);
514 oad(0xe8, -4);
515 ind = saved_ind;
516 /* generate bound check local freeing */
517 o(0x5250); /* save returned value, if any */
518 greloc(cur_text_section, sym_data,
519 ind + 1, R_386_32);
520 oad(0xb8, 0); /* mov %eax, xxx */
521 sym = external_global_sym(TOK___bound_local_delete, &func_old_type, 0);
522 greloc(cur_text_section, sym,
523 ind + 1, R_386_PC32);
524 oad(0xe8, -4);
525 o(0x585a); /* restore returned value, if any */
527 #endif
528 o(0xc9); /* leave */
529 if (func_ret_sub == 0) {
530 o(0xc3); /* ret */
531 } else {
532 o(0xc2); /* ret n */
533 g(func_ret_sub);
534 g(func_ret_sub >> 8);
536 /* align local size to word & save local variables */
538 v = (-loc + 3) & -4;
539 saved_ind = ind;
540 ind = func_sub_sp_offset - FUNC_PROLOG_SIZE;
541 #ifdef TCC_TARGET_PE
542 if (v >= 4096) {
543 Sym *sym = external_global_sym(TOK___chkstk, &func_old_type, 0);
544 oad(0xb8, v); /* mov stacksize, %eax */
545 oad(0xe8, -4); /* call __chkstk, (does the stackframe too) */
546 greloc(cur_text_section, sym, ind-4, R_386_PC32);
547 } else
548 #endif
550 o(0xe58955); /* push %ebp, mov %esp, %ebp */
551 o(0xec81); /* sub esp, stacksize */
552 gen_le32(v);
553 #if FUNC_PROLOG_SIZE == 10
554 o(0x90); /* adjust to FUNC_PROLOG_SIZE */
555 #endif
557 ind = saved_ind;
560 /* generate a jump to a label */
561 int gjmp(int t)
563 return psym(0xe9, t);
566 /* generate a jump to a fixed address */
567 void gjmp_addr(int a)
569 int r;
570 r = a - ind - 2;
571 if (r == (char)r) {
572 g(0xeb);
573 g(r);
574 } else {
575 oad(0xe9, a - ind - 5);
579 /* generate a test. set 'inv' to invert test. Stack entry is popped */
580 int gtst(int inv, int t)
582 int v, *p;
584 v = vtop->r & VT_VALMASK;
585 if (v == VT_CMP) {
586 /* fast case : can jump directly since flags are set */
587 g(0x0f);
588 t = psym((vtop->c.i - 16) ^ inv, t);
589 } else if (v == VT_JMP || v == VT_JMPI) {
590 /* && or || optimization */
591 if ((v & 1) == inv) {
592 /* insert vtop->c jump list in t */
593 p = &vtop->c.i;
594 while (*p != 0)
595 p = (int *)(cur_text_section->data + *p);
596 *p = t;
597 t = vtop->c.i;
598 } else {
599 t = gjmp(t);
600 gsym(vtop->c.i);
602 } else {
603 if (is_float(vtop->type.t) ||
604 (vtop->type.t & VT_BTYPE) == VT_LLONG) {
605 vpushi(0);
606 gen_op(TOK_NE);
608 if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
609 /* constant jmp optimization */
610 if ((vtop->c.i != 0) != inv)
611 t = gjmp(t);
612 } else {
613 v = gv(RC_INT);
614 o(0x85);
615 o(0xc0 + v * 9);
616 g(0x0f);
617 t = psym(0x85 ^ inv, t);
620 vtop--;
621 return t;
624 /* generate an integer binary operation */
625 void gen_opi(int op)
627 int r, fr, opc, c;
629 switch(op) {
630 case '+':
631 case TOK_ADDC1: /* add with carry generation */
632 opc = 0;
633 gen_op8:
634 if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
635 /* constant case */
636 vswap();
637 r = gv(RC_INT);
638 vswap();
639 c = vtop->c.i;
640 if (c == (char)c) {
641 /* XXX: generate inc and dec for smaller code ? */
642 o(0x83);
643 o(0xc0 | (opc << 3) | r);
644 g(c);
645 } else {
646 o(0x81);
647 oad(0xc0 | (opc << 3) | r, c);
649 } else {
650 gv2(RC_INT, RC_INT);
651 r = vtop[-1].r;
652 fr = vtop[0].r;
653 o((opc << 3) | 0x01);
654 o(0xc0 + r + fr * 8);
656 vtop--;
657 if (op >= TOK_ULT && op <= TOK_GT) {
658 vtop->r = VT_CMP;
659 vtop->c.i = op;
661 break;
662 case '-':
663 case TOK_SUBC1: /* sub with carry generation */
664 opc = 5;
665 goto gen_op8;
666 case TOK_ADDC2: /* add with carry use */
667 opc = 2;
668 goto gen_op8;
669 case TOK_SUBC2: /* sub with carry use */
670 opc = 3;
671 goto gen_op8;
672 case '&':
673 opc = 4;
674 goto gen_op8;
675 case '^':
676 opc = 6;
677 goto gen_op8;
678 case '|':
679 opc = 1;
680 goto gen_op8;
681 case '*':
682 gv2(RC_INT, RC_INT);
683 r = vtop[-1].r;
684 fr = vtop[0].r;
685 vtop--;
686 o(0xaf0f); /* imul fr, r */
687 o(0xc0 + fr + r * 8);
688 break;
689 case TOK_SHL:
690 opc = 4;
691 goto gen_shift;
692 case TOK_SHR:
693 opc = 5;
694 goto gen_shift;
695 case TOK_SAR:
696 opc = 7;
697 gen_shift:
698 opc = 0xc0 | (opc << 3);
699 if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
700 /* constant case */
701 vswap();
702 r = gv(RC_INT);
703 vswap();
704 c = vtop->c.i & 0x1f;
705 o(0xc1); /* shl/shr/sar $xxx, r */
706 o(opc | r);
707 g(c);
708 } else {
709 /* we generate the shift in ecx */
710 gv2(RC_INT, RC_ECX);
711 r = vtop[-1].r;
712 o(0xd3); /* shl/shr/sar %cl, r */
713 o(opc | r);
715 vtop--;
716 break;
717 case '/':
718 case TOK_UDIV:
719 case TOK_PDIV:
720 case '%':
721 case TOK_UMOD:
722 case TOK_UMULL:
723 /* first operand must be in eax */
724 /* XXX: need better constraint for second operand */
725 gv2(RC_EAX, RC_ECX);
726 r = vtop[-1].r;
727 fr = vtop[0].r;
728 vtop--;
729 save_reg(TREG_EDX);
730 if (op == TOK_UMULL) {
731 o(0xf7); /* mul fr */
732 o(0xe0 + fr);
733 vtop->r2 = TREG_EDX;
734 r = TREG_EAX;
735 } else {
736 if (op == TOK_UDIV || op == TOK_UMOD) {
737 o(0xf7d231); /* xor %edx, %edx, div fr, %eax */
738 o(0xf0 + fr);
739 } else {
740 o(0xf799); /* cltd, idiv fr, %eax */
741 o(0xf8 + fr);
743 if (op == '%' || op == TOK_UMOD)
744 r = TREG_EDX;
745 else
746 r = TREG_EAX;
748 vtop->r = r;
749 break;
750 default:
751 opc = 7;
752 goto gen_op8;
756 /* generate a floating point operation 'v = t1 op t2' instruction. The
757 two operands are guaranted to have the same floating point type */
758 /* XXX: need to use ST1 too */
759 void gen_opf(int op)
761 int a, ft, fc, swapped, r;
763 /* convert constants to memory references */
764 if ((vtop[-1].r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
765 vswap();
766 gv(RC_FLOAT);
767 vswap();
769 if ((vtop[0].r & (VT_VALMASK | VT_LVAL)) == VT_CONST)
770 gv(RC_FLOAT);
772 /* must put at least one value in the floating point register */
773 if ((vtop[-1].r & VT_LVAL) &&
774 (vtop[0].r & VT_LVAL)) {
775 vswap();
776 gv(RC_FLOAT);
777 vswap();
779 swapped = 0;
780 /* swap the stack if needed so that t1 is the register and t2 is
781 the memory reference */
782 if (vtop[-1].r & VT_LVAL) {
783 vswap();
784 swapped = 1;
786 if (op >= TOK_ULT && op <= TOK_GT) {
787 /* load on stack second operand */
788 load(TREG_ST0, vtop);
789 save_reg(TREG_EAX); /* eax is used by FP comparison code */
790 if (op == TOK_GE || op == TOK_GT)
791 swapped = !swapped;
792 else if (op == TOK_EQ || op == TOK_NE)
793 swapped = 0;
794 if (swapped)
795 o(0xc9d9); /* fxch %st(1) */
796 o(0xe9da); /* fucompp */
797 o(0xe0df); /* fnstsw %ax */
798 if (op == TOK_EQ) {
799 o(0x45e480); /* and $0x45, %ah */
800 o(0x40fC80); /* cmp $0x40, %ah */
801 } else if (op == TOK_NE) {
802 o(0x45e480); /* and $0x45, %ah */
803 o(0x40f480); /* xor $0x40, %ah */
804 op = TOK_NE;
805 } else if (op == TOK_GE || op == TOK_LE) {
806 o(0x05c4f6); /* test $0x05, %ah */
807 op = TOK_EQ;
808 } else {
809 o(0x45c4f6); /* test $0x45, %ah */
810 op = TOK_EQ;
812 vtop--;
813 vtop->r = VT_CMP;
814 vtop->c.i = op;
815 } else {
816 /* no memory reference possible for long double operations */
817 if ((vtop->type.t & VT_BTYPE) == VT_LDOUBLE) {
818 load(TREG_ST0, vtop);
819 swapped = !swapped;
822 switch(op) {
823 default:
824 case '+':
825 a = 0;
826 break;
827 case '-':
828 a = 4;
829 if (swapped)
830 a++;
831 break;
832 case '*':
833 a = 1;
834 break;
835 case '/':
836 a = 6;
837 if (swapped)
838 a++;
839 break;
841 ft = vtop->type.t;
842 fc = vtop->c.ul;
843 if ((ft & VT_BTYPE) == VT_LDOUBLE) {
844 o(0xde); /* fxxxp %st, %st(1) */
845 o(0xc1 + (a << 3));
846 } else {
847 /* if saved lvalue, then we must reload it */
848 r = vtop->r;
849 if ((r & VT_VALMASK) == VT_LLOCAL) {
850 SValue v1;
851 r = get_reg(RC_INT);
852 v1.type.t = VT_INT;
853 v1.r = VT_LOCAL | VT_LVAL;
854 v1.c.ul = fc;
855 load(r, &v1);
856 fc = 0;
859 if ((ft & VT_BTYPE) == VT_DOUBLE)
860 o(0xdc);
861 else
862 o(0xd8);
863 gen_modrm(a, r, vtop->sym, fc);
865 vtop--;
869 /* convert integers to fp 't' type. Must handle 'int', 'unsigned int'
870 and 'long long' cases. */
871 void gen_cvt_itof(int t)
873 save_reg(TREG_ST0);
874 gv(RC_INT);
875 if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
876 /* signed long long to float/double/long double (unsigned case
877 is handled generically) */
878 o(0x50 + vtop->r2); /* push r2 */
879 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
880 o(0x242cdf); /* fildll (%esp) */
881 o(0x08c483); /* add $8, %esp */
882 } else if ((vtop->type.t & (VT_BTYPE | VT_UNSIGNED)) ==
883 (VT_INT | VT_UNSIGNED)) {
884 /* unsigned int to float/double/long double */
885 o(0x6a); /* push $0 */
886 g(0x00);
887 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
888 o(0x242cdf); /* fildll (%esp) */
889 o(0x08c483); /* add $8, %esp */
890 } else {
891 /* int to float/double/long double */
892 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
893 o(0x2404db); /* fildl (%esp) */
894 o(0x04c483); /* add $4, %esp */
896 vtop->r = TREG_ST0;
899 /* convert fp to int 't' type */
900 /* XXX: handle long long case */
901 void gen_cvt_ftoi(int t)
903 int r, r2, size;
904 Sym *sym;
905 CType ushort_type;
907 ushort_type.t = VT_SHORT | VT_UNSIGNED;
909 gv(RC_FLOAT);
910 if (t != VT_INT)
911 size = 8;
912 else
913 size = 4;
915 o(0x2dd9); /* ldcw xxx */
916 sym = external_global_sym(TOK___tcc_int_fpu_control,
917 &ushort_type, VT_LVAL);
918 greloc(cur_text_section, sym,
919 ind, R_386_32);
920 gen_le32(0);
922 oad(0xec81, size); /* sub $xxx, %esp */
923 if (size == 4)
924 o(0x1cdb); /* fistpl */
925 else
926 o(0x3cdf); /* fistpll */
927 o(0x24);
928 o(0x2dd9); /* ldcw xxx */
929 sym = external_global_sym(TOK___tcc_fpu_control,
930 &ushort_type, VT_LVAL);
931 greloc(cur_text_section, sym,
932 ind, R_386_32);
933 gen_le32(0);
935 r = get_reg(RC_INT);
936 o(0x58 + r); /* pop r */
937 if (size == 8) {
938 if (t == VT_LLONG) {
939 vtop->r = r; /* mark reg as used */
940 r2 = get_reg(RC_INT);
941 o(0x58 + r2); /* pop r2 */
942 vtop->r2 = r2;
943 } else {
944 o(0x04c483); /* add $4, %esp */
947 vtop->r = r;
950 /* convert from one floating point type to another */
951 void gen_cvt_ftof(int t)
953 /* all we have to do on i386 is to put the float in a register */
954 gv(RC_FLOAT);
957 /* computed goto support */
958 void ggoto(void)
960 gcall_or_jmp(1);
961 vtop--;
964 /* bound check support functions */
965 #ifdef CONFIG_TCC_BCHECK
967 /* generate a bounded pointer addition */
968 void gen_bounded_ptr_add(void)
970 Sym *sym;
972 /* prepare fast i386 function call (args in eax and edx) */
973 gv2(RC_EAX, RC_EDX);
974 /* save all temporary registers */
975 vtop -= 2;
976 save_regs(0);
977 /* do a fast function call */
978 sym = external_global_sym(TOK___bound_ptr_add, &func_old_type, 0);
979 greloc(cur_text_section, sym,
980 ind + 1, R_386_PC32);
981 oad(0xe8, -4);
982 /* returned pointer is in eax */
983 vtop++;
984 vtop->r = TREG_EAX | VT_BOUNDED;
985 /* address of bounding function call point */
986 vtop->c.ul = (cur_text_section->reloc->data_offset - sizeof(Elf32_Rel));
989 /* patch pointer addition in vtop so that pointer dereferencing is
990 also tested */
991 void gen_bounded_ptr_deref(void)
993 int func;
994 int size, align;
995 Elf32_Rel *rel;
996 Sym *sym;
998 size = 0;
999 /* XXX: put that code in generic part of tcc */
1000 if (!is_float(vtop->type.t)) {
1001 if (vtop->r & VT_LVAL_BYTE)
1002 size = 1;
1003 else if (vtop->r & VT_LVAL_SHORT)
1004 size = 2;
1006 if (!size)
1007 size = type_size(&vtop->type, &align);
1008 switch(size) {
1009 case 1: func = TOK___bound_ptr_indir1; break;
1010 case 2: func = TOK___bound_ptr_indir2; break;
1011 case 4: func = TOK___bound_ptr_indir4; break;
1012 case 8: func = TOK___bound_ptr_indir8; break;
1013 case 12: func = TOK___bound_ptr_indir12; break;
1014 case 16: func = TOK___bound_ptr_indir16; break;
1015 default:
1016 error("unhandled size when derefencing bounded pointer");
1017 func = 0;
1018 break;
1021 /* patch relocation */
1022 /* XXX: find a better solution ? */
1023 rel = (Elf32_Rel *)(cur_text_section->reloc->data + vtop->c.ul);
1024 sym = external_global_sym(func, &func_old_type, 0);
1025 if (!sym->c)
1026 put_extern_sym(sym, NULL, 0, 0);
1027 rel->r_info = ELF32_R_INFO(sym->c, ELF32_R_TYPE(rel->r_info));
1029 #endif
1031 /* end of X86 code generator */
1032 /*************************************************************/