Got test1-3 working on x86-64.
[tinycc.git] / i386-gen.c
blob67123d6bd6a887b5b4f3598aff6fadb874e94b16
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 #ifdef TARGET_DEFS_ONLY
23 /* number of available registers */
24 #define NB_REGS 4
25 #define NB_ASM_REGS 8
27 /* a register can belong to several classes. The classes must be
28 sorted from more general to more precise (see gv2() code which does
29 assumptions on it). */
30 #define RC_INT 0x0001 /* generic integer register */
31 #define RC_FLOAT 0x0002 /* generic float register */
32 #define RC_EAX 0x0004
33 #define RC_ST0 0x0008
34 #define RC_ECX 0x0010
35 #define RC_EDX 0x0020
36 #define RC_IRET RC_EAX /* function return: integer register */
37 #define RC_LRET RC_EDX /* function return: second integer register */
38 #define RC_FRET RC_ST0 /* function return: float register */
40 /* pretty names for the registers */
41 enum {
42 TREG_EAX = 0,
43 TREG_ECX,
44 TREG_EDX,
45 TREG_ST0,
48 /* return registers for function */
49 #define REG_IRET TREG_EAX /* single word int return register */
50 #define REG_LRET TREG_EDX /* second word return register (for long long) */
51 #define REG_FRET TREG_ST0 /* float return register */
53 /* defined if function parameters must be evaluated in reverse order */
54 #define INVERT_FUNC_PARAMS
56 /* defined if structures are passed as pointers. Otherwise structures
57 are directly pushed on stack. */
58 /* #define FUNC_STRUCT_PARAM_AS_PTR */
60 /* pointer size, in bytes */
61 #define PTR_SIZE 4
63 /* long double size and alignment, in bytes */
64 #define LDOUBLE_SIZE 12
65 #define LDOUBLE_ALIGN 4
66 /* maximum alignment (for aligned attribute support) */
67 #define MAX_ALIGN 8
70 #define psym oad
72 /******************************************************/
73 /* ELF defines */
75 #define EM_TCC_TARGET EM_386
77 /* relocation type for 32 bit data relocation */
78 #define R_DATA_32 R_386_32
79 #define R_DATA_PTR 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 /******************************************************/
87 #else /* ! TARGET_DEFS_ONLY */
88 /******************************************************/
89 #include "tcc.h"
91 ST_DATA const int reg_classes[NB_REGS] = {
92 /* eax */ RC_INT | RC_EAX,
93 /* ecx */ RC_INT | RC_ECX,
94 /* edx */ RC_INT | RC_EDX,
95 /* st0 */ RC_FLOAT | RC_ST0,
98 static unsigned long func_sub_sp_offset;
99 static int func_ret_sub;
100 #ifdef CONFIG_TCC_BCHECK
101 static unsigned long func_bound_offset;
102 #endif
104 /* XXX: make it faster ? */
105 ST_FUNC void g(int c)
107 int ind1;
108 ind1 = ind + 1;
109 if (ind1 > cur_text_section->data_allocated)
110 section_realloc(cur_text_section, ind1);
111 cur_text_section->data[ind] = c;
112 ind = ind1;
115 ST_FUNC void o(unsigned int c)
117 while (c) {
118 g(c);
119 c = c >> 8;
123 ST_FUNC void gen_le16(int v)
125 g(v);
126 g(v >> 8);
129 ST_FUNC void gen_le32(int c)
131 g(c);
132 g(c >> 8);
133 g(c >> 16);
134 g(c >> 24);
137 /* output a symbol and patch all calls to it */
138 ST_FUNC void gsym_addr(int t, int a)
140 int n, *ptr;
141 while (t) {
142 ptr = (int *)(cur_text_section->data + t);
143 n = *ptr; /* next value */
144 *ptr = a - t - 4;
145 t = n;
149 ST_FUNC void gsym(int t)
151 gsym_addr(t, ind);
154 /* psym is used to put an instruction with a data field which is a
155 reference to a symbol. It is in fact the same as oad ! */
156 #define psym oad
158 /* instruction + 4 bytes data. Return the address of the data */
159 ST_FUNC int oad(int c, int s)
161 int ind1;
163 o(c);
164 ind1 = ind + 4;
165 if (ind1 > cur_text_section->data_allocated)
166 section_realloc(cur_text_section, ind1);
167 *(int *)(cur_text_section->data + ind) = s;
168 s = ind;
169 ind = ind1;
170 return s;
173 /* output constant with relocation if 'r & VT_SYM' is true */
174 ST_FUNC void gen_addr32(int r, Sym *sym, int c)
176 if (r & VT_SYM)
177 greloc(cur_text_section, sym, ind, R_386_32);
178 gen_le32(c);
181 ST_FUNC void gen_addrpc32(int r, Sym *sym, int c)
183 if (r & VT_SYM)
184 greloc(cur_text_section, sym, ind, R_386_PC32);
185 gen_le32(c - 4);
188 /* generate a modrm reference. 'op_reg' contains the addtionnal 3
189 opcode bits */
190 static void gen_modrm(int op_reg, int r, Sym *sym, int c)
192 op_reg = op_reg << 3;
193 if ((r & VT_VALMASK) == VT_CONST) {
194 /* constant memory reference */
195 o(0x05 | op_reg);
196 gen_addr32(r, sym, c);
197 } else if ((r & VT_VALMASK) == VT_LOCAL) {
198 /* currently, we use only ebp as base */
199 if (c == (char)c) {
200 /* short reference */
201 o(0x45 | op_reg);
202 g(c);
203 } else {
204 oad(0x85 | op_reg, c);
206 } else {
207 g(0x00 | op_reg | (r & VT_VALMASK));
211 /* load 'r' from value 'sv' */
212 ST_FUNC void load(int r, SValue *sv)
214 int v, t, ft, fc, fr;
215 SValue v1;
217 #ifdef TCC_TARGET_PE
218 SValue v2;
219 sv = pe_getimport(sv, &v2);
220 #endif
222 fr = sv->r;
223 ft = sv->type.t;
224 fc = sv->c.ul;
226 v = fr & VT_VALMASK;
227 if (fr & VT_LVAL) {
228 if (v == VT_LLOCAL) {
229 v1.type.t = VT_INT;
230 v1.r = VT_LOCAL | VT_LVAL;
231 v1.c.ul = fc;
232 fr = r;
233 if (!(reg_classes[fr] & RC_INT))
234 fr = get_reg(RC_INT);
235 load(fr, &v1);
237 if ((ft & VT_BTYPE) == VT_FLOAT) {
238 o(0xd9); /* flds */
239 r = 0;
240 } else if ((ft & VT_BTYPE) == VT_DOUBLE) {
241 o(0xdd); /* fldl */
242 r = 0;
243 } else if ((ft & VT_BTYPE) == VT_LDOUBLE) {
244 o(0xdb); /* fldt */
245 r = 5;
246 } else if ((ft & VT_TYPE) == VT_BYTE) {
247 o(0xbe0f); /* movsbl */
248 } else if ((ft & VT_TYPE) == (VT_BYTE | VT_UNSIGNED)) {
249 o(0xb60f); /* movzbl */
250 } else if ((ft & VT_TYPE) == VT_SHORT) {
251 o(0xbf0f); /* movswl */
252 } else if ((ft & VT_TYPE) == (VT_SHORT | VT_UNSIGNED)) {
253 o(0xb70f); /* movzwl */
254 } else {
255 o(0x8b); /* movl */
257 gen_modrm(r, fr, sv->sym, fc);
258 } else {
259 if (v == VT_CONST) {
260 o(0xb8 + r); /* mov $xx, r */
261 gen_addr32(fr, sv->sym, fc);
262 } else if (v == VT_LOCAL) {
263 if (fc) {
264 o(0x8d); /* lea xxx(%ebp), r */
265 gen_modrm(r, VT_LOCAL, sv->sym, fc);
266 } else {
267 o(0x89);
268 o(0xe8 + r); /* mov %ebp, r */
270 } else if (v == VT_CMP) {
271 oad(0xb8 + r, 0); /* mov $0, r */
272 o(0x0f); /* setxx %br */
273 o(fc);
274 o(0xc0 + r);
275 } else if (v == VT_JMP || v == VT_JMPI) {
276 t = v & 1;
277 oad(0xb8 + r, t); /* mov $1, r */
278 o(0x05eb); /* jmp after */
279 gsym(fc);
280 oad(0xb8 + r, t ^ 1); /* mov $0, r */
281 } else if (v != r) {
282 o(0x89);
283 o(0xc0 + r + v * 8); /* mov v, r */
288 /* store register 'r' in lvalue 'v' */
289 ST_FUNC void store(int r, SValue *v)
291 int fr, bt, ft, fc;
293 #ifdef TCC_TARGET_PE
294 SValue v2;
295 v = pe_getimport(v, &v2);
296 #endif
298 ft = v->type.t;
299 fc = v->c.ul;
300 fr = v->r & VT_VALMASK;
301 bt = ft & VT_BTYPE;
302 /* XXX: incorrect if float reg to reg */
303 if (bt == VT_FLOAT) {
304 o(0xd9); /* fsts */
305 r = 2;
306 } else if (bt == VT_DOUBLE) {
307 o(0xdd); /* fstpl */
308 r = 2;
309 } else if (bt == VT_LDOUBLE) {
310 o(0xc0d9); /* fld %st(0) */
311 o(0xdb); /* fstpt */
312 r = 7;
313 } else {
314 if (bt == VT_SHORT)
315 o(0x66);
316 if (bt == VT_BYTE || bt == VT_BOOL)
317 o(0x88);
318 else
319 o(0x89);
321 if (fr == VT_CONST ||
322 fr == VT_LOCAL ||
323 (v->r & VT_LVAL)) {
324 gen_modrm(r, v->r, v->sym, fc);
325 } else if (fr != r) {
326 o(0xc0 + fr + r * 8); /* mov r, fr */
330 static void gadd_sp(int val)
332 if (val == (char)val) {
333 o(0xc483);
334 g(val);
335 } else {
336 oad(0xc481, val); /* add $xxx, %esp */
340 /* 'is_jmp' is '1' if it is a jump */
341 static void gcall_or_jmp(int is_jmp)
343 int r;
344 if ((vtop->r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
345 /* constant case */
346 if (vtop->r & VT_SYM) {
347 /* relocation case */
348 greloc(cur_text_section, vtop->sym,
349 ind + 1, R_386_PC32);
350 } else {
351 /* put an empty PC32 relocation */
352 put_elf_reloc(symtab_section, cur_text_section,
353 ind + 1, R_386_PC32, 0);
355 oad(0xe8 + is_jmp, vtop->c.ul - 4); /* call/jmp im */
356 } else {
357 /* otherwise, indirect call */
358 r = gv(RC_INT);
359 o(0xff); /* call/jmp *r */
360 o(0xd0 + r + (is_jmp << 4));
364 static uint8_t fastcall_regs[3] = { TREG_EAX, TREG_EDX, TREG_ECX };
365 static uint8_t fastcallw_regs[2] = { TREG_ECX, TREG_EDX };
367 /* Return 1 if this function returns via an sret pointer, 0 otherwise */
368 ST_FUNC int gfunc_sret(CType *vt, CType *ret, int *ret_align) {
369 *ret_align = 1; // Never have to re-align return values for x86
370 #ifdef TCC_TARGET_PE
371 int size, align;
372 size = type_size(vt, &align);
373 if (size > 8) {
374 return 1;
375 } else if (size > 4) {
376 ret->ref = NULL;
377 ret->t = VT_LLONG;
378 return 0;
379 } else {
380 ret->ref = NULL;
381 ret->t = VT_INT;
382 return 0;
384 #else
385 return 1;
386 #endif
389 /* Generate function call. The function address is pushed first, then
390 all the parameters in call order. This functions pops all the
391 parameters and the function address. */
392 ST_FUNC void gfunc_call(int nb_args)
394 int size, align, r, args_size, i, func_call;
395 Sym *func_sym;
397 args_size = 0;
398 for(i = 0;i < nb_args; i++) {
399 if ((vtop->type.t & VT_BTYPE) == VT_STRUCT) {
400 size = type_size(&vtop->type, &align);
401 /* align to stack align size */
402 size = (size + 3) & ~3;
403 /* allocate the necessary size on stack */
404 oad(0xec81, size); /* sub $xxx, %esp */
405 /* generate structure store */
406 r = get_reg(RC_INT);
407 o(0x89); /* mov %esp, r */
408 o(0xe0 + r);
409 vset(&vtop->type, r | VT_LVAL, 0);
410 vswap();
411 vstore();
412 args_size += size;
413 } else if (is_float(vtop->type.t)) {
414 gv(RC_FLOAT); /* only one float register */
415 if ((vtop->type.t & VT_BTYPE) == VT_FLOAT)
416 size = 4;
417 else if ((vtop->type.t & VT_BTYPE) == VT_DOUBLE)
418 size = 8;
419 else
420 size = 12;
421 oad(0xec81, size); /* sub $xxx, %esp */
422 if (size == 12)
423 o(0x7cdb);
424 else
425 o(0x5cd9 + size - 4); /* fstp[s|l] 0(%esp) */
426 g(0x24);
427 g(0x00);
428 args_size += size;
429 } else {
430 /* simple type (currently always same size) */
431 /* XXX: implicit cast ? */
432 r = gv(RC_INT);
433 if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
434 size = 8;
435 o(0x50 + vtop->r2); /* push r */
436 } else {
437 size = 4;
439 o(0x50 + r); /* push r */
440 args_size += size;
442 vtop--;
444 save_regs(0); /* save used temporary registers */
445 func_sym = vtop->type.ref;
446 func_call = FUNC_CALL(func_sym->r);
447 /* fast call case */
448 if ((func_call >= FUNC_FASTCALL1 && func_call <= FUNC_FASTCALL3) ||
449 func_call == FUNC_FASTCALLW) {
450 int fastcall_nb_regs;
451 uint8_t *fastcall_regs_ptr;
452 if (func_call == FUNC_FASTCALLW) {
453 fastcall_regs_ptr = fastcallw_regs;
454 fastcall_nb_regs = 2;
455 } else {
456 fastcall_regs_ptr = fastcall_regs;
457 fastcall_nb_regs = func_call - FUNC_FASTCALL1 + 1;
459 for(i = 0;i < fastcall_nb_regs; i++) {
460 if (args_size <= 0)
461 break;
462 o(0x58 + fastcall_regs_ptr[i]); /* pop r */
463 /* XXX: incorrect for struct/floats */
464 args_size -= 4;
467 gcall_or_jmp(0);
469 if (args_size && func_call != FUNC_STDCALL)
470 gadd_sp(args_size);
471 vtop--;
474 #ifdef TCC_TARGET_PE
475 #define FUNC_PROLOG_SIZE 10
476 #else
477 #define FUNC_PROLOG_SIZE 9
478 #endif
480 /* generate function prolog of type 't' */
481 ST_FUNC void gfunc_prolog(CType *func_type)
483 int addr, align, size, func_call, fastcall_nb_regs;
484 int param_index, param_addr;
485 uint8_t *fastcall_regs_ptr;
486 Sym *sym;
487 CType *type;
489 sym = func_type->ref;
490 func_call = FUNC_CALL(sym->r);
491 addr = 8;
492 loc = 0;
493 func_vc = 0;
495 if (func_call >= FUNC_FASTCALL1 && func_call <= FUNC_FASTCALL3) {
496 fastcall_nb_regs = func_call - FUNC_FASTCALL1 + 1;
497 fastcall_regs_ptr = fastcall_regs;
498 } else if (func_call == FUNC_FASTCALLW) {
499 fastcall_nb_regs = 2;
500 fastcall_regs_ptr = fastcallw_regs;
501 } else {
502 fastcall_nb_regs = 0;
503 fastcall_regs_ptr = NULL;
505 param_index = 0;
507 ind += FUNC_PROLOG_SIZE;
508 func_sub_sp_offset = ind;
509 /* if the function returns a structure, then add an
510 implicit pointer parameter */
511 func_vt = sym->type;
512 #ifdef TCC_TARGET_PE
513 size = type_size(&func_vt,&align);
514 if (((func_vt.t & VT_BTYPE) == VT_STRUCT) && (size > 8)) {
515 #else
516 if ((func_vt.t & VT_BTYPE) == VT_STRUCT) {
517 #endif
518 /* XXX: fastcall case ? */
519 func_vc = addr;
520 addr += 4;
521 param_index++;
523 /* define parameters */
524 while ((sym = sym->next) != NULL) {
525 type = &sym->type;
526 size = type_size(type, &align);
527 size = (size + 3) & ~3;
528 #ifdef FUNC_STRUCT_PARAM_AS_PTR
529 /* structs are passed as pointer */
530 if ((type->t & VT_BTYPE) == VT_STRUCT) {
531 size = 4;
533 #endif
534 if (param_index < fastcall_nb_regs) {
535 /* save FASTCALL register */
536 loc -= 4;
537 o(0x89); /* movl */
538 gen_modrm(fastcall_regs_ptr[param_index], VT_LOCAL, NULL, loc);
539 param_addr = loc;
540 } else {
541 param_addr = addr;
542 addr += size;
544 sym_push(sym->v & ~SYM_FIELD, type,
545 VT_LOCAL | lvalue_type(type->t), param_addr);
546 param_index++;
548 func_ret_sub = 0;
549 /* pascal type call ? */
550 if (func_call == FUNC_STDCALL)
551 func_ret_sub = addr - 8;
553 #ifdef CONFIG_TCC_BCHECK
554 /* leave some room for bound checking code */
555 if (tcc_state->do_bounds_check) {
556 oad(0xb8, 0); /* lbound section pointer */
557 oad(0xb8, 0); /* call to function */
558 func_bound_offset = lbounds_section->data_offset;
560 #endif
563 /* generate function epilog */
564 ST_FUNC void gfunc_epilog(void)
566 int v, saved_ind;
568 #ifdef CONFIG_TCC_BCHECK
569 if (tcc_state->do_bounds_check
570 && func_bound_offset != lbounds_section->data_offset) {
571 int saved_ind;
572 int *bounds_ptr;
573 Sym *sym, *sym_data;
574 /* add end of table info */
575 bounds_ptr = section_ptr_add(lbounds_section, sizeof(int));
576 *bounds_ptr = 0;
577 /* generate bound local allocation */
578 saved_ind = ind;
579 ind = func_sub_sp_offset;
580 sym_data = get_sym_ref(&char_pointer_type, lbounds_section,
581 func_bound_offset, lbounds_section->data_offset);
582 greloc(cur_text_section, sym_data,
583 ind + 1, R_386_32);
584 oad(0xb8, 0); /* mov %eax, xxx */
585 sym = external_global_sym(TOK___bound_local_new, &func_old_type, 0);
586 greloc(cur_text_section, sym,
587 ind + 1, R_386_PC32);
588 oad(0xe8, -4);
589 ind = saved_ind;
590 /* generate bound check local freeing */
591 o(0x5250); /* save returned value, if any */
592 greloc(cur_text_section, sym_data,
593 ind + 1, R_386_32);
594 oad(0xb8, 0); /* mov %eax, xxx */
595 sym = external_global_sym(TOK___bound_local_delete, &func_old_type, 0);
596 greloc(cur_text_section, sym,
597 ind + 1, R_386_PC32);
598 oad(0xe8, -4);
599 o(0x585a); /* restore returned value, if any */
601 #endif
602 o(0xc9); /* leave */
603 if (func_ret_sub == 0) {
604 o(0xc3); /* ret */
605 } else {
606 o(0xc2); /* ret n */
607 g(func_ret_sub);
608 g(func_ret_sub >> 8);
610 /* align local size to word & save local variables */
612 v = (-loc + 3) & -4;
613 saved_ind = ind;
614 ind = func_sub_sp_offset - FUNC_PROLOG_SIZE;
615 #ifdef TCC_TARGET_PE
616 if (v >= 4096) {
617 Sym *sym = external_global_sym(TOK___chkstk, &func_old_type, 0);
618 oad(0xb8, v); /* mov stacksize, %eax */
619 oad(0xe8, -4); /* call __chkstk, (does the stackframe too) */
620 greloc(cur_text_section, sym, ind-4, R_386_PC32);
621 } else
622 #endif
624 o(0xe58955); /* push %ebp, mov %esp, %ebp */
625 o(0xec81); /* sub esp, stacksize */
626 gen_le32(v);
627 #if FUNC_PROLOG_SIZE == 10
628 o(0x90); /* adjust to FUNC_PROLOG_SIZE */
629 #endif
631 ind = saved_ind;
634 /* generate a jump to a label */
635 ST_FUNC int gjmp(int t)
637 return psym(0xe9, t);
640 /* generate a jump to a fixed address */
641 ST_FUNC void gjmp_addr(int a)
643 int r;
644 r = a - ind - 2;
645 if (r == (char)r) {
646 g(0xeb);
647 g(r);
648 } else {
649 oad(0xe9, a - ind - 5);
653 /* generate a test. set 'inv' to invert test. Stack entry is popped */
654 ST_FUNC int gtst(int inv, int t)
656 int v, *p;
658 v = vtop->r & VT_VALMASK;
659 if (v == VT_CMP) {
660 /* fast case : can jump directly since flags are set */
661 g(0x0f);
662 t = psym((vtop->c.i - 16) ^ inv, t);
663 } else if (v == VT_JMP || v == VT_JMPI) {
664 /* && or || optimization */
665 if ((v & 1) == inv) {
666 /* insert vtop->c jump list in t */
667 p = &vtop->c.i;
668 while (*p != 0)
669 p = (int *)(cur_text_section->data + *p);
670 *p = t;
671 t = vtop->c.i;
672 } else {
673 t = gjmp(t);
674 gsym(vtop->c.i);
676 } else {
677 if (is_float(vtop->type.t) ||
678 (vtop->type.t & VT_BTYPE) == VT_LLONG) {
679 vpushi(0);
680 gen_op(TOK_NE);
682 if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
683 /* constant jmp optimization */
684 if ((vtop->c.i != 0) != inv)
685 t = gjmp(t);
686 } else {
687 v = gv(RC_INT);
688 o(0x85);
689 o(0xc0 + v * 9);
690 g(0x0f);
691 t = psym(0x85 ^ inv, t);
694 vtop--;
695 return t;
698 /* generate an integer binary operation */
699 ST_FUNC void gen_opi(int op)
701 int r, fr, opc, c;
703 switch(op) {
704 case '+':
705 case TOK_ADDC1: /* add with carry generation */
706 opc = 0;
707 gen_op8:
708 if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
709 /* constant case */
710 vswap();
711 r = gv(RC_INT);
712 vswap();
713 c = vtop->c.i;
714 if (c == (char)c) {
715 /* generate inc and dec for smaller code */
716 if (c==1 && opc==0) {
717 o (0x40 | r); // inc
718 } else if (c==1 && opc==5) {
719 o (0x48 | r); // dec
720 } else {
721 o(0x83);
722 o(0xc0 | (opc << 3) | r);
723 g(c);
725 } else {
726 o(0x81);
727 oad(0xc0 | (opc << 3) | r, c);
729 } else {
730 gv2(RC_INT, RC_INT);
731 r = vtop[-1].r;
732 fr = vtop[0].r;
733 o((opc << 3) | 0x01);
734 o(0xc0 + r + fr * 8);
736 vtop--;
737 if (op >= TOK_ULT && op <= TOK_GT) {
738 vtop->r = VT_CMP;
739 vtop->c.i = op;
741 break;
742 case '-':
743 case TOK_SUBC1: /* sub with carry generation */
744 opc = 5;
745 goto gen_op8;
746 case TOK_ADDC2: /* add with carry use */
747 opc = 2;
748 goto gen_op8;
749 case TOK_SUBC2: /* sub with carry use */
750 opc = 3;
751 goto gen_op8;
752 case '&':
753 opc = 4;
754 goto gen_op8;
755 case '^':
756 opc = 6;
757 goto gen_op8;
758 case '|':
759 opc = 1;
760 goto gen_op8;
761 case '*':
762 gv2(RC_INT, RC_INT);
763 r = vtop[-1].r;
764 fr = vtop[0].r;
765 vtop--;
766 o(0xaf0f); /* imul fr, r */
767 o(0xc0 + fr + r * 8);
768 break;
769 case TOK_SHL:
770 opc = 4;
771 goto gen_shift;
772 case TOK_SHR:
773 opc = 5;
774 goto gen_shift;
775 case TOK_SAR:
776 opc = 7;
777 gen_shift:
778 opc = 0xc0 | (opc << 3);
779 if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
780 /* constant case */
781 vswap();
782 r = gv(RC_INT);
783 vswap();
784 c = vtop->c.i & 0x1f;
785 o(0xc1); /* shl/shr/sar $xxx, r */
786 o(opc | r);
787 g(c);
788 } else {
789 /* we generate the shift in ecx */
790 gv2(RC_INT, RC_ECX);
791 r = vtop[-1].r;
792 o(0xd3); /* shl/shr/sar %cl, r */
793 o(opc | r);
795 vtop--;
796 break;
797 case '/':
798 case TOK_UDIV:
799 case TOK_PDIV:
800 case '%':
801 case TOK_UMOD:
802 case TOK_UMULL:
803 /* first operand must be in eax */
804 /* XXX: need better constraint for second operand */
805 gv2(RC_EAX, RC_ECX);
806 r = vtop[-1].r;
807 fr = vtop[0].r;
808 vtop--;
809 save_reg(TREG_EDX);
810 if (op == TOK_UMULL) {
811 o(0xf7); /* mul fr */
812 o(0xe0 + fr);
813 vtop->r2 = TREG_EDX;
814 r = TREG_EAX;
815 } else {
816 if (op == TOK_UDIV || op == TOK_UMOD) {
817 o(0xf7d231); /* xor %edx, %edx, div fr, %eax */
818 o(0xf0 + fr);
819 } else {
820 o(0xf799); /* cltd, idiv fr, %eax */
821 o(0xf8 + fr);
823 if (op == '%' || op == TOK_UMOD)
824 r = TREG_EDX;
825 else
826 r = TREG_EAX;
828 vtop->r = r;
829 break;
830 default:
831 opc = 7;
832 goto gen_op8;
836 /* generate a floating point operation 'v = t1 op t2' instruction. The
837 two operands are guaranted to have the same floating point type */
838 /* XXX: need to use ST1 too */
839 ST_FUNC void gen_opf(int op)
841 int a, ft, fc, swapped, r;
843 /* convert constants to memory references */
844 if ((vtop[-1].r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
845 vswap();
846 gv(RC_FLOAT);
847 vswap();
849 if ((vtop[0].r & (VT_VALMASK | VT_LVAL)) == VT_CONST)
850 gv(RC_FLOAT);
852 /* must put at least one value in the floating point register */
853 if ((vtop[-1].r & VT_LVAL) &&
854 (vtop[0].r & VT_LVAL)) {
855 vswap();
856 gv(RC_FLOAT);
857 vswap();
859 swapped = 0;
860 /* swap the stack if needed so that t1 is the register and t2 is
861 the memory reference */
862 if (vtop[-1].r & VT_LVAL) {
863 vswap();
864 swapped = 1;
866 if (op >= TOK_ULT && op <= TOK_GT) {
867 /* load on stack second operand */
868 load(TREG_ST0, vtop);
869 save_reg(TREG_EAX); /* eax is used by FP comparison code */
870 if (op == TOK_GE || op == TOK_GT)
871 swapped = !swapped;
872 else if (op == TOK_EQ || op == TOK_NE)
873 swapped = 0;
874 if (swapped)
875 o(0xc9d9); /* fxch %st(1) */
876 o(0xe9da); /* fucompp */
877 o(0xe0df); /* fnstsw %ax */
878 if (op == TOK_EQ) {
879 o(0x45e480); /* and $0x45, %ah */
880 o(0x40fC80); /* cmp $0x40, %ah */
881 } else if (op == TOK_NE) {
882 o(0x45e480); /* and $0x45, %ah */
883 o(0x40f480); /* xor $0x40, %ah */
884 op = TOK_NE;
885 } else if (op == TOK_GE || op == TOK_LE) {
886 o(0x05c4f6); /* test $0x05, %ah */
887 op = TOK_EQ;
888 } else {
889 o(0x45c4f6); /* test $0x45, %ah */
890 op = TOK_EQ;
892 vtop--;
893 vtop->r = VT_CMP;
894 vtop->c.i = op;
895 } else {
896 /* no memory reference possible for long double operations */
897 if ((vtop->type.t & VT_BTYPE) == VT_LDOUBLE) {
898 load(TREG_ST0, vtop);
899 swapped = !swapped;
902 switch(op) {
903 default:
904 case '+':
905 a = 0;
906 break;
907 case '-':
908 a = 4;
909 if (swapped)
910 a++;
911 break;
912 case '*':
913 a = 1;
914 break;
915 case '/':
916 a = 6;
917 if (swapped)
918 a++;
919 break;
921 ft = vtop->type.t;
922 fc = vtop->c.ul;
923 if ((ft & VT_BTYPE) == VT_LDOUBLE) {
924 o(0xde); /* fxxxp %st, %st(1) */
925 o(0xc1 + (a << 3));
926 } else {
927 /* if saved lvalue, then we must reload it */
928 r = vtop->r;
929 if ((r & VT_VALMASK) == VT_LLOCAL) {
930 SValue v1;
931 r = get_reg(RC_INT);
932 v1.type.t = VT_INT;
933 v1.r = VT_LOCAL | VT_LVAL;
934 v1.c.ul = fc;
935 load(r, &v1);
936 fc = 0;
939 if ((ft & VT_BTYPE) == VT_DOUBLE)
940 o(0xdc);
941 else
942 o(0xd8);
943 gen_modrm(a, r, vtop->sym, fc);
945 vtop--;
949 /* convert integers to fp 't' type. Must handle 'int', 'unsigned int'
950 and 'long long' cases. */
951 ST_FUNC void gen_cvt_itof(int t)
953 save_reg(TREG_ST0);
954 gv(RC_INT);
955 if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
956 /* signed long long to float/double/long double (unsigned case
957 is handled generically) */
958 o(0x50 + vtop->r2); /* push r2 */
959 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
960 o(0x242cdf); /* fildll (%esp) */
961 o(0x08c483); /* add $8, %esp */
962 } else if ((vtop->type.t & (VT_BTYPE | VT_UNSIGNED)) ==
963 (VT_INT | VT_UNSIGNED)) {
964 /* unsigned int to float/double/long double */
965 o(0x6a); /* push $0 */
966 g(0x00);
967 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
968 o(0x242cdf); /* fildll (%esp) */
969 o(0x08c483); /* add $8, %esp */
970 } else {
971 /* int to float/double/long double */
972 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
973 o(0x2404db); /* fildl (%esp) */
974 o(0x04c483); /* add $4, %esp */
976 vtop->r = TREG_ST0;
979 /* convert fp to int 't' type */
980 /* XXX: handle long long case */
981 ST_FUNC void gen_cvt_ftoi(int t)
983 int r, r2, size;
984 Sym *sym;
985 CType ushort_type;
987 ushort_type.t = VT_SHORT | VT_UNSIGNED;
988 ushort_type.ref = 0;
990 gv(RC_FLOAT);
991 if (t != VT_INT)
992 size = 8;
993 else
994 size = 4;
996 o(0x2dd9); /* ldcw xxx */
997 sym = external_global_sym(TOK___tcc_int_fpu_control,
998 &ushort_type, VT_LVAL);
999 greloc(cur_text_section, sym,
1000 ind, R_386_32);
1001 gen_le32(0);
1003 oad(0xec81, size); /* sub $xxx, %esp */
1004 if (size == 4)
1005 o(0x1cdb); /* fistpl */
1006 else
1007 o(0x3cdf); /* fistpll */
1008 o(0x24);
1009 o(0x2dd9); /* ldcw xxx */
1010 sym = external_global_sym(TOK___tcc_fpu_control,
1011 &ushort_type, VT_LVAL);
1012 greloc(cur_text_section, sym,
1013 ind, R_386_32);
1014 gen_le32(0);
1016 r = get_reg(RC_INT);
1017 o(0x58 + r); /* pop r */
1018 if (size == 8) {
1019 if (t == VT_LLONG) {
1020 vtop->r = r; /* mark reg as used */
1021 r2 = get_reg(RC_INT);
1022 o(0x58 + r2); /* pop r2 */
1023 vtop->r2 = r2;
1024 } else {
1025 o(0x04c483); /* add $4, %esp */
1028 vtop->r = r;
1031 /* convert from one floating point type to another */
1032 ST_FUNC void gen_cvt_ftof(int t)
1034 /* all we have to do on i386 is to put the float in a register */
1035 gv(RC_FLOAT);
1038 /* computed goto support */
1039 ST_FUNC void ggoto(void)
1041 gcall_or_jmp(1);
1042 vtop--;
1045 /* bound check support functions */
1046 #ifdef CONFIG_TCC_BCHECK
1048 /* generate a bounded pointer addition */
1049 ST_FUNC void gen_bounded_ptr_add(void)
1051 Sym *sym;
1053 /* prepare fast i386 function call (args in eax and edx) */
1054 gv2(RC_EAX, RC_EDX);
1055 /* save all temporary registers */
1056 vtop -= 2;
1057 save_regs(0);
1058 /* do a fast function call */
1059 sym = external_global_sym(TOK___bound_ptr_add, &func_old_type, 0);
1060 greloc(cur_text_section, sym,
1061 ind + 1, R_386_PC32);
1062 oad(0xe8, -4);
1063 /* returned pointer is in eax */
1064 vtop++;
1065 vtop->r = TREG_EAX | VT_BOUNDED;
1066 /* address of bounding function call point */
1067 vtop->c.ul = (cur_text_section->reloc->data_offset - sizeof(Elf32_Rel));
1070 /* patch pointer addition in vtop so that pointer dereferencing is
1071 also tested */
1072 ST_FUNC void gen_bounded_ptr_deref(void)
1074 int func;
1075 int size, align;
1076 Elf32_Rel *rel;
1077 Sym *sym;
1079 size = 0;
1080 /* XXX: put that code in generic part of tcc */
1081 if (!is_float(vtop->type.t)) {
1082 if (vtop->r & VT_LVAL_BYTE)
1083 size = 1;
1084 else if (vtop->r & VT_LVAL_SHORT)
1085 size = 2;
1087 if (!size)
1088 size = type_size(&vtop->type, &align);
1089 switch(size) {
1090 case 1: func = TOK___bound_ptr_indir1; break;
1091 case 2: func = TOK___bound_ptr_indir2; break;
1092 case 4: func = TOK___bound_ptr_indir4; break;
1093 case 8: func = TOK___bound_ptr_indir8; break;
1094 case 12: func = TOK___bound_ptr_indir12; break;
1095 case 16: func = TOK___bound_ptr_indir16; break;
1096 default:
1097 tcc_error("unhandled size when dereferencing bounded pointer");
1098 func = 0;
1099 break;
1102 /* patch relocation */
1103 /* XXX: find a better solution ? */
1104 rel = (Elf32_Rel *)(cur_text_section->reloc->data + vtop->c.ul);
1105 sym = external_global_sym(func, &func_old_type, 0);
1106 if (!sym->c)
1107 put_extern_sym(sym, NULL, 0, 0);
1108 rel->r_info = ELF32_R_INFO(sym->c, ELF32_R_TYPE(rel->r_info));
1110 #endif
1112 /* end of X86 code generator */
1113 /*************************************************************/
1114 #endif
1115 /*************************************************************/