mirrors/tinycc
1
0
Fork 0
mirror of https://github.com/frida/tinycc synced 2024-12-25 22:46:49 +03:00
Code Issues Packages Projects Releases Wiki Activity
1339d04759
tinycc/i386-gen.c
seyko a37f8cfc80 short_call_convention patch from tcc bugzilla 
BUGZILLA:
    interfacing with other compilers

    extend the return value to the whole register if necessary.
    visual studio and gcc do not always set the whole eax register
    when assigning the return value of a function.

    We've encountered wrong execution results on i386 platforms with an
    application that uses both code compiled with TCC and code compiled
    with other compilers (namely: Visual Studio on Windows, and GCC on
    Linux).

    When calling a function that returns an integer value shorter than 32
    bits, TCC reads the return value from the whole EAX register,
    although the code generated by the other compilers can only sets AL
    for 8 bit values or AX for 16 bits values, and the rest of EAX can be
    anything.

    We worked around this with the attached patch on i386 for the version
    0.9.26, but we did not look at other platforms to find if there are
    similar issues.
2016-05-15 21:10:06 +03:00

1163 lines
32 KiB
C
Raw Blame History

/*
* X86 code generator for TCC
*
* Copyright (c) 2001-2004 Fabrice Bellard
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#ifdef TARGET_DEFS_ONLY
/* number of available registers */
#define NB_REGS 4
#define NB_ASM_REGS 8
/* a register can belong to several classes. The classes must be
sorted from more general to more precise (see gv2() code which does
assumptions on it). */
#define RC_INT 0x0001 /* generic integer register */
#define RC_FLOAT 0x0002 /* generic float register */
#define RC_EAX 0x0004
#define RC_ST0 0x0008
#define RC_ECX 0x0010
#define RC_EDX 0x0020
#define RC_IRET RC_EAX /* function return: integer register */
#define RC_LRET RC_EDX /* function return: second integer register */
#define RC_FRET RC_ST0 /* function return: float register */
/* pretty names for the registers */
enum {
TREG_EAX = 0,
TREG_ECX,
TREG_EDX,
TREG_ST0,
TREG_ESP = 4
};
/* return registers for function */
#define REG_IRET TREG_EAX /* single word int return register */
#define REG_LRET TREG_EDX /* second word return register (for long long) */
#define REG_FRET TREG_ST0 /* float return register */
/* defined if function parameters must be evaluated in reverse order */
#define INVERT_FUNC_PARAMS
/* defined if structures are passed as pointers. Otherwise structures
are directly pushed on stack. */
/* #define FUNC_STRUCT_PARAM_AS_PTR */
/* pointer size, in bytes */
#define PTR_SIZE 4
/* long double size and alignment, in bytes */
#define LDOUBLE_SIZE 12
#define LDOUBLE_ALIGN 4
/* maximum alignment (for aligned attribute support) */
#define MAX_ALIGN 8
#define psym oad
/******************************************************/
/* ELF defines */
#define EM_TCC_TARGET EM_386
/* relocation type for 32 bit data relocation */
#define R_DATA_32 R_386_32
#define R_DATA_PTR R_386_32
#define R_JMP_SLOT R_386_JMP_SLOT
#define R_COPY R_386_COPY
#define ELF_START_ADDR 0x08048000
#define ELF_PAGE_SIZE 0x1000
/******************************************************/
#else /* ! TARGET_DEFS_ONLY */
/******************************************************/
#include "tcc.h"
ST_DATA const int reg_classes[NB_REGS] = {
/* eax */ RC_INT | RC_EAX,
/* ecx */ RC_INT | RC_ECX,
/* edx */ RC_INT | RC_EDX,
/* st0 */ RC_FLOAT | RC_ST0,
};
static unsigned long func_sub_sp_offset;
static int func_ret_sub;
#ifdef CONFIG_TCC_BCHECK
static addr_t func_bound_offset;
#endif
/* XXX: make it faster ? */
ST_FUNC void g(int c)
{
int ind1;
ind1 = ind + 1;
if (ind1 > cur_text_section->data_allocated)
section_realloc(cur_text_section, ind1);
cur_text_section->data[ind] = c;
ind = ind1;
}
ST_FUNC void o(unsigned int c)
{
while (c) {
g(c);
c = c >> 8;
}
}
ST_FUNC void gen_le16(int v)
{
g(v);
g(v >> 8);
}
ST_FUNC void gen_le32(int c)
{
g(c);
g(c >> 8);
g(c >> 16);
g(c >> 24);
}
/* output a symbol and patch all calls to it */
ST_FUNC void gsym_addr(int t, int a)
{
while (t) {
unsigned char *ptr = cur_text_section->data + t;
uint32_t n = read32le(ptr); /* next value */
write32le(ptr, a - t - 4);
t = n;
}
}
ST_FUNC void gsym(int t)
{
gsym_addr(t, ind);
}
/* psym is used to put an instruction with a data field which is a
reference to a symbol. It is in fact the same as oad ! */
#define psym oad
/* instruction + 4 bytes data. Return the address of the data */
ST_FUNC int oad(int c, int s)
{
int ind1;
o(c);
ind1 = ind + 4;
if (ind1 > cur_text_section->data_allocated)
section_realloc(cur_text_section, ind1);
write32le(cur_text_section->data + ind, s);
s = ind;
ind = ind1;
return s;
}
/* output constant with relocation if 'r & VT_SYM' is true */
ST_FUNC void gen_addr32(int r, Sym *sym, int c)
{
if (r & VT_SYM)
greloc(cur_text_section, sym, ind, R_386_32);
gen_le32(c);
}
ST_FUNC void gen_addrpc32(int r, Sym *sym, int c)
{
if (r & VT_SYM)
greloc(cur_text_section, sym, ind, R_386_PC32);
gen_le32(c - 4);
}
/* generate a modrm reference. 'op_reg' contains the addtionnal 3
opcode bits */
static void gen_modrm(int op_reg, int r, Sym *sym, int c)
{
op_reg = op_reg << 3;
if ((r & VT_VALMASK) == VT_CONST) {
/* constant memory reference */
o(0x05 | op_reg);
gen_addr32(r, sym, c);
} else if ((r & VT_VALMASK) == VT_LOCAL) {
/* currently, we use only ebp as base */
if (c == (char)c) {
/* short reference */
o(0x45 | op_reg);
g(c);
} else {
oad(0x85 | op_reg, c);
}
} else {
g(0x00 | op_reg | (r & VT_VALMASK));
}
}
/* load 'r' from value 'sv' */
ST_FUNC void load(int r, SValue *sv)
{
int v, t, ft, fc, fr;
SValue v1;
#ifdef TCC_TARGET_PE
SValue v2;
sv = pe_getimport(sv, &v2);
#endif
fr = sv->r;
ft = sv->type.t;
fc = sv->c.i;
ft &= ~(VT_VOLATILE | VT_CONSTANT);
v = fr & VT_VALMASK;
if (fr & VT_LVAL) {
if (v == VT_LLOCAL) {
v1.type.t = VT_INT;
v1.r = VT_LOCAL | VT_LVAL;
v1.c.i = fc;
fr = r;
if (!(reg_classes[fr] & RC_INT))
fr = get_reg(RC_INT);
load(fr, &v1);
}
if ((ft & VT_BTYPE) == VT_FLOAT) {
o(0xd9); /* flds */
r = 0;
} else if ((ft & VT_BTYPE) == VT_DOUBLE) {
o(0xdd); /* fldl */
r = 0;
} else if ((ft & VT_BTYPE) == VT_LDOUBLE) {
o(0xdb); /* fldt */
r = 5;
} else if ((ft & VT_TYPE) == VT_BYTE || (ft & VT_TYPE) == VT_BOOL) {
o(0xbe0f); /* movsbl */
} else if ((ft & VT_TYPE) == (VT_BYTE | VT_UNSIGNED)) {
o(0xb60f); /* movzbl */
} else if ((ft & VT_TYPE) == VT_SHORT) {
o(0xbf0f); /* movswl */
} else if ((ft & VT_TYPE) == (VT_SHORT | VT_UNSIGNED)) {
o(0xb70f); /* movzwl */
} else {
o(0x8b); /* movl */
}
gen_modrm(r, fr, sv->sym, fc);
} else {
if (v == VT_CONST) {
o(0xb8 + r); /* mov $xx, r */
gen_addr32(fr, sv->sym, fc);
} else if (v == VT_LOCAL) {
if (fc) {
o(0x8d); /* lea xxx(%ebp), r */
gen_modrm(r, VT_LOCAL, sv->sym, fc);
} else {
o(0x89);
o(0xe8 + r); /* mov %ebp, r */
}
} else if (v == VT_CMP) {
oad(0xb8 + r, 0); /* mov $0, r */
o(0x0f); /* setxx %br */
o(fc);
o(0xc0 + r);
} else if (v == VT_JMP || v == VT_JMPI) {
t = v & 1;
oad(0xb8 + r, t); /* mov $1, r */
o(0x05eb); /* jmp after */
gsym(fc);
oad(0xb8 + r, t ^ 1); /* mov $0, r */
} else if (v != r) {
o(0x89);
o(0xc0 + r + v * 8); /* mov v, r */
}
}
}
/* store register 'r' in lvalue 'v' */
ST_FUNC void store(int r, SValue *v)
{
int fr, bt, ft, fc;
#ifdef TCC_TARGET_PE
SValue v2;
v = pe_getimport(v, &v2);
#endif
ft = v->type.t;
fc = v->c.i;
fr = v->r & VT_VALMASK;
ft &= ~(VT_VOLATILE | VT_CONSTANT);
bt = ft & VT_BTYPE;
/* XXX: incorrect if float reg to reg */
if (bt == VT_FLOAT) {
o(0xd9); /* fsts */
r = 2;
} else if (bt == VT_DOUBLE) {
o(0xdd); /* fstpl */
r = 2;
} else if (bt == VT_LDOUBLE) {
o(0xc0d9); /* fld %st(0) */
o(0xdb); /* fstpt */
r = 7;
} else {
if (bt == VT_SHORT)
o(0x66);
if (bt == VT_BYTE || bt == VT_BOOL)
o(0x88);
else
o(0x89);
}
if (fr == VT_CONST ||
fr == VT_LOCAL ||
(v->r & VT_LVAL)) {
gen_modrm(r, v->r, v->sym, fc);
} else if (fr != r) {
o(0xc0 + fr + r * 8); /* mov r, fr */
}
}
static void gadd_sp(int val)
{
if (val == (char)val) {
o(0xc483);
g(val);
} else {
oad(0xc481, val); /* add $xxx, %esp */
}
}
static void gen_static_call(int v)
{
Sym *sym;
sym = external_global_sym(v, &func_old_type, 0);
oad(0xe8, -4);
greloc(cur_text_section, sym, ind-4, R_386_PC32);
}
/* 'is_jmp' is '1' if it is a jump */
static void gcall_or_jmp(int is_jmp)
{
int r;
if ((vtop->r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
int rt;
/* constant case */
if (vtop->r & VT_SYM) {
/* relocation case */
greloc(cur_text_section, vtop->sym,
ind + 1, R_386_PC32);
} else {
/* put an empty PC32 relocation */
put_elf_reloc(symtab_section, cur_text_section,
ind + 1, R_386_PC32, 0);
}
oad(0xe8 + is_jmp, vtop->c.i - 4); /* call/jmp im */
/* extend the return value to the whole register if necessary
visual studio and gcc do not always set the whole eax register
when assigning the return value of a function */
rt = vtop->type.ref->type.t;
switch (rt & VT_BTYPE) {
case VT_BYTE:
if (rt & VT_UNSIGNED) {
o(0xc0b60f); /* movzx %al, %eax */
}
else {
o(0xc0be0f); /* movsx %al, %eax */
}
break;
case VT_SHORT:
if (rt & VT_UNSIGNED) {
o(0xc0b70f); /* movzx %ax, %eax */
}
else {
o(0xc0bf0f); /* movsx %ax, %eax */
}
break;
default:
break;
}
} else {
/* otherwise, indirect call */
r = gv(RC_INT);
o(0xff); /* call/jmp *r */
o(0xd0 + r + (is_jmp << 4));
}
}
static uint8_t fastcall_regs[3] = { TREG_EAX, TREG_EDX, TREG_ECX };
static uint8_t fastcallw_regs[2] = { TREG_ECX, TREG_EDX };
/* Return the number of registers needed to return the struct, or 0 if
returning via struct pointer. */
ST_FUNC int gfunc_sret(CType *vt, int variadic, CType *ret, int *ret_align, int *regsize)
{
#ifdef TCC_TARGET_PE
int size, align;
*ret_align = 1; // Never have to re-align return values for x86
*regsize = 4;
size = type_size(vt, &align);
if (size > 8) {
return 0;
} else if (size > 4) {
ret->ref = NULL;
ret->t = VT_LLONG;
return 1;
} else {
ret->ref = NULL;
ret->t = VT_INT;
return 1;
}
#else
*ret_align = 1; // Never have to re-align return values for x86
return 0;
#endif
}
/* Generate function call. The function address is pushed first, then
all the parameters in call order. This functions pops all the
parameters and the function address. */
ST_FUNC void gfunc_call(int nb_args)
{
int size, align, r, args_size, i, func_call;
Sym *func_sym;
args_size = 0;
for(i = 0;i < nb_args; i++) {
if ((vtop->type.t & VT_BTYPE) == VT_STRUCT) {
size = type_size(&vtop->type, &align);
/* align to stack align size */
size = (size + 3) & ~3;
/* allocate the necessary size on stack */
oad(0xec81, size); /* sub $xxx, %esp */
/* generate structure store */
r = get_reg(RC_INT);
o(0x89); /* mov %esp, r */
o(0xe0 + r);
vset(&vtop->type, r | VT_LVAL, 0);
vswap();
vstore();
args_size += size;
} else if (is_float(vtop->type.t)) {
gv(RC_FLOAT); /* only one float register */
if ((vtop->type.t & VT_BTYPE) == VT_FLOAT)
size = 4;
else if ((vtop->type.t & VT_BTYPE) == VT_DOUBLE)
size = 8;
else
size = 12;
oad(0xec81, size); /* sub $xxx, %esp */
if (size == 12)
o(0x7cdb);
else
o(0x5cd9 + size - 4); /* fstp[s|l] 0(%esp) */
g(0x24);
g(0x00);
args_size += size;
} else {
/* simple type (currently always same size) */
/* XXX: implicit cast ? */
r = gv(RC_INT);
if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
size = 8;
o(0x50 + vtop->r2); /* push r */
} else {
size = 4;
}
o(0x50 + r); /* push r */
args_size += size;
}
vtop--;
}
save_regs(0); /* save used temporary registers */
func_sym = vtop->type.ref;
func_call = func_sym->a.func_call;
/* fast call case */
if ((func_call >= FUNC_FASTCALL1 && func_call <= FUNC_FASTCALL3) ||
func_call == FUNC_FASTCALLW) {
int fastcall_nb_regs;
uint8_t *fastcall_regs_ptr;
if (func_call == FUNC_FASTCALLW) {
fastcall_regs_ptr = fastcallw_regs;
fastcall_nb_regs = 2;
} else {
fastcall_regs_ptr = fastcall_regs;
fastcall_nb_regs = func_call - FUNC_FASTCALL1 + 1;
}
for(i = 0;i < fastcall_nb_regs; i++) {
if (args_size <= 0)
break;
o(0x58 + fastcall_regs_ptr[i]); /* pop r */
/* XXX: incorrect for struct/floats */
args_size -= 4;
}
}
#ifndef TCC_TARGET_PE
else if ((vtop->type.ref->type.t & VT_BTYPE) == VT_STRUCT)
args_size -= 4;
#endif
gcall_or_jmp(0);
if (args_size && func_call != FUNC_STDCALL)
gadd_sp(args_size);
vtop--;
}
#ifdef TCC_TARGET_PE
#define FUNC_PROLOG_SIZE 10
#else
#define FUNC_PROLOG_SIZE 9
#endif
/* generate function prolog of type 't' */
ST_FUNC void gfunc_prolog(CType *func_type)
{
int addr, align, size, func_call, fastcall_nb_regs;
int param_index, param_addr;
uint8_t *fastcall_regs_ptr;
Sym *sym;
CType *type;
sym = func_type->ref;
func_call = sym->a.func_call;
addr = 8;
loc = 0;
func_vc = 0;
if (func_call >= FUNC_FASTCALL1 && func_call <= FUNC_FASTCALL3) {
fastcall_nb_regs = func_call - FUNC_FASTCALL1 + 1;
fastcall_regs_ptr = fastcall_regs;
} else if (func_call == FUNC_FASTCALLW) {
fastcall_nb_regs = 2;
fastcall_regs_ptr = fastcallw_regs;
} else {
fastcall_nb_regs = 0;
fastcall_regs_ptr = NULL;
}
param_index = 0;
ind += FUNC_PROLOG_SIZE;
func_sub_sp_offset = ind;
/* if the function returns a structure, then add an
implicit pointer parameter */
func_vt = sym->type;
func_var = (sym->c == FUNC_ELLIPSIS);
#ifdef TCC_TARGET_PE
size = type_size(&func_vt,&align);
if (((func_vt.t & VT_BTYPE) == VT_STRUCT) && (size > 8)) {
#else
if ((func_vt.t & VT_BTYPE) == VT_STRUCT) {
#endif
/* XXX: fastcall case ? */
func_vc = addr;
addr += 4;
param_index++;
}
/* define parameters */
while ((sym = sym->next) != NULL) {
type = &sym->type;
size = type_size(type, &align);
size = (size + 3) & ~3;
#ifdef FUNC_STRUCT_PARAM_AS_PTR
/* structs are passed as pointer */
if ((type->t & VT_BTYPE) == VT_STRUCT) {
size = 4;
}
#endif
if (param_index < fastcall_nb_regs) {
/* save FASTCALL register */
loc -= 4;
o(0x89); /* movl */
gen_modrm(fastcall_regs_ptr[param_index], VT_LOCAL, NULL, loc);
param_addr = loc;
} else {
param_addr = addr;
addr += size;
}
sym_push(sym->v & ~SYM_FIELD, type,
VT_LOCAL | lvalue_type(type->t), param_addr);
param_index++;
}
func_ret_sub = 0;
/* pascal type call ? */
if (func_call == FUNC_STDCALL)
func_ret_sub = addr - 8;
#ifndef TCC_TARGET_PE
else if (func_vc)
func_ret_sub = 4;
#endif
#ifdef CONFIG_TCC_BCHECK
/* leave some room for bound checking code */
if (tcc_state->do_bounds_check) {
oad(0xb8, 0); /* lbound section pointer */
oad(0xb8, 0); /* call to function */
func_bound_offset = lbounds_section->data_offset;
}
#endif
}
/* generate function epilog */
ST_FUNC void gfunc_epilog(void)
{
addr_t v, saved_ind;
#ifdef CONFIG_TCC_BCHECK
if (tcc_state->do_bounds_check
&& func_bound_offset != lbounds_section->data_offset) {
addr_t saved_ind;
addr_t *bounds_ptr;
Sym *sym_data;
/* add end of table info */
bounds_ptr = section_ptr_add(lbounds_section, sizeof(addr_t));
*bounds_ptr = 0;
/* generate bound local allocation */
saved_ind = ind;
ind = func_sub_sp_offset;
sym_data = get_sym_ref(&char_pointer_type, lbounds_section,
func_bound_offset, lbounds_section->data_offset);
greloc(cur_text_section, sym_data,
ind + 1, R_386_32);
oad(0xb8, 0); /* mov %eax, xxx */
gen_static_call(TOK___bound_local_new);
ind = saved_ind;
/* generate bound check local freeing */
o(0x5250); /* save returned value, if any */
greloc(cur_text_section, sym_data,
ind + 1, R_386_32);
oad(0xb8, 0); /* mov %eax, xxx */
gen_static_call(TOK___bound_local_delete);
o(0x585a); /* restore returned value, if any */
}
#endif
o(0xc9); /* leave */
if (func_ret_sub == 0) {
o(0xc3); /* ret */
} else {
o(0xc2); /* ret n */
g(func_ret_sub);
g(func_ret_sub >> 8);
}
/* align local size to word & save local variables */
v = (-loc + 3) & -4;
saved_ind = ind;
ind = func_sub_sp_offset - FUNC_PROLOG_SIZE;
#ifdef TCC_TARGET_PE
if (v >= 4096) {
oad(0xb8, v); /* mov stacksize, %eax */
gen_static_call(TOK___chkstk); /* call __chkstk, (does the stackframe too) */
} else
#endif
{
o(0xe58955); /* push %ebp, mov %esp, %ebp */
o(0xec81); /* sub esp, stacksize */
gen_le32(v);
#if FUNC_PROLOG_SIZE == 10
o(0x90); /* adjust to FUNC_PROLOG_SIZE */
#endif
}
ind = saved_ind;
}
/* generate a jump to a label */
ST_FUNC int gjmp(int t)
{
return psym(0xe9, t);
}
/* generate a jump to a fixed address */
ST_FUNC void gjmp_addr(int a)
{
int r;
r = a - ind - 2;
if (r == (char)r) {
g(0xeb);
g(r);
} else {
oad(0xe9, a - ind - 5);
}
}
/* generate a test. set 'inv' to invert test. Stack entry is popped */
ST_FUNC int gtst(int inv, int t)
{
int v = vtop->r & VT_VALMASK;
if (v == VT_CMP) {
/* fast case : can jump directly since flags are set */
g(0x0f);
t = psym((vtop->c.i - 16) ^ inv, t);
} else if (v == VT_JMP || v == VT_JMPI) {
/* && or || optimization */
if ((v & 1) == inv) {
/* insert vtop->c jump list in t */
uint32_t n1, n = vtop->c.i;
if (n) {
while ((n1 = read32le(cur_text_section->data + n)))
n = n1;
write32le(cur_text_section->data + n, t);
t = vtop->c.i;
}
} else {
t = gjmp(t);
gsym(vtop->c.i);
}
}
vtop--;
return t;
}
/* generate an integer binary operation */
ST_FUNC void gen_opi(int op)
{
int r, fr, opc, c;
switch(op) {
case '+':
case TOK_ADDC1: /* add with carry generation */
opc = 0;
gen_op8:
if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
/* constant case */
vswap();
r = gv(RC_INT);
vswap();
c = vtop->c.i;
if (c == (char)c) {
/* generate inc and dec for smaller code */
if (c==1 && opc==0) {
o (0x40 | r); // inc
} else if (c==1 && opc==5) {
o (0x48 | r); // dec
} else {
o(0x83);
o(0xc0 | (opc << 3) | r);
g(c);
}
} else {
o(0x81);
oad(0xc0 | (opc << 3) | r, c);
}
} else {
gv2(RC_INT, RC_INT);
r = vtop[-1].r;
fr = vtop[0].r;
o((opc << 3) | 0x01);
o(0xc0 + r + fr * 8);
}
vtop--;
if (op >= TOK_ULT && op <= TOK_GT) {
vtop->r = VT_CMP;
vtop->c.i = op;
}
break;
case '-':
case TOK_SUBC1: /* sub with carry generation */
opc = 5;
goto gen_op8;
case TOK_ADDC2: /* add with carry use */
opc = 2;
goto gen_op8;
case TOK_SUBC2: /* sub with carry use */
opc = 3;
goto gen_op8;
case '&':
opc = 4;
goto gen_op8;
case '^':
opc = 6;
goto gen_op8;
case '|':
opc = 1;
goto gen_op8;
case '*':
gv2(RC_INT, RC_INT);
r = vtop[-1].r;
fr = vtop[0].r;
vtop--;
o(0xaf0f); /* imul fr, r */
o(0xc0 + fr + r * 8);
break;
case TOK_SHL:
opc = 4;
goto gen_shift;
case TOK_SHR:
opc = 5;
goto gen_shift;
case TOK_SAR:
opc = 7;
gen_shift:
opc = 0xc0 | (opc << 3);
if ((vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST) {
/* constant case */
vswap();
r = gv(RC_INT);
vswap();
c = vtop->c.i & 0x1f;
o(0xc1); /* shl/shr/sar $xxx, r */
o(opc | r);
g(c);
} else {
/* we generate the shift in ecx */
gv2(RC_INT, RC_ECX);
r = vtop[-1].r;
o(0xd3); /* shl/shr/sar %cl, r */
o(opc | r);
}
vtop--;
break;
case '/':
case TOK_UDIV:
case TOK_PDIV:
case '%':
case TOK_UMOD:
case TOK_UMULL:
/* first operand must be in eax */
/* XXX: need better constraint for second operand */
gv2(RC_EAX, RC_ECX);
r = vtop[-1].r;
fr = vtop[0].r;
vtop--;
save_reg(TREG_EDX);
if (op == TOK_UMULL) {
o(0xf7); /* mul fr */
o(0xe0 + fr);
vtop->r2 = TREG_EDX;
r = TREG_EAX;
} else {
if (op == TOK_UDIV || op == TOK_UMOD) {
o(0xf7d231); /* xor %edx, %edx, div fr, %eax */
o(0xf0 + fr);
} else {
o(0xf799); /* cltd, idiv fr, %eax */
o(0xf8 + fr);
}
if (op == '%' || op == TOK_UMOD)
r = TREG_EDX;
else
r = TREG_EAX;
}
vtop->r = r;
break;
default:
opc = 7;
goto gen_op8;
}
}
/* generate a floating point operation 'v = t1 op t2' instruction. The
two operands are guaranted to have the same floating point type */
/* XXX: need to use ST1 too */
ST_FUNC void gen_opf(int op)
{
int a, ft, fc, swapped, r;
/* convert constants to memory references */
if ((vtop[-1].r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
vswap();
gv(RC_FLOAT);
vswap();
}
if ((vtop[0].r & (VT_VALMASK | VT_LVAL)) == VT_CONST)
gv(RC_FLOAT);
/* must put at least one value in the floating point register */
if ((vtop[-1].r & VT_LVAL) &&
(vtop[0].r & VT_LVAL)) {
vswap();
gv(RC_FLOAT);
vswap();
}
swapped = 0;
/* swap the stack if needed so that t1 is the register and t2 is
the memory reference */
if (vtop[-1].r & VT_LVAL) {
vswap();
swapped = 1;
}
if (op >= TOK_ULT && op <= TOK_GT) {
/* load on stack second operand */
load(TREG_ST0, vtop);
save_reg(TREG_EAX); /* eax is used by FP comparison code */
if (op == TOK_GE || op == TOK_GT)
swapped = !swapped;
else if (op == TOK_EQ || op == TOK_NE)
swapped = 0;
if (swapped)
o(0xc9d9); /* fxch %st(1) */
if (op == TOK_EQ || op == TOK_NE)
o(0xe9da); /* fucompp */
else
o(0xd9de); /* fcompp */
o(0xe0df); /* fnstsw %ax */
if (op == TOK_EQ) {
o(0x45e480); /* and $0x45, %ah */
o(0x40fC80); /* cmp $0x40, %ah */
} else if (op == TOK_NE) {
o(0x45e480); /* and $0x45, %ah */
o(0x40f480); /* xor $0x40, %ah */
op = TOK_NE;
} else if (op == TOK_GE || op == TOK_LE) {
o(0x05c4f6); /* test $0x05, %ah */
op = TOK_EQ;
} else {
o(0x45c4f6); /* test $0x45, %ah */
op = TOK_EQ;
}
vtop--;
vtop->r = VT_CMP;
vtop->c.i = op;
} else {
/* no memory reference possible for long double operations */
if ((vtop->type.t & VT_BTYPE) == VT_LDOUBLE) {
load(TREG_ST0, vtop);
swapped = !swapped;
}
switch(op) {
default:
case '+':
a = 0;
break;
case '-':
a = 4;
if (swapped)
a++;
break;
case '*':
a = 1;
break;
case '/':
a = 6;
if (swapped)
a++;
break;
}
ft = vtop->type.t;
fc = vtop->c.i;
if ((ft & VT_BTYPE) == VT_LDOUBLE) {
o(0xde); /* fxxxp %st, %st(1) */
o(0xc1 + (a << 3));
} else {
/* if saved lvalue, then we must reload it */
r = vtop->r;
if ((r & VT_VALMASK) == VT_LLOCAL) {
SValue v1;
r = get_reg(RC_INT);
v1.type.t = VT_INT;
v1.r = VT_LOCAL | VT_LVAL;
v1.c.i = fc;
load(r, &v1);
fc = 0;
}
if ((ft & VT_BTYPE) == VT_DOUBLE)
o(0xdc);
else
o(0xd8);
gen_modrm(a, r, vtop->sym, fc);
}
vtop--;
}
}
/* convert integers to fp 't' type. Must handle 'int', 'unsigned int'
and 'long long' cases. */
ST_FUNC void gen_cvt_itof(int t)
{
save_reg(TREG_ST0);
gv(RC_INT);
if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
/* signed long long to float/double/long double (unsigned case
is handled generically) */
o(0x50 + vtop->r2); /* push r2 */
o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
o(0x242cdf); /* fildll (%esp) */
o(0x08c483); /* add $8, %esp */
} else if ((vtop->type.t & (VT_BTYPE | VT_UNSIGNED)) ==
(VT_INT | VT_UNSIGNED)) {
/* unsigned int to float/double/long double */
o(0x6a); /* push $0 */
g(0x00);
o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
o(0x242cdf); /* fildll (%esp) */
o(0x08c483); /* add $8, %esp */
} else {
/* int to float/double/long double */
o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
o(0x2404db); /* fildl (%esp) */
o(0x04c483); /* add $4, %esp */
}
vtop->r = TREG_ST0;
}
/* convert fp to int 't' type */
ST_FUNC void gen_cvt_ftoi(int t)
{
#ifndef COMMIT_4ad186c5ef61_IS_FIXED
/* a good version but it takes a more time to execute */
gv(RC_FLOAT);
save_reg(TREG_EAX);
save_reg(TREG_EDX);
gen_static_call(TOK___tcc_cvt_ftol);
vtop->r = TREG_EAX; /* mark reg as used */
if (t == VT_LLONG)
vtop->r2 = TREG_EDX;
#else
/* a new version with a bug: t2a = 44100312 */
/*
#include<stdio.h>
int main() {
int t1 = 176401255;
float f = 0.25f;
int t2a = (int)(t1 * f); // must be 44100313
int t2b = (int)(t1 * (float)0.25f);
printf("t2a=%d t2b=%d \n",t2a,t2b);
return 0;
}
*/
int bt = vtop->type.t & VT_BTYPE;
if (bt == VT_FLOAT)
vpush_global_sym(&func_old_type, TOK___fixsfdi);
else if (bt == VT_LDOUBLE)
vpush_global_sym(&func_old_type, TOK___fixxfdi);
else
vpush_global_sym(&func_old_type, TOK___fixdfdi);
vswap();
gfunc_call(1);
vpushi(0);
vtop->r = REG_IRET;
vtop->r2 = REG_LRET;
#endif
}
/* convert from one floating point type to another */
ST_FUNC void gen_cvt_ftof(int t)
{
/* all we have to do on i386 is to put the float in a register */
gv(RC_FLOAT);
}
/* computed goto support */
ST_FUNC void ggoto(void)
{
gcall_or_jmp(1);
vtop--;
}
/* bound check support functions */
#ifdef CONFIG_TCC_BCHECK
/* generate a bounded pointer addition */
ST_FUNC void gen_bounded_ptr_add(void)
{
/* prepare fast i386 function call (args in eax and edx) */
gv2(RC_EAX, RC_EDX);
/* save all temporary registers */
vtop -= 2;
save_regs(0);
/* do a fast function call */
gen_static_call(TOK___bound_ptr_add);
/* returned pointer is in eax */
vtop++;
vtop->r = TREG_EAX | VT_BOUNDED;
/* address of bounding function call point */
vtop->c.i = (cur_text_section->reloc->data_offset - sizeof(Elf32_Rel));
}
/* patch pointer addition in vtop so that pointer dereferencing is
also tested */
ST_FUNC void gen_bounded_ptr_deref(void)
{
addr_t func;
int size, align;
Elf32_Rel *rel;
Sym *sym;
size = 0;
/* XXX: put that code in generic part of tcc */
if (!is_float(vtop->type.t)) {
if (vtop->r & VT_LVAL_BYTE)
size = 1;
else if (vtop->r & VT_LVAL_SHORT)
size = 2;
}
if (!size)
size = type_size(&vtop->type, &align);
switch(size) {
case 1: func = TOK___bound_ptr_indir1; break;
case 2: func = TOK___bound_ptr_indir2; break;
case 4: func = TOK___bound_ptr_indir4; break;
case 8: func = TOK___bound_ptr_indir8; break;
case 12: func = TOK___bound_ptr_indir12; break;
case 16: func = TOK___bound_ptr_indir16; break;
default:
tcc_error("unhandled size when dereferencing bounded pointer");
func = 0;
break;
}
/* patch relocation */
/* XXX: find a better solution ? */
rel = (Elf32_Rel *)(cur_text_section->reloc->data + vtop->c.i);
sym = external_global_sym(func, &func_old_type, 0);
if (!sym->c)
put_extern_sym(sym, NULL, 0, 0);
rel->r_info = ELF32_R_INFO(sym->c, ELF32_R_TYPE(rel->r_info));
}
#endif
/* Save the stack pointer onto the stack */
ST_FUNC void gen_vla_sp_save(int addr) {
/* mov %esp,addr(%ebp)*/
o(0x89);
gen_modrm(TREG_ESP, VT_LOCAL, NULL, addr);
}
/* Restore the SP from a location on the stack */
ST_FUNC void gen_vla_sp_restore(int addr) {
o(0x8b);
gen_modrm(TREG_ESP, VT_LOCAL, NULL, addr);
}
/* Subtract from the stack pointer, and push the resulting value onto the stack */
ST_FUNC void gen_vla_alloc(CType *type, int align) {
#ifdef TCC_TARGET_PE
/* alloca does more than just adjust %rsp on Windows */
vpush_global_sym(&func_old_type, TOK_alloca);
vswap(); /* Move alloca ref past allocation size */
gfunc_call(1);
#else
int r;
r = gv(RC_INT); /* allocation size */
/* sub r,%rsp */
o(0x2b);
o(0xe0 | r);
/* We align to 16 bytes rather than align */
/* and ~15, %esp */
o(0xf0e483);
vpop();
#endif
}
/* end of X86 code generator */
/*************************************************************/
#endif
/*************************************************************/
Reference in New Issue View Git Blame Copy Permalink