[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Dotgnu-pnet-commits] CVS: pnet/libffi/src/x86 ffi64.c,NONE,1.1 unix64.
|
From: |
Rhys Weatherley <address@hidden> |
|
Subject: |
[Dotgnu-pnet-commits] CVS: pnet/libffi/src/x86 ffi64.c,NONE,1.1 unix64.S,NONE,1.1 ffi.c,1.2,1.3 sysv.S,1.1,1.2 win32.S,1.1,1.2 |
|
Date: |
Wed, 05 Feb 2003 20:15:41 -0500 |
Update of /cvsroot/dotgnu-pnet/pnet/libffi/src/x86
In directory subversions:/tmp/cvs-serv21586/libffi/src/x86
Modified Files:
ffi.c sysv.S win32.S
Added Files:
ffi64.c unix64.S
Log Message:
Update the libffi library from the gcc CVS tree.
--- NEW FILE ---
/* -----------------------------------------------------------------------
ffi.c - Copyright (c) 2002 Bo Thorsen <address@hidden>
x86-64 Foreign Function Interface
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
``Software''), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
The above copyright notice and this permission notice shall be included
in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
IN NO EVENT SHALL CYGNUS SOLUTIONS BE LIABLE FOR ANY CLAIM, DAMAGES OR
OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
OTHER DEALINGS IN THE SOFTWARE.
----------------------------------------------------------------------- */
#include <ffi.h>
#include <ffi_common.h>
#include <stdlib.h>
#include <stdarg.h>
/* ffi_prep_args is called by the assembly routine once stack space
has been allocated for the function's arguments */
#ifdef __x86_64__
#define MAX_GPR_REGS 6
#define MAX_SSE_REGS 8
typedef struct
{
/* Registers for argument passing. */
long gpr[MAX_GPR_REGS];
__int128_t sse[MAX_SSE_REGS];
/* Stack space for arguments. */
char argspace[0];
} stackLayout;
/* All reference to register classes here is identical to the code in
gcc/config/i386/i386.c. Do *not* change one without the other. */
/* Register class used for passing given 64bit part of the argument.
These represent classes as documented by the PS ABI, with the exception
of SSESF, SSEDF classes, that are basically SSE class, just gcc will
use SF or DFmode move instead of DImode to avoid reformating penalties.
Similary we play games with INTEGERSI_CLASS to use cheaper SImode moves
whenever possible (upper half does contain padding).
*/
enum x86_64_reg_class
{
X86_64_NO_CLASS,
X86_64_INTEGER_CLASS,
X86_64_INTEGERSI_CLASS,
X86_64_SSE_CLASS,
X86_64_SSESF_CLASS,
X86_64_SSEDF_CLASS,
X86_64_SSEUP_CLASS,
X86_64_X87_CLASS,
X86_64_X87UP_CLASS,
X86_64_MEMORY_CLASS
};
#define MAX_CLASSES 4
/* x86-64 register passing implementation. See x86-64 ABI for details. Goal
of this code is to classify each 8bytes of incoming argument by the register
class and assign registers accordingly. */
/* Return the union class of CLASS1 and CLASS2.
See the x86-64 PS ABI for details. */
static enum x86_64_reg_class
merge_classes (enum x86_64_reg_class class1, enum x86_64_reg_class class2)
{
/* Rule #1: If both classes are equal, this is the resulting class. */
if (class1 == class2)
return class1;
/* Rule #2: If one of the classes is NO_CLASS, the resulting class is
the other class. */
if (class1 == X86_64_NO_CLASS)
return class2;
if (class2 == X86_64_NO_CLASS)
return class1;
/* Rule #3: If one of the classes is MEMORY, the result is MEMORY. */
if (class1 == X86_64_MEMORY_CLASS || class2 == X86_64_MEMORY_CLASS)
return X86_64_MEMORY_CLASS;
/* Rule #4: If one of the classes is INTEGER, the result is INTEGER. */
if ((class1 == X86_64_INTEGERSI_CLASS && class2 == X86_64_SSESF_CLASS)
|| (class2 == X86_64_INTEGERSI_CLASS && class1 == X86_64_SSESF_CLASS))
return X86_64_INTEGERSI_CLASS;
if (class1 == X86_64_INTEGER_CLASS || class1 == X86_64_INTEGERSI_CLASS
|| class2 == X86_64_INTEGER_CLASS || class2 == X86_64_INTEGERSI_CLASS)
return X86_64_INTEGER_CLASS;
/* Rule #5: If one of the classes is X87 or X87UP class, MEMORY is used. */
if (class1 == X86_64_X87_CLASS || class1 == X86_64_X87UP_CLASS
|| class2 == X86_64_X87_CLASS || class2 == X86_64_X87UP_CLASS)
return X86_64_MEMORY_CLASS;
/* Rule #6: Otherwise class SSE is used. */
return X86_64_SSE_CLASS;
}
/* Classify the argument of type TYPE and mode MODE.
CLASSES will be filled by the register class used to pass each word
of the operand. The number of words is returned. In case the parameter
should be passed in memory, 0 is returned. As a special case for zero
sized containers, classes[0] will be NO_CLASS and 1 is returned.
See the x86-64 PS ABI for details.
*/
static int
classify_argument (ffi_type *type, enum x86_64_reg_class classes[],
int *byte_offset)
{
/* First, align to the right place. */
*byte_offset = ALIGN(*byte_offset, type->alignment);
switch (type->type)
{
case FFI_TYPE_UINT8:
case FFI_TYPE_SINT8:
case FFI_TYPE_UINT16:
case FFI_TYPE_SINT16:
case FFI_TYPE_UINT32:
case FFI_TYPE_SINT32:
case FFI_TYPE_UINT64:
case FFI_TYPE_SINT64:
case FFI_TYPE_POINTER:
if (((*byte_offset) % 8 + type->size) <= 4)
classes[0] = X86_64_INTEGERSI_CLASS;
else
classes[0] = X86_64_INTEGER_CLASS;
return 1;
case FFI_TYPE_FLOAT:
if (((*byte_offset) % 8) == 0)
classes[0] = X86_64_SSESF_CLASS;
else
classes[0] = X86_64_SSE_CLASS;
return 1;
case FFI_TYPE_DOUBLE:
classes[0] = X86_64_SSEDF_CLASS;
return 1;
case FFI_TYPE_LONGDOUBLE:
classes[0] = X86_64_X87_CLASS;
classes[1] = X86_64_X87UP_CLASS;
return 2;
case FFI_TYPE_STRUCT:
{
const int UNITS_PER_WORD = 8;
int words = (type->size + UNITS_PER_WORD - 1) / UNITS_PER_WORD;
ffi_type **ptr;
int i;
enum x86_64_reg_class subclasses[MAX_CLASSES];
/* If the struct is larger than 16 bytes, pass it on the stack. */
if (type->size > 16)
return 0;
for (i = 0; i < words; i++)
classes[i] = X86_64_NO_CLASS;
/* Merge the fields of structure. */
for (ptr=type->elements; (*ptr)!=NULL; ptr++)
{
int num;
num = classify_argument (*ptr, subclasses, byte_offset);
if (num == 0)
return 0;
for (i = 0; i < num; i++)
{
int pos = *byte_offset / 8;
classes[i + pos] =
merge_classes (subclasses[i], classes[i + pos]);
}
if ((*ptr)->type != FFI_TYPE_STRUCT)
*byte_offset += (*ptr)->size;
}
/* Final merger cleanup. */
for (i = 0; i < words; i++)
{
/* If one class is MEMORY, everything should be passed in
memory. */
if (classes[i] == X86_64_MEMORY_CLASS)
return 0;
/* The X86_64_SSEUP_CLASS should be always preceded by
X86_64_SSE_CLASS. */
if (classes[i] == X86_64_SSEUP_CLASS
&& (i == 0 || classes[i - 1] != X86_64_SSE_CLASS))
classes[i] = X86_64_SSE_CLASS;
/* X86_64_X87UP_CLASS should be preceded by X86_64_X87_CLASS. */
if (classes[i] == X86_64_X87UP_CLASS
&& (i == 0 || classes[i - 1] != X86_64_X87_CLASS))
classes[i] = X86_64_SSE_CLASS;
}
return words;
}
default:
FFI_ASSERT(0);
}
return 0; /* Never reached. */
}
/* Examine the argument and return set number of register required in each
class. Return 0 iff parameter should be passed in memory. */
static int
examine_argument (ffi_type *type, int in_return, int *int_nregs,int *sse_nregs)
{
enum x86_64_reg_class class[MAX_CLASSES];
int offset = 0;
int n;
n = classify_argument (type, class, &offset);
if (n == 0)
return 0;
*int_nregs = 0;
*sse_nregs = 0;
for (n--; n>=0; n--)
switch (class[n])
{
case X86_64_INTEGER_CLASS:
case X86_64_INTEGERSI_CLASS:
(*int_nregs)++;
break;
case X86_64_SSE_CLASS:
case X86_64_SSESF_CLASS:
case X86_64_SSEDF_CLASS:
(*sse_nregs)++;
break;
case X86_64_NO_CLASS:
case X86_64_SSEUP_CLASS:
break;
case X86_64_X87_CLASS:
case X86_64_X87UP_CLASS:
if (!in_return)
return 0;
break;
default:
abort ();
}
return 1;
}
/* Functions to load floats and double to an SSE register placeholder. */
extern void float2sse (float, __int128_t *);
extern void double2sse (double, __int128_t *);
extern void floatfloat2sse (void *, __int128_t *);
/* Functions to put the floats and doubles back. */
extern float sse2float (__int128_t *);
extern double sse2double (__int128_t *);
extern void sse2floatfloat(__int128_t *, void *);
/address@hidden@*/
void
ffi_prep_args (stackLayout *stack, extended_cif *ecif)
/address@hidden@*/
{
int gprcount, ssecount, i, g, s;
void **p_argv;
void *argp = &stack->argspace;
ffi_type **p_arg;
/* First check if the return value should be passed in memory. If so,
pass the pointer as the first argument. */
gprcount = ssecount = 0;
if (ecif->cif->rtype->type != FFI_TYPE_VOID
&& examine_argument (ecif->cif->rtype, 1, &g, &s) == 0)
(void *)stack->gpr[gprcount++] = ecif->rvalue;
for (i=ecif->cif->nargs, p_arg=ecif->cif->arg_types, p_argv = ecif->avalue;
i!=0; i--, p_arg++, p_argv++)
{
int in_register = 0;
switch ((*p_arg)->type)
{
case FFI_TYPE_SINT8:
case FFI_TYPE_SINT16:
case FFI_TYPE_SINT32:
case FFI_TYPE_SINT64:
case FFI_TYPE_UINT8:
case FFI_TYPE_UINT16:
case FFI_TYPE_UINT32:
case FFI_TYPE_UINT64:
case FFI_TYPE_POINTER:
if (gprcount < MAX_GPR_REGS)
{
stack->gpr[gprcount] = 0;
stack->gpr[gprcount++] = *(long long *)(*p_argv);
in_register = 1;
}
break;
case FFI_TYPE_FLOAT:
if (ssecount < MAX_SSE_REGS)
{
float2sse (*(float *)(*p_argv), &stack->sse[ssecount++]);
in_register = 1;
}
break;
case FFI_TYPE_DOUBLE:
if (ssecount < MAX_SSE_REGS)
{
double2sse (*(double *)(*p_argv), &stack->sse[ssecount++]);
in_register = 1;
}
break;
}
if (in_register)
continue;
/* Either all places in registers where filled, or this is a
type that potentially goes into a memory slot. */
if (examine_argument (*p_arg, 0, &g, &s) == 0
|| gprcount + g > MAX_GPR_REGS || ssecount + s > MAX_SSE_REGS)
{
/* Pass this argument in memory. */
argp = (void *)ALIGN(argp, (*p_arg)->alignment);
memcpy (argp, *p_argv, (*p_arg)->size);
argp += (*p_arg)->size;
}
else
{
/* All easy cases are eliminated. Now fire the big guns. */
enum x86_64_reg_class classes[MAX_CLASSES];
int offset = 0, j, num;
void *a;
num = classify_argument (*p_arg, classes, &offset);
for (j=0, a=*p_argv; j<num; j++, a+=8)
{
switch (classes[j])
{
case X86_64_INTEGER_CLASS:
case X86_64_INTEGERSI_CLASS:
stack->gpr[gprcount++] = *(long long *)a;
break;
case X86_64_SSE_CLASS:
floatfloat2sse (a, &stack->sse[ssecount++]);
break;
case X86_64_SSESF_CLASS:
float2sse (*(float *)a, &stack->sse[ssecount++]);
break;
case X86_64_SSEDF_CLASS:
double2sse (*(double *)a, &stack->sse[ssecount++]);
break;
default:
abort();
}
}
}
}
}
/* Perform machine dependent cif processing. */
ffi_status
ffi_prep_cif_machdep (ffi_cif *cif)
{
int gprcount, ssecount, i, g, s;
gprcount = ssecount = 0;
/* Reset the byte count. We handle this size estimation here. */
cif->bytes = 0;
/* If the return value should be passed in memory, pass the pointer
as the first argument. The actual memory isn't allocated here. */
if (cif->rtype->type != FFI_TYPE_VOID
&& examine_argument (cif->rtype, 1, &g, &s) == 0)
gprcount = 1;
/* Go over all arguments and determine the way they should be passed.
If it's in a register and there is space for it, let that be so. If
not, add it's size to the stack byte count. */
for (i=0; i<cif->nargs; i++)
{
if (examine_argument (cif->arg_types[i], 0, &g, &s) == 0
|| gprcount + g > MAX_GPR_REGS || ssecount + s > MAX_SSE_REGS)
{
/* This is passed in memory. First align to the basic type. */
cif->bytes = ALIGN(cif->bytes, cif->arg_types[i]->alignment);
/* Stack arguments are *always* at least 8 byte aligned. */
cif->bytes = ALIGN(cif->bytes, 8);
/* Now add the size of this argument. */
cif->bytes += cif->arg_types[i]->size;
}
else
{
gprcount += g;
ssecount += s;
}
}
/* Set the flag for the closures return. */
switch (cif->rtype->type)
{
case FFI_TYPE_VOID:
case FFI_TYPE_STRUCT:
case FFI_TYPE_SINT64:
case FFI_TYPE_FLOAT:
case FFI_TYPE_DOUBLE:
case FFI_TYPE_LONGDOUBLE:
cif->flags = (unsigned) cif->rtype->type;
break;
case FFI_TYPE_UINT64:
cif->flags = FFI_TYPE_SINT64;
break;
default:
cif->flags = FFI_TYPE_INT;
break;
}
return FFI_OK;
}
typedef struct
{
long gpr[2];
__int128_t sse[2];
long double st0;
} return_value;
void
ffi_fill_return_value (return_value *rv, extended_cif *ecif)
{
enum x86_64_reg_class classes[MAX_CLASSES];
int i = 0, num;
long *gpr = rv->gpr;
__int128_t *sse = rv->sse;
signed char sc;
signed short ss;
/* This is needed because of the way x86-64 handles signed short
integers. */
switch (ecif->cif->rtype->type)
{
case FFI_TYPE_SINT8:
sc = *(signed char *)gpr;
*(long long *)ecif->rvalue = (long long)sc;
return;
case FFI_TYPE_SINT16:
ss = *(signed short *)gpr;
*(long long *)ecif->rvalue = (long long)ss;
return;
default:
/* Just continue. */
;
}
num = classify_argument (ecif->cif->rtype, classes, &i);
if (num == 0)
/* Return in memory. */
ecif->rvalue = (void *) rv->gpr[0];
else if (num == 2 && classes[0] == X86_64_X87_CLASS &&
classes[1] == X86_64_X87UP_CLASS)
/* This is a long double (this is easiest to handle this way instead
of an eightbyte at a time as in the loop below. */
*((long double *)ecif->rvalue) = rv->st0;
else
{
void *a;
for (i=0, a=ecif->rvalue; i<num; i++, a+=8)
{
switch (classes[i])
{
case X86_64_INTEGER_CLASS:
case X86_64_INTEGERSI_CLASS:
*(long long *)a = *gpr;
gpr++;
break;
case X86_64_SSE_CLASS:
sse2floatfloat (sse++, a);
break;
case X86_64_SSESF_CLASS:
*(float *)a = sse2float (sse++);
break;
case X86_64_SSEDF_CLASS:
*(double *)a = sse2double (sse++);
break;
default:
abort();
}
}
}
}
/address@hidden@*/
/address@hidden@*/
extern void ffi_call_UNIX64(void (*)(stackLayout *, extended_cif *),
void (*) (return_value *, extended_cif *),
/address@hidden@*/ extended_cif *,
unsigned, /address@hidden@*/ unsigned *, void
(*fn)());
/address@hidden@*/
/address@hidden@*/
void ffi_call(/address@hidden@*/ ffi_cif *cif,
void (*fn)(),
/address@hidden@*/ void *rvalue,
/address@hidden@*/ void **avalue)
{
extended_cif ecif;
int dummy;
ecif.cif = cif;
ecif.avalue = avalue;
/* If the return value is a struct and we don't have a return */
/* value address then we need to make one */
if ((rvalue == NULL) &&
(examine_argument (cif->rtype, 1, &dummy, &dummy) == 0))
{
/address@hidden@*/
ecif.rvalue = alloca(cif->rtype->size);
/address@hidden@*/
}
else
ecif.rvalue = rvalue;
/* Stack must always be 16byte aligned. Make it so. */
cif->bytes = ALIGN(cif->bytes, 16);
switch (cif->abi)
{
case FFI_SYSV:
/* Calling 32bit code from 64bit is not possible */
FFI_ASSERT(0);
break;
case FFI_UNIX64:
/address@hidden@*/
ffi_call_UNIX64 (ffi_prep_args, ffi_fill_return_value, &ecif,
cif->bytes, ecif.rvalue, fn);
/address@hidden@*/
break;
default:
FFI_ASSERT(0);
break;
}
}
extern void ffi_closure_UNIX64(void);
ffi_status
ffi_prep_closure (ffi_closure* closure,
ffi_cif* cif,
void (*fun)(ffi_cif*, void*, void**, void*),
void *user_data)
{
volatile unsigned short *tramp;
/* FFI_ASSERT (cif->abi == FFI_OSF); */
tramp = (volatile unsigned short *) &closure->tramp[0];
tramp[0] = 0xbb49; /* mov <code>, %r11 */
tramp[5] = 0xba49; /* mov <data>, %r10 */
tramp[10] = 0xff49; /* jmp *%r11 */
tramp[11] = 0x00e3;
*(void * volatile *) &tramp[1] = ffi_closure_UNIX64;
*(void * volatile *) &tramp[6] = closure;
closure->cif = cif;
closure->fun = fun;
closure->user_data = user_data;
return FFI_OK;
}
int
ffi_closure_UNIX64_inner(ffi_closure *closure, va_list l, void *rp)
{
ffi_cif *cif;
void **avalue;
ffi_type **arg_types;
long i, avn, argn;
cif = closure->cif;
avalue = alloca(cif->nargs * sizeof(void *));
argn = 0;
i = 0;
avn = cif->nargs;
arg_types = cif->arg_types;
/* Grab the addresses of the arguments from the stack frame. */
while (i < avn)
{
switch (arg_types[i]->type)
{
case FFI_TYPE_SINT8:
case FFI_TYPE_UINT8:
case FFI_TYPE_SINT16:
case FFI_TYPE_UINT16:
case FFI_TYPE_SINT32:
case FFI_TYPE_UINT32:
case FFI_TYPE_SINT64:
case FFI_TYPE_UINT64:
case FFI_TYPE_POINTER:
{
if (l->gp_offset > 48-8)
{
avalue[i] = l->overflow_arg_area;
l->overflow_arg_area = (char *)l->overflow_arg_area + 8;
}
else
{
avalue[i] = (char *)l->reg_save_area + l->gp_offset;
l->gp_offset += 8;
}
}
break;
case FFI_TYPE_STRUCT:
/* FIXME */
FFI_ASSERT(0);
break;
case FFI_TYPE_DOUBLE:
{
if (l->fp_offset > 176-16)
{
avalue[i] = l->overflow_arg_area;
l->overflow_arg_area = (char *)l->overflow_arg_area + 8;
}
else
{
avalue[i] = (char *)l->reg_save_area + l->fp_offset;
l->fp_offset += 16;
}
}
#if DEBUG_FFI
fprintf (stderr, "double arg %d = %g\n", i, *(double *)avalue[i]);
#endif
break;
case FFI_TYPE_FLOAT:
{
if (l->fp_offset > 176-16)
{
avalue[i] = l->overflow_arg_area;
l->overflow_arg_area = (char *)l->overflow_arg_area + 8;
}
else
{
avalue[i] = (char *)l->reg_save_area + l->fp_offset;
l->fp_offset += 16;
}
}
#if DEBUG_FFI
fprintf (stderr, "float arg %d = %g\n", i, *(float *)avalue[i]);
#endif
break;
default:
FFI_ASSERT(0);
}
argn += ALIGN(arg_types[i]->size, SIZEOF_ARG) / SIZEOF_ARG;
i++;
}
/* Invoke the closure. */
(closure->fun) (cif, rp, avalue, closure->user_data);
/* FIXME: Structs not supported. */
FFI_ASSERT(cif->rtype->type != FFI_TYPE_STRUCT);
/* Tell ffi_closure_UNIX64 how to perform return type promotions. */
return cif->rtype->type;
}
#endif /* ifndef __x86_64__ */
--- NEW FILE ---
/* -----------------------------------------------------------------------
unix64.S - Copyright (c) 2002 Bo Thorsen <address@hidden>
x86-64 Foreign Function Interface
Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
``Software''), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:
The above copyright notice and this permission notice shall be included
in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED ``AS IS'', WITHOUT WARRANTY OF ANY KIND, EXPRESS
OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
IN NO EVENT SHALL CYGNUS SOLUTIONS BE LIABLE FOR ANY CLAIM, DAMAGES OR
OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
OTHER DEALINGS IN THE SOFTWARE.
----------------------------------------------------------------------- */
#ifdef __x86_64__
#define LIBFFI_ASM
#include <ffi.h>
.section .rodata
.LC0:
.string "asm in progress %lld\n"
.LC1:
.string "asm in progress\n"
.text
.align 2
.globl ffi_call_UNIX64
.type ffi_call_UNIX64,@function
ffi_call_UNIX64:
.LFB1:
pushq %rbp
.LCFI0:
movq %rsp, %rbp
.LCFI1:
/* Save all arguments */
subq $48, %rsp
.LCFI2:
movq %rdi, -8(%rbp) /* ffi_prep_args */
movq %rsi, -16(%rbp) /* ffi_fill_return_value */
movq %rdx, -24(%rbp) /* ecif */
movq %rcx, -32(%rbp) /* cif->bytes */
movq %r8, -40(%rbp) /* ecif.rvalue */
movq %r9, -48(%rbp) /* fn */
/* Make room for all of the new args and the register args */
addl $176, %ecx
.LCFI3:
subq %rcx, %rsp
.LCFI4:
/* Setup the call to ffi_prep_args. */
movq %rdi, %rax /* &ffi_prep_args */
movq %rsp, %rdi /* stackLayout */
movq %rdx, %rsi /* ecif */
call *%rax /* ffi_prep_args(stackLayout, ecif);*/
/* ffi_prep_args have put all the register contents into the */
/* stackLayout struct. Now put the register values in place. */
movq (%rsp), %rdi
movq 8(%rsp), %rsi
movq 16(%rsp), %rdx
movq 24(%rsp), %rcx
movq 32(%rsp), %r8
movq 40(%rsp), %r9
movaps 48(%rsp), %xmm0
movaps 64(%rsp), %xmm1
movaps 80(%rsp), %xmm2
movaps 96(%rsp), %xmm3
movaps 112(%rsp), %xmm4
movaps 128(%rsp), %xmm5
movaps 144(%rsp), %xmm6
movaps 160(%rsp), %xmm7
/* Remove space for stackLayout so stack arguments are placed
correctly for the call. */
.LCFI5:
addq $176, %rsp
.LCFI6:
/* Call the user function. */
call *-48(%rbp)
/* Make stack space for the return_value struct. */
subq $64, %rsp
/* Fill in all potential return values to this struct. */
movq %rax, (%rsp)
movq %rdx, 8(%rsp)
movaps %xmm0, 16(%rsp)
movaps %xmm1, 32(%rsp)
fstpt 48(%rsp)
/* Now call ffi_fill_return_value. */
movq %rsp, %rdi /* struct return_value */
movq -24(%rbp), %rsi /* ecif */
movq -16(%rbp), %rax /* &ffi_fill_return_value */
call *%rax /* call it */
/* And the work is done. */
leave
ret
.LFE1:
.ffi_call_UNIX64_end:
.size ffi_call_UNIX64,.ffi_call_UNIX64_end-ffi_call_UNIX64
.text
.align 2
.globl float2sse
.type float2sse,@function
float2sse:
/* Save the contents of this sse-float in a pointer. */
movaps %xmm0, (%rdi)
ret
.align 2
.globl floatfloat2sse
.type floatfloat2sse,@function
floatfloat2sse:
/* Save the contents of these two sse-floats in a pointer. */
movq (%rdi), %xmm0
movaps %xmm0, (%rsi)
ret
.align 2
.globl double2sse
.type double2sse,@function
double2sse:
/* Save the contents of this sse-double in a pointer. */
movaps %xmm0, (%rdi)
ret
.align 2
.globl sse2float
.type sse2float,@function
sse2float:
/* Save the contents of this sse-float in a pointer. */
movaps (%rdi), %xmm0
ret
.align 2
.globl sse2double
.type sse2double,@function
sse2double:
/* Save the contents of this pointer in a sse-double. */
movaps (%rdi), %xmm0
ret
.align 2
.globl sse2floatfloat
.type sse2floatfloat,@function
sse2floatfloat:
/* Save the contents of this pointer in two sse-floats. */
movaps (%rdi), %xmm0
movq %xmm0, (%rsi)
ret
.align 2
.globl ffi_closure_UNIX64
.type ffi_closure_UNIX64,@function
ffi_closure_UNIX64:
.LFB2:
pushq %rbp
.LCFI10:
movq %rsp, %rbp
.LCFI11:
subq $240, %rsp
.LCFI12:
movq %rdi, -176(%rbp)
movq %rsi, -168(%rbp)
movq %rdx, -160(%rbp)
movq %rcx, -152(%rbp)
movq %r8, -144(%rbp)
movq %r9, -136(%rbp)
/* FIXME: We can avoid all this stashing of XMM registers by
(in ffi_prep_closure) computing the number of
floating-point args and moving it into %rax before calling
this function. Once this is done, uncomment the next few
lines and only the essential XMM registers will be written
to memory. This is a significant saving. */
/* movzbl %al, %eax */
/* movq %rax, %rdx */
/* leaq 0(,%rdx,4), %rax */
/* leaq 2f(%rip), %rdx */
/* subq %rax, %rdx */
leaq -1(%rbp), %rax
/* jmp *%rdx */
movaps %xmm7, -15(%rax)
movaps %xmm6, -31(%rax)
movaps %xmm5, -47(%rax)
movaps %xmm4, -63(%rax)
movaps %xmm3, -79(%rax)
movaps %xmm2, -95(%rax)
movaps %xmm1, -111(%rax)
movaps %xmm0, -127(%rax)
2:
movl %edi, -180(%rbp)
movl $0, -224(%rbp)
movl $48, -220(%rbp)
leaq 16(%rbp), %rax
movq %rax, -216(%rbp)
leaq -176(%rbp), %rdx
movq %rdx, -208(%rbp)
leaq -224(%rbp), %rsi
movq %r10, %rdi
movq %rsp, %rdx
call address@hidden
cmpl $FFI_TYPE_FLOAT, %eax
je 1f
cmpl $FFI_TYPE_DOUBLE, %eax
je 2f
cmpl $FFI_TYPE_LONGDOUBLE, %eax
je 3f
cmpl $FFI_TYPE_STRUCT, %eax
je 4f
popq %rax
leave
ret
1:
2:
3:
movaps -240(%rbp), %xmm0
leave
ret
4:
leave
ret
.LFE2:
.section .eh_frame,"a",@progbits
.Lframe0:
.long .LECIE1-.LSCIE1
.LSCIE1:
.long 0x0
.byte 0x1
.string "zR"
.uleb128 0x1
.sleb128 -8
.byte 0x10
.uleb128 0x1
.byte 0x1b
.byte 0xc
.uleb128 0x7
.uleb128 0x8
.byte 0x90
.uleb128 0x1
.align 8
.LECIE1:
.LSFDE1:
.long .LEFDE1-.LASFDE1
.LASFDE1:
.long .LASFDE1-.Lframe0
.long .LFB1-.
.long .LFE1-.LFB1
.uleb128 0x0
.byte 0x4 # DW_CFA_advance_loc4
.long .LCFI0-.LFB1
.byte 0xe # DW_CFA_def_cfa_offset
.uleb128 0x10
.byte 0x86 # DW_CFA_offset: r6 at cfa-16
.uleb128 0x2
.byte 0x4 # DW_CFA_advance_loc4
.long .LCFI1-.LCFI0
.byte 0x86 # DW_CFA_offset: r6 at cfa-16
.uleb128 0x2
.byte 0xd # DW_CFA_def_cfa_reg: r6
.uleb128 0x6
.align 8
.LEFDE1:
.LSFDE3:
.long .LEFDE3-.LASFDE3 # FDE Length
.LASFDE3:
.long .LASFDE3-.Lframe0 # FDE CIE offset
.long .LFB2-. # FDE initial location
.long .LFE2-.LFB2 # FDE address range
.uleb128 0x0 # Augmentation size
.byte 0x4 # DW_CFA_advance_loc4
.long .LCFI10-.LFB2
.byte 0xe # DW_CFA_def_cfa_offset
.uleb128 0x10
.byte 0x86 # DW_CFA_offset, column 0x6
.uleb128 0x2
.byte 0x4 # DW_CFA_advance_loc4
.long .LCFI11-.LCFI10
.byte 0xd # DW_CFA_def_cfa_register
.uleb128 0x6
.align 8
.LEFDE3:
#endif /* __x86_64__ */
Index: ffi.c
===================================================================
RCS file: /cvsroot/dotgnu-pnet/pnet/libffi/src/x86/ffi.c,v
retrieving revision 1.2
retrieving revision 1.3
diff -C2 -r1.2 -r1.3
*** ffi.c 11 Jun 2002 01:57:18 -0000 1.2
--- ffi.c 6 Feb 2003 01:15:39 -0000 1.3
***************
*** 1,4 ****
/* -----------------------------------------------------------------------
! ffi.c - Copyright (c) 1996, 1998, 1999 Cygnus Solutions
x86 Foreign Function Interface
--- 1,7 ----
/* -----------------------------------------------------------------------
! ffi.c - Copyright (c) 1996, 1998, 1999, 2001 Red Hat, Inc.
! Copyright (c) 2002 Ranjit Mathew
! Copyright (c) 2002 Bo Thorsen
! Copyright (c) 2002 Roger Sayle
x86 Foreign Function Interface
***************
*** 24,27 ****
--- 27,32 ----
----------------------------------------------------------------------- */
+ #ifndef __x86_64__
+
#include <ffi.h>
#include <ffi_common.h>
***************
*** 37,46 ****
{
register unsigned int i;
- register int tmp;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
- tmp = 0;
argp = stack;
--- 42,49 ----
***************
*** 149,152 ****
--- 152,167 ----
/address@hidden@*/
+ #ifdef X86_WIN32
+ /address@hidden@*/
+ /address@hidden@*/
+ extern void ffi_call_STDCALL(void (*)(char *, extended_cif *),
+ /address@hidden@*/ extended_cif *,
+ unsigned, unsigned,
+ /address@hidden@*/ unsigned *,
+ void (*fn)());
+ /address@hidden@*/
+ /address@hidden@*/
+ #endif /* X86_WIN32 */
+
void ffi_call(/address@hidden@*/ ffi_cif *cif,
void (*fn)(),
***************
*** 181,184 ****
--- 196,207 ----
/address@hidden@*/
break;
+ #ifdef X86_WIN32
+ case FFI_STDCALL:
+ /address@hidden@*/
+ ffi_call_STDCALL(ffi_prep_args, &ecif, cif->bytes,
+ cif->flags, ecif.rvalue, fn);
+ /address@hidden@*/
+ break;
+ #endif /* X86_WIN32 */
default:
FFI_ASSERT(0);
***************
*** 267,276 ****
{
register unsigned int i;
- register int tmp;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
- tmp = 0;
argp = stack;
--- 290,297 ----
***************
*** 451,454 ****
--- 472,484 ----
void (*fn)());
+ #ifdef X86_WIN32
+ extern void
+ ffi_call_STDCALL(void (*)(char *, extended_cif *),
+ /address@hidden@*/ extended_cif *,
+ unsigned, unsigned,
+ /address@hidden@*/ unsigned *,
+ void (*fn)());
+ #endif /* X86_WIN32 */
+
void
ffi_raw_call(/address@hidden@*/ ffi_cif *cif,
***************
*** 485,488 ****
--- 515,526 ----
/address@hidden@*/
break;
+ #ifdef X86_WIN32
+ case FFI_STDCALL:
+ /address@hidden@*/
+ ffi_call_STDCALL(ffi_prep_args_raw, &ecif, cif->bytes,
+ cif->flags, ecif.rvalue, fn);
+ /address@hidden@*/
+ break;
+ #endif /* X86_WIN32 */
default:
FFI_ASSERT(0);
***************
*** 492,493 ****
--- 530,533 ----
#endif
+
+ #endif /* __x86_64__ */
Index: sysv.S
===================================================================
RCS file: /cvsroot/dotgnu-pnet/pnet/libffi/src/x86/sysv.S,v
retrieving revision 1.1
retrieving revision 1.2
diff -C2 -r1.1 -r1.2
*** sysv.S 10 May 2002 23:04:40 -0000 1.1
--- sysv.S 6 Feb 2003 01:15:39 -0000 1.2
***************
*** 1,4 ****
/* -----------------------------------------------------------------------
! sysv.S - Copyright (c) 1996, 1998, 2001 Cygnus Solutions
X86 Foreign Function Interface
--- 1,4 ----
/* -----------------------------------------------------------------------
! sysv.S - Copyright (c) 1996, 1998, 2001, 2002 Cygnus Solutions
X86 Foreign Function Interface
***************
*** 24,27 ****
--- 24,29 ----
----------------------------------------------------------------------- */
+ #ifndef __x86_64__
+
#define LIBFFI_ASM
#include <ffi.h>
***************
*** 164,165 ****
--- 166,169 ----
.LEFDE1:
.set .LLFDE1,.LEFDE1-.LSFDE1
+
+ #endif /* ifndef __x86_64__ */
Index: win32.S
===================================================================
RCS file: /cvsroot/dotgnu-pnet/pnet/libffi/src/x86/win32.S,v
retrieving revision 1.1
retrieving revision 1.2
diff -C2 -r1.1 -r1.2
*** win32.S 10 May 2002 23:04:40 -0000 1.1
--- win32.S 6 Feb 2003 01:15:39 -0000 1.2
***************
*** 1,4 ****
/* -----------------------------------------------------------------------
! win32.S - Copyright (c) 1996, 1998, 2001 Cygnus Solutions
X86 Foreign Function Interface
--- 1,7 ----
/* -----------------------------------------------------------------------
! win32.S - Copyright (c) 1996, 1998, 2001, 2002 Red Hat, Inc.
! Copyright (c) 2001 John Beniton
! Copyright (c) 2002 Ranjit Mathew
!
X86 Foreign Function Interface
***************
*** 53,57 ****
addl $8,%esp
! call *28(%ebp)
# Remove the space we pushed for the args
--- 56,63 ----
addl $8,%esp
! # FIXME: Align the stack to a 128-bit boundary to avoid
! # potential performance hits.
!
! call *28(%ebp)
# Remove the space we pushed for the args
***************
*** 124,125 ****
--- 130,226 ----
.ffi_call_SYSV_end:
+
+ # This assumes we are using gas.
+ .balign 16
+ .globl _ffi_call_STDCALL
+
+ _ffi_call_STDCALL:
+ pushl %ebp
+ movl %esp,%ebp
+
+ # Make room for all of the new args.
+ movl 16(%ebp),%ecx
+ subl %ecx,%esp
+
+ movl %esp,%eax
+
+ # Place all of the ffi_prep_args in position
+ pushl 12(%ebp)
+ pushl %eax
+ call *8(%ebp)
+
+ # Return stack to previous state and call the function
+ addl $8,%esp
+
+ # FIXME: Align the stack to a 128-bit boundary to avoid
+ # potential performance hits.
+
+ call *28(%ebp)
+
+ # stdcall functions pop arguments off the stack themselves
+
+ # Load %ecx with the return type code
+ movl 20(%ebp),%ecx
+
+ # If the return value pointer is NULL, assume no return value.
+ cmpl $0,24(%ebp)
+ jne sc_retint
+
+ # Even if there is no space for the return value, we are
+ # obliged to handle floating-point values.
+ cmpl $FFI_TYPE_FLOAT,%ecx
+ jne sc_noretval
+ fstp %st(0)
+
+ jmp sc_epilogue
+
+ sc_retint:
+ cmpl $FFI_TYPE_INT,%ecx
+ jne sc_retfloat
+ # Load %ecx with the pointer to storage for the return value
+ movl 24(%ebp),%ecx
+ movl %eax,0(%ecx)
+ jmp sc_epilogue
+
+ sc_retfloat:
+ cmpl $FFI_TYPE_FLOAT,%ecx
+ jne sc_retdouble
+ # Load %ecx with the pointer to storage for the return value
+ movl 24(%ebp),%ecx
+ fstps (%ecx)
+ jmp sc_epilogue
+
+ sc_retdouble:
+ cmpl $FFI_TYPE_DOUBLE,%ecx
+ jne sc_retlongdouble
+ # Load %ecx with the pointer to storage for the return value
+ movl 24(%ebp),%ecx
+ fstpl (%ecx)
+ jmp sc_epilogue
+
+ sc_retlongdouble:
+ cmpl $FFI_TYPE_LONGDOUBLE,%ecx
+ jne sc_retint64
+ # Load %ecx with the pointer to storage for the return value
+ movl 24(%ebp),%ecx
+ fstpt (%ecx)
+ jmp sc_epilogue
+
+ sc_retint64:
+ cmpl $FFI_TYPE_SINT64,%ecx
+ jne sc_retstruct
+ # Load %ecx with the pointer to storage for the return value
+ movl 24(%ebp),%ecx
+ movl %eax,0(%ecx)
+ movl %edx,4(%ecx)
+
+ sc_retstruct:
+ # Nothing to do!
+
+ sc_noretval:
+ sc_epilogue:
+ movl %ebp,%esp
+ popl %ebp
+ ret
+
+ .ffi_call_STDCALL_end:
| [Prev in Thread] |
Current Thread |
[Next in Thread] |
- [Dotgnu-pnet-commits] CVS: pnet/libffi/src/x86 ffi64.c,NONE,1.1 unix64.S,NONE,1.1 ffi.c,1.2,1.3 sysv.S,1.1,1.2 win32.S,1.1,1.2,
Rhys Weatherley <address@hidden> <=