ag/cpython-source-deps
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix-comments patch

Browse Source
This commit is contained in:
Anthony Green
2010-03-15 05:57:24 -04:00
parent c3813b6d7f
commit 3de1eb36d3
11 changed files with 6161 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.pc/applied-patches
View File

@@ -2,3 +2,4 @@ stand-alone
closure-api-example-doc
sunstudio
msvc
fix-comments

0
.pc/fix-comments/.timestamp Normal file
View File

4140
.pc/fix-comments/ChangeLog Normal file
View File

File diff suppressed because it is too large Load Diff

665
.pc/fix-comments/src/x86/ffi.c Normal file
View File

@@ -0,0 +1,665 @@
/* -----------------------------------------------------------------------
ffi.c - Copyright (c) 1996, 1998, 1999, 2001, 2007, 2008 Red Hat, Inc.
Copyright (c) 2002 Ranjit Mathew
Copyright (c) 2002 Bo Thorsen
Copyright (c) 2002 Roger Sayle
Copyright (C) 2008 Free Software Foundation, Inc.
x86 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 THE AUTHORS OR COPYRIGHT
HOLDERS 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.
----------------------------------------------------------------------- */
#if !defined(__x86_64__) || defined(_WIN64)
#ifdef _WIN64
#include <windows.h>
#endif
#include <ffi.h>
#include <ffi_common.h>
#include <stdlib.h>
/* ffi_prep_args is called by the assembly routine once stack space
has been allocated for the function's arguments */
void ffi_prep_args(char *stack, extended_cif *ecif)
{
register unsigned int i;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
argp = stack;
if (ecif->cif->flags == FFI_TYPE_STRUCT
#ifdef X86_WIN64
&& (ecif->cif->rtype->size != 1 && ecif->cif->rtype->size != 2
&& ecif->cif->rtype->size != 4 && ecif->cif->rtype->size != 8)
#endif
)
{
*(void **) argp = ecif->rvalue;
argp += sizeof(void*);
}
p_argv = ecif->avalue;
for (i = ecif->cif->nargs, p_arg = ecif->cif->arg_types;
i != 0;
i--, p_arg++)
{
size_t z;
/* Align if necessary */
if ((sizeof(void*) - 1) & (size_t) argp)
argp = (char *) ALIGN(argp, sizeof(void*));
z = (*p_arg)->size;
#ifdef X86_WIN64
if (z > sizeof(ffi_arg)
|| ((*p_arg)->type == FFI_TYPE_STRUCT
&& (z != 1 && z != 2 && z != 4 && z != 8))
#if FFI_TYPE_DOUBLE != FFI_TYPE_LONGDOUBLE
|| ((*p_arg)->type == FFI_TYPE_LONGDOUBLE)
#endif
)
{
z = sizeof(ffi_arg);
*(void **)argp = *p_argv;
}
else if ((*p_arg)->type == FFI_TYPE_FLOAT)
{
memcpy(argp, *p_argv, z);
}
else
#endif
if (z < sizeof(ffi_arg))
{
z = sizeof(ffi_arg);
switch ((*p_arg)->type)
{
case FFI_TYPE_SINT8:
*(ffi_sarg *) argp = (ffi_sarg)*(SINT8 *)(* p_argv);
break;
case FFI_TYPE_UINT8:
*(ffi_arg *) argp = (ffi_arg)*(UINT8 *)(* p_argv);
break;
case FFI_TYPE_SINT16:
*(ffi_sarg *) argp = (ffi_sarg)*(SINT16 *)(* p_argv);
break;
case FFI_TYPE_UINT16:
*(ffi_arg *) argp = (ffi_arg)*(UINT16 *)(* p_argv);
break;
case FFI_TYPE_SINT32:
*(ffi_sarg *) argp = (ffi_sarg)*(SINT32 *)(* p_argv);
break;
case FFI_TYPE_UINT32:
*(ffi_arg *) argp = (ffi_arg)*(UINT32 *)(* p_argv);
break;
case FFI_TYPE_STRUCT:
*(ffi_arg *) argp = *(ffi_arg *)(* p_argv);
break;
default:
FFI_ASSERT(0);
}
}
else
{
memcpy(argp, *p_argv, z);
}
p_argv++;
#ifdef X86_WIN64
argp += (z + sizeof(void*) - 1) & ~(sizeof(void*) - 1);
#else
argp += z;
#endif
}
return;
}
/* Perform machine dependent cif processing */
ffi_status ffi_prep_cif_machdep(ffi_cif *cif)
{
unsigned int i;
ffi_type **ptr;
/* Set the return type flag */
switch (cif->rtype->type)
{
case FFI_TYPE_VOID:
#if defined(X86) || defined (X86_WIN32) || defined(X86_FREEBSD) || defined(X86_DARWIN) || defined(X86_WIN64)
case FFI_TYPE_UINT8:
case FFI_TYPE_UINT16:
case FFI_TYPE_SINT8:
case FFI_TYPE_SINT16:
#endif
#ifdef X86_WIN64
case FFI_TYPE_UINT32:
case FFI_TYPE_SINT32:
#endif
case FFI_TYPE_SINT64:
case FFI_TYPE_FLOAT:
case FFI_TYPE_DOUBLE:
#ifndef X86_WIN64
#if FFI_TYPE_DOUBLE != FFI_TYPE_LONGDOUBLE
case FFI_TYPE_LONGDOUBLE:
#endif
#endif
cif->flags = (unsigned) cif->rtype->type;
break;
case FFI_TYPE_UINT64:
#ifdef X86_WIN64
case FFI_TYPE_POINTER:
#endif
cif->flags = FFI_TYPE_SINT64;
break;
case FFI_TYPE_STRUCT:
#ifndef X86
if (cif->rtype->size == 1)
{
cif->flags = FFI_TYPE_SMALL_STRUCT_1B; /* same as char size */
}
else if (cif->rtype->size == 2)
{
cif->flags = FFI_TYPE_SMALL_STRUCT_2B; /* same as short size */
}
else if (cif->rtype->size == 4)
{
#ifdef X86_WIN64
cif->flags = FFI_TYPE_SMALL_STRUCT_4B;
#else
cif->flags = FFI_TYPE_INT; /* same as int type */
#endif
}
else if (cif->rtype->size == 8)
{
cif->flags = FFI_TYPE_SINT64; /* same as int64 type */
}
else
#endif
{
cif->flags = FFI_TYPE_STRUCT;
// allocate space for return value pointer
cif->bytes += ALIGN(sizeof(void*), FFI_SIZEOF_ARG);
}
break;
default:
#ifdef X86_WIN64
cif->flags = FFI_TYPE_SINT64;
break;
case FFI_TYPE_INT:
cif->flags = FFI_TYPE_SINT32;
#else
cif->flags = FFI_TYPE_INT;
#endif
break;
}
for (ptr = cif->arg_types, i = cif->nargs; i > 0; i--, ptr++)
{
if (((*ptr)->alignment - 1) & cif->bytes)
cif->bytes = ALIGN(cif->bytes, (*ptr)->alignment);
cif->bytes += ALIGN((*ptr)->size, FFI_SIZEOF_ARG);
}
#ifdef X86_WIN64
// ensure space for storing four registers
cif->bytes += 4 * sizeof(ffi_arg);
#endif
#ifdef X86_DARWIN
cif->bytes = (cif->bytes + 15) & ~0xF;
#endif
return FFI_OK;
}
#ifdef X86_WIN64
extern int
ffi_call_win64(void (*)(char *, extended_cif *), extended_cif *,
unsigned, unsigned, unsigned *, void (*fn)(void));
#elif defined(X86_WIN32)
extern void
ffi_call_win32(void (*)(char *, extended_cif *), extended_cif *,
unsigned, unsigned, unsigned *, void (*fn)(void));
#else
extern void ffi_call_SYSV(void (*)(char *, extended_cif *), extended_cif *,
unsigned, unsigned, unsigned *, void (*fn)(void));
#endif
void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
{
extended_cif ecif;
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 */
#ifdef X86_WIN64
if (rvalue == NULL
&& cif->flags == FFI_TYPE_STRUCT
&& cif->rtype->size != 1 && cif->rtype->size != 2
&& cif->rtype->size != 4 && cif->rtype->size != 8)
{
ecif.rvalue = alloca((cif->rtype->size + 0xF) & ~0xF);
}
#else
if (rvalue == NULL
&& cif->flags == FFI_TYPE_STRUCT)
{
ecif.rvalue = alloca(cif->rtype->size);
}
#endif
else
ecif.rvalue = rvalue;
switch (cif->abi)
{
#ifdef X86_WIN64
case FFI_WIN64:
{
// Make copies of all struct arguments
// NOTE: not sure if responsibility should be here or in caller
unsigned int i;
for (i=0; i < cif->nargs;i++) {
size_t size = cif->arg_types[i]->size;
if ((cif->arg_types[i]->type == FFI_TYPE_STRUCT
&& (size != 1 && size != 2 && size != 4 && size != 8))
#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
|| cif->arg_types[i]->type == FFI_TYPE_LONGDOUBLE
#endif
)
{
void *local = alloca(size);
memcpy(local, avalue[i], size);
avalue[i] = local;
}
}
ffi_call_win64(ffi_prep_args, &ecif, cif->bytes,
cif->flags, ecif.rvalue, fn);
}
break;
#elif defined(X86_WIN32)
case FFI_SYSV:
case FFI_STDCALL:
ffi_call_win32(ffi_prep_args, &ecif, cif->bytes, cif->flags,
ecif.rvalue, fn);
break;
#else
case FFI_SYSV:
ffi_call_SYSV(ffi_prep_args, &ecif, cif->bytes, cif->flags, ecif.rvalue,
fn);
break;
#endif
default:
FFI_ASSERT(0);
break;
}
}
/** private members **/
/* The following __attribute__((regparm(1))) decorations will have no effect
on MSVC - standard cdecl convention applies. */
static void ffi_prep_incoming_args_SYSV (char *stack, void **ret,
void** args, ffi_cif* cif);
void FFI_HIDDEN ffi_closure_SYSV (ffi_closure *)
__attribute__ ((regparm(1)));
unsigned int FFI_HIDDEN ffi_closure_SYSV_inner (ffi_closure *, void **, void *)
__attribute__ ((regparm(1)));
void FFI_HIDDEN ffi_closure_raw_SYSV (ffi_raw_closure *)
__attribute__ ((regparm(1)));
#ifdef X86_WIN32
void FFI_HIDDEN ffi_closure_STDCALL (ffi_closure *)
__attribute__ ((regparm(1)));
#endif
#ifdef X86_WIN64
void FFI_HIDDEN ffi_closure_win64 (ffi_closure *);
#endif
/* This function is jumped to by the trampoline */
#ifdef X86_WIN64
void * FFI_HIDDEN
ffi_closure_win64_inner (ffi_closure *closure, void *args) {
ffi_cif *cif;
void **arg_area;
void *result;
void *resp = &result;
cif = closure->cif;
arg_area = (void**) alloca (cif->nargs * sizeof (void*));
/* this call will initialize ARG_AREA, such that each
* element in that array points to the corresponding
* value on the stack; and if the function returns
* a structure, it will change RESP to point to the
* structure return address. */
ffi_prep_incoming_args_SYSV(args, &resp, arg_area, cif);
(closure->fun) (cif, resp, arg_area, closure->user_data);
/* The result is returned in rax. This does the right thing for
result types except for floats; we have to 'mov xmm0, rax' in the
caller to correct this.
TODO: structure sizes of 3 5 6 7 are returned by reference, too!!!
*/
return cif->rtype->size > sizeof(void *) ? resp : *(void **)resp;
}
#else
unsigned int FFI_HIDDEN __attribute__ ((regparm(1)))
ffi_closure_SYSV_inner (ffi_closure *closure, void **respp, void *args)
{
/* our various things... */
ffi_cif *cif;
void **arg_area;
cif = closure->cif;
arg_area = (void**) alloca (cif->nargs * sizeof (void*));
/* this call will initialize ARG_AREA, such that each
* element in that array points to the corresponding
* value on the stack; and if the function returns
* a structure, it will change RESP to point to the
* structure return address. */
ffi_prep_incoming_args_SYSV(args, respp, arg_area, cif);
(closure->fun) (cif, *respp, arg_area, closure->user_data);
return cif->flags;
}
#endif /* !X86_WIN64 */
static void
ffi_prep_incoming_args_SYSV(char *stack, void **rvalue, void **avalue,
ffi_cif *cif)
{
register unsigned int i;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
argp = stack;
#ifdef X86_WIN64
if (cif->rtype->size > sizeof(ffi_arg)
|| (cif->flags == FFI_TYPE_STRUCT
&& (cif->rtype->size != 1 && cif->rtype->size != 2
&& cif->rtype->size != 4 && cif->rtype->size != 8))) {
*rvalue = *(void **) argp;
argp += sizeof(void *);
}
#else
if ( cif->flags == FFI_TYPE_STRUCT ) {
*rvalue = *(void **) argp;
argp += sizeof(void *);
}
#endif
p_argv = avalue;
for (i = cif->nargs, p_arg = cif->arg_types; (i != 0); i--, p_arg++)
{
size_t z;
/* Align if necessary */
if ((sizeof(void*) - 1) & (size_t) argp) {
argp = (char *) ALIGN(argp, sizeof(void*));
}
#ifdef X86_WIN64
if ((*p_arg)->size > sizeof(ffi_arg)
|| ((*p_arg)->type == FFI_TYPE_STRUCT
&& ((*p_arg)->size != 1 && (*p_arg)->size != 2
&& (*p_arg)->size != 4 && (*p_arg)->size != 8)))
{
z = sizeof(void *);
*p_argv = *(void **)argp;
}
else
#endif
{
z = (*p_arg)->size;
/* because we're little endian, this is what it turns into. */
*p_argv = (void*) argp;
}
p_argv++;
#ifdef X86_WIN64
argp += (z + sizeof(void*) - 1) & ~(sizeof(void*) - 1);
#else
argp += z;
#endif
}
return;
}
#define FFI_INIT_TRAMPOLINE_WIN64(TRAMP,FUN,CTX,MASK) \
{ unsigned char *__tramp = (unsigned char*)(TRAMP); \
void* __fun = (void*)(FUN); \
void* __ctx = (void*)(CTX); \
*(unsigned char*) &__tramp[0] = 0x41; \
*(unsigned char*) &__tramp[1] = 0xbb; \
*(unsigned int*) &__tramp[2] = MASK; /* mov $mask, %r11 */ \
*(unsigned char*) &__tramp[6] = 0x48; \
*(unsigned char*) &__tramp[7] = 0xb8; \
*(void**) &__tramp[8] = __ctx; /* mov __ctx, %rax */ \
*(unsigned char *) &__tramp[16] = 0x49; \
*(unsigned char *) &__tramp[17] = 0xba; \
*(void**) &__tramp[18] = __fun; /* mov __fun, %r10 */ \
*(unsigned char *) &__tramp[26] = 0x41; \
*(unsigned char *) &__tramp[27] = 0xff; \
*(unsigned char *) &__tramp[28] = 0xe2; /* jmp %r10 */ \
}
/* How to make a trampoline. Derived from gcc/config/i386/i386.c. */
#define FFI_INIT_TRAMPOLINE(TRAMP,FUN,CTX) \
{ unsigned char *__tramp = (unsigned char*)(TRAMP); \
unsigned int __fun = (unsigned int)(FUN); \
unsigned int __ctx = (unsigned int)(CTX); \
unsigned int __dis = __fun - (__ctx + 10); \
*(unsigned char*) &__tramp[0] = 0xb8; \
*(unsigned int*) &__tramp[1] = __ctx; /* movl __ctx, %eax */ \
*(unsigned char *) &__tramp[5] = 0xe9; \
*(unsigned int*) &__tramp[6] = __dis; /* jmp __fun */ \
}
#define FFI_INIT_TRAMPOLINE_STDCALL(TRAMP,FUN,CTX,SIZE) \
{ unsigned char *__tramp = (unsigned char*)(TRAMP); \
unsigned int __fun = (unsigned int)(FUN); \
unsigned int __ctx = (unsigned int)(CTX); \
unsigned int __dis = __fun - (__ctx + 10); \
unsigned short __size = (unsigned short)(SIZE); \
*(unsigned char*) &__tramp[0] = 0xb8; \
*(unsigned int*) &__tramp[1] = __ctx; /* movl __ctx, %eax */ \
*(unsigned char *) &__tramp[5] = 0xe8; \
*(unsigned int*) &__tramp[6] = __dis; /* call __fun */ \
*(unsigned char *) &__tramp[10] = 0xc2; \
*(unsigned short*) &__tramp[11] = __size; /* ret __size */ \
}
/* the cif must already be prep'ed */
ffi_status
ffi_prep_closure_loc (ffi_closure* closure,
ffi_cif* cif,
void (*fun)(ffi_cif*,void*,void**,void*),
void *user_data,
void *codeloc)
{
#ifdef X86_WIN64
#define ISFLOAT(IDX) (cif->arg_types[IDX]->type == FFI_TYPE_FLOAT || cif->arg_types[IDX]->type == FFI_TYPE_DOUBLE)
#define FLAG(IDX) (cif->nargs>(IDX)&&ISFLOAT(IDX)?(1<<(IDX)):0)
if (cif->abi == FFI_WIN64)
{
int mask = FLAG(0)|FLAG(1)|FLAG(2)|FLAG(3);
FFI_INIT_TRAMPOLINE_WIN64 (&closure->tramp[0],
&ffi_closure_win64,
codeloc, mask);
/* make sure we can execute here */
}
#else
if (cif->abi == FFI_SYSV)
{
FFI_INIT_TRAMPOLINE (&closure->tramp[0],
&ffi_closure_SYSV,
(void*)codeloc);
}
#ifdef X86_WIN32
else if (cif->abi == FFI_STDCALL)
{
FFI_INIT_TRAMPOLINE_STDCALL (&closure->tramp[0],
&ffi_closure_STDCALL,
(void*)codeloc, cif->bytes);
}
#endif /* X86_WIN32 */
#endif /* !X86_WIN64 */
else
{
return FFI_BAD_ABI;
}
closure->cif = cif;
closure->user_data = user_data;
closure->fun = fun;
return FFI_OK;
}
/* ------- Native raw API support -------------------------------- */
#if !FFI_NO_RAW_API
ffi_status
ffi_prep_raw_closure_loc (ffi_raw_closure* closure,
ffi_cif* cif,
void (*fun)(ffi_cif*,void*,ffi_raw*,void*),
void *user_data,
void *codeloc)
{
int i;
if (cif->abi != FFI_SYSV) {
return FFI_BAD_ABI;
}
// we currently don't support certain kinds of arguments for raw
// closures. This should be implemented by a separate assembly language
// routine, since it would require argument processing, something we
// don't do now for performance.
for (i = cif->nargs-1; i >= 0; i--)
{
FFI_ASSERT (cif->arg_types[i]->type != FFI_TYPE_STRUCT);
FFI_ASSERT (cif->arg_types[i]->type != FFI_TYPE_LONGDOUBLE);
}
FFI_INIT_TRAMPOLINE (&closure->tramp[0], &ffi_closure_raw_SYSV,
codeloc);
closure->cif = cif;
closure->user_data = user_data;
closure->fun = fun;
return FFI_OK;
}
static void
ffi_prep_args_raw(char *stack, extended_cif *ecif)
{
memcpy (stack, ecif->avalue, ecif->cif->bytes);
}
/* we borrow this routine from libffi (it must be changed, though, to
* actually call the function passed in the first argument. as of
* libffi-1.20, this is not the case.)
*/
void
ffi_raw_call(ffi_cif *cif, void (*fn)(void), void *rvalue, ffi_raw *fake_avalue)
{
extended_cif ecif;
void **avalue = (void **)fake_avalue;
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) &&
(cif->rtype->type == FFI_TYPE_STRUCT))
{
ecif.rvalue = alloca(cif->rtype->size);
}
else
ecif.rvalue = rvalue;
switch (cif->abi)
{
#ifdef X86_WIN32
case FFI_SYSV:
case FFI_STDCALL:
ffi_call_win32(ffi_prep_args_raw, &ecif, cif->bytes, cif->flags,
ecif.rvalue, fn);
break;
#else
case FFI_SYSV:
ffi_call_SYSV(ffi_prep_args_raw, &ecif, cif->bytes, cif->flags,
ecif.rvalue, fn);
break;
#endif
default:
FFI_ASSERT(0);
break;
}
}
#endif
#endif /* !__x86_64__ || X86_WIN64 */

626
.pc/fix-comments/src/x86/ffi64.c Normal file
View File

@@ -0,0 +1,626 @@
/* -----------------------------------------------------------------------
ffi64.c - Copyright (c) 2002, 2007 Bo Thorsen <bo@suse.de>
Copyright (c) 2008 Red Hat, Inc.
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 THE AUTHORS OR COPYRIGHT
HOLDERS 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>
#ifdef __x86_64__
#define MAX_GPR_REGS 6
#define MAX_SSE_REGS 8
struct register_args
{
/* Registers for argument passing. */
UINT64 gpr[MAX_GPR_REGS];
__int128_t sse[MAX_SSE_REGS];
};
extern void ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags,
void *raddr, void (*fnaddr)(void), unsigned ssecount);
/* 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_COMPLEX_X87_CLASS,
X86_64_MEMORY_CLASS
};
#define MAX_CLASSES 4
#define SSE_CLASS_P(X) ((X) >= X86_64_SSE_CLASS && X <= X86_64_SSEUP_CLASS)
/* 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, X87UP, or COMPLEX_X87 class,
MEMORY is used. */
if (class1 == X86_64_X87_CLASS
|| class1 == X86_64_X87UP_CLASS
|| class1 == X86_64_COMPLEX_X87_CLASS
|| class2 == X86_64_X87_CLASS
|| class2 == X86_64_X87UP_CLASS
|| class2 == X86_64_COMPLEX_X87_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[],
size_t byte_offset)
{
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:
{
int size = byte_offset + type->size;
if (size <= 4)
{
classes[0] = X86_64_INTEGERSI_CLASS;
return 1;
}
else if (size <= 8)
{
classes[0] = X86_64_INTEGER_CLASS;
return 1;
}
else if (size <= 12)
{
classes[0] = X86_64_INTEGER_CLASS;
classes[1] = X86_64_INTEGERSI_CLASS;
return 2;
}
else if (size <= 16)
{
classes[0] = classes[1] = X86_64_INTEGERSI_CLASS;
return 2;
}
else
FFI_ASSERT (0);
}
case FFI_TYPE_FLOAT:
if (!(byte_offset % 8))
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 32 bytes, pass it on the stack. */
if (type->size > 32)
return 0;
for (i = 0; i < words; i++)
classes[i] = X86_64_NO_CLASS;
/* Zero sized arrays or structures are NO_CLASS. We return 0 to
signalize memory class, so handle it as special case. */
if (!words)
{
classes[0] = X86_64_NO_CLASS;
return 1;
}
/* Merge the fields of structure. */
for (ptr = type->elements; *ptr != NULL; ptr++)
{
int num;
byte_offset = ALIGN (byte_offset, (*ptr)->alignment);
num = classify_argument (*ptr, subclasses, byte_offset % 8);
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]);
}
byte_offset += (*ptr)->size;
}
if (words > 2)
{
/* When size > 16 bytes, if the first one isn't
X86_64_SSE_CLASS or any other ones aren't
X86_64_SSEUP_CLASS, everything should be passed in
memory. */
if (classes[0] != X86_64_SSE_CLASS)
return 0;
for (i = 1; i < words; i++)
if (classes[i] != X86_64_SSEUP_CLASS)
return 0;
}
/* 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 or X86_64_SSEUP_CLASS. */
if (classes[i] == X86_64_SSEUP_CLASS
&& classes[i - 1] != X86_64_SSE_CLASS
&& classes[i - 1] != X86_64_SSEUP_CLASS)
{
/* The first one should never be X86_64_SSEUP_CLASS. */
FFI_ASSERT (i != 0);
classes[i] = X86_64_SSE_CLASS;
}
/* If X86_64_X87UP_CLASS isn't preceded by X86_64_X87_CLASS,
everything should be passed in memory. */
if (classes[i] == X86_64_X87UP_CLASS
&& (classes[i - 1] != X86_64_X87_CLASS))
{
/* The first one should never be X86_64_X87UP_CLASS. */
FFI_ASSERT (i != 0);
return 0;
}
}
return words;
}
default:
FFI_ASSERT(0);
}
return 0; /* Never reached. */
}
/* Examine the argument and return set number of register required in each
class. Return zero iff parameter should be passed in memory, otherwise
the number of registers. */
static int
examine_argument (ffi_type *type, enum x86_64_reg_class classes[MAX_CLASSES],
_Bool in_return, int *pngpr, int *pnsse)
{
int i, n, ngpr, nsse;
n = classify_argument (type, classes, 0);
if (n == 0)
return 0;
ngpr = nsse = 0;
for (i = 0; i < n; ++i)
switch (classes[i])
{
case X86_64_INTEGER_CLASS:
case X86_64_INTEGERSI_CLASS:
ngpr++;
break;
case X86_64_SSE_CLASS:
case X86_64_SSESF_CLASS:
case X86_64_SSEDF_CLASS:
nsse++;
break;
case X86_64_NO_CLASS:
case X86_64_SSEUP_CLASS:
break;
case X86_64_X87_CLASS:
case X86_64_X87UP_CLASS:
case X86_64_COMPLEX_X87_CLASS:
return in_return != 0;
default:
abort ();
}
*pngpr = ngpr;
*pnsse = nsse;
return n;
}
/* Perform machine dependent cif processing. */
ffi_status
ffi_prep_cif_machdep (ffi_cif *cif)
{
int gprcount, ssecount, i, avn, n, ngpr, nsse, flags;
enum x86_64_reg_class classes[MAX_CLASSES];
size_t bytes;
gprcount = ssecount = 0;
flags = cif->rtype->type;
if (flags != FFI_TYPE_VOID)
{
n = examine_argument (cif->rtype, classes, 1, &ngpr, &nsse);
if (n == 0)
{
/* The return value is passed in memory. A pointer to that
memory is the first argument. Allocate a register for it. */
gprcount++;
/* We don't have to do anything in asm for the return. */
flags = FFI_TYPE_VOID;
}
else if (flags == FFI_TYPE_STRUCT)
{
/* Mark which registers the result appears in. */
_Bool sse0 = SSE_CLASS_P (classes[0]);
_Bool sse1 = n == 2 && SSE_CLASS_P (classes[1]);
if (sse0 && !sse1)
flags |= 1 << 8;
else if (!sse0 && sse1)
flags |= 1 << 9;
else if (sse0 && sse1)
flags |= 1 << 10;
/* Mark the true size of the structure. */
flags |= cif->rtype->size << 12;
}
}
/* 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 (bytes = 0, i = 0, avn = cif->nargs; i < avn; i++)
{
if (examine_argument (cif->arg_types[i], classes, 0, &ngpr, &nsse) == 0
|| gprcount + ngpr > MAX_GPR_REGS
|| ssecount + nsse > MAX_SSE_REGS)
{
long align = cif->arg_types[i]->alignment;
if (align < 8)
align = 8;
bytes = ALIGN(bytes, align);
bytes += cif->arg_types[i]->size;
}
else
{
gprcount += ngpr;
ssecount += nsse;
}
}
if (ssecount)
flags |= 1 << 11;
cif->flags = flags;
cif->bytes = bytes;
return FFI_OK;
}
void
ffi_call (ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
{
enum x86_64_reg_class classes[MAX_CLASSES];
char *stack, *argp;
ffi_type **arg_types;
int gprcount, ssecount, ngpr, nsse, i, avn;
_Bool ret_in_memory;
struct register_args *reg_args;
/* Can't call 32-bit mode from 64-bit mode. */
FFI_ASSERT (cif->abi == FFI_UNIX64);
/* If the return value is a struct and we don't have a return value
address then we need to make one. Note the setting of flags to
VOID above in ffi_prep_cif_machdep. */
ret_in_memory = (cif->rtype->type == FFI_TYPE_STRUCT
&& (cif->flags & 0xff) == FFI_TYPE_VOID);
if (rvalue == NULL && ret_in_memory)
rvalue = alloca (cif->rtype->size);
/* Allocate the space for the arguments, plus 4 words of temp space. */
stack = alloca (sizeof (struct register_args) + cif->bytes + 4*8);
reg_args = (struct register_args *) stack;
argp = stack + sizeof (struct register_args);
gprcount = ssecount = 0;
/* If the return value is passed in memory, add the pointer as the
first integer argument. */
if (ret_in_memory)
reg_args->gpr[gprcount++] = (long) rvalue;
avn = cif->nargs;
arg_types = cif->arg_types;
for (i = 0; i < avn; ++i)
{
size_t size = arg_types[i]->size;
int n;
n = examine_argument (arg_types[i], classes, 0, &ngpr, &nsse);
if (n == 0
|| gprcount + ngpr > MAX_GPR_REGS
|| ssecount + nsse > MAX_SSE_REGS)
{
long align = arg_types[i]->alignment;
/* Stack arguments are *always* at least 8 byte aligned. */
if (align < 8)
align = 8;
/* Pass this argument in memory. */
argp = (void *) ALIGN (argp, align);
memcpy (argp, avalue[i], size);
argp += size;
}
else
{
/* The argument is passed entirely in registers. */
char *a = (char *) avalue[i];
int j;
for (j = 0; j < n; j++, a += 8, size -= 8)
{
switch (classes[j])
{
case X86_64_INTEGER_CLASS:
case X86_64_INTEGERSI_CLASS:
reg_args->gpr[gprcount] = 0;
memcpy (&reg_args->gpr[gprcount], a, size < 8 ? size : 8);
gprcount++;
break;
case X86_64_SSE_CLASS:
case X86_64_SSEDF_CLASS:
reg_args->sse[ssecount++] = *(UINT64 *) a;
break;
case X86_64_SSESF_CLASS:
reg_args->sse[ssecount++] = *(UINT32 *) a;
break;
default:
abort();
}
}
}
}
ffi_call_unix64 (stack, cif->bytes + sizeof (struct register_args),
cif->flags, rvalue, fn, ssecount);
}
extern void ffi_closure_unix64(void);
ffi_status
ffi_prep_closure_loc (ffi_closure* closure,
ffi_cif* cif,
void (*fun)(ffi_cif*, void*, void**, void*),
void *user_data,
void *codeloc)
{
volatile unsigned short *tramp;
tramp = (volatile unsigned short *) &closure->tramp[0];
tramp[0] = 0xbb49; /* mov <code>, %r11 */
*(void * volatile *) &tramp[1] = ffi_closure_unix64;
tramp[5] = 0xba49; /* mov <data>, %r10 */
*(void * volatile *) &tramp[6] = codeloc;
/* Set the carry bit iff the function uses any sse registers.
This is clc or stc, together with the first byte of the jmp. */
tramp[10] = cif->flags & (1 << 11) ? 0x49f9 : 0x49f8;
tramp[11] = 0xe3ff; /* jmp *%r11 */
closure->cif = cif;
closure->fun = fun;
closure->user_data = user_data;
return FFI_OK;
}
int
ffi_closure_unix64_inner(ffi_closure *closure, void *rvalue,
struct register_args *reg_args, char *argp)
{
ffi_cif *cif;
void **avalue;
ffi_type **arg_types;
long i, avn;
int gprcount, ssecount, ngpr, nsse;
int ret;
cif = closure->cif;
avalue = alloca(cif->nargs * sizeof(void *));
gprcount = ssecount = 0;
ret = cif->rtype->type;
if (ret != FFI_TYPE_VOID)
{
enum x86_64_reg_class classes[MAX_CLASSES];
int n = examine_argument (cif->rtype, classes, 1, &ngpr, &nsse);
if (n == 0)
{
/* The return value goes in memory. Arrange for the closure
return value to go directly back to the original caller. */
rvalue = (void *) reg_args->gpr[gprcount++];
/* We don't have to do anything in asm for the return. */
ret = FFI_TYPE_VOID;
}
else if (ret == FFI_TYPE_STRUCT && n == 2)
{
/* Mark which register the second word of the structure goes in. */
_Bool sse0 = SSE_CLASS_P (classes[0]);
_Bool sse1 = SSE_CLASS_P (classes[1]);
if (!sse0 && sse1)
ret |= 1 << 8;
else if (sse0 && !sse1)
ret |= 1 << 9;
}
}
avn = cif->nargs;
arg_types = cif->arg_types;
for (i = 0; i < avn; ++i)
{
enum x86_64_reg_class classes[MAX_CLASSES];
int n;
n = examine_argument (arg_types[i], classes, 0, &ngpr, &nsse);
if (n == 0
|| gprcount + ngpr > MAX_GPR_REGS
|| ssecount + nsse > MAX_SSE_REGS)
{
long align = arg_types[i]->alignment;
/* Stack arguments are *always* at least 8 byte aligned. */
if (align < 8)
align = 8;
/* Pass this argument in memory. */
argp = (void *) ALIGN (argp, align);
avalue[i] = argp;
argp += arg_types[i]->size;
}
/* If the argument is in a single register, or two consecutive
integer registers, then we can use that address directly. */
else if (n == 1
|| (n == 2 && !(SSE_CLASS_P (classes[0])
|| SSE_CLASS_P (classes[1]))))
{
/* The argument is in a single register. */
if (SSE_CLASS_P (classes[0]))
{
avalue[i] = &reg_args->sse[ssecount];
ssecount += n;
}
else
{
avalue[i] = &reg_args->gpr[gprcount];
gprcount += n;
}
}
/* Otherwise, allocate space to make them consecutive. */
else
{
char *a = alloca (16);
int j;
avalue[i] = a;
for (j = 0; j < n; j++, a += 8)
{
if (SSE_CLASS_P (classes[j]))
memcpy (a, &reg_args->sse[ssecount++], 8);
else
memcpy (a, &reg_args->gpr[gprcount++], 8);
}
}
}
/* Invoke the closure. */
closure->fun (cif, rvalue, avalue, closure->user_data);
/* Tell assembly how to perform return type promotions. */
return ret;
}
#endif /* __x86_64__ */

5
ChangeLog
View File

@@ -1,3 +1,8 @@
2010-03-14 Matthias Klose <doko@ubuntu.com>
* src/x86/ffi64.c: Fix typo in comment.
* src/x86/ffi.c: Use /* ... */ comment style.
2010-01-07 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
PR libffi/40701

50
patches/fix-comments Normal file
View File

@@ -0,0 +1,50 @@
Index: libffi/ChangeLog
===================================================================
--- libffi.orig/ChangeLog
+++ libffi/ChangeLog
@@ -1,3 +1,8 @@
+2010-03-14 Matthias Klose <doko@ubuntu.com>
+
+ * src/x86/ffi64.c: Fix typo in comment.
+ * src/x86/ffi.c: Use /* ... */ comment style.
+
2010-01-07 Rainer Orth <ro@CeBiTec.Uni-Bielefeld.DE>
PR libffi/40701
Index: libffi/src/x86/ffi.c
===================================================================
--- libffi.orig/src/x86/ffi.c
+++ libffi/src/x86/ffi.c
@@ -585,10 +585,10 @@ ffi_prep_raw_closure_loc (ffi_raw_closur
return FFI_BAD_ABI;
}
- // we currently don't support certain kinds of arguments for raw
- // closures. This should be implemented by a separate assembly language
- // routine, since it would require argument processing, something we
- // don't do now for performance.
+ /* we currently don't support certain kinds of arguments for raw
+ closures. This should be implemented by a separate assembly
+ language routine, since it would require argument processing,
+ something we don't do now for performance. */
for (i = cif->nargs-1; i >= 0; i--)
{
Index: libffi/src/x86/ffi64.c
===================================================================
--- libffi.orig/src/x86/ffi64.c
+++ libffi/src/x86/ffi64.c
@@ -50,9 +50,10 @@ extern void ffi_call_unix64 (void *args,
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.
+ 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
+ reformatting penalties.
Similary we play games with INTEGERSI_CLASS to use cheaper SImode moves
whenever possible (upper half does contain padding). */

1
patches/series
View File

@@ -2,3 +2,4 @@ stand-alone
closure-api-example-doc
sunstudio
msvc
fix-comments

8
src/x86/ffi.c
View File

@@ -585,10 +585,10 @@ ffi_prep_raw_closure_loc (ffi_raw_closure* closure,
return FFI_BAD_ABI;
}
// we currently don't support certain kinds of arguments for raw
// closures. This should be implemented by a separate assembly language
// routine, since it would require argument processing, something we
// don't do now for performance.
/* we currently don't support certain kinds of arguments for raw
closures. This should be implemented by a separate assembly
language routine, since it would require argument processing,
something we don't do now for performance. */
for (i = cif->nargs-1; i >= 0; i--)
{

665
src/x86/ffi.c.orig Normal file
View File

@@ -0,0 +1,665 @@
/* -----------------------------------------------------------------------
ffi.c - Copyright (c) 1996, 1998, 1999, 2001, 2007, 2008 Red Hat, Inc.
Copyright (c) 2002 Ranjit Mathew
Copyright (c) 2002 Bo Thorsen
Copyright (c) 2002 Roger Sayle
Copyright (C) 2008 Free Software Foundation, Inc.
x86 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 THE AUTHORS OR COPYRIGHT
HOLDERS 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.
----------------------------------------------------------------------- */
#if !defined(__x86_64__) || defined(_WIN64)
#ifdef _WIN64
#include <windows.h>
#endif
#include <ffi.h>
#include <ffi_common.h>
#include <stdlib.h>
/* ffi_prep_args is called by the assembly routine once stack space
has been allocated for the function's arguments */
void ffi_prep_args(char *stack, extended_cif *ecif)
{
register unsigned int i;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
argp = stack;
if (ecif->cif->flags == FFI_TYPE_STRUCT
#ifdef X86_WIN64
&& (ecif->cif->rtype->size != 1 && ecif->cif->rtype->size != 2
&& ecif->cif->rtype->size != 4 && ecif->cif->rtype->size != 8)
#endif
)
{
*(void **) argp = ecif->rvalue;
argp += sizeof(void*);
}
p_argv = ecif->avalue;
for (i = ecif->cif->nargs, p_arg = ecif->cif->arg_types;
i != 0;
i--, p_arg++)
{
size_t z;
/* Align if necessary */
if ((sizeof(void*) - 1) & (size_t) argp)
argp = (char *) ALIGN(argp, sizeof(void*));
z = (*p_arg)->size;
#ifdef X86_WIN64
if (z > sizeof(ffi_arg)
|| ((*p_arg)->type == FFI_TYPE_STRUCT
&& (z != 1 && z != 2 && z != 4 && z != 8))
#if FFI_TYPE_DOUBLE != FFI_TYPE_LONGDOUBLE
|| ((*p_arg)->type == FFI_TYPE_LONGDOUBLE)
#endif
)
{
z = sizeof(ffi_arg);
*(void **)argp = *p_argv;
}
else if ((*p_arg)->type == FFI_TYPE_FLOAT)
{
memcpy(argp, *p_argv, z);
}
else
#endif
if (z < sizeof(ffi_arg))
{
z = sizeof(ffi_arg);
switch ((*p_arg)->type)
{
case FFI_TYPE_SINT8:
*(ffi_sarg *) argp = (ffi_sarg)*(SINT8 *)(* p_argv);
break;
case FFI_TYPE_UINT8:
*(ffi_arg *) argp = (ffi_arg)*(UINT8 *)(* p_argv);
break;
case FFI_TYPE_SINT16:
*(ffi_sarg *) argp = (ffi_sarg)*(SINT16 *)(* p_argv);
break;
case FFI_TYPE_UINT16:
*(ffi_arg *) argp = (ffi_arg)*(UINT16 *)(* p_argv);
break;
case FFI_TYPE_SINT32:
*(ffi_sarg *) argp = (ffi_sarg)*(SINT32 *)(* p_argv);
break;
case FFI_TYPE_UINT32:
*(ffi_arg *) argp = (ffi_arg)*(UINT32 *)(* p_argv);
break;
case FFI_TYPE_STRUCT:
*(ffi_arg *) argp = *(ffi_arg *)(* p_argv);
break;
default:
FFI_ASSERT(0);
}
}
else
{
memcpy(argp, *p_argv, z);
}
p_argv++;
#ifdef X86_WIN64
argp += (z + sizeof(void*) - 1) & ~(sizeof(void*) - 1);
#else
argp += z;
#endif
}
return;
}
/* Perform machine dependent cif processing */
ffi_status ffi_prep_cif_machdep(ffi_cif *cif)
{
unsigned int i;
ffi_type **ptr;
/* Set the return type flag */
switch (cif->rtype->type)
{
case FFI_TYPE_VOID:
#if defined(X86) || defined (X86_WIN32) || defined(X86_FREEBSD) || defined(X86_DARWIN) || defined(X86_WIN64)
case FFI_TYPE_UINT8:
case FFI_TYPE_UINT16:
case FFI_TYPE_SINT8:
case FFI_TYPE_SINT16:
#endif
#ifdef X86_WIN64
case FFI_TYPE_UINT32:
case FFI_TYPE_SINT32:
#endif
case FFI_TYPE_SINT64:
case FFI_TYPE_FLOAT:
case FFI_TYPE_DOUBLE:
#ifndef X86_WIN64
#if FFI_TYPE_DOUBLE != FFI_TYPE_LONGDOUBLE
case FFI_TYPE_LONGDOUBLE:
#endif
#endif
cif->flags = (unsigned) cif->rtype->type;
break;
case FFI_TYPE_UINT64:
#ifdef X86_WIN64
case FFI_TYPE_POINTER:
#endif
cif->flags = FFI_TYPE_SINT64;
break;
case FFI_TYPE_STRUCT:
#ifndef X86
if (cif->rtype->size == 1)
{
cif->flags = FFI_TYPE_SMALL_STRUCT_1B; /* same as char size */
}
else if (cif->rtype->size == 2)
{
cif->flags = FFI_TYPE_SMALL_STRUCT_2B; /* same as short size */
}
else if (cif->rtype->size == 4)
{
#ifdef X86_WIN64
cif->flags = FFI_TYPE_SMALL_STRUCT_4B;
#else
cif->flags = FFI_TYPE_INT; /* same as int type */
#endif
}
else if (cif->rtype->size == 8)
{
cif->flags = FFI_TYPE_SINT64; /* same as int64 type */
}
else
#endif
{
cif->flags = FFI_TYPE_STRUCT;
// allocate space for return value pointer
cif->bytes += ALIGN(sizeof(void*), FFI_SIZEOF_ARG);
}
break;
default:
#ifdef X86_WIN64
cif->flags = FFI_TYPE_SINT64;
break;
case FFI_TYPE_INT:
cif->flags = FFI_TYPE_SINT32;
#else
cif->flags = FFI_TYPE_INT;
#endif
break;
}
for (ptr = cif->arg_types, i = cif->nargs; i > 0; i--, ptr++)
{
if (((*ptr)->alignment - 1) & cif->bytes)
cif->bytes = ALIGN(cif->bytes, (*ptr)->alignment);
cif->bytes += ALIGN((*ptr)->size, FFI_SIZEOF_ARG);
}
#ifdef X86_WIN64
// ensure space for storing four registers
cif->bytes += 4 * sizeof(ffi_arg);
#endif
#ifdef X86_DARWIN
cif->bytes = (cif->bytes + 15) & ~0xF;
#endif
return FFI_OK;
}
#ifdef X86_WIN64
extern int
ffi_call_win64(void (*)(char *, extended_cif *), extended_cif *,
unsigned, unsigned, unsigned *, void (*fn)(void));
#elif defined(X86_WIN32)
extern void
ffi_call_win32(void (*)(char *, extended_cif *), extended_cif *,
unsigned, unsigned, unsigned *, void (*fn)(void));
#else
extern void ffi_call_SYSV(void (*)(char *, extended_cif *), extended_cif *,
unsigned, unsigned, unsigned *, void (*fn)(void));
#endif
void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
{
extended_cif ecif;
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 */
#ifdef X86_WIN64
if (rvalue == NULL
&& cif->flags == FFI_TYPE_STRUCT
&& cif->rtype->size != 1 && cif->rtype->size != 2
&& cif->rtype->size != 4 && cif->rtype->size != 8)
{
ecif.rvalue = alloca((cif->rtype->size + 0xF) & ~0xF);
}
#else
if (rvalue == NULL
&& cif->flags == FFI_TYPE_STRUCT)
{
ecif.rvalue = alloca(cif->rtype->size);
}
#endif
else
ecif.rvalue = rvalue;
switch (cif->abi)
{
#ifdef X86_WIN64
case FFI_WIN64:
{
// Make copies of all struct arguments
// NOTE: not sure if responsibility should be here or in caller
unsigned int i;
for (i=0; i < cif->nargs;i++) {
size_t size = cif->arg_types[i]->size;
if ((cif->arg_types[i]->type == FFI_TYPE_STRUCT
&& (size != 1 && size != 2 && size != 4 && size != 8))
#if FFI_TYPE_LONGDOUBLE != FFI_TYPE_DOUBLE
|| cif->arg_types[i]->type == FFI_TYPE_LONGDOUBLE
#endif
)
{
void *local = alloca(size);
memcpy(local, avalue[i], size);
avalue[i] = local;
}
}
ffi_call_win64(ffi_prep_args, &ecif, cif->bytes,
cif->flags, ecif.rvalue, fn);
}
break;
#elif defined(X86_WIN32)
case FFI_SYSV:
case FFI_STDCALL:
ffi_call_win32(ffi_prep_args, &ecif, cif->bytes, cif->flags,
ecif.rvalue, fn);
break;
#else
case FFI_SYSV:
ffi_call_SYSV(ffi_prep_args, &ecif, cif->bytes, cif->flags, ecif.rvalue,
fn);
break;
#endif
default:
FFI_ASSERT(0);
break;
}
}
/** private members **/
/* The following __attribute__((regparm(1))) decorations will have no effect
on MSVC - standard cdecl convention applies. */
static void ffi_prep_incoming_args_SYSV (char *stack, void **ret,
void** args, ffi_cif* cif);
void FFI_HIDDEN ffi_closure_SYSV (ffi_closure *)
__attribute__ ((regparm(1)));
unsigned int FFI_HIDDEN ffi_closure_SYSV_inner (ffi_closure *, void **, void *)
__attribute__ ((regparm(1)));
void FFI_HIDDEN ffi_closure_raw_SYSV (ffi_raw_closure *)
__attribute__ ((regparm(1)));
#ifdef X86_WIN32
void FFI_HIDDEN ffi_closure_STDCALL (ffi_closure *)
__attribute__ ((regparm(1)));
#endif
#ifdef X86_WIN64
void FFI_HIDDEN ffi_closure_win64 (ffi_closure *);
#endif
/* This function is jumped to by the trampoline */
#ifdef X86_WIN64
void * FFI_HIDDEN
ffi_closure_win64_inner (ffi_closure *closure, void *args) {
ffi_cif *cif;
void **arg_area;
void *result;
void *resp = &result;
cif = closure->cif;
arg_area = (void**) alloca (cif->nargs * sizeof (void*));
/* this call will initialize ARG_AREA, such that each
* element in that array points to the corresponding
* value on the stack; and if the function returns
* a structure, it will change RESP to point to the
* structure return address. */
ffi_prep_incoming_args_SYSV(args, &resp, arg_area, cif);
(closure->fun) (cif, resp, arg_area, closure->user_data);
/* The result is returned in rax. This does the right thing for
result types except for floats; we have to 'mov xmm0, rax' in the
caller to correct this.
TODO: structure sizes of 3 5 6 7 are returned by reference, too!!!
*/
return cif->rtype->size > sizeof(void *) ? resp : *(void **)resp;
}
#else
unsigned int FFI_HIDDEN __attribute__ ((regparm(1)))
ffi_closure_SYSV_inner (ffi_closure *closure, void **respp, void *args)
{
/* our various things... */
ffi_cif *cif;
void **arg_area;
cif = closure->cif;
arg_area = (void**) alloca (cif->nargs * sizeof (void*));
/* this call will initialize ARG_AREA, such that each
* element in that array points to the corresponding
* value on the stack; and if the function returns
* a structure, it will change RESP to point to the
* structure return address. */
ffi_prep_incoming_args_SYSV(args, respp, arg_area, cif);
(closure->fun) (cif, *respp, arg_area, closure->user_data);
return cif->flags;
}
#endif /* !X86_WIN64 */
static void
ffi_prep_incoming_args_SYSV(char *stack, void **rvalue, void **avalue,
ffi_cif *cif)
{
register unsigned int i;
register void **p_argv;
register char *argp;
register ffi_type **p_arg;
argp = stack;
#ifdef X86_WIN64
if (cif->rtype->size > sizeof(ffi_arg)
|| (cif->flags == FFI_TYPE_STRUCT
&& (cif->rtype->size != 1 && cif->rtype->size != 2
&& cif->rtype->size != 4 && cif->rtype->size != 8))) {
*rvalue = *(void **) argp;
argp += sizeof(void *);
}
#else
if ( cif->flags == FFI_TYPE_STRUCT ) {
*rvalue = *(void **) argp;
argp += sizeof(void *);
}
#endif
p_argv = avalue;
for (i = cif->nargs, p_arg = cif->arg_types; (i != 0); i--, p_arg++)
{
size_t z;
/* Align if necessary */
if ((sizeof(void*) - 1) & (size_t) argp) {
argp = (char *) ALIGN(argp, sizeof(void*));
}
#ifdef X86_WIN64
if ((*p_arg)->size > sizeof(ffi_arg)
|| ((*p_arg)->type == FFI_TYPE_STRUCT
&& ((*p_arg)->size != 1 && (*p_arg)->size != 2
&& (*p_arg)->size != 4 && (*p_arg)->size != 8)))
{
z = sizeof(void *);
*p_argv = *(void **)argp;
}
else
#endif
{
z = (*p_arg)->size;
/* because we're little endian, this is what it turns into. */
*p_argv = (void*) argp;
}
p_argv++;
#ifdef X86_WIN64
argp += (z + sizeof(void*) - 1) & ~(sizeof(void*) - 1);
#else
argp += z;
#endif
}
return;
}
#define FFI_INIT_TRAMPOLINE_WIN64(TRAMP,FUN,CTX,MASK) \
{ unsigned char *__tramp = (unsigned char*)(TRAMP); \
void* __fun = (void*)(FUN); \
void* __ctx = (void*)(CTX); \
*(unsigned char*) &__tramp[0] = 0x41; \
*(unsigned char*) &__tramp[1] = 0xbb; \
*(unsigned int*) &__tramp[2] = MASK; /* mov $mask, %r11 */ \
*(unsigned char*) &__tramp[6] = 0x48; \
*(unsigned char*) &__tramp[7] = 0xb8; \
*(void**) &__tramp[8] = __ctx; /* mov __ctx, %rax */ \
*(unsigned char *) &__tramp[16] = 0x49; \
*(unsigned char *) &__tramp[17] = 0xba; \
*(void**) &__tramp[18] = __fun; /* mov __fun, %r10 */ \
*(unsigned char *) &__tramp[26] = 0x41; \
*(unsigned char *) &__tramp[27] = 0xff; \
*(unsigned char *) &__tramp[28] = 0xe2; /* jmp %r10 */ \
}
/* How to make a trampoline. Derived from gcc/config/i386/i386.c. */
#define FFI_INIT_TRAMPOLINE(TRAMP,FUN,CTX) \
{ unsigned char *__tramp = (unsigned char*)(TRAMP); \
unsigned int __fun = (unsigned int)(FUN); \
unsigned int __ctx = (unsigned int)(CTX); \
unsigned int __dis = __fun - (__ctx + 10); \
*(unsigned char*) &__tramp[0] = 0xb8; \
*(unsigned int*) &__tramp[1] = __ctx; /* movl __ctx, %eax */ \
*(unsigned char *) &__tramp[5] = 0xe9; \
*(unsigned int*) &__tramp[6] = __dis; /* jmp __fun */ \
}
#define FFI_INIT_TRAMPOLINE_STDCALL(TRAMP,FUN,CTX,SIZE) \
{ unsigned char *__tramp = (unsigned char*)(TRAMP); \
unsigned int __fun = (unsigned int)(FUN); \
unsigned int __ctx = (unsigned int)(CTX); \
unsigned int __dis = __fun - (__ctx + 10); \
unsigned short __size = (unsigned short)(SIZE); \
*(unsigned char*) &__tramp[0] = 0xb8; \
*(unsigned int*) &__tramp[1] = __ctx; /* movl __ctx, %eax */ \
*(unsigned char *) &__tramp[5] = 0xe8; \
*(unsigned int*) &__tramp[6] = __dis; /* call __fun */ \
*(unsigned char *) &__tramp[10] = 0xc2; \
*(unsigned short*) &__tramp[11] = __size; /* ret __size */ \
}
/* the cif must already be prep'ed */
ffi_status
ffi_prep_closure_loc (ffi_closure* closure,
ffi_cif* cif,
void (*fun)(ffi_cif*,void*,void**,void*),
void *user_data,
void *codeloc)
{
#ifdef X86_WIN64
#define ISFLOAT(IDX) (cif->arg_types[IDX]->type == FFI_TYPE_FLOAT || cif->arg_types[IDX]->type == FFI_TYPE_DOUBLE)
#define FLAG(IDX) (cif->nargs>(IDX)&&ISFLOAT(IDX)?(1<<(IDX)):0)
if (cif->abi == FFI_WIN64)
{
int mask = FLAG(0)|FLAG(1)|FLAG(2)|FLAG(3);
FFI_INIT_TRAMPOLINE_WIN64 (&closure->tramp[0],
&ffi_closure_win64,
codeloc, mask);
/* make sure we can execute here */
}
#else
if (cif->abi == FFI_SYSV)
{
FFI_INIT_TRAMPOLINE (&closure->tramp[0],
&ffi_closure_SYSV,
(void*)codeloc);
}
#ifdef X86_WIN32
else if (cif->abi == FFI_STDCALL)
{
FFI_INIT_TRAMPOLINE_STDCALL (&closure->tramp[0],
&ffi_closure_STDCALL,
(void*)codeloc, cif->bytes);
}
#endif /* X86_WIN32 */
#endif /* !X86_WIN64 */
else
{
return FFI_BAD_ABI;
}
closure->cif = cif;
closure->user_data = user_data;
closure->fun = fun;
return FFI_OK;
}
/* ------- Native raw API support -------------------------------- */
#if !FFI_NO_RAW_API
ffi_status
ffi_prep_raw_closure_loc (ffi_raw_closure* closure,
ffi_cif* cif,
void (*fun)(ffi_cif*,void*,ffi_raw*,void*),
void *user_data,
void *codeloc)
{
int i;
if (cif->abi != FFI_SYSV) {
return FFI_BAD_ABI;
}
// we currently don't support certain kinds of arguments for raw
// closures. This should be implemented by a separate assembly language
// routine, since it would require argument processing, something we
// don't do now for performance.
for (i = cif->nargs-1; i >= 0; i--)
{
FFI_ASSERT (cif->arg_types[i]->type != FFI_TYPE_STRUCT);
FFI_ASSERT (cif->arg_types[i]->type != FFI_TYPE_LONGDOUBLE);
}
FFI_INIT_TRAMPOLINE (&closure->tramp[0], &ffi_closure_raw_SYSV,
codeloc);
closure->cif = cif;
closure->user_data = user_data;
closure->fun = fun;
return FFI_OK;
}
static void
ffi_prep_args_raw(char *stack, extended_cif *ecif)
{
memcpy (stack, ecif->avalue, ecif->cif->bytes);
}
/* we borrow this routine from libffi (it must be changed, though, to
* actually call the function passed in the first argument. as of
* libffi-1.20, this is not the case.)
*/
void
ffi_raw_call(ffi_cif *cif, void (*fn)(void), void *rvalue, ffi_raw *fake_avalue)
{
extended_cif ecif;
void **avalue = (void **)fake_avalue;
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) &&
(cif->rtype->type == FFI_TYPE_STRUCT))
{
ecif.rvalue = alloca(cif->rtype->size);
}
else
ecif.rvalue = rvalue;
switch (cif->abi)
{
#ifdef X86_WIN32
case FFI_SYSV:
case FFI_STDCALL:
ffi_call_win32(ffi_prep_args_raw, &ecif, cif->bytes, cif->flags,
ecif.rvalue, fn);
break;
#else
case FFI_SYSV:
ffi_call_SYSV(ffi_prep_args_raw, &ecif, cif->bytes, cif->flags,
ecif.rvalue, fn);
break;
#endif
default:
FFI_ASSERT(0);
break;
}
}
#endif
#endif /* !__x86_64__ || X86_WIN64 */

7
src/x86/ffi64.c
View File

@@ -50,9 +50,10 @@ extern void ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags,
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.
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
reformatting penalties.
Similary we play games with INTEGERSI_CLASS to use cheaper SImode moves
whenever possible (upper half does contain padding). */