ffi 1.0.0 → 1.17.1

data/ext/ffi_c/libffi/src/x86/sysv.S

@@ -1,5 +1,7 @@
 /* -----------------------------------------------------------------------
-   sysv.S - Copyright (c) 1996, 1998, 2001-2003, 2005, 2008  Red Hat, Inc.
+   sysv.S - Copyright (c) 2017  Anthony Green
+          - Copyright (c) 2013  The Written Word, Inc.
+          - Copyright (c) 1996,1998,2001-2003,2005,2008,2010  Red Hat, Inc.
    
    X86 Foreign Function Interface 
 
@@ -24,430 +26,1246 @@
    DEALINGS IN THE SOFTWARE.
    ----------------------------------------------------------------------- */
 
-#ifndef __x86_64__
+#ifdef __i386__
+#ifndef _MSC_VER
 
 #define LIBFFI_ASM	
 #include <fficonfig.h>
 #include <ffi.h>
+#include "internal.h"
 
-.text
-
-.globl ffi_prep_args
-
-	.align 4
-.globl ffi_call_SYSV
-        .type    ffi_call_SYSV,@function
-
-ffi_call_SYSV:
-.LFB1:
-        pushl %ebp
-.LCFI0:
-        movl  %esp,%ebp
-.LCFI1:
-	/* 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	
-
-	call  *28(%ebp)
-
-	/* Load %ecx with the return type code  */
-	movl  20(%ebp),%ecx	
-
-	/* Protect %esi.  We're going to pop it in the epilogue.  */
-	pushl %esi
-
-	/* If the return value pointer is NULL, assume no return value.  */
-	cmpl  $0,24(%ebp)
-	jne  0f
-
-	/* Even if there is no space for the return value, we are 
-	   obliged to handle floating-point values.  */
-	cmpl  $FFI_TYPE_FLOAT,%ecx
-	jne   noretval
-	fstp  %st(0)
-
-        jmp   epilogue
-
-0:
-	call  1f
-
-.Lstore_table:
-	.long	noretval-.Lstore_table	/* FFI_TYPE_VOID */
-	.long	retint-.Lstore_table	/* FFI_TYPE_INT */
-	.long	retfloat-.Lstore_table	/* FFI_TYPE_FLOAT */
-	.long	retdouble-.Lstore_table	/* FFI_TYPE_DOUBLE */
-	.long	retlongdouble-.Lstore_table	/* FFI_TYPE_LONGDOUBLE */
-	.long	retuint8-.Lstore_table	/* FFI_TYPE_UINT8 */
-	.long	retsint8-.Lstore_table	/* FFI_TYPE_SINT8 */
-	.long	retuint16-.Lstore_table	/* FFI_TYPE_UINT16 */
-	.long	retsint16-.Lstore_table	/* FFI_TYPE_SINT16 */
-	.long	retint-.Lstore_table	/* FFI_TYPE_UINT32 */
-	.long	retint-.Lstore_table	/* FFI_TYPE_SINT32 */
-	.long	retint64-.Lstore_table	/* FFI_TYPE_UINT64 */
-	.long	retint64-.Lstore_table	/* FFI_TYPE_SINT64 */
-	.long	retstruct-.Lstore_table	/* FFI_TYPE_STRUCT */
-	.long	retint-.Lstore_table	/* FFI_TYPE_POINTER */
+#define C2(X, Y)  X ## Y
+#define C1(X, Y)  C2(X, Y)
+#ifdef __USER_LABEL_PREFIX__
+# define C(X)     C1(__USER_LABEL_PREFIX__, X)
+#else
+# define C(X)     X
+#endif
 
-1:
-	pop  %esi
-	add  (%esi, %ecx, 4), %esi
-	jmp  *%esi
+#ifdef X86_DARWIN
+# define L(X)     C1(L, X)
+#else
+# define L(X)     C1(.L, X)
+#endif
 
-	/* Sign/zero extend as appropriate.  */
-retsint8:
-	movsbl  %al, %eax
-	jmp  retint
+#ifdef __ELF__
+# define ENDF(X)  .type	X,@function; .size X, . - X
+#else
+# define ENDF(X)
+#endif
 
-retsint16:
-	movswl  %ax, %eax
-	jmp  retint
+/* Handle win32 fastcall name mangling.  */
+#ifdef X86_WIN32
+# define ffi_call_i386		"@ffi_call_i386@8"
+# define ffi_closure_inner	"@ffi_closure_inner@8"
+#else
+# define ffi_call_i386		C(ffi_call_i386)
+# define ffi_closure_inner	C(ffi_closure_inner)
+#endif
 
-retuint8:
-	movzbl  %al, %eax
-	jmp  retint
+/* This macro allows the safe creation of jump tables without an
+   actual table.  The entry points into the table are all 8 bytes.
+   The use of ORG asserts that we're at the correct location.  */
+/* ??? The clang assembler doesn't handle .org with symbolic expressions.  */
+#if defined(__clang__) || defined(__APPLE__) || (defined (__sun__) && defined(__svr4__))
+# define E(BASE, X)	.balign 8
+#else
+# define E(BASE, X)	.balign 8; .org BASE + X * 8
+#endif
 
-retuint16:
-	movzwl  %ax, %eax
-	jmp  retint
-
-retfloat:
-	/* Load %ecx with the pointer to storage for the return value  */
-	movl  24(%ebp),%ecx	
-	fstps (%ecx)
-	jmp   epilogue
-
-retdouble:
-	/* Load %ecx with the pointer to storage for the return value  */
-	movl  24(%ebp),%ecx	
-	fstpl (%ecx)
-	jmp   epilogue
-
-retlongdouble:
-	/* Load %ecx with the pointer to storage for the return value  */
-	movl  24(%ebp),%ecx	
-	fstpt (%ecx)
-	jmp   epilogue
-	
-retint64:	
-	/* Load %ecx with the pointer to storage for the return value  */
-	movl  24(%ebp),%ecx	
-	movl  %eax,0(%ecx)
-	movl  %edx,4(%ecx)
-	jmp   epilogue
-	
-retint:
-	/* Load %ecx with the pointer to storage for the return value  */
-	movl  24(%ebp),%ecx	
-	movl  %eax,0(%ecx)
-
-retstruct:
-	/* Nothing to do!  */
-
-noretval:
-epilogue:
-        popl %esi
-        movl %ebp,%esp
-        popl %ebp
-        ret
-.LFE1:
-.ffi_call_SYSV_end:
-        .size    ffi_call_SYSV,.ffi_call_SYSV_end-ffi_call_SYSV
+	.text
+	.balign	16
+	.globl	ffi_call_i386
+	FFI_HIDDEN(ffi_call_i386)
 
-	.align	4
-FFI_HIDDEN (ffi_closure_SYSV)
-.globl ffi_closure_SYSV
-	.type	ffi_closure_SYSV, @function
-
-ffi_closure_SYSV:
-.LFB2:
-	pushl	%ebp
-.LCFI2:
-	movl	%esp, %ebp
-.LCFI3:
-	subl	$40, %esp
-	leal	-24(%ebp), %edx
-	movl	%edx, -12(%ebp)	/* resp */
-	leal	8(%ebp), %edx
-	movl	%edx, 4(%esp)	/* args = __builtin_dwarf_cfa () */
-	leal	-12(%ebp), %edx
-	movl	%edx, (%esp)	/* &resp */
-#if defined HAVE_HIDDEN_VISIBILITY_ATTRIBUTE || !defined __PIC__
-	call	ffi_closure_SYSV_inner
+/* This is declared as
+
+   void ffi_call_i386(struct call_frame *frame, char *argp)
+        __attribute__((fastcall));
+
+   Thus the arguments are present in
+
+        ecx: frame
+        edx: argp
+*/
+
+ffi_call_i386:
+L(UW0):
+	# cfi_startproc
+	_CET_ENDBR
+#if !HAVE_FASTCALL
+	movl	4(%esp), %ecx
+	movl	8(%esp), %edx
+#endif
+	movl	(%esp), %eax		/* move the return address */
+	movl	%ebp, (%ecx)		/* store %ebp into local frame */
+	movl	%eax, 4(%ecx)		/* store retaddr into local frame */
+
+	/* New stack frame based off ebp.  This is a itty bit of unwind
+	   trickery in that the CFA *has* changed.  There is no easy way
+	   to describe it correctly on entry to the function.  Fortunately,
+	   it doesn't matter too much since at all points we can correctly
+	   unwind back to ffi_call.  Note that the location to which we
+	   moved the return address is (the new) CFA-4, so from the
+	   perspective of the unwind info, it hasn't moved.  */
+	movl	%ecx, %ebp
+L(UW1):
+	# cfi_def_cfa(%ebp, 8)
+	# cfi_rel_offset(%ebp, 0)
+
+	movl	%edx, %esp		/* set outgoing argument stack */
+	movl	20+R_EAX*4(%ebp), %eax	/* set register arguments */
+	movl	20+R_EDX*4(%ebp), %edx
+	movl	20+R_ECX*4(%ebp), %ecx
+
+	call	*8(%ebp)
+
+	movl	12(%ebp), %ecx		/* load return type code */
+	movl	%ebx, 8(%ebp)		/* preserve %ebx */
+L(UW2):
+	# cfi_rel_offset(%ebx, 8)
+
+	andl	$X86_RET_TYPE_MASK, %ecx
+#ifdef __PIC__
+	call	C(__x86.get_pc_thunk.bx)
+L(pc1):
+	leal	L(store_table)-L(pc1)(%ebx, %ecx, 8), %ebx
 #else
-	movl	%ebx, 8(%esp)
-.LCFI7:
-	call	1f
-1:	popl	%ebx
-	addl	$_GLOBAL_OFFSET_TABLE_+[.-1b], %ebx
-	call	ffi_closure_SYSV_inner@PLT
-	movl	8(%esp), %ebx
+	leal	L(store_table)(,%ecx, 8), %ebx
 #endif
-	movl	-12(%ebp), %ecx
-	cmpl	$FFI_TYPE_INT, %eax
-	je	.Lcls_retint
-
-	/* Handle FFI_TYPE_UINT8, FFI_TYPE_SINT8, FFI_TYPE_UINT16,
-	   FFI_TYPE_SINT16, FFI_TYPE_UINT32, FFI_TYPE_SINT32.  */
-	cmpl	$FFI_TYPE_UINT64, %eax
-	jge	0f
-	cmpl	$FFI_TYPE_UINT8, %eax
-	jge	.Lcls_retint
-	
-0:	cmpl	$FFI_TYPE_FLOAT, %eax
-	je	.Lcls_retfloat
-	cmpl	$FFI_TYPE_DOUBLE, %eax
-	je	.Lcls_retdouble
-	cmpl	$FFI_TYPE_LONGDOUBLE, %eax
-	je	.Lcls_retldouble
-	cmpl	$FFI_TYPE_SINT64, %eax
-	je	.Lcls_retllong
-	cmpl	$FFI_TYPE_STRUCT, %eax
-	je	.Lcls_retstruct
-.Lcls_epilogue:
+	movl	16(%ebp), %ecx		/* load result address */
+	_CET_NOTRACK jmp *%ebx
+
+	.balign	8
+L(store_table):
+E(L(store_table), X86_RET_FLOAT)
+	fstps	(%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_DOUBLE)
+	fstpl	(%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_LDOUBLE)
+	fstpt	(%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_SINT8)
+	movsbl	%al, %eax
+	mov	%eax, (%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_SINT16)
+	movswl	%ax, %eax
+	mov	%eax, (%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_UINT8)
+	movzbl	%al, %eax
+	mov	%eax, (%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_UINT16)
+	movzwl	%ax, %eax
+	mov	%eax, (%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_INT64)
+	movl	%edx, 4(%ecx)
+	/* fallthru */
+E(L(store_table), X86_RET_INT32)
+	movl	%eax, (%ecx)
+	/* fallthru */
+E(L(store_table), X86_RET_VOID)
+L(e1):
+	movl	8(%ebp), %ebx
 	movl	%ebp, %esp
 	popl	%ebp
+L(UW3):
+	# cfi_remember_state
+	# cfi_def_cfa(%esp, 4)
+	# cfi_restore(%ebx)
+	# cfi_restore(%ebp)
 	ret
-.Lcls_retint:
-	movl	(%ecx), %eax
-	jmp	.Lcls_epilogue
-.Lcls_retfloat:
-	flds	(%ecx)
-	jmp	.Lcls_epilogue
-.Lcls_retdouble:
-	fldl	(%ecx)
-	jmp	.Lcls_epilogue
-.Lcls_retldouble:
-	fldt	(%ecx)
-	jmp	.Lcls_epilogue
-.Lcls_retllong:
-	movl	(%ecx), %eax
-	movl	4(%ecx), %edx
-	jmp	.Lcls_epilogue
-.Lcls_retstruct:
-	movl	%ebp, %esp
-	popl	%ebp
-	ret	$4
-.LFE2:
-	.size	ffi_closure_SYSV, .-ffi_closure_SYSV
+L(UW4):
+	# cfi_restore_state
 
-#if !FFI_NO_RAW_API
+E(L(store_table), X86_RET_STRUCTPOP)
+	jmp	L(e1)
+E(L(store_table), X86_RET_STRUCTARG)
+	jmp	L(e1)
+E(L(store_table), X86_RET_STRUCT_1B)
+	movb	%al, (%ecx)
+	jmp	L(e1)
+E(L(store_table), X86_RET_STRUCT_2B)
+	movw	%ax, (%ecx)
+	jmp	L(e1)
+
+	/* Fill out the table so that bad values are predictable.  */
+E(L(store_table), X86_RET_UNUSED14)
+	ud2
+E(L(store_table), X86_RET_UNUSED15)
+	ud2
+
+L(UW5):
+	# cfi_endproc
+ENDF(ffi_call_i386)
+
+/* The inner helper is declared as
 
-/* Precalculate for e.g. the Solaris 10/x86 assembler.  */
-#if FFI_TRAMPOLINE_SIZE == 10
-#define RAW_CLOSURE_CIF_OFFSET 12
-#define RAW_CLOSURE_FUN_OFFSET 16
-#define RAW_CLOSURE_USER_DATA_OFFSET 20
-#elif FFI_TRAMPOLINE_SIZE == 24
-#define RAW_CLOSURE_CIF_OFFSET 24
-#define RAW_CLOSURE_FUN_OFFSET 28
-#define RAW_CLOSURE_USER_DATA_OFFSET 32
+   void ffi_closure_inner(struct closure_frame *frame, char *argp)
+	__attribute_((fastcall))
+
+   Thus the arguments are placed in
+
+	ecx:	frame
+	edx:	argp
+*/
+
+/* Macros to help setting up the closure_data structure.  */
+
+#if HAVE_FASTCALL
+# define closure_FS	(40 + 4)
+# define closure_CF	0
 #else
-#define RAW_CLOSURE_CIF_OFFSET ((FFI_TRAMPOLINE_SIZE + 3) & ~3)
-#define RAW_CLOSURE_FUN_OFFSET (RAW_CLOSURE_CIF_OFFSET + 4)
-#define RAW_CLOSURE_USER_DATA_OFFSET (RAW_CLOSURE_FUN_OFFSET + 4)
+# define closure_FS	(8 + 40 + 12)
+# define closure_CF	8
 #endif
-#define CIF_FLAGS_OFFSET 20
 
-	.align	4
-FFI_HIDDEN (ffi_closure_raw_SYSV)
-.globl ffi_closure_raw_SYSV
-	.type	ffi_closure_raw_SYSV, @function
-
-ffi_closure_raw_SYSV:
-.LFB3:
-	pushl	%ebp
-.LCFI4:
-	movl	%esp, %ebp
-.LCFI5:
-	pushl	%esi
-.LCFI6:
-	subl	$36, %esp
-	movl	RAW_CLOSURE_CIF_OFFSET(%eax), %esi	 /* closure->cif */
-	movl	RAW_CLOSURE_USER_DATA_OFFSET(%eax), %edx /* closure->user_data */
-	movl	%edx, 12(%esp)	/* user_data */
-	leal	8(%ebp), %edx	/* __builtin_dwarf_cfa () */
-	movl	%edx, 8(%esp)	/* raw_args */
-	leal	-24(%ebp), %edx
-	movl	%edx, 4(%esp)	/* &res */
-	movl	%esi, (%esp)	/* cif */
-	call	*RAW_CLOSURE_FUN_OFFSET(%eax)		 /* closure->fun */
-	movl	CIF_FLAGS_OFFSET(%esi), %eax		 /* rtype */
-	cmpl	$FFI_TYPE_INT, %eax
-	je	.Lrcls_retint
-
-	/* Handle FFI_TYPE_UINT8, FFI_TYPE_SINT8, FFI_TYPE_UINT16,
-	   FFI_TYPE_SINT16, FFI_TYPE_UINT32, FFI_TYPE_SINT32.  */
-	cmpl	$FFI_TYPE_UINT64, %eax
-	jge	0f
-	cmpl	$FFI_TYPE_UINT8, %eax
-	jge	.Lrcls_retint
-0:
-	cmpl	$FFI_TYPE_FLOAT, %eax
-	je	.Lrcls_retfloat
-	cmpl	$FFI_TYPE_DOUBLE, %eax
-	je	.Lrcls_retdouble
-	cmpl	$FFI_TYPE_LONGDOUBLE, %eax
-	je	.Lrcls_retldouble
-	cmpl	$FFI_TYPE_SINT64, %eax
-	je	.Lrcls_retllong
-.Lrcls_epilogue:
-	addl	$36, %esp
-	popl	%esi
-	popl	%ebp
+#define FFI_CLOSURE_SAVE_REGS		\
+	movl	%eax, closure_CF+16+R_EAX*4(%esp);	\
+	movl	%edx, closure_CF+16+R_EDX*4(%esp);	\
+	movl	%ecx, closure_CF+16+R_ECX*4(%esp)
+
+#define FFI_CLOSURE_COPY_TRAMP_DATA					\
+	movl	FFI_TRAMPOLINE_SIZE(%eax), %edx;	/* copy cif */	\
+	movl	FFI_TRAMPOLINE_SIZE+4(%eax), %ecx;	/* copy fun */	\
+	movl	FFI_TRAMPOLINE_SIZE+8(%eax), %eax;	/* copy user_data */ \
+	movl	%edx, closure_CF+28(%esp);				\
+	movl	%ecx, closure_CF+32(%esp);				\
+	movl	%eax, closure_CF+36(%esp)
+
+#if HAVE_FASTCALL
+# define FFI_CLOSURE_PREP_CALL						\
+	movl	%esp, %ecx;			/* load closure_data */	\
+	leal	closure_FS+4(%esp), %edx;	/* load incoming stack */
+#else
+# define FFI_CLOSURE_PREP_CALL						\
+	leal	closure_CF(%esp), %ecx;		/* load closure_data */	\
+	leal	closure_FS+4(%esp), %edx;	/* load incoming stack */ \
+	movl	%ecx, (%esp);						\
+	movl	%edx, 4(%esp)
+#endif
+
+#define FFI_CLOSURE_CALL_INNER(UWN) \
+	call	ffi_closure_inner
+
+#define FFI_CLOSURE_MASK_AND_JUMP(N, UW)				\
+	andl	$X86_RET_TYPE_MASK, %eax;				\
+	leal	L(C1(load_table,N))(, %eax, 8), %edx;			\
+	movl	closure_CF(%esp), %eax;		/* optimiztic load */	\
+	_CET_NOTRACK jmp *%edx
+
+#ifdef __PIC__
+# if defined X86_DARWIN || defined HAVE_HIDDEN_VISIBILITY_ATTRIBUTE
+#  undef FFI_CLOSURE_MASK_AND_JUMP
+#  define FFI_CLOSURE_MASK_AND_JUMP(N, UW)				\
+	andl	$X86_RET_TYPE_MASK, %eax;				\
+	call	C(__x86.get_pc_thunk.dx);				\
+L(C1(pc,N)):								\
+	leal	L(C1(load_table,N))-L(C1(pc,N))(%edx, %eax, 8), %edx;	\
+	movl	closure_CF(%esp), %eax;		/* optimiztic load */	\
+	_CET_NOTRACK jmp *%edx
+# else
+#  define FFI_CLOSURE_CALL_INNER_SAVE_EBX
+#  undef FFI_CLOSURE_CALL_INNER
+#  define FFI_CLOSURE_CALL_INNER(UWN)					\
+	movl	%ebx, 40(%esp);			/* save ebx */		\
+L(C1(UW,UWN)):								\
+	/* cfi_rel_offset(%ebx, 40); */					\
+	call	C(__x86.get_pc_thunk.bx);	/* load got register */	\
+	addl	$C(_GLOBAL_OFFSET_TABLE_), %ebx;			\
+	call	ffi_closure_inner@PLT
+#  undef FFI_CLOSURE_MASK_AND_JUMP
+#  define FFI_CLOSURE_MASK_AND_JUMP(N, UWN)				\
+	andl	$X86_RET_TYPE_MASK, %eax;				\
+	leal	L(C1(load_table,N))@GOTOFF(%ebx, %eax, 8), %edx;	\
+	movl	40(%esp), %ebx;			/* restore ebx */	\
+L(C1(UW,UWN)):								\
+	/* cfi_restore(%ebx); */					\
+	movl	closure_CF(%esp), %eax;		/* optimiztic load */	\
+	_CET_NOTRACK jmp *%edx
+# endif /* DARWIN || HIDDEN */
+#endif /* __PIC__ */
+
+	.balign	16
+	.globl	C(ffi_go_closure_EAX)
+	FFI_HIDDEN(C(ffi_go_closure_EAX))
+C(ffi_go_closure_EAX):
+L(UW6):
+	# cfi_startproc
+	_CET_ENDBR
+	subl	$closure_FS, %esp
+L(UW7):
+	# cfi_def_cfa_offset(closure_FS + 4)
+	FFI_CLOSURE_SAVE_REGS
+	movl	4(%eax), %edx			/* copy cif */
+	movl	8(%eax), %ecx			/* copy fun */
+	movl	%edx, closure_CF+28(%esp)
+	movl	%ecx, closure_CF+32(%esp)
+	movl	%eax, closure_CF+36(%esp)	/* closure is user_data */
+	jmp	L(do_closure_i386)
+L(UW8):
+	# cfi_endproc
+ENDF(C(ffi_go_closure_EAX))
+
+	.balign	16
+	.globl	C(ffi_go_closure_ECX)
+	FFI_HIDDEN(C(ffi_go_closure_ECX))
+C(ffi_go_closure_ECX):
+L(UW9):
+	# cfi_startproc
+	_CET_ENDBR
+	subl	$closure_FS, %esp
+L(UW10):
+	# cfi_def_cfa_offset(closure_FS + 4)
+	FFI_CLOSURE_SAVE_REGS
+	movl	4(%ecx), %edx			/* copy cif */
+	movl	8(%ecx), %eax			/* copy fun */
+	movl	%edx, closure_CF+28(%esp)
+	movl	%eax, closure_CF+32(%esp)
+	movl	%ecx, closure_CF+36(%esp)	/* closure is user_data */
+	jmp	L(do_closure_i386)
+L(UW11):
+	# cfi_endproc
+ENDF(C(ffi_go_closure_ECX))
+
+/* The closure entry points are reached from the ffi_closure trampoline.
+   On entry, %eax contains the address of the ffi_closure.  */
+
+	.balign	16
+	.globl	C(ffi_closure_i386)
+	FFI_HIDDEN(C(ffi_closure_i386))
+
+C(ffi_closure_i386):
+L(UW12):
+	# cfi_startproc
+	_CET_ENDBR
+	subl	$closure_FS, %esp
+L(UW13):
+	# cfi_def_cfa_offset(closure_FS + 4)
+
+	FFI_CLOSURE_SAVE_REGS
+	FFI_CLOSURE_COPY_TRAMP_DATA
+
+	/* Entry point from preceeding Go closures.  */
+L(do_closure_i386):
+
+	FFI_CLOSURE_PREP_CALL
+	FFI_CLOSURE_CALL_INNER(14)
+	FFI_CLOSURE_MASK_AND_JUMP(2, 15)
+
+	.balign	8
+L(load_table2):
+E(L(load_table2), X86_RET_FLOAT)
+	flds	closure_CF(%esp)
+	jmp	L(e2)
+E(L(load_table2), X86_RET_DOUBLE)
+	fldl	closure_CF(%esp)
+	jmp	L(e2)
+E(L(load_table2), X86_RET_LDOUBLE)
+	fldt	closure_CF(%esp)
+	jmp	L(e2)
+E(L(load_table2), X86_RET_SINT8)
+	movsbl	%al, %eax
+	jmp	L(e2)
+E(L(load_table2), X86_RET_SINT16)
+	movswl	%ax, %eax
+	jmp	L(e2)
+E(L(load_table2), X86_RET_UINT8)
+	movzbl	%al, %eax
+	jmp	L(e2)
+E(L(load_table2), X86_RET_UINT16)
+	movzwl	%ax, %eax
+	jmp	L(e2)
+E(L(load_table2), X86_RET_INT64)
+	movl	closure_CF+4(%esp), %edx
+	jmp	L(e2)
+E(L(load_table2), X86_RET_INT32)
+	nop
+	/* fallthru */
+E(L(load_table2), X86_RET_VOID)
+L(e2):
+	addl	$closure_FS, %esp
+L(UW16):
+	# cfi_adjust_cfa_offset(-closure_FS)
 	ret
-.Lrcls_retint:
-	movl	-24(%ebp), %eax
-	jmp	.Lrcls_epilogue
-.Lrcls_retfloat:
-	flds	-24(%ebp)
-	jmp	.Lrcls_epilogue
-.Lrcls_retdouble:
-	fldl	-24(%ebp)
-	jmp	.Lrcls_epilogue
-.Lrcls_retldouble:
-	fldt	-24(%ebp)
-	jmp	.Lrcls_epilogue
-.Lrcls_retllong:
-	movl	-24(%ebp), %eax
-	movl	-20(%ebp), %edx
-	jmp	.Lrcls_epilogue
-.LFE3:
-	.size	ffi_closure_raw_SYSV, .-ffi_closure_raw_SYSV
+L(UW17):
+	# cfi_adjust_cfa_offset(closure_FS)
+E(L(load_table2), X86_RET_STRUCTPOP)
+	addl	$closure_FS, %esp
+L(UW18):
+	# cfi_adjust_cfa_offset(-closure_FS)
+	ret	$4
+L(UW19):
+	# cfi_adjust_cfa_offset(closure_FS)
+E(L(load_table2), X86_RET_STRUCTARG)
+	jmp	L(e2)
+E(L(load_table2), X86_RET_STRUCT_1B)
+	movzbl	%al, %eax
+	jmp	L(e2)
+E(L(load_table2), X86_RET_STRUCT_2B)
+	movzwl	%ax, %eax
+	jmp	L(e2)
+
+	/* Fill out the table so that bad values are predictable.  */
+E(L(load_table2), X86_RET_UNUSED14)
+	ud2
+E(L(load_table2), X86_RET_UNUSED15)
+	ud2
+
+L(UW20):
+	# cfi_endproc
+ENDF(C(ffi_closure_i386))
+
+	.balign	16
+	.globl	C(ffi_go_closure_STDCALL)
+	FFI_HIDDEN(C(ffi_go_closure_STDCALL))
+C(ffi_go_closure_STDCALL):
+L(UW21):
+	# cfi_startproc
+	_CET_ENDBR
+	subl	$closure_FS, %esp
+L(UW22):
+	# cfi_def_cfa_offset(closure_FS + 4)
+	FFI_CLOSURE_SAVE_REGS
+	movl	4(%ecx), %edx			/* copy cif */
+	movl	8(%ecx), %eax			/* copy fun */
+	movl	%edx, closure_CF+28(%esp)
+	movl	%eax, closure_CF+32(%esp)
+	movl	%ecx, closure_CF+36(%esp)	/* closure is user_data */
+	jmp	L(do_closure_STDCALL)
+L(UW23):
+	# cfi_endproc
+ENDF(C(ffi_go_closure_STDCALL))
+
+/* For REGISTER, we have no available parameter registers, and so we
+   enter here having pushed the closure onto the stack.  */
+
+	.balign	16
+	.globl	C(ffi_closure_REGISTER)
+	FFI_HIDDEN(C(ffi_closure_REGISTER))
+C(ffi_closure_REGISTER):
+L(UW24):
+	# cfi_startproc
+	# cfi_def_cfa(%esp, 8)
+	# cfi_offset(%eip, -8)
+	_CET_ENDBR
+	subl	$closure_FS-4, %esp
+L(UW25):
+	# cfi_def_cfa_offset(closure_FS + 4)
+	FFI_CLOSURE_SAVE_REGS
+	movl	closure_FS-4(%esp), %ecx	/* load retaddr */
+	movl	closure_FS(%esp), %eax		/* load closure */
+	movl	%ecx, closure_FS(%esp)		/* move retaddr */
+	jmp	L(do_closure_REGISTER)
+L(UW26):
+	# cfi_endproc
+ENDF(C(ffi_closure_REGISTER))
+
+/* For STDCALL (and others), we need to pop N bytes of arguments off
+   the stack following the closure.  The amount needing to be popped
+   is returned to us from ffi_closure_inner.  */
+
+	.balign	16
+	.globl	C(ffi_closure_STDCALL)
+	FFI_HIDDEN(C(ffi_closure_STDCALL))
+C(ffi_closure_STDCALL):
+L(UW27):
+	# cfi_startproc
+	_CET_ENDBR
+	subl	$closure_FS, %esp
+L(UW28):
+	# cfi_def_cfa_offset(closure_FS + 4)
+
+	FFI_CLOSURE_SAVE_REGS
+
+	/* Entry point from ffi_closure_REGISTER.  */
+L(do_closure_REGISTER):
+
+	FFI_CLOSURE_COPY_TRAMP_DATA
+
+	/* Entry point from preceeding Go closure.  */
+L(do_closure_STDCALL):
+
+	FFI_CLOSURE_PREP_CALL
+	FFI_CLOSURE_CALL_INNER(29)
+
+	movl	%eax, %ecx
+	shrl	$X86_RET_POP_SHIFT, %ecx	/* isolate pop count */
+	leal	closure_FS(%esp, %ecx), %ecx	/* compute popped esp */
+	movl	closure_FS(%esp), %edx		/* move return address */
+	movl	%edx, (%ecx)
+
+	/* From this point on, the value of %esp upon return is %ecx+4,
+	   and we've copied the return address to %ecx to make return easy.
+	   There's no point in representing this in the unwind info, as
+	   there is always a window between the mov and the ret which
+	   will be wrong from one point of view or another.  */
+
+	FFI_CLOSURE_MASK_AND_JUMP(3, 30)
+
+	.balign	8
+L(load_table3):
+E(L(load_table3), X86_RET_FLOAT)
+	flds    closure_CF(%esp)
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_DOUBLE)
+	fldl    closure_CF(%esp)
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_LDOUBLE)
+	fldt    closure_CF(%esp)
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_SINT8)
+	movsbl  %al, %eax
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_SINT16)
+	movswl  %ax, %eax
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_UINT8)
+	movzbl  %al, %eax
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_UINT16)
+	movzwl  %ax, %eax
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_INT64)
+	movl	closure_CF+4(%esp), %edx
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_INT32)
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_VOID)
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_STRUCTPOP)
+	movl    %ecx, %esp
+	ret
+E(L(load_table3), X86_RET_STRUCTARG)
+	movl	%ecx, %esp
+	ret
+E(L(load_table3), X86_RET_STRUCT_1B)
+	movzbl	%al, %eax
+	movl	%ecx, %esp
+	ret
+E(L(load_table3), X86_RET_STRUCT_2B)
+	movzwl	%ax, %eax
+	movl	%ecx, %esp
+	ret
+
+	/* Fill out the table so that bad values are predictable.  */
+E(L(load_table3), X86_RET_UNUSED14)
+	ud2
+E(L(load_table3), X86_RET_UNUSED15)
+	ud2
+
+L(UW31):
+	# cfi_endproc
+ENDF(C(ffi_closure_STDCALL))
+
+#if defined(FFI_EXEC_STATIC_TRAMP)
+	.balign	16
+	.globl	C(ffi_closure_i386_alt)
+	FFI_HIDDEN(C(ffi_closure_i386_alt))
+C(ffi_closure_i386_alt):
+	/* See the comments above trampoline_code_table. */
+	_CET_ENDBR
+	movl	4(%esp), %eax			/* Load closure in eax */
+	add	$8, %esp			/* Restore the stack */
+	jmp	C(ffi_closure_i386)
+ENDF(C(ffi_closure_i386_alt))
+
+	.balign	16
+	.globl	C(ffi_closure_REGISTER_alt)
+	FFI_HIDDEN(C(ffi_closure_REGISTER_alt))
+C(ffi_closure_REGISTER_alt):
+	/* See the comments above trampoline_code_table. */
+	_CET_ENDBR
+	movl	(%esp), %eax			/* Restore eax */
+	add	$4, %esp			/* Leave closure on stack */
+	jmp	C(ffi_closure_REGISTER)
+ENDF(C(ffi_closure_REGISTER_alt))
+
+	.balign	16
+	.globl	C(ffi_closure_STDCALL_alt)
+	FFI_HIDDEN(C(ffi_closure_STDCALL_alt))
+C(ffi_closure_STDCALL_alt):
+	/* See the comments above trampoline_code_table. */
+	_CET_ENDBR
+	movl	4(%esp), %eax			/* Load closure in eax */
+	add	$8, %esp			/* Restore the stack */
+	jmp	C(ffi_closure_STDCALL)
+ENDF(C(ffi_closure_STDCALL_alt))
+
+/*
+ * Below is the definition of the trampoline code table. Each element in
+ * the code table is a trampoline.
+ *
+ * Because we jump to the trampoline, we place a _CET_ENDBR at the
+ * beginning of the trampoline to mark it as a valid branch target. This is
+ * part of the the Intel CET (Control Flow Enforcement Technology).
+ */
+/*
+ * The trampoline uses register eax.  It saves the original value of eax on
+ * the stack.
+ *
+ * The trampoline has two parameters - target code to jump to and data for
+ * the target code. The trampoline extracts the parameters from its parameter
+ * block (see tramp_table_map()). The trampoline saves the data address on
+ * the stack. Finally, it jumps to the target code.
+ *
+ * The target code can choose to:
+ *
+ * - restore the value of eax
+ * - load the data address in a register
+ * - restore the stack pointer to what it was when the trampoline was invoked.
+ */
+#ifdef ENDBR_PRESENT
+#define X86_DATA_OFFSET		4081
+#define X86_CODE_OFFSET		4070
+#else
+#define X86_DATA_OFFSET		4085
+#define X86_CODE_OFFSET		4074
 #endif
 
-	.section	.eh_frame,EH_FRAME_FLAGS,@progbits
-.Lframe1:
-	.long	.LECIE1-.LSCIE1	/* Length of Common Information Entry */
-.LSCIE1:
-	.long	0x0	/* CIE Identifier Tag */
-	.byte	0x1	/* CIE Version */
+	.align	X86_TRAMP_MAP_SIZE
+	.globl	C(trampoline_code_table)
+	FFI_HIDDEN(C(trampoline_code_table))
+C(trampoline_code_table):
+	.rept	X86_TRAMP_MAP_SIZE / X86_TRAMP_SIZE
+	_CET_ENDBR
+	sub	$8, %esp
+	movl	%eax, (%esp)			/* Save %eax on stack */
+	call	1f				/* Get next PC into %eax */
+	movl	X86_DATA_OFFSET(%eax), %eax	/* Copy data into %eax */
+	movl	%eax, 4(%esp)			/* Save data on stack */
+	call	1f				/* Get next PC into %eax */
+	movl	X86_CODE_OFFSET(%eax), %eax	/* Copy code into %eax */
+	jmp	*%eax				/* Jump to code */
+1:
+	mov	(%esp), %eax
+	ret
+	.align	4
+	.endr
+ENDF(C(trampoline_code_table))
+	.align	X86_TRAMP_MAP_SIZE
+#endif /* FFI_EXEC_STATIC_TRAMP */
+
+#if !FFI_NO_RAW_API
+
+#define raw_closure_S_FS	(16+16+12)
+
+	.balign	16
+	.globl	C(ffi_closure_raw_SYSV)
+	FFI_HIDDEN(C(ffi_closure_raw_SYSV))
+C(ffi_closure_raw_SYSV):
+L(UW32):
+	# cfi_startproc
+	_CET_ENDBR
+	subl	$raw_closure_S_FS, %esp
+L(UW33):
+	# cfi_def_cfa_offset(raw_closure_S_FS + 4)
+	movl	%ebx, raw_closure_S_FS-4(%esp)
+L(UW34):
+	# cfi_rel_offset(%ebx, raw_closure_S_FS-4)
+
+	movl	FFI_TRAMPOLINE_SIZE+8(%eax), %edx	/* load cl->user_data */
+	movl	%edx, 12(%esp)
+	leal	raw_closure_S_FS+4(%esp), %edx		/* load raw_args */
+	movl	%edx, 8(%esp)
+	leal	16(%esp), %edx				/* load &res */
+	movl	%edx, 4(%esp)
+	movl	FFI_TRAMPOLINE_SIZE(%eax), %ebx		/* load cl->cif */
+	movl	%ebx, (%esp)
+	call	*FFI_TRAMPOLINE_SIZE+4(%eax)		/* call cl->fun */
+
+	movl	20(%ebx), %eax				/* load cif->flags */
+	andl	$X86_RET_TYPE_MASK, %eax
 #ifdef __PIC__
-	.ascii "zR\0"	/* CIE Augmentation */
+	call	C(__x86.get_pc_thunk.bx)
+L(pc4):
+	leal	L(load_table4)-L(pc4)(%ebx, %eax, 8), %ecx
 #else
-	.ascii "\0"	/* CIE Augmentation */
+	leal	L(load_table4)(,%eax, 8), %ecx
 #endif
-	.byte	0x1	/* .uleb128 0x1; CIE Code Alignment Factor */
-	.byte	0x7c	/* .sleb128 -4; CIE Data Alignment Factor */
-	.byte	0x8	/* CIE RA Column */
+	movl	raw_closure_S_FS-4(%esp), %ebx
+L(UW35):
+	# cfi_restore(%ebx)
+	movl	16(%esp), %eax				/* Optimistic load */
+	jmp	*%ecx
+
+	.balign	8
+L(load_table4):
+E(L(load_table4), X86_RET_FLOAT)
+	flds	16(%esp)
+	jmp	L(e4)
+E(L(load_table4), X86_RET_DOUBLE)
+	fldl	16(%esp)
+	jmp	L(e4)
+E(L(load_table4), X86_RET_LDOUBLE)
+	fldt	16(%esp)
+	jmp	L(e4)
+E(L(load_table4), X86_RET_SINT8)
+	movsbl	%al, %eax
+	jmp	L(e4)
+E(L(load_table4), X86_RET_SINT16)
+	movswl	%ax, %eax
+	jmp	L(e4)
+E(L(load_table4), X86_RET_UINT8)
+	movzbl	%al, %eax
+	jmp	L(e4)
+E(L(load_table4), X86_RET_UINT16)
+	movzwl	%ax, %eax
+	jmp	L(e4)
+E(L(load_table4), X86_RET_INT64)
+	movl	16+4(%esp), %edx
+	jmp	L(e4)
+E(L(load_table4), X86_RET_INT32)
+	nop
+	/* fallthru */
+E(L(load_table4), X86_RET_VOID)
+L(e4):
+	addl	$raw_closure_S_FS, %esp
+L(UW36):
+	# cfi_adjust_cfa_offset(-raw_closure_S_FS)
+	ret
+L(UW37):
+	# cfi_adjust_cfa_offset(raw_closure_S_FS)
+E(L(load_table4), X86_RET_STRUCTPOP)
+	addl	$raw_closure_S_FS, %esp
+L(UW38):
+	# cfi_adjust_cfa_offset(-raw_closure_S_FS)
+	ret	$4
+L(UW39):
+	# cfi_adjust_cfa_offset(raw_closure_S_FS)
+E(L(load_table4), X86_RET_STRUCTARG)
+	jmp	L(e4)
+E(L(load_table4), X86_RET_STRUCT_1B)
+	movzbl	%al, %eax
+	jmp	L(e4)
+E(L(load_table4), X86_RET_STRUCT_2B)
+	movzwl	%ax, %eax
+	jmp	L(e4)
+
+	/* Fill out the table so that bad values are predictable.  */
+E(L(load_table4), X86_RET_UNUSED14)
+	ud2
+E(L(load_table4), X86_RET_UNUSED15)
+	ud2
+
+L(UW40):
+	# cfi_endproc
+ENDF(C(ffi_closure_raw_SYSV))
+
+#define raw_closure_T_FS	(16+16+8)
+
+	.balign	16
+	.globl	C(ffi_closure_raw_THISCALL)
+	FFI_HIDDEN(C(ffi_closure_raw_THISCALL))
+C(ffi_closure_raw_THISCALL):
+L(UW41):
+	# cfi_startproc
+	_CET_ENDBR
+	/* Rearrange the stack such that %ecx is the first argument.
+	   This means moving the return address.  */
+	popl	%edx
+L(UW42):
+	# cfi_def_cfa_offset(0)
+	# cfi_register(%eip, %edx)
+	pushl	%ecx
+L(UW43):
+	# cfi_adjust_cfa_offset(4)
+	pushl	%edx
+L(UW44):
+	# cfi_adjust_cfa_offset(4)
+	# cfi_rel_offset(%eip, 0)
+	subl	$raw_closure_T_FS, %esp
+L(UW45):
+	# cfi_adjust_cfa_offset(raw_closure_T_FS)
+	movl	%ebx, raw_closure_T_FS-4(%esp)
+L(UW46):
+	# cfi_rel_offset(%ebx, raw_closure_T_FS-4)
+
+	movl	FFI_TRAMPOLINE_SIZE+8(%eax), %edx	/* load cl->user_data */
+	movl	%edx, 12(%esp)
+	leal	raw_closure_T_FS+4(%esp), %edx		/* load raw_args */
+	movl	%edx, 8(%esp)
+	leal	16(%esp), %edx				/* load &res */
+	movl	%edx, 4(%esp)
+	movl	FFI_TRAMPOLINE_SIZE(%eax), %ebx		/* load cl->cif */
+	movl	%ebx, (%esp)
+	call	*FFI_TRAMPOLINE_SIZE+4(%eax)		/* call cl->fun */
+
+	movl	20(%ebx), %eax				/* load cif->flags */
+	andl	$X86_RET_TYPE_MASK, %eax
 #ifdef __PIC__
-	.byte	0x1	/* .uleb128 0x1; Augmentation size */
-	.byte	0x1b	/* FDE Encoding (pcrel sdata4) */
+	call	C(__x86.get_pc_thunk.bx)
+L(pc5):
+	leal	L(load_table5)-L(pc5)(%ebx, %eax, 8), %ecx
+#else
+	leal	L(load_table5)(,%eax, 8), %ecx
 #endif
-	.byte	0xc	/* DW_CFA_def_cfa */
-	.byte	0x4	/* .uleb128 0x4 */
-	.byte	0x4	/* .uleb128 0x4 */
-	.byte	0x88	/* DW_CFA_offset, column 0x8 */
-	.byte	0x1	/* .uleb128 0x1 */
-	.align 4
-.LECIE1:
-.LSFDE1:
-	.long	.LEFDE1-.LASFDE1	/* FDE Length */
-.LASFDE1:
-	.long	.LASFDE1-.Lframe1	/* FDE CIE offset */
-#if defined __PIC__ && defined HAVE_AS_X86_PCREL
-	.long	.LFB1-.	/* FDE initial location */
-#elif defined __PIC__
-	.long	.LFB1@rel
+	movl	raw_closure_T_FS-4(%esp), %ebx
+L(UW47):
+	# cfi_restore(%ebx)
+	movl	16(%esp), %eax				/* Optimistic load */
+	jmp	*%ecx
+
+	.balign	8
+L(load_table5):
+E(L(load_table5), X86_RET_FLOAT)
+	flds	16(%esp)
+	jmp	L(e5)
+E(L(load_table5), X86_RET_DOUBLE)
+	fldl	16(%esp)
+	jmp	L(e5)
+E(L(load_table5), X86_RET_LDOUBLE)
+	fldt	16(%esp)
+	jmp	L(e5)
+E(L(load_table5), X86_RET_SINT8)
+	movsbl	%al, %eax
+	jmp	L(e5)
+E(L(load_table5), X86_RET_SINT16)
+	movswl	%ax, %eax
+	jmp	L(e5)
+E(L(load_table5), X86_RET_UINT8)
+	movzbl	%al, %eax
+	jmp	L(e5)
+E(L(load_table5), X86_RET_UINT16)
+	movzwl	%ax, %eax
+	jmp	L(e5)
+E(L(load_table5), X86_RET_INT64)
+	movl	16+4(%esp), %edx
+	jmp	L(e5)
+E(L(load_table5), X86_RET_INT32)
+	nop
+	/* fallthru */
+E(L(load_table5), X86_RET_VOID)
+L(e5):
+	addl	$raw_closure_T_FS, %esp
+L(UW48):
+	# cfi_adjust_cfa_offset(-raw_closure_T_FS)
+	/* Remove the extra %ecx argument we pushed.  */
+	ret	$4
+L(UW49):
+	# cfi_adjust_cfa_offset(raw_closure_T_FS)
+E(L(load_table5), X86_RET_STRUCTPOP)
+	addl	$raw_closure_T_FS, %esp
+L(UW50):
+	# cfi_adjust_cfa_offset(-raw_closure_T_FS)
+	ret	$8
+L(UW51):
+	# cfi_adjust_cfa_offset(raw_closure_T_FS)
+E(L(load_table5), X86_RET_STRUCTARG)
+	jmp	L(e5)
+E(L(load_table5), X86_RET_STRUCT_1B)
+	movzbl	%al, %eax
+	jmp	L(e5)
+E(L(load_table5), X86_RET_STRUCT_2B)
+	movzwl	%ax, %eax
+	jmp	L(e5)
+
+	/* Fill out the table so that bad values are predictable.  */
+E(L(load_table5), X86_RET_UNUSED14)
+	ud2
+E(L(load_table5), X86_RET_UNUSED15)
+	ud2
+
+L(UW52):
+	# cfi_endproc
+ENDF(C(ffi_closure_raw_THISCALL))
+
+#endif /* !FFI_NO_RAW_API */
+
+#ifdef X86_DARWIN
+/* The linker in use on earlier Darwin needs weak definitions to be
+   placed in a coalesced section.  That section should not be called
+   __TEXT,__text since that would be re-defining the attributes of the
+   .text section (which is an error for earlier tools). Here we use
+   '__textcoal_nt' which is what GCC emits for this.
+   Later linker versions are happy to use a normal section and, after
+   Darwin12 / OSX 10.8, the tools warn that using coalesced sections
+   for this is deprecated so we must switch to avoid build fails and/or
+   deprecation warnings.  */
+# if defined(__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__) &&		\
+   __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ < 1080
+#  define COMDAT(X)							\
+        .section __TEXT,__textcoal_nt,coalesced,pure_instructions;	\
+        .weak_definition X;						\
+        FFI_HIDDEN(X)
+# else
+#  define COMDAT(X)							\
+        .text;								\
+        .weak_definition X;						\
+        FFI_HIDDEN(X)
+# endif
+#elif defined __ELF__ && !(defined(__sun__) && defined(__svr4__))
+# define COMDAT(X)							\
+	.section .text.X,"axG",@progbits,X,comdat;			\
+	.globl	X;							\
+	FFI_HIDDEN(X)
 #else
-	.long	.LFB1
+# define COMDAT(X)
 #endif
-	.long	.LFE1-.LFB1	/* FDE address range */
-#ifdef __PIC__
-	.byte	0x0	/* .uleb128 0x0; Augmentation size */
+
+#if defined(__PIC__)
+	COMDAT(C(__x86.get_pc_thunk.bx))
+C(__x86.get_pc_thunk.bx):
+	movl	(%esp), %ebx
+	ret
+ENDF(C(__x86.get_pc_thunk.bx))
+# if defined X86_DARWIN || defined HAVE_HIDDEN_VISIBILITY_ATTRIBUTE
+	COMDAT(C(__x86.get_pc_thunk.dx))
+C(__x86.get_pc_thunk.dx):
+	movl	(%esp), %edx
+	ret
+ENDF(C(__x86.get_pc_thunk.dx))
+#endif /* DARWIN || HIDDEN */
+#endif /* __PIC__ */
+
+/* Sadly, OSX cctools-as does not understand .cfi directives at all so
+   we build an eh frame by hand.  */
+
+#ifdef __APPLE__
+/* The cctools assembler will try to make a difference between two local
+   symbols into a relocation against, which will not work in the eh (produces
+   link-time fails).
+   To avoid this, we compute the symbol difference with a .set directive and
+   then substitute this value.  */
+# define LEN(N, P)	.set Llen$N$P,L(N)-L(P); .long Llen$N$P
+/* Note, this assume DW_CFA_advance_loc1 fits into 7 bits.  */
+# define ADV(N, P)	.set Ladv$N$P,L(N)-L(P); .byte 2, Ladv$N$P
+/* For historical reasons, the EH reg numbers for SP and FP are swapped from
+   the DWARF ones for 32b Darwin.  */
+# define SP 5
+# define FP 4
+# define ENC 0x10
+#else
+# define LEN(N, P)	.long L(N)-L(P)
+/* Assume DW_CFA_advance_loc1 fits.  */
+# define ADV(N, P)	.byte 2, L(N)-L(P)
+# define SP 4
+# define FP 5
+# define ENC 0x1b
 #endif
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI0-.LFB1
-	.byte	0xe	/* DW_CFA_def_cfa_offset */
-	.byte	0x8	/* .uleb128 0x8 */
-	.byte	0x85	/* DW_CFA_offset, column 0x5 */
-	.byte	0x2	/* .uleb128 0x2 */
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI1-.LCFI0
-	.byte	0xd	/* DW_CFA_def_cfa_register */
-	.byte	0x5	/* .uleb128 0x5 */
-	.align 4
-.LEFDE1:
-.LSFDE2:
-	.long	.LEFDE2-.LASFDE2	/* FDE Length */
-.LASFDE2:
-	.long	.LASFDE2-.Lframe1	/* FDE CIE offset */
-#if defined __PIC__ && defined HAVE_AS_X86_PCREL
-	.long	.LFB2-.	/* FDE initial location */
-#elif defined __PIC__
-	.long	.LFB2@rel
+
+#ifdef HAVE_AS_X86_PCREL
+# define PCREL(X)	X-.
 #else
-	.long	.LFB2
+# define PCREL(X)	X@rel
 #endif
-	.long	.LFE2-.LFB2	/* FDE address range */
-#ifdef __PIC__
-	.byte	0x0	/* .uleb128 0x0; Augmentation size */
+
+#ifdef __APPLE__
+.section __TEXT,__eh_frame,coalesced,no_toc+strip_static_syms+live_support
+EHFrame0:
+#elif defined(X86_WIN32)
+.section .eh_frame,"r"
+#elif defined(HAVE_AS_X86_64_UNWIND_SECTION_TYPE)
+.section .eh_frame,EH_FRAME_FLAGS,@unwind
+#else
+.section .eh_frame,EH_FRAME_FLAGS,@progbits
 #endif
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI2-.LFB2
-	.byte	0xe	/* DW_CFA_def_cfa_offset */
-	.byte	0x8	/* .uleb128 0x8 */
-	.byte	0x85	/* DW_CFA_offset, column 0x5 */
-	.byte	0x2	/* .uleb128 0x2 */
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI3-.LCFI2
-	.byte	0xd	/* DW_CFA_def_cfa_register */
-	.byte	0x5	/* .uleb128 0x5 */
-#if !defined HAVE_HIDDEN_VISIBILITY_ATTRIBUTE && defined __PIC__
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI7-.LCFI3
-	.byte	0x83	/* DW_CFA_offset, column 0x3 */
-	.byte	0xa	/* .uleb128 0xa */
+#ifndef __APPLE__
+/* EH sections are already suitably aligned on Darwin.  */
+	.balign 4
 #endif
-	.align 4
-.LEFDE2:
 
-#if !FFI_NO_RAW_API
+L(CIE):
+	.set	L(set0),L(ECIE)-L(SCIE)
+	.long	L(set0)			/* CIE Length */
+L(SCIE):
+	.long	0			/* CIE Identifier Tag */
+	.byte	1			/* CIE Version */
+	.ascii	"zR\0"			/* CIE Augmentation */
+	.byte	1			/* CIE Code Alignment Factor */
+	.byte	0x7c			/* CIE Data Alignment Factor */
+	.byte	0x8			/* CIE RA Column */
+	.byte	1			/* Augmentation size */
+	.byte	ENC			/* FDE Encoding (pcrel abs/4byte) */
+	.byte	0xc, SP, 4		/* DW_CFA_def_cfa, %esp offset 4 */
+	.byte	0x80+8, 1		/* DW_CFA_offset, %eip offset 1*-4 */
+	.balign 4
+L(ECIE):
+
+	.set	L(set1),L(EFDE1)-L(SFDE1)
+	.long	L(set1)			/* FDE Length */
+L(SFDE1):
+	LEN(SFDE1, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW0))		/* Initial location */
+	LEN(UW5, UW0)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW1, UW0)
+	.byte	0xc, FP, 8		/* DW_CFA_def_cfa, %ebp 8 */
+	.byte	0x80+FP, 2		/* DW_CFA_offset, %ebp 2*-4 */
+	ADV(UW2, UW1)
+	.byte	0x80+3, 0		/* DW_CFA_offset, %ebx 0*-4 */
+	ADV(UW3, UW2)
+	.byte	0xa			/* DW_CFA_remember_state */
+	.byte	0xc, SP, 4		/* DW_CFA_def_cfa, %esp 4 */
+	.byte	0xc0+3			/* DW_CFA_restore, %ebx */
+	.byte	0xc0+FP			/* DW_CFA_restore, %ebp */
+	ADV(UW4, UW3)
+	.byte	0xb			/* DW_CFA_restore_state */
+	.balign	4
+L(EFDE1):
 
-.LSFDE3:
-	.long	.LEFDE3-.LASFDE3	/* FDE Length */
-.LASFDE3:
-	.long	.LASFDE3-.Lframe1	/* FDE CIE offset */
-#if defined __PIC__ && defined HAVE_AS_X86_PCREL
-	.long	.LFB3-.	/* FDE initial location */
-#elif defined __PIC__
-	.long	.LFB3@rel
+	.set	L(set2),L(EFDE2)-L(SFDE2)
+	.long	L(set2)			/* FDE Length */
+L(SFDE2):
+	LEN(SFDE2, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW6))		/* Initial location */
+	LEN(UW8,UW6)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW7, UW6)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE2):
+
+	.set	L(set3),L(EFDE3)-L(SFDE3)
+	.long	L(set3)			/* FDE Length */
+L(SFDE3):
+	LEN(SFDE3, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW9))		/* Initial location */
+	LEN(UW11, UW9)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW10, UW9)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE3):
+
+	.set	L(set4),L(EFDE4)-L(SFDE4)
+	.long	L(set4)			/* FDE Length */
+L(SFDE4):
+	LEN(SFDE4, CIE)		/* FDE CIE offset */
+	.long	PCREL(L(UW12))		/* Initial location */
+	LEN(UW20, UW12)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW13, UW12)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+#ifdef FFI_CLOSURE_CALL_INNER_SAVE_EBX
+	ADV(UW14, UW13)
+	.byte	0x80+3, (40-(closure_FS+4))/-4  /* DW_CFA_offset %ebx */
+	ADV(UW15, UW14)
+	.byte	0xc0+3			/* DW_CFA_restore %ebx */
+	ADV(UW16, UW15)
 #else
-	.long	.LFB3
+	ADV(UW16, UW13)
 #endif
-	.long	.LFE3-.LFB3	/* FDE address range */
-#ifdef __PIC__
-	.byte	0x0	/* .uleb128 0x0; Augmentation size */
+	.byte	0xe, 4			/* DW_CFA_def_cfa_offset */
+	ADV(UW17, UW16)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+	ADV(UW18, UW17)
+	.byte	0xe, 4			/* DW_CFA_def_cfa_offset */
+	ADV(UW19, UW18)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE4):
+
+	.set	L(set5),L(EFDE5)-L(SFDE5)
+	.long	L(set5)			/* FDE Length */
+L(SFDE5):
+	LEN(SFDE5, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW21))		/* Initial location */
+	LEN(UW23, UW21)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW22, UW21)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE5):
+
+	.set	L(set6),L(EFDE6)-L(SFDE6)
+	.long	L(set6)			/* FDE Length */
+L(SFDE6):
+	LEN(SFDE6, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW24))		/* Initial location */
+	LEN(UW26, UW24)			/* Address range */
+	.byte	0			/* Augmentation size */
+	.byte	0xe, 8			/* DW_CFA_def_cfa_offset */
+	.byte	0x80+8, 2		/* DW_CFA_offset %eip, 2*-4 */
+	ADV(UW25, UW24)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE6):
+
+	.set	L(set7),L(EFDE7)-L(SFDE7)
+	.long	L(set7)			/* FDE Length */
+L(SFDE7):
+	LEN(SFDE7, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW27))		/* Initial location */
+	LEN(UW31, UW27)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW28, UW27)
+	.byte	0xe, closure_FS+4	/* DW_CFA_def_cfa_offset */
+#ifdef FFI_CLOSURE_CALL_INNER_SAVE_EBX
+	ADV(UW29, UW28)
+	.byte	0x80+3, (40-(closure_FS+4))/-4  /* DW_CFA_offset %ebx */
+	ADV(UW30, UW29)
+	.byte	0xc0+3			/* DW_CFA_restore %ebx */
 #endif
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI4-.LFB3
-	.byte	0xe	/* DW_CFA_def_cfa_offset */
-	.byte	0x8	/* .uleb128 0x8 */
-	.byte	0x85	/* DW_CFA_offset, column 0x5 */
-	.byte	0x2	/* .uleb128 0x2 */
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI5-.LCFI4
-	.byte	0xd	/* DW_CFA_def_cfa_register */
-	.byte	0x5	/* .uleb128 0x5 */
-	.byte	0x4	/* DW_CFA_advance_loc4 */
-	.long	.LCFI6-.LCFI5
-	.byte	0x86	/* DW_CFA_offset, column 0x6 */
-	.byte	0x3	/* .uleb128 0x3 */
-	.align 4
-.LEFDE3:
+	.balign	4
+L(EFDE7):
+#if !FFI_NO_RAW_API
+	.set	L(set8),L(EFDE8)-L(SFDE8)
+	.long	L(set8)			/* FDE Length */
+L(SFDE8):
+	LEN(SFDE8, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW32))		/* Initial location */
+	LEN(UW40, UW32)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW33, UW32)
+	.byte	0xe, raw_closure_S_FS+4	/* DW_CFA_def_cfa_offset */
+	ADV(UW34, UW33)
+	.byte	0x80+3, 2		/* DW_CFA_offset %ebx 2*-4 */
+	ADV(UW35, UW34)
+	.byte	0xc0+3			/* DW_CFA_restore %ebx */
+	ADV(UW36, UW35)
+	.byte	0xe, 4			/* DW_CFA_def_cfa_offset */
+	ADV(UW37, UW36)
+	.byte	0xe, raw_closure_S_FS+4	/* DW_CFA_def_cfa_offset */
+	ADV(UW38, UW37)
+	.byte	0xe, 4			/* DW_CFA_def_cfa_offset */
+	ADV(UW39, UW38)
+	.byte	0xe, raw_closure_S_FS+4	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE8):
+
+	.set	L(set9),L(EFDE9)-L(SFDE9)
+	.long	L(set9)			/* FDE Length */
+L(SFDE9):
+	LEN(SFDE9, CIE)			/* FDE CIE offset */
+	.long	PCREL(L(UW41))		/* Initial location */
+	LEN(UW52, UW41)			/* Address range */
+	.byte	0			/* Augmentation size */
+	ADV(UW42, UW41)
+	.byte	0xe, 0			/* DW_CFA_def_cfa_offset */
+	.byte	0x9, 8, 2		/* DW_CFA_register %eip, %edx */
+	ADV(UW43, UW42)
+	.byte	0xe, 4			/* DW_CFA_def_cfa_offset */
+	ADV(UW44, UW43)
+	.byte	0xe, 8			/* DW_CFA_def_cfa_offset */
+	.byte	0x80+8, 2		/* DW_CFA_offset %eip 2*-4 */
+	ADV(UW45, UW44)
+	.byte	0xe, raw_closure_T_FS+8	/* DW_CFA_def_cfa_offset */
+	ADV(UW46, UW45)
+	.byte	0x80+3, 3		/* DW_CFA_offset %ebx 3*-4 */
+	ADV(UW47, UW46)
+	.byte	0xc0+3			/* DW_CFA_restore %ebx */
+	ADV(UW48, UW47)
+	.byte	0xe, 8			/* DW_CFA_def_cfa_offset */
+	ADV(UW49, UW48)
+	.byte	0xe, raw_closure_T_FS+8	/* DW_CFA_def_cfa_offset */
+	ADV(UW50, UW49)
+	.byte	0xe, 8			/* DW_CFA_def_cfa_offset */
+	ADV(UW51, UW50)
+	.byte	0xe, raw_closure_T_FS+8	/* DW_CFA_def_cfa_offset */
+	.balign	4
+L(EFDE9):
+#endif /* !FFI_NO_RAW_API */
 
+#ifdef _WIN32
+	.def	 @feat.00;
+	.scl	3;
+	.type	0;
+	.endef
+	.globl	@feat.00
+@feat.00 = 1
 #endif
 
-#endif /* ifndef __x86_64__ */
+#if defined(__APPLE__)
+    .subsections_via_symbols
+# if defined(__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__) &&		\
+   __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ >= 1070 && __clang__
+/* compact unwind is not used with GCC at present, was not present before 10.6
+   but has some bugs there, so do not emit until 10.7.  */
+    .section __LD,__compact_unwind,regular,debug
+
+    /* compact unwind for ffi_call_i386 */
+    .long    C(ffi_call_i386)
+    .set     L1,L(UW5)-L(UW0)
+    .long    L1
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_go_closure_EAX */
+    .long    C(ffi_go_closure_EAX)
+    .set     L2,L(UW8)-L(UW6)
+    .long    L2
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_go_closure_ECX */
+    .long    C(ffi_go_closure_ECX)
+    .set     L3,L(UW11)-L(UW9)
+    .long    L3
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_closure_i386 */
+    .long    C(ffi_closure_i386)
+    .set     L4,L(UW20)-L(UW12)
+    .long    L4
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_go_closure_STDCALL */
+    .long    C(ffi_go_closure_STDCALL)
+    .set     L5,L(UW23)-L(UW21)
+    .long    L5
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_closure_REGISTER */
+    .long    C(ffi_closure_REGISTER)
+    .set     L6,L(UW26)-L(UW24)
+    .long    L6
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_closure_STDCALL */
+    .long    C(ffi_closure_STDCALL)
+    .set     L7,L(UW31)-L(UW27)
+    .long    L7
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_closure_raw_SYSV */
+    .long    C(ffi_closure_raw_SYSV)
+    .set     L8,L(UW40)-L(UW32)
+    .long    L8
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+
+    /* compact unwind for ffi_closure_raw_THISCALL */
+    .long    C(ffi_closure_raw_THISCALL)
+    .set     L9,L(UW52)-L(UW41)
+    .long    L9
+    .long    0x04000000 /* use dwarf unwind info */
+    .long    0
+    .long    0
+#endif /* use compact unwind */
+#endif /* __APPLE__ */
+
+#endif /* ifndef _MSC_VER */
+
+#endif /* ifdef __i386__ */
 
 #if defined __ELF__ && defined __linux__
 	.section	.note.GNU-stack,"",@progbits