ffi 1.9.3-x64-mingw32

data/ext/ffi_c/libffi/src/x86/darwin64.S

@@ -0,0 +1,416 @@
+/* -----------------------------------------------------------------------
+   darwin64.S - Copyright (c) 2006 Free Software Foundation, Inc.
+	        Copyright (c) 2008 Red Hat, Inc.
+   derived from unix64.S
+
+   x86-64 Foreign Function Interface for Darwin.
+
+   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 AUTHOR BE LIABLE FOR ANY CLAIM, DAMAGES OR
+   OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+   ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+   OTHER DEALINGS IN THE SOFTWARE.
+   ----------------------------------------------------------------------- */
+
+#ifdef __x86_64__
+#define LIBFFI_ASM
+#include <fficonfig.h>
+#include <ffi.h>
+
+	.file "darwin64.S"
+.text
+
+/* ffi_call_unix64 (void *args, unsigned long bytes, unsigned flags,
+		    void *raddr, void (*fnaddr)(void));
+
+   Bit o trickiness here -- ARGS+BYTES is the base of the stack frame
+   for this function.  This has been allocated by ffi_call.  We also
+   deallocate some of the stack that has been alloca'd.  */
+
+	.align	3
+	.globl	_ffi_call_unix64
+
+_ffi_call_unix64:
+LUW0:
+	movq	(%rsp), %r10		/* Load return address.  */
+	leaq	(%rdi, %rsi), %rax	/* Find local stack base.  */
+	movq	%rdx, (%rax)		/* Save flags.  */
+	movq	%rcx, 8(%rax)		/* Save raddr.  */
+	movq	%rbp, 16(%rax)		/* Save old frame pointer.  */
+	movq	%r10, 24(%rax)		/* Relocate return address.  */
+	movq	%rax, %rbp		/* Finalize local stack frame.  */
+LUW1:
+	movq	%rdi, %r10		/* Save a copy of the register area. */
+	movq	%r8, %r11		/* Save a copy of the target fn.  */
+	movl	%r9d, %eax		/* Set number of SSE registers.  */
+
+	/* Load up all argument registers.  */
+	movq	(%r10), %rdi
+	movq	8(%r10), %rsi
+	movq	16(%r10), %rdx
+	movq	24(%r10), %rcx
+	movq	32(%r10), %r8
+	movq	40(%r10), %r9
+	testl	%eax, %eax
+	jnz	Lload_sse
+Lret_from_load_sse:
+
+	/* Deallocate the reg arg area.  */
+	leaq	176(%r10), %rsp
+
+	/* Call the user function.  */
+	call	*%r11
+
+	/* Deallocate stack arg area; local stack frame in redzone.  */
+	leaq	24(%rbp), %rsp
+
+	movq	0(%rbp), %rcx		/* Reload flags.  */
+	movq	8(%rbp), %rdi		/* Reload raddr.  */
+	movq	16(%rbp), %rbp		/* Reload old frame pointer.  */
+LUW2:
+
+	/* The first byte of the flags contains the FFI_TYPE.  */
+	movzbl	%cl, %r10d
+	leaq	Lstore_table(%rip), %r11
+	movslq	(%r11, %r10, 4), %r10
+	addq	%r11, %r10
+	jmp	*%r10
+
+Lstore_table:
+	.long	Lst_void-Lstore_table		/* FFI_TYPE_VOID */
+	.long	Lst_sint32-Lstore_table		/* FFI_TYPE_INT */
+	.long	Lst_float-Lstore_table		/* FFI_TYPE_FLOAT */
+	.long	Lst_double-Lstore_table		/* FFI_TYPE_DOUBLE */
+	.long	Lst_ldouble-Lstore_table	/* FFI_TYPE_LONGDOUBLE */
+	.long	Lst_uint8-Lstore_table		/* FFI_TYPE_UINT8 */
+	.long	Lst_sint8-Lstore_table		/* FFI_TYPE_SINT8 */
+	.long	Lst_uint16-Lstore_table		/* FFI_TYPE_UINT16 */
+	.long	Lst_sint16-Lstore_table		/* FFI_TYPE_SINT16 */
+	.long	Lst_uint32-Lstore_table		/* FFI_TYPE_UINT32 */
+	.long	Lst_sint32-Lstore_table		/* FFI_TYPE_SINT32 */
+	.long	Lst_int64-Lstore_table		/* FFI_TYPE_UINT64 */
+	.long	Lst_int64-Lstore_table		/* FFI_TYPE_SINT64 */
+	.long	Lst_struct-Lstore_table		/* FFI_TYPE_STRUCT */
+	.long	Lst_int64-Lstore_table		/* FFI_TYPE_POINTER */
+
+	.text
+	.align	3
+Lst_void:
+	ret
+	.align	3
+Lst_uint8:
+	movzbq	%al, %rax
+	movq	%rax, (%rdi)
+	ret
+	.align	3
+Lst_sint8:
+	movsbq	%al, %rax
+	movq	%rax, (%rdi)
+	ret
+	.align	3
+Lst_uint16:
+	movzwq	%ax, %rax
+	movq	%rax, (%rdi)
+	.align	3
+Lst_sint16:
+	movswq	%ax, %rax
+	movq	%rax, (%rdi)
+	ret
+	.align	3
+Lst_uint32:
+	movl	%eax, %eax
+	movq	%rax, (%rdi)
+	.align	3
+Lst_sint32:
+	cltq
+	movq	%rax, (%rdi)
+	ret
+	.align	3
+Lst_int64:
+	movq	%rax, (%rdi)
+	ret
+	.align	3
+Lst_float:
+	movss	%xmm0, (%rdi)
+	ret
+	.align	3
+Lst_double:
+	movsd	%xmm0, (%rdi)
+	ret
+Lst_ldouble:
+	fstpt	(%rdi)
+	ret
+	.align	3
+Lst_struct:
+	leaq	-20(%rsp), %rsi		/* Scratch area in redzone.  */
+
+	/* We have to locate the values now, and since we don't want to
+	   write too much data into the user's return value, we spill the
+	   value to a 16 byte scratch area first.  Bits 8, 9, and 10
+	   control where the values are located.  Only one of the three
+	   bits will be set; see ffi_prep_cif_machdep for the pattern.  */
+	movd	%xmm0, %r10
+	movd	%xmm1, %r11
+	testl	$0x100, %ecx
+	cmovnz	%rax, %rdx
+	cmovnz	%r10, %rax
+	testl	$0x200, %ecx
+	cmovnz	%r10, %rdx
+	testl	$0x400, %ecx
+	cmovnz	%r10, %rax
+	cmovnz	%r11, %rdx
+	movq	%rax, (%rsi)
+	movq	%rdx, 8(%rsi)
+
+	/* Bits 12-31 contain the true size of the structure.  Copy from
+	   the scratch area to the true destination.  */
+	shrl	$12, %ecx
+	rep movsb
+	ret
+
+	/* Many times we can avoid loading any SSE registers at all.
+	   It's not worth an indirect jump to load the exact set of
+	   SSE registers needed; zero or all is a good compromise.  */
+	.align	3
+LUW3:
+Lload_sse:
+	movdqa	48(%r10), %xmm0
+	movdqa	64(%r10), %xmm1
+	movdqa	80(%r10), %xmm2
+	movdqa	96(%r10), %xmm3
+	movdqa	112(%r10), %xmm4
+	movdqa	128(%r10), %xmm5
+	movdqa	144(%r10), %xmm6
+	movdqa	160(%r10), %xmm7
+	jmp	Lret_from_load_sse
+
+LUW4:
+	.align	3
+	.globl	_ffi_closure_unix64
+
+_ffi_closure_unix64:
+LUW5:
+	/* The carry flag is set by the trampoline iff SSE registers
+	   are used.  Don't clobber it before the branch instruction.  */
+	leaq    -200(%rsp), %rsp
+LUW6:
+	movq	%rdi, (%rsp)
+	movq    %rsi, 8(%rsp)
+	movq    %rdx, 16(%rsp)
+	movq    %rcx, 24(%rsp)
+	movq    %r8, 32(%rsp)
+	movq    %r9, 40(%rsp)
+	jc      Lsave_sse
+Lret_from_save_sse:
+
+	movq	%r10, %rdi
+	leaq	176(%rsp), %rsi
+	movq	%rsp, %rdx
+	leaq	208(%rsp), %rcx
+	call	_ffi_closure_unix64_inner
+
+	/* Deallocate stack frame early; return value is now in redzone.  */
+	addq	$200, %rsp
+LUW7:
+
+	/* The first byte of the return value contains the FFI_TYPE.  */
+	movzbl	%al, %r10d
+	leaq	Lload_table(%rip), %r11
+	movslq	(%r11, %r10, 4), %r10
+	addq	%r11, %r10
+	jmp	*%r10
+
+Lload_table:
+	.long	Lld_void-Lload_table		/* FFI_TYPE_VOID */
+	.long	Lld_int32-Lload_table		/* FFI_TYPE_INT */
+	.long	Lld_float-Lload_table		/* FFI_TYPE_FLOAT */
+	.long	Lld_double-Lload_table		/* FFI_TYPE_DOUBLE */
+	.long	Lld_ldouble-Lload_table		/* FFI_TYPE_LONGDOUBLE */
+	.long	Lld_int8-Lload_table		/* FFI_TYPE_UINT8 */
+	.long	Lld_int8-Lload_table		/* FFI_TYPE_SINT8 */
+	.long	Lld_int16-Lload_table		/* FFI_TYPE_UINT16 */
+	.long	Lld_int16-Lload_table		/* FFI_TYPE_SINT16 */
+	.long	Lld_int32-Lload_table		/* FFI_TYPE_UINT32 */
+	.long	Lld_int32-Lload_table		/* FFI_TYPE_SINT32 */
+	.long	Lld_int64-Lload_table		/* FFI_TYPE_UINT64 */
+	.long	Lld_int64-Lload_table		/* FFI_TYPE_SINT64 */
+	.long	Lld_struct-Lload_table		/* FFI_TYPE_STRUCT */
+	.long	Lld_int64-Lload_table		/* FFI_TYPE_POINTER */
+
+	.text
+	.align	3
+Lld_void:
+	ret
+	.align	3
+Lld_int8:
+	movzbl	-24(%rsp), %eax
+	ret
+	.align	3
+Lld_int16:
+	movzwl	-24(%rsp), %eax
+	ret
+	.align	3
+Lld_int32:
+	movl	-24(%rsp), %eax
+	ret
+	.align	3
+Lld_int64:
+	movq	-24(%rsp), %rax
+	ret
+	.align	3
+Lld_float:
+	movss	-24(%rsp), %xmm0
+	ret
+	.align	3
+Lld_double:
+	movsd	-24(%rsp), %xmm0
+	ret
+	.align	3
+Lld_ldouble:
+	fldt	-24(%rsp)
+	ret
+	.align	3
+Lld_struct:
+	/* There are four possibilities here, %rax/%rdx, %xmm0/%rax,
+	   %rax/%xmm0, %xmm0/%xmm1.  We collapse two by always loading
+	   both rdx and xmm1 with the second word.  For the remaining,
+	   bit 8 set means xmm0 gets the second word, and bit 9 means
+	   that rax gets the second word.  */
+	movq	-24(%rsp), %rcx
+	movq	-16(%rsp), %rdx
+	movq	-16(%rsp), %xmm1
+	testl	$0x100, %eax
+	cmovnz	%rdx, %rcx
+	movd	%rcx, %xmm0
+	testl	$0x200, %eax
+	movq	-24(%rsp), %rax
+	cmovnz	%rdx, %rax
+	ret
+
+	/* See the comment above Lload_sse; the same logic applies here.  */
+	.align	3
+LUW8:
+Lsave_sse:
+	movdqa	%xmm0, 48(%rsp)
+	movdqa	%xmm1, 64(%rsp)
+	movdqa	%xmm2, 80(%rsp)
+	movdqa	%xmm3, 96(%rsp)
+	movdqa	%xmm4, 112(%rsp)
+	movdqa	%xmm5, 128(%rsp)
+	movdqa	%xmm6, 144(%rsp)
+	movdqa	%xmm7, 160(%rsp)
+	jmp	Lret_from_save_sse
+
+LUW9:
+.section __TEXT,__eh_frame,coalesced,no_toc+strip_static_syms+live_support
+EH_frame1:
+	.set	L$set$0,LECIE1-LSCIE1		/* CIE Length */
+	.long	L$set$0
+LSCIE1:
+	.long	0x0		/* CIE Identifier Tag */
+	.byte	0x1		/* CIE Version */
+	.ascii	"zR\0"		/* CIE Augmentation */
+	.byte	0x1		/* uleb128 0x1; CIE Code Alignment Factor */
+	.byte	0x78		/* sleb128 -8; CIE Data Alignment Factor */
+	.byte	0x10		/* CIE RA Column */
+	.byte	0x1		/* uleb128 0x1; Augmentation size */
+	.byte	0x10		/* FDE Encoding (pcrel sdata4) */
+	.byte	0xc		/* DW_CFA_def_cfa, %rsp offset 8 */
+	.byte	0x7		/* uleb128 0x7 */
+	.byte	0x8		/* uleb128 0x8 */
+	.byte	0x90		/* DW_CFA_offset, column 0x10 */
+	.byte	0x1
+	.align	3
+LECIE1:
+	.globl _ffi_call_unix64.eh
+_ffi_call_unix64.eh:
+LSFDE1:
+	.set	L$set$1,LEFDE1-LASFDE1	/* FDE Length */
+	.long	L$set$1
+LASFDE1:
+	.long	LASFDE1-EH_frame1	/* FDE CIE offset */
+	.quad	LUW0-.			/* FDE initial location */
+	.set	L$set$2,LUW4-LUW0	/* FDE address range */
+	.quad	L$set$2
+	.byte	0x0			/* Augmentation size */
+	.byte	0x4			/* DW_CFA_advance_loc4 */
+	.set	L$set$3,LUW1-LUW0
+	.long	L$set$3
+
+	/* New stack frame based off rbp.  This is a itty bit of unwind
+	   trickery in that the CFA *has* changed.  There is no easy way
+	   to describe it correctly on entry to the function.  Fortunately,
+	   it doesn't matter too much since at all points we can correctly
+	   unwind back to ffi_call.  Note that the location to which we
+	   moved the return address is (the new) CFA-8, so from the
+	   perspective of the unwind info, it hasn't moved.  */
+	.byte	0xc			/* DW_CFA_def_cfa, %rbp offset 32 */
+	.byte	0x6
+	.byte	0x20
+	.byte	0x80+6			/* DW_CFA_offset, %rbp offset 2*-8 */
+	.byte	0x2
+	.byte	0xa			/* DW_CFA_remember_state */
+
+	.byte	0x4			/* DW_CFA_advance_loc4 */
+	.set	L$set$4,LUW2-LUW1
+	.long	L$set$4
+	.byte	0xc			/* DW_CFA_def_cfa, %rsp offset 8 */
+	.byte	0x7
+	.byte	0x8
+	.byte	0xc0+6			/* DW_CFA_restore, %rbp */
+
+	.byte	0x4			/* DW_CFA_advance_loc4 */
+	.set	L$set$5,LUW3-LUW2
+	.long	L$set$5
+	.byte	0xb			/* DW_CFA_restore_state */
+
+	.align	3
+LEFDE1:
+	.globl _ffi_closure_unix64.eh
+_ffi_closure_unix64.eh:
+LSFDE3:
+	.set	L$set$6,LEFDE3-LASFDE3	/* FDE Length */
+	.long	L$set$6
+LASFDE3:
+	.long	LASFDE3-EH_frame1	/* FDE CIE offset */
+	.quad	LUW5-.			/* FDE initial location */
+	.set	L$set$7,LUW9-LUW5	/* FDE address range */
+	.quad	L$set$7
+	.byte	0x0			/* Augmentation size */
+
+	.byte	0x4			/* DW_CFA_advance_loc4 */
+	.set	L$set$8,LUW6-LUW5
+	.long	L$set$8
+	.byte	0xe			/* DW_CFA_def_cfa_offset */
+	.byte	208,1			/* uleb128 208 */
+	.byte	0xa			/* DW_CFA_remember_state */
+
+	.byte	0x4			/* DW_CFA_advance_loc4 */
+	.set	L$set$9,LUW7-LUW6
+	.long	L$set$9
+	.byte	0xe			/* DW_CFA_def_cfa_offset */
+	.byte	0x8
+
+	.byte	0x4			/* DW_CFA_advance_loc4 */
+	.set	L$set$10,LUW8-LUW7
+	.long	L$set$10
+	.byte	0xb			/* DW_CFA_restore_state */
+
+	.align	3
+LEFDE3:
+	.subsections_via_symbols
+
+#endif /* __x86_64__ */

data/ext/ffi_c/libffi/src/x86/ffi.c

@@ -0,0 +1,644 @@
+/* -----------------------------------------------------------------------
+   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, 2010  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:
+    case FFI_TYPE_UINT8:
+    case FFI_TYPE_UINT16:
+    case FFI_TYPE_SINT8:
+    case FFI_TYPE_SINT16:
+#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:
+      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 */
+