c3-lang/libffi

diff --git a/src/x86/ffi.c b/src/x86/ffi.c
index 1dd00eb..cb5f634 100644
--- a/src/x86/ffi.c
+++ b/src/x86/ffi.c
@@ -419,6 +419,8 @@ 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)));
+unsigned int FFI_HIDDEN ffi_closure_WIN32_inner (ffi_closure *, void **, void *)
+     __attribute__ ((regparm(1)));
 void FFI_HIDDEN ffi_closure_raw_SYSV (ffi_raw_closure *)
      __attribute__ ((regparm(1)));
 #ifdef X86_WIN32
@@ -426,12 +428,10 @@ void FFI_HIDDEN ffi_closure_raw_THISCALL (ffi_raw_closure *)
      __attribute__ ((regparm(1)));
 #endif
 #ifndef X86_WIN64
-void FFI_HIDDEN ffi_closure_STDCALL (ffi_closure *)
-     __attribute__ ((regparm(1)));
-void FFI_HIDDEN ffi_closure_THISCALL (ffi_closure *)
-     __attribute__ ((regparm(1)));
-void FFI_HIDDEN ffi_closure_FASTCALL (ffi_closure *)
-     __attribute__ ((regparm(1)));
+void FFI_HIDDEN ffi_closure_STDCALL (ffi_closure *);
+void FFI_HIDDEN ffi_closure_THISCALL (ffi_closure *);
+void FFI_HIDDEN ffi_closure_FASTCALL (ffi_closure *);
+void FFI_HIDDEN ffi_closure_REGISTER (ffi_closure *);
 #else
 void FFI_HIDDEN ffi_closure_win64 (ffi_closure *);
 #endif
@@ -490,6 +490,29 @@ ffi_closure_SYSV_inner (ffi_closure *closure, void **respp, void *args)
 
   return cif->flags;
 }
+
+unsigned int FFI_HIDDEN __attribute__ ((regparm(1)))
+ffi_closure_WIN32_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->bytes;
+}
 #endif /* !X86_WIN64 */
 
 static void
@@ -587,7 +610,7 @@ ffi_prep_incoming_args_SYSV(char *stack, void **rvalue, void **avalue,
    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 char*) &__tramp[5] = 0xe9; \
    *(unsigned int*)  &__tramp[6] = __dis; /* jmp __fun  */ \
  }
 
@@ -618,15 +641,15 @@ ffi_prep_incoming_args_SYSV(char *stack, void **rvalue, void **avalue,
    *(unsigned short*)  &__tramp[50] = (__size + 8); /* ret (__size + 8)  */ \
  }
 
-#define FFI_INIT_TRAMPOLINE_STDCALL(TRAMP,FUN,CTX)  \
+#define FFI_INIT_TRAMPOLINE_WIN32(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] = 0xe8; \
-   *(unsigned int*)  &__tramp[6] = __dis; /* call __fun  */ \
+   *(unsigned char*) &__tramp[0] = 0x68; \
+   *(unsigned int*)  &__tramp[1] = __ctx; /* push __ctx */ \
+   *(unsigned char*) &__tramp[5] = 0xe9; \
+   *(unsigned int*)  &__tramp[6] = __dis; /* jmp __fun  */ \
  }
 
 /* the cif must already be prep'ed */
@@ -656,21 +679,27 @@ ffi_prep_closure_loc (ffi_closure* closure,
                            &ffi_closure_SYSV,
                            (void*)codeloc);
     }
+  else if (cif->abi == FFI_REGISTER)
+    {
+      FFI_INIT_TRAMPOLINE_WIN32 (&closure->tramp[0],
+                                   &ffi_closure_REGISTER,
+                                   (void*)codeloc);
+    }
   else if (cif->abi == FFI_FASTCALL)
     {
-      FFI_INIT_TRAMPOLINE_STDCALL (&closure->tramp[0],
+      FFI_INIT_TRAMPOLINE_WIN32 (&closure->tramp[0],
                                    &ffi_closure_FASTCALL,
                                    (void*)codeloc);
     }
   else if (cif->abi == FFI_THISCALL)
     {
-      FFI_INIT_TRAMPOLINE_STDCALL (&closure->tramp[0],
+      FFI_INIT_TRAMPOLINE_WIN32 (&closure->tramp[0],
                                    &ffi_closure_THISCALL,
                                    (void*)codeloc);
     }
-  else if (cif->abi == FFI_STDCALL)
+  else if (cif->abi == FFI_STDCALL || cif->abi == FFI_PASCAL)
     {
-      FFI_INIT_TRAMPOLINE_STDCALL (&closure->tramp[0],
+      FFI_INIT_TRAMPOLINE_WIN32 (&closure->tramp[0],
                                    &ffi_closure_STDCALL,
                                    (void*)codeloc);
     }
diff --git a/src/x86/win32.S b/src/x86/win32.S
index 1104ead..96e27df 100644
--- a/src/x86/win32.S
+++ b/src/x86/win32.S
@@ -34,8 +34,8 @@
 #include <fficonfig.h>
 #include <ffi.h>
 
-#define CIF_ABI_OFFSET 0
 #define CIF_BYTES_OFFSET 16
+#define CIF_FLAGS_OFFSET 20
 
 #ifdef _MSC_VER
 
@@ -45,6 +45,7 @@
 .MODEL FLAT, C
 
 EXTRN ffi_closure_SYSV_inner:NEAR
+EXTRN ffi_closure_WIN32_inner:NEAR
 
 _TEXT SEGMENT
 
@@ -215,7 +216,7 @@ ffi_closure_THISCALL PROC NEAR
 ffi_closure_THISCALL ENDP
 
 ffi_closure_FASTCALL PROC NEAR
-        ;; Insert the register argument on the stack as the first argument
+        ;; Insert the 2 register arguments on the stack as the first argument
         xchg	DWORD PTR [esp+4], edx
         xchg	DWORD PTR [esp], ecx
         push	edx
@@ -223,6 +224,16 @@ ffi_closure_FASTCALL PROC NEAR
         jmp	ffi_closure_STDCALL
 ffi_closure_FASTCALL ENDP
 
+ffi_closure_REGISTER PROC NEAR
+        ;; Insert the 3 register arguments on the stack as the first argument
+        push	eax
+        xchg	DWORD PTR [esp+8], ecx
+        xchg	DWORD PTR [esp+4], edx
+        push	ecx
+        push	edx
+        jmp	ffi_closure_STDCALL
+ffi_closure_FASTCALL ENDP
+
 ffi_closure_SYSV PROC NEAR FORCEFRAME
     ;; the ffi_closure ctx is passed in eax by the trampoline.
 
@@ -320,7 +331,6 @@ ffi_closure_SYSV ENDP
 #define RAW_CLOSURE_CIF_OFFSET ((FFI_TRAMPOLINE_SIZE + 3) AND NOT 3)
 #define RAW_CLOSURE_FUN_OFFSET (RAW_CLOSURE_CIF_OFFSET + 4)
 #define RAW_CLOSURE_USER_DATA_OFFSET (RAW_CLOSURE_FUN_OFFSET + 4)
-#define CIF_FLAGS_OFFSET 20
 
 ffi_closure_raw_THISCALL PROC NEAR USES esi FORCEFRAME
         sub esp, 36
@@ -417,7 +427,7 @@ ffi_closure_raw_SYSV ENDP
 #endif /* !FFI_NO_RAW_API */
 
 ffi_closure_STDCALL PROC NEAR FORCEFRAME
-    ;; the ffi_closure ctx is passed in eax by the trampoline.
+        mov  eax, [esp] ;; the ffi_closure ctx passed by the trampoline.
 
         sub  esp, 40
         lea  edx, [ebp - 24]
@@ -427,9 +437,13 @@ ffi_closure_STDCALL PROC NEAR FORCEFRAME
         lea  edx, [ebp - 12]
         mov  [esp + 4], edx          ;; &resp
         mov  [esp], eax              ;; closure
-        call ffi_closure_SYSV_inner
+        call ffi_closure_WIN32_inner
         mov  ecx, [ebp - 12]
 
+        xchg [ebp + 4], eax          ;;xchg size of stack parameters and ffi_closure ctx
+        mov  eax, DWORD PTR [eax + CLOSURE_CIF_OFFSET]
+        mov  eax, DWORD PTR [eax + CIF_FLAGS_OFFSET]
+		
 cd_jumptable:
         jmp  [cd_jumpdata + 4 * eax]
 cd_jumpdata:
@@ -493,21 +507,10 @@ cd_retlongdouble:
 cd_epilogue:
         mov   esp, ebp
         pop   ebp
-        pop   ecx
-        pop   edx
-        mov   ecx, DWORD PTR [ecx + (CLOSURE_CIF_OFFSET-10)]
-        add   esp, DWORD PTR [ecx + CIF_BYTES_OFFSET]
-        mov   ecx, DWORD PTR [ecx + CIF_ABI_OFFSET]
-        cmp   ecx, 3
-        je    cd_thiscall
-        cmp   ecx, 4
-        jne   cd_not_fastcall
-
-        add   esp, 4
-cd_thiscall:
-        add   esp, 4
-cd_not_fastcall:
-        jmp   edx
+        mov   ecx, [esp + 4]  ;; Return address
+        add   esp, [esp]      ;; Parameters stack size
+		add   esp, 8
+        jmp   ecx
 ffi_closure_STDCALL ENDP
 
 _TEXT ENDS
@@ -728,14 +731,27 @@ FFI_HIDDEN(ffi_closure_FASTCALL)
         .def	_ffi_closure_FASTCALL;	.scl	2;	.type	32;	.endef
 #endif
 USCORE_SYMBOL(ffi_closure_FASTCALL):
-        /* Insert the register arguments on the stack as the first two arguments */
+        /* Insert the 2 register arguments on the stack as the first two arguments */
         xchg	%edx, 4(%esp)
         xchg	%ecx, (%esp)
         push	%edx
         push	%ecx
         jmp	.ffi_closure_STDCALL_internal
+FFI_HIDDEN(ffi_closure_REGISTER)
+        .globl	USCORE_SYMBOL(ffi_closure_REGISTER)
+#if defined(X86_WIN32) && !defined(__OS2__)
+        .def	_ffi_closure_REGISTER;	.scl	2;	.type	32;	.endef
+#endif
+USCORE_SYMBOL(ffi_closure_REGISTER):
+        /* Insert the 3 register arguments on the stack as the first two arguments */
+        push	%eax
+        xchg	%ecx, 8(%esp)
+        xchg	%edx, 4(%esp)
+        push	%ecx
+        push	%edx
+        jmp	.ffi_closure_STDCALL_internal
+		
 .LFE1:
-
         # This assumes we are using gas.
         .balign 16
 FFI_HIDDEN(ffi_closure_SYSV)
@@ -879,7 +895,6 @@ USCORE_SYMBOL(ffi_closure_SYSV):
 #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 CIF_FLAGS_OFFSET 20
 
 #ifdef X86_WIN32
         .balign 16
@@ -1032,6 +1047,8 @@ FFI_HIDDEN(ffi_closure_STDCALL)
 #endif
 USCORE_SYMBOL(ffi_closure_STDCALL):
 .ffi_closure_STDCALL_internal:
+        /* ffi_closure ctx is at top of the stack */
+        movl	(%esp), %eax
 .LFB5:
         pushl	%ebp
 .LCFI9:
@@ -1045,19 +1062,23 @@ USCORE_SYMBOL(ffi_closure_STDCALL):
         leal	-12(%ebp), %edx
         movl	%edx, (%esp)	/* &resp */
 #if defined(HAVE_HIDDEN_VISIBILITY_ATTRIBUTE) || !defined(__PIC__)
-        call	USCORE_SYMBOL(ffi_closure_SYSV_inner)
+        call	USCORE_SYMBOL(ffi_closure_WIN32_inner)
 #elif defined(X86_DARWIN)
-        calll	L_ffi_closure_SYSV_inner$stub
+        calll	L_ffi_closure_WIN32_inner$stub
 #else
         movl	%ebx, 8(%esp)
         call	1f
-1:	popl	%ebx
+1:      popl	%ebx
         addl	$_GLOBAL_OFFSET_TABLE_+[.-1b], %ebx
-        call	ffi_closure_SYSV_inner@PLT
+        call	ffi_closure_WIN32_inner@PLT
         movl	8(%esp), %ebx
 #endif
         movl	-12(%ebp), %ecx
 0:
+		xchgl	4(%ebp), %eax /* xchg size of stack parameters and ffi_closure ctx */
+        movl	CLOSURE_CIF_OFFSET(%eax), %eax
+        movl	CIF_FLAGS_OFFSET(%eax), %eax
+		
         call	1f
         # Do not insert anything here between the call and the jump table.
 .Lscls_store_table:
@@ -1144,19 +1165,10 @@ USCORE_SYMBOL(ffi_closure_STDCALL):
 .Lscls_epilogue:
         movl	%ebp, %esp
         popl	%ebp
-        popl	%ecx
-        popl	%edx
-        movl	(CLOSURE_CIF_OFFSET-10)(%ecx), %ecx
-        addl	CIF_BYTES_OFFSET(%ecx), %esp
-        movl	CIF_ABI_OFFSET(%ecx), %ecx
-        cmpl	$3, %ecx /* FFI_THISCALL */
-        je	1f
-        cmpl	$4, %ecx /* FFI_FASTCALL */
-        jne	2f
-
-        addl	$4, %esp
-1:	addl	$4, %esp
-2:	jmp	*%edx
+		movl	4(%esp), %ecx /* Return address */
+		addl	(%esp), %esp  /* Parameters stack size */
+		addl	$8, %esp
+        jmp	*%ecx
 .ffi_closure_STDCALL_end:
 .LFE5:
 
@@ -1165,6 +1177,9 @@ USCORE_SYMBOL(ffi_closure_STDCALL):
 L_ffi_closure_SYSV_inner$stub:
         .indirect_symbol _ffi_closure_SYSV_inner
         hlt ; hlt ; hlt ; hlt ; hlt
+L_ffi_closure_WIN32_inner$stub:
+        .indirect_symbol _ffi_closure_WIN32_inner
+        hlt ; hlt ; hlt ; hlt ; hlt
 #endif
 
 #if defined(X86_WIN32) && !defined(__OS2__)