ffi.c (ffi_call_win32): Add new argument to prototype for specify calling-convention.

        * src/libffi/src/x86/ffi.c (ffi_call_win32): Add new
        argument to prototype for specify calling-convention.
        (ffi_call): Add support for stdcall/thiscall convention.
        (ffi_prep_args): Likewise.
        (ffi_raw_call): Likewise.
        * src/x86/ffitarget.h (ffi_abi): Add FFI_THISCALL and
        FFI_FASTCALL.
        * src/x86/win32.S (_ffi_call_win32): Add support for
        fastcall/thiscall calling-convention calls.
        * testsuite/libffi.call/fastthis1_win32.c: New test.
        * testsuite/libffi.call/fastthis2_win32.c: New test.
        * testsuite/libffi.call/fastthis3_win32.c: New test.
        * testsuite/libffi.call/strlen2_win32.c: New test.
        * testsuite/libffi.call/many2_win32.c: New test.
        * testsuite/libffi.call/struct1_win32.c: New test.
        * testsuite/libffi.call/struct2_win32.c: New test.

From-SVN: r183676

3 files changed, 181 insertions, 19 deletions

diff --git a/libffi/src/x86/ffi.c b/libffi/src/x86/ffi.c
index fea9d6d..bbfd95e 100644
--- a/libffi/src/x86/ffi.c
+++ b/libffi/src/x86/ffi.c
@@ -48,6 +48,13 @@ void ffi_prep_args(char *stack, extended_cif *ecif)
   register void **p_argv;
   register char *argp;
   register ffi_type **p_arg;
+#ifdef X86_WIN32
+  size_t p_stack_args[2];
+  void *p_stack_data[2];
+  char *argp2 = stack;
+  int stack_args_count = 0;
+  int cabi = ecif->cif->abi;
+#endif
 
   argp = stack;
 
@@ -59,6 +66,16 @@ void ffi_prep_args(char *stack, extended_cif *ecif)
       )
     {
       *(void **) argp = ecif->rvalue;
+#ifdef X86_WIN32
+      /* For fastcall/thiscall this is first register-passed
+         argument.  */
+      if (cabi == FFI_THISCALL || cabi == FFI_FASTCALL)
+	{
+	  p_stack_args[stack_args_count] = sizeof (void*);
+	  p_stack_data[stack_args_count] = argp;
+	  ++stack_args_count;
+	}
+#endif
       argp += sizeof(void*);
     }
 
@@ -134,6 +151,24 @@ void ffi_prep_args(char *stack, extended_cif *ecif)
         {
           memcpy(argp, *p_argv, z);
         }
+
+#ifdef X86_WIN32
+    /* For thiscall/fastcall convention register-passed arguments
+       are the first two none-floating-point arguments with a size
+       smaller or equal to sizeof (void*).  */
+    if ((cabi == FFI_THISCALL && stack_args_count < 1)
+        || (cabi == FFI_FASTCALL && stack_args_count < 2))
+      {
+	if (z <= 4
+	    && ((*p_arg)->type != FFI_TYPE_FLOAT
+	        && (*p_arg)->type != FFI_TYPE_STRUCT))
+	  {
+	    p_stack_args[stack_args_count] = z;
+	    p_stack_data[stack_args_count] = argp;
+	    ++stack_args_count;
+	  }
+      }
+#endif
       p_argv++;
 #ifdef X86_WIN64
       argp += (z + sizeof(void*) - 1) & ~(sizeof(void*) - 1);
@@ -141,7 +176,45 @@ void ffi_prep_args(char *stack, extended_cif *ecif)
       argp += z;
 #endif
     }
-  
+
+#ifdef X86_WIN32
+  /* We need to move the register-passed arguments for thiscall/fastcall
+     on top of stack, so that those can be moved to registers ecx/edx by
+     call-handler.  */
+  if (stack_args_count > 0)
+    {
+      size_t zz = (p_stack_args[0] + 3) & ~3;
+      char *h;
+
+      /* Move first argument to top-stack position.  */
+      if (p_stack_data[0] != argp2)
+	{
+	  h = alloca (zz + 1);
+	  memcpy (h, p_stack_data[0], zz);
+	  memmove (argp2 + zz, argp2,
+	           (size_t) ((char *) p_stack_data[0] - (char*)argp2));
+	  memcpy (argp2, h, zz);
+	}
+
+      argp2 += zz;
+      --stack_args_count;
+      if (zz > 4)
+	stack_args_count = 0;
+
+      /* If we have a second argument, then move it on top
+         after the first one.  */
+      if (stack_args_count > 0 && p_stack_data[1] != argp2)
+	{
+	  zz = p_stack_args[1];
+	  zz = (zz + 3) & ~3;
+	  h = alloca (zz + 1);
+	  h = alloca (zz + 1);
+	  memcpy (h, p_stack_data[1], zz);
+	  memmove (argp2 + zz, argp2, (size_t) ((char*) p_stack_data[1] - (char*)argp2));
+	  memcpy (argp2, h, zz);
+	}
+    }
+#endif
   return;
 }
 
@@ -252,7 +325,7 @@ ffi_call_win64(void (*)(char *, extended_cif *), extended_cif *,
 #elif defined(X86_WIN32)
 extern void
 ffi_call_win32(void (*)(char *, extended_cif *), extended_cif *,
-               unsigned, unsigned, unsigned *, void (*fn)(void));
+               unsigned, unsigned, unsigned, unsigned *, void (*fn)(void));
 #else
 extern void ffi_call_SYSV(void (*)(char *, extended_cif *), extended_cif *,
                           unsigned, unsigned, unsigned *, void (*fn)(void));
@@ -316,8 +389,37 @@ void ffi_call(ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
 #elif defined(X86_WIN32)
     case FFI_SYSV:
     case FFI_STDCALL:
-      ffi_call_win32(ffi_prep_args, &ecif, cif->bytes, cif->flags,
-                     ecif.rvalue, fn);
+      ffi_call_win32(ffi_prep_args, &ecif, cif->abi, cif->bytes, cif->flags,
+		     ecif.rvalue, fn);
+      break;
+    case FFI_THISCALL:
+    case FFI_FASTCALL:
+      {
+	unsigned int abi = cif->abi;
+	unsigned int i, passed_regs = 0;
+
+	if (cif->flags == FFI_TYPE_STRUCT)
+	  ++passed_regs;
+
+	for (i=0; i < cif->nargs && passed_regs < 2;i++)
+	  {
+	    size_t sz;
+
+	    if (cif->arg_types[i]->type == FFI_TYPE_FLOAT
+	        || cif->arg_types[i]->type == FFI_TYPE_STRUCT)
+	      continue;
+	    sz = (cif->arg_types[i]->size + 3) & ~3;
+	    if (sz == 0 || sz > 4)
+	      continue;
+	    ++passed_regs;
+	  }
+	if (passed_regs < 2 && abi == FFI_FASTCALL)
+	  abi = FFI_THISCALL;
+	if (passed_regs < 1 && abi == FFI_THISCALL)
+	  abi = FFI_STDCALL;
+        ffi_call_win32(ffi_prep_args, &ecif, abi, cif->bytes, cif->flags,
+                       ecif.rvalue, fn);
+      }
       break;
 #else
     case FFI_SYSV:
@@ -644,8 +746,37 @@ ffi_raw_call(ffi_cif *cif, void (*fn)(void), void *rvalue, ffi_raw *fake_avalue)
 #ifdef X86_WIN32
     case FFI_SYSV:
     case FFI_STDCALL:
-      ffi_call_win32(ffi_prep_args_raw, &ecif, cif->bytes, cif->flags,
-                     ecif.rvalue, fn);
+      ffi_call_win32(ffi_prep_args, &ecif, cif->abi, cif->bytes, cif->flags,
+		     ecif.rvalue, fn);
+      break;
+    case FFI_THISCALL:
+    case FFI_FASTCALL:
+      {
+	unsigned int abi = cif->abi;
+	unsigned int i, passed_regs = 0;
+
+	if (cif->flags == FFI_TYPE_STRUCT)
+	  ++passed_regs;
+
+	for (i=0; i < cif->nargs && passed_regs < 2;i++)
+	  {
+	    size_t sz;
+
+	    if (cif->arg_types[i]->type == FFI_TYPE_FLOAT
+	        || cif->arg_types[i]->type == FFI_TYPE_STRUCT)
+	      continue;
+	    sz = (cif->arg_types[i]->size + 3) & ~3;
+	    if (sz == 0 || sz > 4)
+	      continue;
+	    ++passed_regs;
+	  }
+	if (passed_regs < 2 && abi == FFI_FASTCALL)
+	  cif->abi = abi = FFI_THISCALL;
+	if (passed_regs < 1 && abi == FFI_THISCALL)
+	  cif->abi = abi = FFI_STDCALL;
+        ffi_call_win32(ffi_prep_args, &ecif, abi, cif->bytes, cif->flags,
+                       ecif.rvalue, fn);
+      }
       break;
 #else
     case FFI_SYSV:
diff --git a/libffi/src/x86/ffitarget.h b/libffi/src/x86/ffitarget.h
index b85016c..4154762 100644
--- a/libffi/src/x86/ffitarget.h
+++ b/libffi/src/x86/ffitarget.h
@@ -64,6 +64,8 @@ typedef enum ffi_abi {
 #ifdef X86_WIN32
   FFI_SYSV,
   FFI_STDCALL,
+  FFI_THISCALL,
+  FFI_FASTCALL,
   /* TODO: Add fastcall support for the sake of completeness */
   FFI_DEFAULT_ABI = FFI_SYSV,
 #endif
diff --git a/libffi/src/x86/win32.S b/libffi/src/x86/win32.S
index 34ec0fd..e472d62 100644
--- a/libffi/src/x86/win32.S
+++ b/libffi/src/x86/win32.S
@@ -45,6 +45,7 @@ _TEXT SEGMENT
 ffi_call_win32 PROC NEAR,
     ffi_prep_args : NEAR PTR DWORD,
     ecif          : NEAR PTR DWORD,
+    cif_abi       : DWORD,
     cif_bytes     : DWORD,
     cif_flags     : DWORD,
     rvalue        : NEAR PTR DWORD,
@@ -64,6 +65,19 @@ ffi_call_win32 PROC NEAR,
         ;; Return stack to previous state and call the function
         add  esp, 8
 
+	;; Handle thiscall and fastcall
+	cmp cif_abi, 3 ;; FFI_THISCALL
+	jz do_thiscall
+	cmp cif_abi, 4 ;; FFI_FASTCALL
+	jnz do_stdcall
+	mov ecx, DWORD PTR [esp]
+	mov edx, DWORD PTR [esp+4]
+	add esp, 8
+	jmp do_stdcall
+do_thiscall:
+	mov ecx, DWORD PTR [esp]
+	add esp, 4
+do_stdcall:
         call fn
 
         ;; cdecl:   we restore esp in the epilogue, so there's no need to
@@ -405,7 +419,7 @@ _ffi_call_win32:
         movl  %esp,%ebp
 .LCFI1:
         # Make room for all of the new args.
-        movl  16(%ebp),%ecx                                                     
+        movl  20(%ebp),%ecx                                                     
         subl  %ecx,%esp
  
         movl  %esp,%eax
@@ -417,19 +431,34 @@ _ffi_call_win32:
  
         # Return stack to previous state and call the function
         addl  $8,%esp
- 
+
+	# Handle fastcall and thiscall
+	cmpl $3, 16(%ebp)  # FFI_THISCALL
+	jz .do_thiscall
+	cmpl $4, 16(%ebp) # FFI_FASTCALL
+	jnz .do_fncall
+	movl (%esp), %ecx
+	movl 4(%esp), %edx
+	addl $8, %esp
+	jmp .do_fncall
+.do_thiscall:
+	movl (%esp), %ecx
+	addl $4, %esp
+
+.do_fncall:
+	 
         # FIXME: Align the stack to a 128-bit boundary to avoid
         # potential performance hits.
 
-        call  *28(%ebp)
+        call  *32(%ebp)
  
         # stdcall functions pop arguments off the stack themselves
 
         # Load %ecx with the return type code
-        movl  20(%ebp),%ecx
+        movl  24(%ebp),%ecx
  
         # If the return value pointer is NULL, assume no return value.
-        cmpl  $0,24(%ebp)
+        cmpl  $0,28(%ebp)
         jne   0f
  
         # Even if there is no space for the return value, we are
@@ -488,50 +517,50 @@ _ffi_call_win32:
 
 .Lretint:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         movl  %eax,0(%ecx)
         jmp   .Lepilogue
  
 .Lretfloat:
          # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         fstps (%ecx)
         jmp   .Lepilogue
  
 .Lretdouble:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         fstpl (%ecx)
         jmp   .Lepilogue
  
 .Lretlongdouble:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         fstpt (%ecx)
         jmp   .Lepilogue
  
 .Lretint64:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         movl  %eax,0(%ecx)
         movl  %edx,4(%ecx)
 	jmp   .Lepilogue
 
 .Lretstruct1b:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         movb  %al,0(%ecx)
         jmp   .Lepilogue
  
 .Lretstruct2b:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         movw  %ax,0(%ecx)
         jmp   .Lepilogue
 
 .Lretstruct4b:
         # Load %ecx with the pointer to storage for the return value
-        movl  24(%ebp),%ecx
+        movl  28(%ebp),%ecx
         movl  %eax,0(%ecx)
         jmp   .Lepilogue