Add basic __vectorcall support

This is a fake implementation, which piggy-backs on the x64 standard
calling convention and __fastcall, for x86_64 and x86 respectively,
without any modifications.  It won't work for some vector type
parameters and for any HVA types.  Since SIMD vector types and HVA are
not supported by ext/ffi anyway, only float/double parameters would be
a real issue.

Author: cmb69

src/x86/ffi.c

@@ -74,6 +74,7 @@ ffi_prep_cif_machdep(ffi_cif *cif)
     case FFI_MS_CDECL:
     case FFI_PASCAL:
     case FFI_REGISTER:
+    case FFI_VECTORCALL_PARTIAL:
       break;
     default:
       return FFI_BAD_ABI;
@@ -136,6 +137,7 @@ ffi_prep_cif_machdep(ffi_cif *cif)
 	    case FFI_FASTCALL:
 	    case FFI_STDCALL:
 	    case FFI_MS_CDECL:
+      case FFI_VECTORCALL_PARTIAL:
 	      flags = X86_RET_STRUCTARG;
 	      break;
 	    default:
@@ -240,6 +242,7 @@ static const struct abi_params abi_params[FFI_LAST_ABI] = {
   [FFI_PASCAL] = { -1, R_ECX, 0 },
   /* ??? No defined static chain; gcc does not support REGISTER.  */
   [FFI_REGISTER] = { -1, R_ECX, 3, { R_EAX, R_EDX, R_ECX } },
+  [FFI_VECTORCALL_PARTIAL] = { 1, R_EAX, 2, { R_ECX, R_EDX } },
   [FFI_MS_CDECL] = { 1, R_ECX, 0 }
 };
 
@@ -355,7 +358,7 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *rvalue,
 	  /* Issue 434: For thiscall and fastcall, if the paramter passed
 	     as 64-bit integer or struct, all following integer paramters
 	     will be passed on stack.  */
-	  if ((cabi == FFI_THISCALL || cabi == FFI_FASTCALL)
+	  if ((cabi == FFI_THISCALL || cabi == FFI_FASTCALL || cabi == FFI_VECTORCALL_PARTIAL)
 	      && (t == FFI_TYPE_SINT64
 		  || t == FFI_TYPE_UINT64
 		  || t == FFI_TYPE_STRUCT))
@@ -494,7 +497,7 @@ ffi_closure_inner (struct closure_frame *frame, char *stack)
 	  /* Issue 434: For thiscall and fastcall, if the paramter passed
 	     as 64-bit integer or struct, all following integer paramters
 	     will be passed on stack.  */
-	  if ((cabi == FFI_THISCALL || cabi == FFI_FASTCALL)
+	  if ((cabi == FFI_THISCALL || cabi == FFI_FASTCALL || cabi == FFI_VECTORCALL_PARTIAL)
 	      && (t == FFI_TYPE_SINT64
 		  || t == FFI_TYPE_UINT64
 		  || t == FFI_TYPE_STRUCT))
@@ -543,6 +546,7 @@ ffi_prep_closure_loc (ffi_closure* closure,
     case FFI_THISCALL:
     case FFI_FASTCALL:
     case FFI_MS_CDECL:
+    case FFI_VECTORCALL_PARTIAL:
       dest = ffi_closure_i386;
       break;
     case FFI_STDCALL:
@@ -590,6 +594,7 @@ ffi_prep_go_closure (ffi_go_closure* closure, ffi_cif* cif,
       break;
     case FFI_THISCALL:
     case FFI_FASTCALL:
+    case FFI_VECTORCALL_PARTIAL:
       dest = ffi_go_closure_EAX;
       break;
     case FFI_STDCALL:

src/x86/ffitarget.h

@@ -82,6 +82,7 @@ typedef enum ffi_abi {
   FFI_FIRST_ABI = 0,
   FFI_WIN64,            /* sizeof(long double) == 8  - microsoft compilers */
   FFI_GNUW64,           /* sizeof(long double) == 16 - GNU compilers */
+  FFI_VECTORCALL_PARTIAL,
   FFI_LAST_ABI,
 #ifdef __GNUC__
   FFI_DEFAULT_ABI = FFI_GNUW64
@@ -107,6 +108,7 @@ typedef enum ffi_abi {
   FFI_MS_CDECL  = 5,
   FFI_PASCAL    = 6,
   FFI_REGISTER  = 7,
+  FFI_VECTORCALL_PARTIAL = 8,
   FFI_LAST_ABI,
   FFI_DEFAULT_ABI = FFI_MS_CDECL
 #else

src/x86/ffiw64.c

@@ -57,6 +57,7 @@ EFI64(ffi_prep_cif_machdep)(ffi_cif *cif)
   switch (cif->abi)
     {
     case FFI_WIN64:
+    case FFI_VECTORCALL_PARTIAL:
     case FFI_GNUW64:
       break;
     default:
@@ -116,7 +117,7 @@ ffi_call_int (ffi_cif *cif, void (*fn)(void), void *rvalue,
   size_t rsize;
   struct win64_call_frame *frame;
 
-  FFI_ASSERT(cif->abi == FFI_GNUW64 || cif->abi == FFI_WIN64);
+  FFI_ASSERT(cif->abi == FFI_GNUW64 || cif->abi == FFI_WIN64 || cif->abi == FFI_VECTORCALL_PARTIAL);
 
   flags = cif->flags;
   rsize = 0;
@@ -209,6 +210,7 @@ EFI64(ffi_prep_closure_loc)(ffi_closure* closure,
   switch (cif->abi)
     {
     case FFI_WIN64:
+    case FFI_VECTORCALL_PARTIAL:
     case FFI_GNUW64:
       break;
     default:
@@ -232,6 +234,7 @@ EFI64(ffi_prep_go_closure)(ffi_go_closure* closure, ffi_cif* cif,
   switch (cif->abi)
     {
     case FFI_WIN64:
+    case FFI_VECTORCALL_PARTIAL:
     case FFI_GNUW64:
       break;
     default: