Fix va_list handling for Windows Arm64 ABI

Author: Blackhex

gcc/config/aarch64/aarch64.cc

@@ -21483,6 +21483,17 @@ aarch64_expand_builtin_va_start (tree valist, rtx nextarg ATTRIBUTE_UNUSED)
   expand_expr (t, const0_rtx, VOIDmode, EXPAND_NORMAL);
 }
 
+#ifdef TARGET_AARCH64_MS_ABI
+
+static void
+aarch64_ms_abi_expand_builtin_va_start (tree valist, rtx nextarg)
+{
+  nextarg = plus_constant (Pmode, nextarg, -cfun->gr_saved * UNITS_PER_WORD); // -56
+  std_expand_builtin_va_start (valist, nextarg);
+}
+
+#endif
+
 /* Implement TARGET_GIMPLIFY_VA_ARG_EXPR.  */
 
 static tree
@@ -21787,7 +21798,7 @@ aarch64_setup_incoming_varargs (cumulative_args_t cum_v,
 {
   CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v);
   CUMULATIVE_ARGS local_cum;
-  int gr_saved = cfun->va_list_gpr_size;
+  cfun->gr_saved = cfun->va_list_gpr_size;
   int vr_saved = cfun->va_list_fpr_size;
 
   /* The caller has advanced CUM up to, but not beyond, the last named
@@ -21800,7 +21811,7 @@ aarch64_setup_incoming_varargs (cumulative_args_t cum_v,
   /* Found out how many registers we need to save.
      Honor tree-stdvar analysis results.  */
   if (cfun->va_list_gpr_size)
-    gr_saved = MIN (NUM_ARG_REGS - local_cum.aapcs_ncrn,
+    cfun->gr_saved = MIN (NUM_ARG_REGS - local_cum.aapcs_ncrn,
 		    cfun->va_list_gpr_size / UNITS_PER_WORD);
   if (cfun->va_list_fpr_size)
     vr_saved = MIN (NUM_FP_ARG_REGS - local_cum.aapcs_nvrn,
@@ -21814,18 +21825,18 @@ aarch64_setup_incoming_varargs (cumulative_args_t cum_v,
 
   if (!no_rtl)
     {
-      if (gr_saved > 0)
+      if (cfun->gr_saved > 0)
 	{
 	  rtx ptr, mem;
 
 	  /* virtual_incoming_args_rtx should have been 16-byte aligned.  */
 	  ptr = plus_constant (Pmode, virtual_incoming_args_rtx,
-			       - gr_saved * UNITS_PER_WORD);
+			       - cfun->gr_saved * UNITS_PER_WORD);
 	  mem = gen_frame_mem (BLKmode, ptr);
 	  set_mem_alias_set (mem, get_varargs_alias_set ());
 
 	  move_block_from_reg (local_cum.aapcs_ncrn + R0_REGNUM,
-			       mem, gr_saved);
+			       mem, cfun->gr_saved);
 	}
       if (vr_saved > 0)
 	{
@@ -21837,7 +21848,7 @@ aarch64_setup_incoming_varargs (cumulative_args_t cum_v,
 	  /* Set OFF to the offset from virtual_incoming_args_rtx of
 	     the first vector register.  The VR save area lies below
 	     the GR one, and is aligned to 16 bytes.  */
-	  off = -ROUND_UP (gr_saved * UNITS_PER_WORD,
+	  off = -ROUND_UP (cfun->gr_saved * UNITS_PER_WORD,
 			   STACK_BOUNDARY / BITS_PER_UNIT);
 	  off -= vr_saved * UNITS_PER_VREG;
 
@@ -21858,7 +21869,7 @@ aarch64_setup_incoming_varargs (cumulative_args_t cum_v,
   /* We don't save the size into *PRETEND_SIZE because we want to avoid
      any complication of having crtl->args.pretend_args_size changed.  */
   cfun->machine->frame.saved_varargs_size
-    = (ROUND_UP (gr_saved * UNITS_PER_WORD,
+    = (ROUND_UP (cfun->gr_saved * UNITS_PER_WORD,
 		 STACK_BOUNDARY / BITS_PER_UNIT)
        + vr_saved * UNITS_PER_VREG);
 }
@@ -22634,9 +22645,12 @@ static const char *
 aarch64_mangle_type (const_tree type)
 {
   /* The AArch64 ABI documents say that "__va_list" has to be
-     mangled as if it is in the "std" namespace.  */
+     mangled as if it is in the "std" namespace.
+     The Windows Arm64 ABI uses just an address of the first variadict argument.  */
+#ifndef TARGET_AARCH64_MS_ABI
   if (lang_hooks.types_compatible_p (CONST_CAST_TREE (type), va_list_type))
     return "St9__va_list";
+#endif
 
   /* Half-precision floating point types.  */
   if (SCALAR_FLOAT_TYPE_P (type) && TYPE_PRECISION (type) == 16)
@@ -30986,8 +31000,10 @@ aarch64_run_selftests (void)
 #undef TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY
 #define TARGET_ASM_PRINT_PATCHABLE_FUNCTION_ENTRY aarch64_print_patchable_function_entry
 
+#ifndef TARGET_AARCH64_MS_ABI
 #undef TARGET_BUILD_BUILTIN_VA_LIST
 #define TARGET_BUILD_BUILTIN_VA_LIST aarch64_build_builtin_va_list
+#endif
 
 #undef TARGET_CALLEE_COPIES
 #define TARGET_CALLEE_COPIES hook_bool_CUMULATIVE_ARGS_arg_info_false
@@ -31058,7 +31074,11 @@ aarch64_run_selftests (void)
 #define TARGET_EXPAND_BUILTIN aarch64_expand_builtin
 
 #undef TARGET_EXPAND_BUILTIN_VA_START
+#ifdef TARGET_AARCH64_MS_ABI
+#define TARGET_EXPAND_BUILTIN_VA_START aarch64_ms_abi_expand_builtin_va_start
+#else
 #define TARGET_EXPAND_BUILTIN_VA_START aarch64_expand_builtin_va_start
+#endif
 
 #undef TARGET_FOLD_BUILTIN
 #define TARGET_FOLD_BUILTIN aarch64_fold_builtin
@@ -31098,8 +31118,10 @@ aarch64_run_selftests (void)
 #undef TARGET_GIMPLE_FOLD_BUILTIN
 #define TARGET_GIMPLE_FOLD_BUILTIN aarch64_gimple_fold_builtin
 
+#ifndef TARGET_AARCH64_MS_ABI
 #undef TARGET_GIMPLIFY_VA_ARG_EXPR
 #define TARGET_GIMPLIFY_VA_ARG_EXPR aarch64_gimplify_va_arg_expr
+#endif
 
 #undef  TARGET_INIT_BUILTINS
 #define TARGET_INIT_BUILTINS  aarch64_init_builtins

gcc/function.h

@@ -350,6 +350,8 @@ struct GTY(()) function {
   /* Last assigned dependence info clique.  */
   unsigned short last_clique;
 
+  unsigned int gr_saved;
+
   /* Collected bit flags.  */
 
   /* Number of units of general registers that need saving in stdarg