[RFA] fix calling conventions wrt 32-bit values

[RFA] fix calling conventions wrt 32-bit values

[Richard Henderson]

Ok?


r~


	* alpha-tdep.c (alpha_push_dummy_call): Handle ABI mandated 
	sign-extension of 32-bit values.
	(alpha_store_return_value): Similarly.


--- alpha-tdep.c.5	2003-06-01 22:34:01.000000000 -0700
+++ alpha-tdep.c	2003-06-01 22:42:32.000000000 -0700
@@ -276,9 +276,16 @@ alpha_push_dummy_call (struct gdbarch *g
 	case TYPE_CODE_CHAR:
 	case TYPE_CODE_RANGE:
 	case TYPE_CODE_ENUM:
-	  if (TYPE_LENGTH (arg_type) < TYPE_LENGTH (builtin_type_long))
+	  if (TYPE_LENGTH (arg_type) == 4)
 	    {
-	      arg_type = builtin_type_long;
+	      /* 32-bit values must be sign-extended to 64 bits
+		 even if the base data type is unsigned.  */
+	      arg_type = builtin_type_int32;
+	      arg = value_cast (arg_type, arg);
+	    }
+	  if (TYPE_LENGTH (arg_type) < ALPHA_REGISTER_SIZE)
+	    {
+	      arg_type = builtin_type_int64;
 	      arg = value_cast (arg_type, arg);
 	    }
 	  break;
@@ -541,6 +548,10 @@ alpha_store_return_value (struct type *v
 
     default:
       /* Assume everything else degenerates to an integer.  */
+      /* 32-bit values must be sign-extended to 64 bits
+	 even if the base data type is unsigned.  */
+      if (length == 4)
+	valtype = builtin_type_int32;
       l = unpack_long (valtype, valbuf);
       regcache_cooked_write_unsigned (regcache, ALPHA_V0_REGNUM, l);
       break;

Re: [RFA] fix calling conventions wrt 32-bit values

[Elena Zannoni]

Richard Henderson writes:
 > Ok?
 > 
 > 

The alpha target has no official maintainer, and I think we can
consider most of these patches obvious. I think that if they work for
you, you can commit them yourself, and just post the patch with a
[PATCH] subject.

For patches to files other than the alpha ones, you still need an
[RFA].

elena


 > r~
 > 
 > 
 > 	* alpha-tdep.c (alpha_push_dummy_call): Handle ABI mandated 
 > 	sign-extension of 32-bit values.
 > 	(alpha_store_return_value): Similarly.
 > 
 > 
 > --- alpha-tdep.c.5	2003-06-01 22:34:01.000000000 -0700
 > +++ alpha-tdep.c	2003-06-01 22:42:32.000000000 -0700
 > @@ -276,9 +276,16 @@ alpha_push_dummy_call (struct gdbarch *g
 >  	case TYPE_CODE_CHAR:
 >  	case TYPE_CODE_RANGE:
 >  	case TYPE_CODE_ENUM:
 > -	  if (TYPE_LENGTH (arg_type) < TYPE_LENGTH (builtin_type_long))
 > +	  if (TYPE_LENGTH (arg_type) == 4)
 >  	    {
 > -	      arg_type = builtin_type_long;
 > +	      /* 32-bit values must be sign-extended to 64 bits
 > +		 even if the base data type is unsigned.  */
 > +	      arg_type = builtin_type_int32;
 > +	      arg = value_cast (arg_type, arg);
 > +	    }
 > +	  if (TYPE_LENGTH (arg_type) < ALPHA_REGISTER_SIZE)
 > +	    {
 > +	      arg_type = builtin_type_int64;
 >  	      arg = value_cast (arg_type, arg);
 >  	    }
 >  	  break;
 > @@ -541,6 +548,10 @@ alpha_store_return_value (struct type *v
 >  
 >      default:
 >        /* Assume everything else degenerates to an integer.  */
 > +      /* 32-bit values must be sign-extended to 64 bits
 > +	 even if the base data type is unsigned.  */
 > +      if (length == 4)
 > +	valtype = builtin_type_int32;
 >        l = unpack_long (valtype, valbuf);
 >        regcache_cooked_write_unsigned (regcache, ALPHA_V0_REGNUM, l);
 >        break;

Re: [RFA] fix calling conventions wrt 32-bit values

[Richard Henderson]

On Mon, Jun 02, 2003 at 09:49:12AM -0400, Elena Zannoni wrote:
> The alpha target has no official maintainer, and I think we can
> consider most of these patches obvious. I think that if they work for
> you, you can commit them yourself, and just post the patch with a
> [PATCH] subject.

Thanks.  I'll still specially mark any patch that I'm not 100%
sure I'm using the proper gdb interface, but I think the bulk
of that job is done.  There's only one more pending patch that
removes instances of the word "deprecated".


r~