[PATCH]: Fix PR26734: GCC cannot bootstrap on IA64 HP-UX

[Richard Sandiford]

Maxim Kuvyrkov <mkuvyrkov@ispras.ru> writes:
> GCC fails to bootstrap at ia64-hpux due to wrong computation of the size 
> of DEPS_LIST rtx.  Before my patch any rtx could contain either rtunion 
> or HOST_WIDE_INT, but not both.  DEPS_LIST is defined to hold two 
> rtunions and one HOST_WIDE_INT. As it wasn't mentioned among special 
> cases of rtx_code_size its size calculated assuming all its fields are 
> rtunions.  This problem can be fixed by adding DEPS_LIST to the list of 
> special cases of rtx_code_size.

Having an rtx that uses both u.fld[] and u.hwint[] fields seems very
unsafe to me.

> As I understand, it is assumed that the size of HOST_WIDE_INT is *not 
> less* than the size of void*.  If this is not necessarily should be 
> true, than rtx_code_size should be fixed in a little bit more tricky way.

No, the assumption is that every rtx either contains all HOST_WIDE_INTs
or all rtunion fields.  It shouldn't mix between the two.  So...

> @@ -111,7 +111,7 @@ const enum rtx_class rtx_class[NUM_RTX_C
>  
>  const unsigned char rtx_code_size[NUM_RTX_CODE] = {
>  #define DEF_RTL_EXPR(ENUM, NAME, FORMAT, CLASS)				\
> -  ((ENUM) == CONST_INT || (ENUM) == CONST_DOUBLE			\
> +  ((ENUM) == CONST_INT || (ENUM) == CONST_DOUBLE || (ENUM) == DEPS_LIST	\
>     ? RTX_HDR_SIZE + (sizeof FORMAT - 1) * sizeof (HOST_WIDE_INT)	\
>     : RTX_HDR_SIZE + (sizeof FORMAT - 1) * sizeof (rtunion)),

...CONST_INT and CONST_DOUBLE are both all-HOST_WIDE_INT fields.

Does the "degree of weakness" really need to be a HOST_WIDE_INT
rather than a plain int?

Richard