Re: [PATCH] config.gcc: improve description of need_64bit_hwint

[Dave Korn <dave dot korn dot cygwin at googlemail dot com>]

Joseph S. Myers wrote:
> On Fri, 20 Mar 2009, Dave Korn wrote:
> 
>> Bingfeng Mei wrote:
>>
>>> +#                      for this target.  This is true if this target
>>> +#                      supports "long" or "wchar_t" wider than 32 bits,
>>> +#                       or BITS_PER_WORD is wider than 32 bits.
>>   Need to tweak the alignment there before you check it in :)
>>
>>   (BTW, isn't this logically the same anyway, since if BITS_PER_WORD (=
>> SImode) is > 32 bits, then int must be > 32 bits, and long is necessarily at
>> least >= int?  Or can SImode be < BITS_PER_WORD sometimes?  Not that I
>> disagree with making the implications explicit for clarity in either case.)
> 
> int is nothing to do with SImode, and SImode is nothing to do with 
> BITS_PER_WORD.  SImode (if it exists - I don't think it did on c4x) is 
> always four times QImode, and QImode's width is BITS_PER_UNIT.  int's 
> width is INT_TYPE_SIZE which may be more or less than four times 
> BITS_PER_UNIT.  BITS_PER_WORD is a comparatively arbitrary and poorly 
> defined value that need have no relation to the C type sizes.
> 

  Thanks for the clarification.  I thought that SImode generally meant
"machine's natural word size" and was therefore coupled with, albeit not
identical to, BITS_PER_WORD.

    cheers,
      DaveK