Is there a way to get the high half of a DImode on a 32-bit machine?

[gcc@horizon.com]

> I suppose you may go ahead and add it to print_operand_address() and
> print_reg() in gcc/config/i386/i386.c

> Actually, it can't handle CONST_INTs, because they don't have a mode
> to indicate its width.  A patch that checks for non-VOIDmode and
> complains would be appreciated.

Hints taken.  I've never patched gcc before, so I'm kind of nervous,
but whack me hard enough with a clue-by-four and I get it.

> To give a more useful version, try the ones-complement checksum routine

>> asm("add%z0 %2,%0; adc%z0 %3,%0" : "=r" (sum) :
>> 	"r" (sum), "g" (c), "g" ((typeof sum)0));

> Does this fail similarly?  I don't see why.  In any case, I suggest
> using "+r" instead of "=r" and "r" for sum.  If you'd rather not do
> that to support older versions of GCC that didn't support "+" in asm
> constraints, I suggest using "0" in the input `sum' constraint.

Um, oops!  I knew that; I just wasn't thinking so clearly when typing it.

I *didn't* know that "+r" is now supported, but yes, using "0" is mandatory.

However, it does indeed fail:

$ gcc-2.95 -o foo foo.c
foo.c: In function `checksum':
foo.c:14: Internal compiler error:
foo.c:14: Internal compiler error in `print_operand', at config/i386/i386.c:3325
Please submit a full bug report.
See <URL: http://www.gnu.org/software/gcc/bugs.html > for instructions.

$ gcc-3.0 -o foo foo.c
foo.c: In function `checksum':
foo.c:14: Internal compiler error in print_operand, at config/i386/i386.c:3539
Please submit a full bug report,
with preprocessed source if appropriate.
See <URL: http://www.gnu.org/software/gcc/bugs.html > for instructions.