Who is responsible for satisfying assembler constraints?

[Jan Vroonhof <jan dot vroonhof at insignia dot com>]

[Please CC. I am not subsribed]

I have a question about constraints on inline assembler parameters.
If this is in the docs I missed them (Please feel free to point me
to the right location if so).

Consider the following fragment, which is
adapted from code deeply embedded in a large C file:

void    
func (void)
{

  unsigned long ourFpuCw;
  unsigned long	savedFpuCw;
 	{
	  __asm__ __volatile__ (
			 "fstcw	%0	 	\n"	"fldcw	%1		\n"
			 : "=o" (savedFpuCw)
			 : "o" (ourFpuCw)
			 );
	} 	
} 

/*
  Local Variables:
    compile-command: "gcc -O1  -c -o interp.o ibug.c"
  End:
*/

This produces the following output in gcc 2.95.2
ibug.c:12: warning: asm operand 0 probably doesn't match constraints
ibug.c:14: Internal compiler error in `fixup_var_refs_1', at function.c:2071

I reported the crash elsewhere. This question is about the warning.
I can make the warning (and the crash) go away when I make the input and
output variables `volatile'. My hunch is that this forces gcc to put/keep
them in memory instead of a register and thus satisfying the constraints.

However I thought the whole point of constraints was that gcc would make sure
they would be satisfied?

Am I mistaken?

Jan