c/7871: ICE on legal code, global register variables problems

[Richard Henderson]

On Thu, Feb 27, 2003 at 10:23:53PM -0500, Jim Wilson wrote:
> ! 	  /* Calls change all call-used registers.  Calls may or may not
> ! 	     change global registers.  Since this will cause previous stores
> ! 	     to be deleted as dead, we must assume that global registers are
> ! 	     not set in the call.  */
>   	  for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
> ! 	    if (TEST_HARD_REG_BIT (regs_invalidated_by_call, i)
> ! 		&& ! global_regs[i])

I don't think this is right.  The call may change the variable.
If we don't mark the value set, we won't have proper log_links.

The register was *supposed* to be marked used by 

          /* Calls may also reference any of the global registers,
             so they are made live.  */
          for (i = 0; i < FIRST_PSEUDO_REGISTER; i++)
            if (global_regs[i])
              mark_used_reg (pbi, regno_reg_rtx[i], cond, insn);

so I'm not sure where we went wrong...

A test case that I think may fail with your change is

	register int d __asm__("d7");
	void bar()
	{
	  if (d != 1)
	    abort();
	  d = 4;
	}
	void baz()
	{
	  if (d != 6)
	    abort();
	}
	int main()
	{
	  d = 1;
	  bar();
	  d |= 2;
	  baz();
	  return 0;
	}

If log_links are wrong, combine will see d=1 linked with d|=2 and
produce d=3.


r~