FAIL: gcc.c-torture/execute/950628-1.c execution, -O1

[Richard Henderson]

On Wed, Jan 24, 2001 at 04:27:36PM -0500, John David Anglin wrote:
> Then, I expect that only the original problem is fixed.  I just checked
> and the problem with with long doubles isn't fixed (regstack-1 et al).
> I will bet that this is also related to unchanging memory.  Here is some rtl
> from the cfg pass for regstack-1 using -O3:

In that case we really do have invalid rtl.

(insn 34 50 36 (set (mem/u/f:SI (plus:SI (reg/f:SI 3 %r3)
                (const_int 24 [0x18])) 4)
        (subreg:SI (reg:TF 101) 0)) 69 {*pa.md:2099} (nil)
    (insn_list:REG_LIBCALL 65 (nil)))
...
(insn 88 85 90 (set (mem/u/f:SI (plus:SI (reg/f:SI 3 %r3)
                (const_int 24 [0x18])) 4)
        (subreg:SI (reg:TF 108) 0)) 69 {*pa.md:2099} (nil)
    (nil))

This is almost certainly related to the /u sharing bits that
Richard Kenner and I have been discussing.


r~