[Bug ada/13470] 64bits Ada bootstrap failure:xnmake etc. crash generating nmake.adb etc.

["jakub at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From jakub at gcc dot gnu dot org  2005-01-17 14:48 -------
This seems to be related to RTX_UNCHANGING_P, but I'm not sure it is actually
a bug in the compiler itself, or in the library code.
I don't know Ada at all, but e.g. from the comment about it it sounds like
it is doing something that is not kosher and is just being bitten by that.
GCC marks the u argument as /u, so both return U.Reference; lines end up
RTL like:
(insn 23 190 24 3 a-stunau.adb:43 (set (reg:DI 68)
        (mem/s/u/j:DI (plus:DI (reg/v/u/f:DI 60 [ u ])
                (const_int 32 [0x20])) [3 <variable>.reference+0 S8 A128])) -1
(nil)
    (nil))

(insn 24 23 25 3 a-stunau.adb:43 (set (reg:DI 69)
        (mem/s/u/j:DI (plus:DI (reg/v/u/f:DI 60 [ u ])
                (const_int 40 [0x28])) [3 <variable>.reference+8 S8 A64])) -1
(nil)
    (nil))

(insn 25 24 26 3 a-stunau.adb:43 (set (reg:DI 58 [ <result> ])
        (reg:DI 68)) -1 (nil)
    (nil))

(insn 26 25 27 3 a-stunau.adb:43 (set (reg:DI 59 [ <result>+8 ])
        (reg:DI 69)) -1 (nil)

while in the second case the code expects changes done through u_ptr pointer
(set to (reg/v/u/f:DI 60 [ u ])), not done with /u, to propagate to the return
value.  But, as soon as the compiler starts optimizing a little bit more
aggressively, although it will change the U.Reference value, it can very well
return Old; instead, as it was promissed U.Reference is not going to change
during the lifetime of the function.


-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |laurent at guerby dot net,
                   |                            |kenner at vlsi1 dot ultra
                   |                            |dot nyu dot edu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=13470