[Bug optimization/11320] [3.3 regression] [ia64] Reload bug

[ebotcazou at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11320


ebotcazou at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
   Last reconfirmed|0000-00-00 00:00:00         |2003-07-06 11:25:21
               date|                            |


------- Additional Comments From ebotcazou at gcc dot gnu dot org  2003-07-06 11:25 -------
Confirmed on gcc version 3.3.1 20030703 (prerelease).

Preliminary investigations suggest that the culprit is the machine dependent
reorganisation pass, which swaps

(jump_insn 82 88 122 3 0x401b54c0 (set (pc)
        (if_then_else (eq (reg:BI 262 p6 [361])
                (const_int 0 [0x0]))
            (label_ref 100)
            (pc))) 214 {*br_true} (insn_list 81 (nil))
    (expr_list:REG_DEAD (reg:BI 262 p6 [361])
        (expr_list:REG_BR_PROB (const_int 5000 [0x1388])
            (nil))))

(insn 96 122 159 4 0x401b54c0 (set (reg/f:DI 14 r14 [340])
        (lo_sum:DI (reg:DI 14 r14 [365])
            (symbol_ref/f:DI ("*.LC2")))) 12 {*load_symptr_low} (nil)
    (expr_list:REG_EQUAL (symbol_ref/f:DI ("*.LC2"))
        (nil)))

to produce

	.mbb
	ld8 r14 = [r14]
	(p7) br.cond.dptk .L4


I don't know enough of IA-64 to say whether swapping back the instructions in
the assembly file is sufficient, or whether they should not have ended up in the
 same bundle in the first place.