[Bug target/17626] New: Recent peephole2:s may cause internal compiler errors

[hp at gcc dot gnu dot org]

The peephole2s marked "moverside" and "movemside" may cause noncanonical RTX,
which leads to an ICE that looks like this:

x/hws3.c: In function `f0':
x/hws3.c:47: unrecognizable insn:
(insn 196 106 108 (parallel[
            (set (reg:HI 11 r11)
                (mem/u:HI (plus:SI (reg/f:SI 13 r13 [67])
                        (mult:SI (reg/v:SI 1 r1 [23])
                            (const_int 2 [0x2]))) [7 cav1 S2 A8]))
            (set (reg/f:SI 12 r12 [70])
                (plus:SI (reg/f:SI 13 r13 [67])
                    (mult:SI (reg/v:SI 1 r1 [23])
                        (const_int 2 [0x2]))))
        ] ) -1 (nil)
    (expr_list:REG_DEAD (reg/v:SI 1 r1 [23])
        (nil)))

The left operand of the PLUS should have been the MULT, not the register.
No testcase for trunk, as this was only observed with a 3.2.1-derivate
with the peephole2-patch applied, and the failing code "works" on trunk.

-- 
           Summary: Recent peephole2:s may cause internal compiler errors
           Product: gcc
           Version: 4.0.0
            Status: UNCONFIRMED
          Keywords: ice-on-valid-code
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: hp at gcc dot gnu dot org
        ReportedBy: hp at gcc dot gnu dot org
                CC: gcc-bugs at gcc dot gnu dot org
GCC target triplet: cris-*


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17626