[Bug debug/52727] [4.7/4.8 Regression] internal compiler error at dwarf2cfi.c2:685

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

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=52727

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |rth at gcc dot gnu.org

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-03-27 16:49:37 UTC ---
Looks like a problem during CSA.  Before *.csa we have (insn 18 is a stack
adjustment after a call):
(insn 18 17 19 2 (parallel [
            (set (reg/f:SI 7 sp)
                (plus:SI (reg/f:SI 7 sp)
                    (const_int 4 [0x4])))
            (clobber (reg:CC 17 flags))
        ]) pr52727.ii:29 251 {*addsi_1}
     (expr_list:REG_UNUSED (reg:CC 17 flags)
        (expr_list:REG_ARGS_SIZE (const_int 0 [0])
            (nil))))
...
(insn 205 27 206 2 (set (mem:SI (pre_dec:SI (reg/f:SI 7 sp)) [0 S4 A8])
        (reg:SI 0 ax [115])) pr52727.ii:31 43 {*pushsi2}
     (expr_list:REG_DEAD (reg:SI 0 ax [115])
        (nil)))

(insn 206 205 207 2 (set (reg:DF 8 st [112])
        (plus:DF (float:DF (mem:SI (reg/f:SI 7 sp) [0 S4 A8]))
            (reg:DF 8 st [112]))) pr52727.ii:31 765 {*fop_df_2_i387}
     (nil))

(insn 207 206 171 2 (set (reg/f:SI 7 sp)
        (plus:SI (reg/f:SI 7 sp)
            (const_int 4 [0x4]))) pr52727.ii:31 245 {*leasi_2}
     (nil))

but then CSA decides to remove insn 18 and adjust insn 205, so that it no
longer pushes the value, but stores it to *sp.  Unfortunately it means
REG_ARGS_SIZE is dropped, there is no REG_ARGS_SIZE note on insn 207.  Insns
205/206/207 were introduced during split2.  Not sure what to do though, adding
REG_ARGS_SIZE notes in the ix86_free_from_memory and ix86_force_to_memory would
mean we'd have to find out what the current REG_ARGS_SIZE depth at that point
is.  And to solve this in csa pass, we'd need to find out that if we are
removing stack deallocation with REG_ARGS_SIZE note against later stack
allocation without REG_ARGS_SIZE, there must be some following stack
deallocation on which we would force REG_ARGS_SIZE note.  Richard?