[Bug middle-end/85878] [6/7/8/9 Regression] ICE in convert_mode_scalar, at expr.c:287

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85878

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |ice-on-valid-code
           Priority|P3                          |P2
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2018-05-23
          Component|fortran                     |middle-end
   Target Milestone|---                         |6.5
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Looks like a middle-end issue to me.  We expand

MEM[(c_char * {ref-all})&transfer.2] = _23;

where transfer.2 is expanded to

 (concat:SC (reg:SF 91 [ transfer.2 ])
    (reg:SF 92 [ transfer.2+4 ])

and _23 is

 <ssa_name 0x7ffff6a66120
    type <integer_type 0x7ffff68a8690 character(kind=4) public unsigned SI
        size <integer_cst 0x7ffff6894e28 constant 32>
        unit-size <integer_cst 0x7ffff6894e40 constant 4>
        align:32 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff68a8690 precision:32 min <integer_cst 0x7ffff6894e58 0> max <integer_cst
0x7ffff6894e10 4294967295>
        pointer_to_this <pointer_type 0x7ffff68b4690>>
    visited
    def_stmt _23 = VIEW_CONVERT_EXPR<character(kind=4)>(a_24(D));
    version:23>

and we call convert_move with

(reg:SF 91 [ transfer.2 ])
(subreg:SI (reg/v:SF 90 [ a ]) 0)

it might go wrong in expand_assignment which "loses" the punning on the LHS.