https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99308
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
CC| |burnus at gcc dot gnu.org
--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> ---
I think there are two errors:
(a) The value can be simplified at compile time as the size (8 or 2) is known
at compile time, but isn't.
I wonder whether this only a problem for 'size' or also for 'ubound','lbound'
or when doing
var(:)%idx = 0 – for the bounds of the generated do loop.
(b) The run-time version is wrong.
In my dump, (and output '2 2') I see for either variable
parm.10.span = spcs->_vptr->_size;
parm.10.dtype = {.elem_len=4, .rank=1, .type=5};
D.3995 = stride.5;
parm.10.dim[0].lbound = 1;
parm.10.dim[0].ubound = 2;
parm.10.dim[0].stride = NON_LVALUE_EXPR <D.3995>;
parm.10.data = 0B;
parm.10.offset = -NON_LVALUE_EXPR <D.3995>;
D.3996 = &parm.10;
D.3997 = (integer(kind=4)) MAX_EXPR <(D.3996->dim[0].ubound -
D.3996->dim[0].lbound) + 1, 0>;
Obviously, 'ubound = 2' is only valid for 'fuel' and not for 'spcs'.
