https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122373
H.J. Lu <hjl.tools at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Component|middle-end |fortran
Ever confirmed|0 |1
Resolution|INVALID |---
Status|RESOLVED |REOPENED
Last reconfirmed| |2025-10-22
--- Comment #2 from H.J. Lu <hjl.tools at gmail dot com> ---
For gfortran.dg/deferred_character_23.f90, with -O0, middle-end passes
(gdb) call debug (max_size_exp)
(const_int 32 [0x20])
(gdb) call debug (min_size_exp)
(const_int 32 [0x20])
(gdb) call debug (count_exp)
(subreg:SI (reg:DI 118 [ _21 ]) 0)
to movmem expander from
.strings_267 = 32;
...
...
<bb 14> :
_288 = .strings_267;
_14 = S.10_214 + _286;
_15 = _283 + _14;
_16 = strings.span;
_17 = _15 * _16;
_18 = (sizetype) _17;
_289 = _282 + _18;
_290 = A.8[S.10_214];
_291 = _288 + -32;
if (_288 > 0)
goto <bb 15>; [INV]
else
goto <bb 18>; [INV]
<bb 15> :
if (_288 > 32)
goto <bb 16>; [INV]
else
goto <bb 17>; [INV]
<bb 16> :
__builtin_memmove (_289, _290, 32);
_19 = (unsigned long) _291;
_20 = _289 + 32;
__builtin_memset (_20, 32, _19);
goto <bb 18>; [INV]
<bb 17> :
_21 = (unsigned long) _288;
__builtin_memmove (_289, _290, _21);
<bb 18> :
S.10_295 = S.10_214 + 1;
goto <bb 13>; [INV]
...
_27 = _gfortran_string_len_trim (.strings_267, _26);
At run-time, 2 is passed to memmove, instead of 32. Does Fortran frontend let
the middle-end know that _gfortran_string_len_trim may change .strings_267?
