https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122502
--- Comment #16 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Andrew Pinski from comment #14)
> (In reply to Sergei Trofimovich from comment #13)
> > Another example seemingly not fixed by this change from `ffmpeg-8.0`:
> >
> > // $ cat softfloat.c.c
> > typedef struct {
> > int mant;
> > int exp
> > } SoftFloat;
> > SoftFloat __trans_tmp_8, main___trans_tmp_5;
> > static SoftFloat av_normalize_sf(SoftFloat a) {
> > while (a.mant + 536870911 < 1073741823) {
> > a.mant += a.mant;
> > a.exp -= 1;
> > }
> > return a;
> > }
> > void main() {
> > main___trans_tmp_5 = av_normalize_sf((SoftFloat){1, 29 + 1});
> > SoftFloat sf1 = main___trans_tmp_5;
> > for (;;) {
> > int t = main___trans_tmp_5.exp - sf1.exp;
> > if (t < 2)
> > sf1 = __trans_tmp_8;
> > }
> > }
> >
> > gcc/xgcc -Bgcc -c softfloat.c.c -O2
> > softfloat.c.c:4:1: warning: no semicolon at end of struct or union
> > 4 | } SoftFloat;
> > | ^
> > during GIMPLE pass: sccp
> > softfloat.c.c: In function ‘main’:
> > softfloat.c.c:13:6: internal compiler error: Segmentation fault
> > 13 | void main() {
> > | ^~~~
> > 0x26506c4 diagnostics::context::diagnostic_impl(rich_location*,
> > diagnostics::metadata const*, diagnostics::option_id, char const*,
> > __va_list_tag (*) [1], diagnostics::kind)
>
> This is the similar issue but in a different location. This time it is back
> in replace_uses_by which does:
> ```
> FOR_EACH_IMM_USE_STMT (stmt, imm_iter, name)
> {
> /* Mark the block if we change the last stmt in it. */
> if (cfgcleanup_altered_bbs
> && stmt_ends_bb_p (stmt))
> bitmap_set_bit (cfgcleanup_altered_bbs, gimple_bb (stmt)->index);
>
> FOR_EACH_IMM_USE_ON_STMT (use, imm_iter)
> {
> ...
> if (fold_stmt (&gsi))
> stmt = gsi_stmt (gsi);
> ...
> ```
> I remember running into this exact same thing before; I remember doing the
> folding after the FOR_EACH loop in replace_uses_by but I can't remember what
> I did to hit it nor why I decided I was doing something else wrong.
The proposed patch at
https://gcc.gnu.org/pipermail/gcc-patches/2025-October/699165.html fixes this
as well.
I'll note this "pattern" is what is currently documented as way to fold
uses.
