https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105215
Frank Scheiner <frank.scheiner at web dot de> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |frank.scheiner at web dot de
--- Comment #4 from Frank Scheiner <frank.scheiner at web dot de> ---
(In reply to Andrew Pinski from comment #3)
> This is not a dup of bug 85410 as there is no loop there.
This problem seems to be gone or hidden with gcc-15-20240630 ([1]):
```
$ md5sum reduced-reproducer.c
0b62444c75e58e96e1defd20de71f82a reduced-reproducer.c
$ /opt/gcc-15-20240630-lra-nolibc/ia64-linux/bin/ia64-linux-gcc -std=gnu17
-Wall -O3 -c reduced-reproducer.c -o bug.o -g2
$ echo $?
0
$ /opt/gcc-15-20240630-nolibc/ia64-linux/bin/ia64-linux-gcc -std=gnu17 -Wall
-O3 -c reduced-reproducer.c -o bug.o -g2
$ echo $?
0
```
[1]:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=30ad2fafa9ab2497cc12df62a3240cff6ef25d00
Out of luck I still had cross-compilers from the previous snapshot
(gcc-15-20240623 ([2])) installed, and with these, the problem is still
present. Shown in the following is the stock version w/o LRA enabled, but the
LRA enabled one yields the same (except a changed line number due to the LRA
enablement):
```
$ /opt/gcc-15-20240623-nolibc/ia64-linux/bin/ia64-linux-gcc -std=gnu17 -Wall
-O3 -c reduced-reproducer.c -o bug.o -g2
during RTL pass: mach
reduced-reproducer.c: In function 'X509_CRL_add0_revoked':
reduced-reproducer.c:24:1: internal compiler error: in
create_block_for_bookkeeping, at sel-sched.cc:4557
24 | }
| ^
0x5f4493 create_block_for_bookkeeping
/usr/src/gcc/gcc/sel-sched.cc:4557
0x5f4493 find_place_for_bookkeeping
/usr/src/gcc/gcc/sel-sched.cc:4694
0x5f4493 generate_bookkeeping_insn
/usr/src/gcc/gcc/sel-sched.cc:4794
0x5f4493 move_op_at_first_insn
/usr/src/gcc/gcc/sel-sched.cc:6071
0xbc6545 code_motion_path_driver
/usr/src/gcc/gcc/sel-sched.cc:6665
0xbc6936 code_motion_process_successors
/usr/src/gcc/gcc/sel-sched.cc:6350
0xbc6936 code_motion_path_driver
/usr/src/gcc/gcc/sel-sched.cc:6616
0xbc8adb move_op
/usr/src/gcc/gcc/sel-sched.cc:6710
0xbc8adb move_exprs_to_boundary
/usr/src/gcc/gcc/sel-sched.cc:5231
0xbc8adb schedule_expr_on_boundary
/usr/src/gcc/gcc/sel-sched.cc:5444
0xbca9d7 fill_insns
/usr/src/gcc/gcc/sel-sched.cc:5586
0xbca9d7 schedule_on_fences
/usr/src/gcc/gcc/sel-sched.cc:7361
0xbca9d7 sel_sched_region_2
/usr/src/gcc/gcc/sel-sched.cc:7499
0xbcc17b sel_sched_region_1
/usr/src/gcc/gcc/sel-sched.cc:7541
0xbcdb0b sel_sched_region(int)
/usr/src/gcc/gcc/sel-sched.cc:7642
0xbcdb0b sel_sched_region(int)
/usr/src/gcc/gcc/sel-sched.cc:7627
0xbce2e1 run_selective_scheduling()
/usr/src/gcc/gcc/sel-sched.cc:7728
0xf31e2e ia64_reorg
/usr/src/gcc/gcc/config/ia64/ia64.cc:9870
0xb9fff9 execute
/usr/src/gcc/gcc/reorg.cc:3931
```
[2]:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=eb647daa87b466d0a71246fad302cd81bfce9be5
****
So some change between [2] and [1] possibly fixed that issue for ia64.
@Andrew: Maybe you can already spot the related/most likely commit in that
range?
****
Going back to `-O2` with gcc-15-20240623 also makes it go away and I also
didn't see any issue for the OpenSSL versions compiled for EPIC Slack, e.g.
[3]. But this is understandable as Slackware (and hence EPIC Slack) compiles
packages with `-O2` by default. I also believe T2 doesn't go beyond `-O2` for
ia64 packages.
[3]:
https://ftp.machine-hall.org/pub/slackware/epic-slack-current/slackware64/n/openssl-3.5.1-ia64-1.txz
****
Otherwise it's still present for me with:
* GCC 12.4.0
* GCC 13.3.0
* GCC 14.2.0
...cross-compilers and `-O3`. I haven't yet built any newer cross-compilers for
these major versions and also didn't test them with `-O2`.
