https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109629
--- Comment #13 from Linus <linus.zhu at mavenir dot com> --- (In reply to Andrew Pinski from comment #12) > (In reply to Linus from comment #10) > > (In reply to Andrew Pinski from comment #5) > > > /* If we have a length attribute, this instruction should have > > > been split in shorten_branches, to ensure that we would have > > > valid length info for the splitees. */ > > > gcc_assert (!HAVE_ATTR_length); > > > > Yes, when I checked source code of gcc 11, I saw them. > > > I just wait for your new response, right? > > Thanks for checking the issue. > > No, I was just pointing out where the ICE is happening. But really it is > something inside the target backend's md file. Still need to reduce/debug it > to figure out which pattern is not being split. > > Also Please FILE this with redhat since this is a redhat modified compiler > first. Okay and thanks. Is there a chance that this issue will be gone if I try the original gcc 9 or gcc 10? By the way, same code didn't survive aarch64 gcc 9.3.1, but did survive X86 gcc 9.3.1.