https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120881
--- Comment #3 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to cuilili from comment #2)
> I think it is an old bug, since shrink wrap, NOTE_INSN_PROLOGUE_END does not
> represent the entry bb.
There are
(note 99 98 100 2 NOTE_INSN_PROLOGUE_END)
(insn 100 99 3 2 (unspec_volatile [
(const_int 0 [0])
] UNSPECV_BLOCKAGE) "x.c":21:1 1490 {blockage}
(nil))
(note 3 100 114 2 NOTE_INSN_FUNCTION_BEG)
But the sched2 pass still moves it around.
