On 5/9/25 2:27 PM, Vineet Gupta wrote:
I don't see how get_last_bb_insn itself inherently needs this behavior. I could believe something that *calls* get_last_bb_insn might depend on this behavior -- but I'd also consider that bogus. The CODE_LABEL is part of the next block, so returning that seems quite wrong given the original comment and users like get_last_bb_insn.gcc/ChangeLog: * emit-rtl.cc (next_nonnote_nondebug_insn): Update comments. Signed-off-by: Vineet Gupta <vine...@rivosinc.com> --- gcc/emit-rtl.cc | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-) diff --git a/gcc/emit-rtl.cc b/gcc/emit-rtl.cc index 3e2c4309dee6..b78b29ecf989 100644 --- a/gcc/emit-rtl.cc +++ b/gcc/emit-rtl.cc @@ -3677,7 +3677,11 @@ next_nonnote_nondebug_insn (rtx_insn *insn)/* Return the next insn after INSN that is not a NOTE nor DEBUG_INSN,but stop the search before we enter another basic block. This - routine does not look inside SEQUENCEs. */ + routine does not look inside SEQUENCEs. + NOTE: This can potentially bleed into next BB. If current insn is + last insn of BB, followed by a code_label before the start of + the next BB, code_label will be returned. But this is the + behavior rest of gcc assumes/relies on e.g. get_last_bb_insn. */
I don't mind adding the comment, but I'd much rather chase down the offenders and fix them.
Jeff
