On Tue, 3 Apr 2018, Andrey Belevantsev wrote:
> Hello,
>
> This issues is about the correct order in which we need to call the
> routines that fix up the control flow for us.
OK with formatting both in the new comment and the Changelog fixed.
> Best,
> Andrey
>
> 2018-04-03 Andrey Belevantsev <a...@ispras.ru>
>
> PR rtl-optimization/83962
>
> * sel-sched-ir.c (tidy_control_flow): Correct the order in which we call
> tidy_fallthru_edge
> and tidy_control_flow.
>
> * gcc.dg/pr83962.c: New test.
>
> diff --git a/gcc/sel-sched-ir.c b/gcc/sel-sched-ir.c
> index a965d2ec42f..f6de96a7f3d 100644
> --- a/gcc/sel-sched-ir.c
> +++ b/gcc/sel-sched-ir.c
> @@ -3839,9 +3839,13 @@ tidy_control_flow (basic_block xbb, bool full_tidying)
> && INSN_SCHED_TIMES (BB_END (xbb)) == 0
> && !IN_CURRENT_FENCE_P (BB_END (xbb)))
> {
> - if (sel_remove_insn (BB_END (xbb), false, false))
> - return true;
> + /* We used to call sel_remove_insn here that can trigger
> tidy_control_flow
> + before we fix up the fallthru edge. Correct that ordering by
> + explicitly doing the latter before the former. */
> + clear_expr (INSN_EXPR (BB_END (xbb)));
> tidy_fallthru_edge (EDGE_SUCC (xbb, 0));
> + if (tidy_control_flow (xbb, false))
> + return true;
> }
>
> first = sel_bb_head (xbb);
> diff --git a/gcc/testsuite/gcc.dg/pr83962.c b/gcc/testsuite/gcc.dg/pr83962.c
> new file mode 100644
> index 00000000000..0547e218715
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/pr83962.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile { target powerpc*-*-* ia64-*-* i?86-*-* x86_64-*-* } } */
> +/* { dg-options "-std=gnu99 -O1 -fselective-scheduling2 -fschedule-insns2
> -fcse-follow-jumps -fno-ssa-phiopt -fno-guess-branch-probability" } */
> +unsigned int ca;
> +
> +void
> +v6 (long long unsigned int as, int p9)
> +{
> + while (p9 < 1)
> + as = (as != ca) || (as > 1);
> +}
>
