[PATCH] Fix rtl_split_edge (PR rtl-optimization/65761)

[PATCH] Fix rtl_split_edge (PR rtl-optimization/65761)

[Jakub Jelinek]

Hi!

On the testcase from the PR (not suitable for testsuite, as it is
-fprofile-use only with *.gcda required and thus hard to reduce),
we ICE because rtl_split_edge decided to insert a new basic block
in between a tablejump instruction and corresponding jump table.
Fixed by using get_last_bb_insn, which skips the jump table and also
in case there are multiple barriers also the extra barriers.

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk/5.1?

2015-04-14  Jakub Jelinek  <jakub@redhat.com>

	PR rtl-optimization/65761
	* cfgrtl.c (rtl_split_edge): For EDGE_CROSSING split, use
	get_last_bb_insn (after) instead of NEXT_INSN (BB_END (after)).

--- gcc/cfgrtl.c.jj	2015-04-12 21:50:18.000000000 +0200
+++ gcc/cfgrtl.c	2015-04-14 12:45:34.127722958 +0200
@@ -1928,7 +1928,7 @@ rtl_split_edge (edge edge_in)
               && (edge_in->flags & EDGE_CROSSING))
             {
               after = last_bb_in_partition (edge_in->src);
-              before = NEXT_INSN (BB_END (after));
+              before = get_last_bb_insn (after);
               /* The instruction following the last bb in partition should
                  be a barrier, since it cannot end in a fall-through.  */
               gcc_checking_assert (BARRIER_P (before));

	Jakub

Re: [PATCH] Fix rtl_split_edge (PR rtl-optimization/65761)

[Steven Bosscher]

On Tue, Apr 14, 2015 at 2:58 PM, Jakub Jelinek wrote:
>         PR rtl-optimization/65761
>         * cfgrtl.c (rtl_split_edge): For EDGE_CROSSING split, use
>         get_last_bb_insn (after) instead of NEXT_INSN (BB_END (after)).
>
> --- gcc/cfgrtl.c.jj     2015-04-12 21:50:18.000000000 +0200
> +++ gcc/cfgrtl.c        2015-04-14 12:45:34.127722958 +0200
> @@ -1928,7 +1928,7 @@ rtl_split_edge (edge edge_in)
>                && (edge_in->flags & EDGE_CROSSING))
>              {
>                after = last_bb_in_partition (edge_in->src);
> -              before = NEXT_INSN (BB_END (after));
> +              before = get_last_bb_insn (after);
>                /* The instruction following the last bb in partition should
>                   be a barrier, since it cannot end in a fall-through.  */
>                gcc_checking_assert (BARRIER_P (before));

This is OK.

Ciao!
Steven