On Fri, 16 Dec 2022, Jakub Jelinek wrote:
> Hi!
>
> The RTL loop passes only request simple preheaders, but don't require
> fallthru preheaders, while move_invariant_reg apparently assumes the
> latter, that it can just append instruction(s) to the end of the preheader
> basic block.
>
> The following patch fixes that by splitting the preheader edge if
> the preheader bb ends with a JUMP_INSN (asm goto in this case).
> Without that we get control flow in the middle of a bb.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK.
> 2022-12-16 Jakub Jelinek <ja...@redhat.com>
>
> PR rtl-optimization/106751
> * loop-invariant.cc (move_invariant_reg): If preheader bb ends
> with a JUMP_INSN, split the preheader edge and emit invariants
> into the new preheader basic block.
>
> * gcc.c-torture/compile/pr106751.c: New test.
>
> --- gcc/loop-invariant.cc.jj 2022-01-18 11:58:59.683980614 +0100
> +++ gcc/loop-invariant.cc 2022-12-15 17:14:32.906883258 +0100
> @@ -1837,6 +1837,8 @@ move_invariant_reg (class loop *loop, un
> else if (dump_file)
> fprintf (dump_file, "Invariant %d moved without introducing a new "
> "temporary register\n", invno);
> + if (JUMP_P (BB_END (preheader)))
> + preheader = split_edge (loop_preheader_edge (loop));
> reorder_insns (inv->insn, inv->insn, BB_END (preheader));
> df_recompute_luids (preheader);
>
> --- gcc/testsuite/gcc.c-torture/compile/pr106751.c.jj 2022-12-15
> 17:44:35.602519711 +0100
> +++ gcc/testsuite/gcc.c-torture/compile/pr106751.c 2022-12-15
> 17:44:21.789721827 +0100
> @@ -0,0 +1,17 @@
> +/* PR rtl-optimization/106751 */
> +
> +int *foo (void);
> +
> +void
> +bar (void)
> +{
> + asm goto ("" : : : : lab);
> + __builtin_unreachable ();
> +lab:
> + while (1)
> + {
> + int o;
> + asm ("" : "=r" (o) : "g" (1));
> + *foo () = o;
> + }
> +}
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Frankenstrasse 146, 90461 Nuernberg,
Germany; GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman;
HRB 36809 (AG Nuernberg)
