Review ping for this patch. IMO it almost counts as obviously correct
but just in case...
Kito Cheng <k...@0xlab.org> writes:
> expand_movstr is work fine when we don't define movstr pattern or
> always expand it successfully, however it's will crash when if movstr
> expand fail since expand_insn expect always expand successfully (it's
> place a gcc_unreachable() when expand fail).
>
> this patch use maybe_expand_insn instead of expand_insn, check it done
> or not and just return NULL_RTX for generate a libcall if fail to
> expand movstr.
>
> 2014-01-11 Kito Cheng <k...@0xlab.org>
>
> * gcc/builtins.c (expand_movstr): Check movstr expand done or fail.
>
> diff --git a/gcc/builtins.c b/gcc/builtins.c
> index e3c32a9..fca51a1 100644
> --- a/gcc/builtins.c
> +++ b/gcc/builtins.c
> @@ -3239,7 +3239,8 @@ expand_movstr (tree dest, tree src, rtx target, int
> endp)
> create_output_operand (&ops[0], endp ? target : NULL_RTX, Pmode);
> create_fixed_operand (&ops[1], dest_mem);
> create_fixed_operand (&ops[2], src_mem);
> - expand_insn (CODE_FOR_movstr, 3, ops);
> + if (!maybe_expand_insn (CODE_FOR_movstr, 3, ops))
> + return NULL_RTX;
>
> if (endp && target != const0_rtx)
> {
