Segher Boessenkool <seg...@kernel.crashing.org> writes:
> Combine will put CLOBBER (with a non-void mode) anywhere in a pattern
> to poison it. reg_overlap_mentioned_p did not handle this. This patch
> fixes that.
>
> Tested on the PR testcase on x86_64-linux; also bootstrapped and
> regression checked on powerpc64-linux {-m32,-m64}. Is this okay for
> trunk and backports?
Yes, thanks.
> What should I do with the testcase? I don't think tool-generated
> testcases are useful at all (instead, we should ship the tool!); but
> if people want it anyway, should this be gcc.target/x86_64, or more
> general somehow?
Seems obvious enough not to need a testcase IMO.
Richard
> Segher
>
>
> 2018-09-18 Segher Boessenkool <seg...@kernel.crashing.org>
>
> PR rtl-optimization/86882
> * rtlanal.c (reg_overlap_mentioned_p): Handle CLOBBER.
>
> ---
> gcc/rtlanal.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/gcc/rtlanal.c b/gcc/rtlanal.c
> index 6c620a4..e8b6b9c 100644
> --- a/gcc/rtlanal.c
> +++ b/gcc/rtlanal.c
> @@ -1815,6 +1815,7 @@ reg_overlap_mentioned_p (const_rtx x, const_rtx in)
> recurse:
> switch (GET_CODE (x))
> {
> + case CLOBBER:
> case STRICT_LOW_PART:
> case ZERO_EXTRACT:
> case SIGN_EXTRACT:
