On Thu, Jan 26, 2017 at 09:42:39PM +0100, Dominik Vogt wrote:
> gcc/ChangeLog-pr79240
>
> PR target/79240
> * config/s390/s390.md ("*r<noxa>sbg_<mode>_srl_bitmask")
> ("*r<noxa>sbg_<mode>_sll_bitmask")
> ("*extzv_<mode>_srl<clobbercc_or_nocc>")
> ("*extzv_<mode>_sll<clobbercc_or_nocc>"):
> Use contiguous_bitmask_nowrap_operand
Missing full stop at the end.
> --- a/gcc/config/s390/s390.md
> +++ b/gcc/config/s390/s390.md
> @@ -4127,7 +4127,7 @@
> (lshiftrt:GPR
> (match_operand:GPR 1 "nonimmediate_operand" "d")
> (match_operand:GPR 3 "nonzero_shift_count_operand" ""))
> - (match_operand:GPR 2 "contiguous_bitmask_operand" ""))
> + (match_operand:GPR 2 "contiguous_bitmask_nowrap_operand" ""))
> (match_operand:GPR 4 "nonimmediate_operand" "0")))
> (clobber (reg:CC CC_REGNUM))]
> "TARGET_Z10
This shows another cleanup possibility for later (GCC8), for
(match_operand... "whatever_operand" "")
one can use just
(match_operand... "whatever_operand")
The empty constraint is implicit. E.g. in i386/*.md Uros has done that some
time ago and keeps redoing it if some of these get in.
One can verify if it doesn't change anything by doing make mddump
before/after and comparing the tmp-mddump.md files.
Jakub
