On Mon, Jan 18, 2021 at 11:04 PM Ilya Leoshkevich via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> Boostrapped and regtested on x86_64-redhat-linux, ppc64le-redhat-linux
> and s390x-redhat-linux. I realize it might be too late for a change
> like this, but it's desirable to have this in conjunction with the
> https://gcc.gnu.org/pipermail/gcc-patches/2021-January/563799.html s390
> regression fix, which otherwise produces unnecessary store/load
> sequences in certain glibc routines, e.g. __ieee754_sqrtl. Ok for
> master?
>
>
>
> Suppose we have:
>
> (set (reg/v:TF 63) (mem/c:TF (reg/v:DI 62)))
> (set (reg:FPRX2 66) (subreg:FPRX2 (reg/v:TF 63) 0))
>
> It is clearly profitable to propagate the first insn into the second
> one and get:
>
> (set (reg:FPRX2 66) (mem/c:FPRX2 (reg/v:DI 62)))
>
> fwprop actually manages to perform this, but doesn't think the result is
> worth it, which results in unnecessary store/load sequences on s390.
> Improve the situation by classifying SUBREG -> MEM changes as
> profitable.
IIRC fwprop also propagates into multiple uses and replacing a non-MEM
with a MEM is only good when the original MEM goes away - is that properly
dealt with here?
Richard.
> gcc/ChangeLog:
>
> 2021-01-15 Ilya Leoshkevich <i...@linux.ibm.com>
>
> * fwprop.c (fwprop_propagation::classify_result): Allow
> (subreg (mem)) simplifications.
>
> gcc/testsuite/ChangeLog:
>
> 2021-01-15 Ilya Leoshkevich <i...@linux.ibm.com>
>
> * gcc.target/s390/vector/long-double-to-i64.c: Expect that
> float-vector moves do *not* happen.
> ---
> gcc/fwprop.c | 5 +++++
> gcc/testsuite/gcc.target/s390/vector/long-double-to-i64.c | 3 +--
> 2 files changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/gcc/fwprop.c b/gcc/fwprop.c
> index eff8f7cc141..46b8ec7eccf 100644
> --- a/gcc/fwprop.c
> +++ b/gcc/fwprop.c
> @@ -262,6 +262,11 @@ fwprop_propagation::classify_result (rtx old_rtx, rtx
> new_rtx)
> && GET_MODE (new_rtx) == GET_MODE_INNER (GET_MODE (from)))
> return PROFITABLE;
>
> + /* Allow (subreg (mem)) -> (mem) simplifications. However, do not allow
> + creating new (mem/v)s, since DCE will not remove the old ones. */
> + if (SUBREG_P (old_rtx) && MEM_P (new_rtx) && !MEM_VOLATILE_P (new_rtx))
> + return PROFITABLE;
> +
> return 0;
> }
>
> diff --git a/gcc/testsuite/gcc.target/s390/vector/long-double-to-i64.c
> b/gcc/testsuite/gcc.target/s390/vector/long-double-to-i64.c
> index 2dbbb5d1c03..8f4e377ed72 100644
> --- a/gcc/testsuite/gcc.target/s390/vector/long-double-to-i64.c
> +++ b/gcc/testsuite/gcc.target/s390/vector/long-double-to-i64.c
> @@ -10,8 +10,7 @@ long_double_to_i64 (long double x)
> return x;
> }
>
> -/* { dg-final { scan-assembler-times {\n\tvpdi\t%v\d+,%v\d+,%v\d+,1\n} 1 } }
> */
> -/* { dg-final { scan-assembler-times {\n\tvpdi\t%v\d+,%v\d+,%v\d+,5\n} 1 } }
> */
> +/* { dg-final { scan-assembler-not {\n\tvpdi\t} } } */
> /* { dg-final { scan-assembler-times {\n\tcgxbr\t} 1 } } */
>
> int
> --
> 2.26.2
>
