Re: [PATCH, i386] Support ANDN in stv pass

[Uros Bizjak <ubizjak at gmail dot com>]

On Fri, Jan 15, 2016 at 4:46 PM, Ilya Enkovich <enkovich.gnu@gmail.com> wrote:
> Hi,
>
> This patch continues resolving andn regression case in stv pass
> (see https://gcc.gnu.org/ml/gcc-patches/2016-01/msg01017.html).
> In this patch a new andn pattern added similar to other bit
> DI patterns we have for stv pass.
>
> This improves performance of 462.libquantum benchmark on Haswell
> (+2.6% on -O2, +1% on -O3 -flto).
>
> Unfortunately this patch doesn't enable generation of pandn in case
> target doesn't have BMI.  Probably peephole may be used for such targets?
> Or we may allow andn and then split it back to and + xor for them.

IMO, we want a splitter here. We should optimize compiler for newer targets.

> Bootstrapped and regtested on x86_64-unknown-linux-gnu.  OK for trunk?
>
> Thanks,
> Ilya
> --
> gcc/
>
> 2016-01-15  Ilya Enkovich  <enkovich.gnu@gmail.com>
>
>         * config/i386/i386.c (scalar_to_vector_candidate_p): Support
>         andnot instruction.
>         (scalar_chain::convert_op): Likewise.
>         * config/i386/i386.md (*andndi3_doubleword): New.
>
> gcc/testsuite/
>
> 2016-01-15  Ilya Enkovich  <enkovich.gnu@gmail.com>
>
>         * gcc.target/i386/pr65105-5.c: Adjust to andn generation.

OK for mainline.

Thanks,
Uros.

>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index de41477..a0b0d68 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -2815,7 +2815,11 @@ scalar_to_vector_candidate_p (rtx_insn *insn)
>        return false;
>      }
>
> -  if (!REG_P (XEXP (src, 0)) && !MEM_P (XEXP (src, 0)))
> +  if (!REG_P (XEXP (src, 0)) && !MEM_P (XEXP (src, 0))
> +      /* Check for andnot case.  */
> +      && (GET_CODE (src) != AND
> +         || GET_CODE (XEXP (src, 0)) != NOT
> +         || !REG_P (XEXP (XEXP (src, 0), 0))))
>        return false;
>
>    if (!REG_P (XEXP (src, 1)) && !MEM_P (XEXP (src, 1)))
> @@ -3383,7 +3387,12 @@ scalar_chain::convert_op (rtx *op, rtx_insn *insn)
>  {
>    *op = copy_rtx_if_shared (*op);
>
> -  if (MEM_P (*op))
> +  if (GET_CODE (*op) == NOT)
> +    {
> +      convert_op (&XEXP (*op, 0), insn);
> +      PUT_MODE (*op, V2DImode);
> +    }
> +  else if (MEM_P (*op))
>      {
>        rtx tmp = gen_reg_rtx (DImode);
>
> diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
> index 71941d0..f16b42a 100644
> --- a/gcc/config/i386/i386.md
> +++ b/gcc/config/i386/i386.md
> @@ -8645,6 +8645,23 @@
>               (clobber (reg:CC FLAGS_REG))])]
>    "split_double_mode (DImode, &operands[0], 3, &operands[0], &operands[3]);")
>
> +(define_insn_and_split "*andndi3_doubleword"
> +  [(set (match_operand:DI 0 "register_operand" "=r,r")
> +       (and:DI
> +         (not:DI (match_operand:DI 1 "register_operand" "r,r"))
> +         (match_operand:DI 2 "nonimmediate_operand" "r,m")))
> +   (clobber (reg:CC FLAGS_REG))]
> +  "TARGET_BMI && !TARGET_64BIT && TARGET_STV && TARGET_SSE"
> +  "#"
> +  "&& reload_completed"
> +  [(parallel [(set (match_dup 0)
> +                  (and:SI (not:SI (match_dup 1)) (match_dup 2)))
> +             (clobber (reg:CC FLAGS_REG))])
> +   (parallel [(set (match_dup 3)
> +                  (and:SI (not:SI (match_dup 4)) (match_dup 5)))
> +             (clobber (reg:CC FLAGS_REG))])]
> +  "split_double_mode (DImode, &operands[0], 3, &operands[0], &operands[3]);")
> +
>  (define_insn "*<code>hi_1"
>    [(set (match_operand:HI 0 "nonimmediate_operand" "=r,rm,!k")
>         (any_or:HI
> diff --git a/gcc/testsuite/gcc.target/i386/pr65105-5.c b/gcc/testsuite/gcc.target/i386/pr65105-5.c
> index 5818c1c..639bbe1 100644
> --- a/gcc/testsuite/gcc.target/i386/pr65105-5.c
> +++ b/gcc/testsuite/gcc.target/i386/pr65105-5.c
> @@ -1,7 +1,7 @@
>  /* PR target/pr65105 */
>  /* { dg-do compile { target { ia32 } } } */
>  /* { dg-options "-O2 -march=core-avx2" } */
> -/* { dg-final { scan-assembler "pand" } } */
> +/* { dg-final { scan-assembler "pandn" } } */
>  /* { dg-final { scan-assembler "pxor" } } */
>  /* { dg-final { scan-assembler "ptest" } } */
>