This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Add zero extending patterns to vptest{,n}m instructions (PR target/88461)
- From: Uros Bizjak <ubizjak at gmail dot com>
- To: Jakub Jelinek <jakub at redhat dot com>
- Cc: "gcc-patches at gcc dot gnu dot org" <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 13 Dec 2018 08:54:37 +0100
- Subject: Re: [PATCH] Add zero extending patterns to vptest{,n}m instructions (PR target/88461)
- References: <20181212223608.GN12380@tucnak>
On Wed, Dec 12, 2018 at 11:36 PM Jakub Jelinek <jakub@redhat.com> wrote:
>
> Hi!
>
> The following patch uses a new mode iterator to avoid 2 almost same
> vptestm and 2 almost same vptestnm patterns, and adds patterns that zero
> extend those results to wider k registers, because that is what the
> instructions actually do.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2018-12-12 Jakub Jelinek <jakub@redhat.com>
>
> PR target/88461
> * config/i386/sse.md (VI1248_AVX512VLBW, AVX512ZEXTMASK): New
> mode iterators.
> (<avx512>_testm<mode>3<mask_scalar_merge_name>,
> <avx512>_testnm<mode>3<mask_scalar_merge_name>): Merge patterns
> with VI12_AVX512VL and VI48_AVX512VL iterators into ones with
> VI1248_AVX512VLBW iterator.
> (*<avx512>_testm<VI1248_AVX512VLBW:mode>3_zext,
> *<avx512>_testm<VI1248_AVX512VLBW:mode>3_zext_mask,
> *<avx512>_testnm<VI1248_AVX512VLBW:mode>3_zext,
> *<avx512>_testnm<VI1248_AVX512VLBW:mode>3_zext_mask): New
> define_insns.
>
> * gcc.target/i386/pr88461.c: New test.
LGTM.
Thanks,
Uros.
> --- gcc/config/i386/sse.md.jj 2018-12-03 21:56:35.252575482 +0100
> +++ gcc/config/i386/sse.md 2018-12-12 16:47:00.370433319 +0100
> @@ -12322,22 +12322,22 @@ (define_insn "*<code><mode>3_bcst"
> (set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
>
> -(define_insn "<avx512>_testm<mode>3<mask_scalar_merge_name>"
> - [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=Yk")
> - (unspec:<avx512fmaskmode>
> - [(match_operand:VI12_AVX512VL 1 "register_operand" "v")
> - (match_operand:VI12_AVX512VL 2 "nonimmediate_operand" "vm")]
> - UNSPEC_TESTM))]
> - "TARGET_AVX512BW"
> - "vptestm<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}"
> - [(set_attr "prefix" "evex")
> - (set_attr "mode" "<sseinsnmode>")])
> +(define_mode_iterator VI1248_AVX512VLBW
> + [(V64QI "TARGET_AVX512BW") (V32QI "TARGET_AVX512VL && TARGET_AVX512BW")
> + (V16QI "TARGET_AVX512VL && TARGET_AVX512BW")
> + (V32HI "TARGET_AVX512BW") (V16HI "TARGET_AVX512VL && TARGET_AVX512BW")
> + (V8HI "TARGET_AVX512VL && TARGET_AVX512BW")
> + V16SI (V8SI "TARGET_AVX512VL") (V4SI "TARGET_AVX512VL")
> + V8DI (V4DI "TARGET_AVX512VL") (V2DI "TARGET_AVX512VL")])
> +
> +(define_mode_iterator AVX512ZEXTMASK
> + [(DI "TARGET_AVX512BW") (SI "TARGET_AVX512BW") HI])
>
> (define_insn "<avx512>_testm<mode>3<mask_scalar_merge_name>"
> [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=Yk")
> (unspec:<avx512fmaskmode>
> - [(match_operand:VI48_AVX512VL 1 "register_operand" "v")
> - (match_operand:VI48_AVX512VL 2 "nonimmediate_operand" "vm")]
> + [(match_operand:VI1248_AVX512VLBW 1 "register_operand" "v")
> + (match_operand:VI1248_AVX512VLBW 2 "nonimmediate_operand" "vm")]
> UNSPEC_TESTM))]
> "TARGET_AVX512F"
> "vptestm<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}"
> @@ -12347,24 +12347,73 @@ (define_insn "<avx512>_testm<mode>3<mask
> (define_insn "<avx512>_testnm<mode>3<mask_scalar_merge_name>"
> [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=Yk")
> (unspec:<avx512fmaskmode>
> - [(match_operand:VI12_AVX512VL 1 "register_operand" "v")
> - (match_operand:VI12_AVX512VL 2 "nonimmediate_operand" "vm")]
> + [(match_operand:VI1248_AVX512VLBW 1 "register_operand" "v")
> + (match_operand:VI1248_AVX512VLBW 2 "nonimmediate_operand" "vm")]
> UNSPEC_TESTNM))]
> - "TARGET_AVX512BW"
> + "TARGET_AVX512F"
> "vptestnm<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}"
> [(set_attr "prefix" "evex")
> (set_attr "mode" "<sseinsnmode>")])
>
> -(define_insn "<avx512>_testnm<mode>3<mask_scalar_merge_name>"
> - [(set (match_operand:<avx512fmaskmode> 0 "register_operand" "=Yk")
> - (unspec:<avx512fmaskmode>
> - [(match_operand:VI48_AVX512VL 1 "register_operand" "v")
> - (match_operand:VI48_AVX512VL 2 "nonimmediate_operand" "vm")]
> - UNSPEC_TESTNM))]
> - "TARGET_AVX512F"
> - "vptestnm<ssemodesuffix>\t{%2, %1, %0<mask_scalar_merge_operand3>|%0<mask_scalar_merge_operand3>, %1, %2}"
> +(define_insn "*<avx512>_testm<VI1248_AVX512VLBW:mode>3_zext"
> + [(set (match_operand:AVX512ZEXTMASK 0 "register_operand" "=Yk")
> + (zero_extend:AVX512ZEXTMASK
> + (unspec:<VI1248_AVX512VLBW:avx512fmaskmode>
> + [(match_operand:VI1248_AVX512VLBW 1 "register_operand" "v")
> + (match_operand:VI1248_AVX512VLBW 2 "nonimmediate_operand" "vm")]
> + UNSPEC_TESTM)))]
> + "TARGET_AVX512BW
> + && (<AVX512ZEXTMASK:MODE_SIZE>
> + > GET_MODE_SIZE (<VI1248_AVX512VLBW:avx512fmaskmode>mode))"
> + "vptestm<VI1248_AVX512VLBW:ssemodesuffix>\t{%2, %1, %0|%0, %1, %2}"
> [(set_attr "prefix" "evex")
> - (set_attr "mode" "<sseinsnmode>")])
> + (set_attr "mode" "<VI1248_AVX512VLBW:sseinsnmode>")])
> +
> +(define_insn "*<avx512>_testm<VI1248_AVX512VLBW:mode>3_zext_mask"
> + [(set (match_operand:AVX512ZEXTMASK 0 "register_operand" "=Yk")
> + (zero_extend:AVX512ZEXTMASK
> + (and:<VI1248_AVX512VLBW:avx512fmaskmode>
> + (unspec:<VI1248_AVX512VLBW:avx512fmaskmode>
> + [(match_operand:VI1248_AVX512VLBW 1 "register_operand" "v")
> + (match_operand:VI1248_AVX512VLBW 2 "nonimmediate_operand" "vm")]
> + UNSPEC_TESTM)
> + (match_operand:<VI1248_AVX512VLBW:avx512fmaskmode> 3 "register_operand" "Yk"))))]
> + "TARGET_AVX512BW
> + && (<AVX512ZEXTMASK:MODE_SIZE>
> + > GET_MODE_SIZE (<VI1248_AVX512VLBW:avx512fmaskmode>mode))"
> + "vptestm<VI1248_AVX512VLBW:ssemodesuffix>\t{%2, %1, %0%{%3%}|%0%{%3%}, %1, %2}"
> + [(set_attr "prefix" "evex")
> + (set_attr "mode" "<VI1248_AVX512VLBW:sseinsnmode>")])
> +
> +(define_insn "*<avx512>_testnm<VI1248_AVX512VLBW:mode>3_zext"
> + [(set (match_operand:AVX512ZEXTMASK 0 "register_operand" "=Yk")
> + (zero_extend:AVX512ZEXTMASK
> + (unspec:<VI1248_AVX512VLBW:avx512fmaskmode>
> + [(match_operand:VI1248_AVX512VLBW 1 "register_operand" "v")
> + (match_operand:VI1248_AVX512VLBW 2 "nonimmediate_operand" "vm")]
> + UNSPEC_TESTNM)))]
> + "TARGET_AVX512BW
> + && (<AVX512ZEXTMASK:MODE_SIZE>
> + > GET_MODE_SIZE (<VI1248_AVX512VLBW:avx512fmaskmode>mode))"
> + "vptestnm<VI1248_AVX512VLBW:ssemodesuffix>\t{%2, %1, %0|%0, %1, %2}"
> + [(set_attr "prefix" "evex")
> + (set_attr "mode" "<VI1248_AVX512VLBW:sseinsnmode>")])
> +
> +(define_insn "*<avx512>_testnm<VI1248_AVX512VLBW:mode>3_zext_mask"
> + [(set (match_operand:AVX512ZEXTMASK 0 "register_operand" "=Yk")
> + (zero_extend:AVX512ZEXTMASK
> + (and:<VI1248_AVX512VLBW:avx512fmaskmode>
> + (unspec:<VI1248_AVX512VLBW:avx512fmaskmode>
> + [(match_operand:VI1248_AVX512VLBW 1 "register_operand" "v")
> + (match_operand:VI1248_AVX512VLBW 2 "nonimmediate_operand" "vm")]
> + UNSPEC_TESTNM)
> + (match_operand:<VI1248_AVX512VLBW:avx512fmaskmode> 3 "register_operand" "Yk"))))]
> + "TARGET_AVX512BW
> + && (<AVX512ZEXTMASK:MODE_SIZE>
> + > GET_MODE_SIZE (<VI1248_AVX512VLBW:avx512fmaskmode>mode))"
> + "vptestnm<VI1248_AVX512VLBW:ssemodesuffix>\t{%2, %1, %0%{%3%}|%0%{%3%}, %1, %2}"
> + [(set_attr "prefix" "evex")
> + (set_attr "mode" "<VI1248_AVX512VLBW:sseinsnmode>")])
>
> ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
> ;;
> --- gcc/testsuite/gcc.target/i386/pr88461.c.jj 2018-12-12 16:54:03.779529123 +0100
> +++ gcc/testsuite/gcc.target/i386/pr88461.c 2018-12-12 16:53:42.308879227 +0100
> @@ -0,0 +1,16 @@
> +/* PR target/88461 */
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mavx512vl -mavx512bw" } */
> +/* { dg-final { scan-assembler-times "kmovw\[ \t]" 2 } } */
> +
> +#include <x86intrin.h>
> +
> +int
> +foo (const __m128i *data, int a)
> +{
> + __m128i v = _mm_load_si128 (data);
> + __mmask16 m = _mm_testn_epi16_mask (v, v);
> + m = _kshiftli_mask16 (m, 1);
> + m = _kandn_mask16 (m, a);
> + return m;
> +}
>
> Jakub