This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Enable scatter vectorization with 128-bit and 256-bit vectors with AVX512VL (PR target/88464)
- 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: Tue, 18 Dec 2018 19:36:15 +0100
- Subject: Re: [PATCH] Enable scatter vectorization with 128-bit and 256-bit vectors with AVX512VL (PR target/88464)
- References: <20181218145735.GO23305@tucnak>
On Tue, Dec 18, 2018 at 3:57 PM Jakub Jelinek <jakub@redhat.com> wrote:
>
> Hi!
>
> We weren't vectorizing with unconditional or masked scatters when
> -mprefered-vector-width={128,256}. While for DI index and DF/DI
> stores or SI index and SF/SI stores we even have the builtins,
> for the remaining combinations I had to add a few alt builtins (with spaces
> in names as in other cases). I've also renamed the other alt builtin
> visible names so that they match the IX86_BUILTIN_* names, they were pretty
> confusing before.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
> 2018-12-18 Jakub Jelinek <jakub@redhat.com>
>
> PR target/88464
> * config/i386/i386-builtin-types.def
> (VOID_FTYPE_PDOUBLE_QI_V8SI_V4DF_INT,
> VOID_FTYPE_PFLOAT_QI_V4DI_V8SF_INT,
> VOID_FTYPE_PLONGLONG_QI_V8SI_V4DI_INT,
> VOID_FTYPE_PINT_QI_V4DI_V8SI_INT,
> VOID_FTYPE_PDOUBLE_QI_V4SI_V2DF_INT,
> VOID_FTYPE_PFLOAT_QI_V2DI_V4SF_INT,
> VOID_FTYPE_PLONGLONG_QI_V4SI_V2DI_INT,
> VOID_FTYPE_PINT_QI_V2DI_V4SI_INT): New builtin types.
> * config/i386/i386.c (enum ix86_builtins): Add
> IX86_BUILTIN_SCATTERALTSIV4DF, IX86_BUILTIN_SCATTERALTDIV8SF,
> IX86_BUILTIN_SCATTERALTSIV4DI, IX86_BUILTIN_SCATTERALTDIV8SI,
> IX86_BUILTIN_SCATTERALTSIV2DF, IX86_BUILTIN_SCATTERALTDIV4SF,
> IX86_BUILTIN_SCATTERALTSIV2DI and IX86_BUILTIN_SCATTERALTDIV4SI.
> (ix86_init_mmx_sse_builtins): Fix up names of IX86_BUILTIN_GATHERALT*,
> IX86_BUILTIN_GATHER3ALT* and IX86_BUILTIN_SCATTERALT* builtins to
> match the IX86_BUILTIN codes. Build IX86_BUILTIN_SCATTERALTSIV4DF,
> IX86_BUILTIN_SCATTERALTDIV8SF, IX86_BUILTIN_SCATTERALTSIV4DI,
> IX86_BUILTIN_SCATTERALTDIV8SI, IX86_BUILTIN_SCATTERALTSIV2DF,
> IX86_BUILTIN_SCATTERALTDIV4SF, IX86_BUILTIN_SCATTERALTSIV2DI and
> IX86_BUILTIN_SCATTERALTDIV4SI decls.
> (ix86_vectorize_builtin_scatter): Expand those new builtins.
>
> * gcc.target/i386/avx512f-pr88464-5.c: New test.
> * gcc.target/i386/avx512f-pr88464-6.c: New test.
> * gcc.target/i386/avx512f-pr88464-7.c: New test.
> * gcc.target/i386/avx512f-pr88464-8.c: New test.
> * gcc.target/i386/avx512vl-pr88464-5.c: New test.
> * gcc.target/i386/avx512vl-pr88464-6.c: New test.
> * gcc.target/i386/avx512vl-pr88464-7.c: New test.
> * gcc.target/i386/avx512vl-pr88464-8.c: New test.
> * gcc.target/i386/avx512vl-pr88464-9.c: New test.
> * gcc.target/i386/avx512vl-pr88464-10.c: New test.
> * gcc.target/i386/avx512vl-pr88464-11.c: New test.
> * gcc.target/i386/avx512vl-pr88464-12.c: New test.
> * gcc.target/i386/avx512vl-pr88464-13.c: New test.
> * gcc.target/i386/avx512vl-pr88464-14.c: New test.
> * gcc.target/i386/avx512vl-pr88464-15.c: New test.
> * gcc.target/i386/avx512vl-pr88464-16.c: New test.
LGTM.
Thanks,
Uros.
> --- gcc/config/i386/i386-builtin-types.def.jj 2018-11-08 18:07:10.298826353 +0100
> +++ gcc/config/i386/i386-builtin-types.def 2018-12-18 11:22:07.965503704 +0100
> @@ -1068,7 +1068,14 @@ DEF_FUNCTION_TYPE (VOID, PFLOAT, HI, V8D
> DEF_FUNCTION_TYPE (VOID, PDOUBLE, QI, V16SI, V8DF, INT)
> DEF_FUNCTION_TYPE (VOID, PINT, HI, V8DI, V16SI, INT)
> DEF_FUNCTION_TYPE (VOID, PLONGLONG, QI, V16SI, V8DI, INT)
> -
> +DEF_FUNCTION_TYPE (VOID, PFLOAT, QI, V4DI, V8SF, INT)
> +DEF_FUNCTION_TYPE (VOID, PDOUBLE, QI, V8SI, V4DF, INT)
> +DEF_FUNCTION_TYPE (VOID, PINT, QI, V4DI, V8SI, INT)
> +DEF_FUNCTION_TYPE (VOID, PLONGLONG, QI, V8SI, V4DI, INT)
> +DEF_FUNCTION_TYPE (VOID, PFLOAT, QI, V2DI, V4SF, INT)
> +DEF_FUNCTION_TYPE (VOID, PDOUBLE, QI, V4SI, V2DF, INT)
> +DEF_FUNCTION_TYPE (VOID, PINT, QI, V2DI, V4SI, INT)
> +DEF_FUNCTION_TYPE (VOID, PLONGLONG, QI, V4SI, V2DI, INT)
>
> DEF_FUNCTION_TYPE (V16SF, V16SF, PCVOID, V16SI, HI, INT)
> DEF_FUNCTION_TYPE (V8DF, V8DF, PCVOID, V8SI, QI, INT)
> --- gcc/config/i386/i386.c.jj 2018-12-18 10:23:58.751164982 +0100
> +++ gcc/config/i386/i386.c 2018-12-18 11:58:18.813311983 +0100
> @@ -30072,6 +30072,14 @@ enum ix86_builtins
> IX86_BUILTIN_SCATTERALTDIV16SF,
> IX86_BUILTIN_SCATTERALTSIV8DI,
> IX86_BUILTIN_SCATTERALTDIV16SI,
> + IX86_BUILTIN_SCATTERALTSIV4DF,
> + IX86_BUILTIN_SCATTERALTDIV8SF,
> + IX86_BUILTIN_SCATTERALTSIV4DI,
> + IX86_BUILTIN_SCATTERALTDIV8SI,
> + IX86_BUILTIN_SCATTERALTSIV2DF,
> + IX86_BUILTIN_SCATTERALTDIV4SF,
> + IX86_BUILTIN_SCATTERALTSIV2DI,
> + IX86_BUILTIN_SCATTERALTDIV4SI,
> IX86_BUILTIN_SCATTERDIV16SF,
> IX86_BUILTIN_SCATTERDIV16SI,
> IX86_BUILTIN_SCATTERDIV8DF,
> @@ -30879,7 +30887,7 @@ ix86_init_mmx_sse_builtins (void)
> V4DF_FTYPE_V4DF_PCDOUBLE_V8SI_V4DF_INT,
> IX86_BUILTIN_GATHERALTSIV4DF);
>
> - def_builtin_pure (OPTION_MASK_ISA_AVX2, "__builtin_ia32_gatheraltdiv4sf256 ",
> + def_builtin_pure (OPTION_MASK_ISA_AVX2, "__builtin_ia32_gatheraltdiv8sf ",
> V8SF_FTYPE_V8SF_PCFLOAT_V4DI_V8SF_INT,
> IX86_BUILTIN_GATHERALTDIV8SF);
>
> @@ -30887,7 +30895,7 @@ ix86_init_mmx_sse_builtins (void)
> V4DI_FTYPE_V4DI_PCINT64_V8SI_V4DI_INT,
> IX86_BUILTIN_GATHERALTSIV4DI);
>
> - def_builtin_pure (OPTION_MASK_ISA_AVX2, "__builtin_ia32_gatheraltdiv4si256 ",
> + def_builtin_pure (OPTION_MASK_ISA_AVX2, "__builtin_ia32_gatheraltdiv8si ",
> V8SI_FTYPE_V8SI_PCINT_V4DI_V8SI_INT,
> IX86_BUILTIN_GATHERALTDIV8SI);
>
> @@ -30924,19 +30932,19 @@ ix86_init_mmx_sse_builtins (void)
> V8DI_FTYPE_V8DI_PCVOID_V8DI_QI_INT,
> IX86_BUILTIN_GATHER3DIV8DI);
>
> - def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gatheraltsiv8df ",
> + def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gather3altsiv8df ",
> V8DF_FTYPE_V8DF_PCDOUBLE_V16SI_QI_INT,
> IX86_BUILTIN_GATHER3ALTSIV8DF);
>
> - def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gatheraltdiv8sf ",
> + def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gather3altdiv16sf ",
> V16SF_FTYPE_V16SF_PCFLOAT_V8DI_HI_INT,
> IX86_BUILTIN_GATHER3ALTDIV16SF);
>
> - def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gatheraltsiv8di ",
> + def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gather3altsiv8di ",
> V8DI_FTYPE_V8DI_PCINT64_V16SI_QI_INT,
> IX86_BUILTIN_GATHER3ALTSIV8DI);
>
> - def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gatheraltdiv8si ",
> + def_builtin_pure (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_gather3altdiv16si ",
> V16SI_FTYPE_V16SI_PCINT_V8DI_HI_INT,
> IX86_BUILTIN_GATHER3ALTDIV16SI);
>
> @@ -31116,11 +31124,12 @@ ix86_init_mmx_sse_builtins (void)
> def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatterdiv2di",
> VOID_FTYPE_PVOID_QI_V2DI_V2DI_INT,
> IX86_BUILTIN_SCATTERDIV2DI);
> +
> def_builtin (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_scatteraltsiv8df ",
> VOID_FTYPE_PDOUBLE_QI_V16SI_V8DF_INT,
> IX86_BUILTIN_SCATTERALTSIV8DF);
>
> - def_builtin (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_scatteraltdiv8sf ",
> + def_builtin (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_scatteraltdiv16sf ",
> VOID_FTYPE_PFLOAT_HI_V8DI_V16SF_INT,
> IX86_BUILTIN_SCATTERALTDIV16SF);
>
> @@ -31128,10 +31137,42 @@ ix86_init_mmx_sse_builtins (void)
> VOID_FTYPE_PLONGLONG_QI_V16SI_V8DI_INT,
> IX86_BUILTIN_SCATTERALTSIV8DI);
>
> - def_builtin (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_scatteraltdiv8si ",
> + def_builtin (OPTION_MASK_ISA_AVX512F, "__builtin_ia32_scatteraltdiv16si ",
> VOID_FTYPE_PINT_HI_V8DI_V16SI_INT,
> IX86_BUILTIN_SCATTERALTDIV16SI);
>
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltsiv4df ",
> + VOID_FTYPE_PDOUBLE_QI_V8SI_V4DF_INT,
> + IX86_BUILTIN_SCATTERALTSIV4DF);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltdiv8sf ",
> + VOID_FTYPE_PFLOAT_QI_V4DI_V8SF_INT,
> + IX86_BUILTIN_SCATTERALTDIV8SF);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltsiv4di ",
> + VOID_FTYPE_PLONGLONG_QI_V8SI_V4DI_INT,
> + IX86_BUILTIN_SCATTERALTSIV4DI);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltdiv8si ",
> + VOID_FTYPE_PINT_QI_V4DI_V8SI_INT,
> + IX86_BUILTIN_SCATTERALTDIV8SI);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltsiv2df ",
> + VOID_FTYPE_PDOUBLE_QI_V4SI_V2DF_INT,
> + IX86_BUILTIN_SCATTERALTSIV2DF);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltdiv4sf ",
> + VOID_FTYPE_PFLOAT_QI_V2DI_V4SF_INT,
> + IX86_BUILTIN_SCATTERALTDIV4SF);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltsiv2di ",
> + VOID_FTYPE_PLONGLONG_QI_V4SI_V2DI_INT,
> + IX86_BUILTIN_SCATTERALTSIV2DI);
> +
> + def_builtin (OPTION_MASK_ISA_AVX512VL, "__builtin_ia32_scatteraltdiv4si ",
> + VOID_FTYPE_PINT_QI_V2DI_V4SI_INT,
> + IX86_BUILTIN_SCATTERALTDIV4SI);
> +
> /* AVX512PF */
> def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdpd",
> VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> @@ -37529,6 +37570,30 @@ rdseed_step:
> case IX86_BUILTIN_SCATTERALTDIV16SI:
> icode = CODE_FOR_avx512f_scatterdiv16si;
> goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTSIV4DF:
> + icode = CODE_FOR_avx512vl_scattersiv4df;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTDIV8SF:
> + icode = CODE_FOR_avx512vl_scatterdiv8sf;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTSIV4DI:
> + icode = CODE_FOR_avx512vl_scattersiv4di;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTDIV8SI:
> + icode = CODE_FOR_avx512vl_scatterdiv8si;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTSIV2DF:
> + icode = CODE_FOR_avx512vl_scattersiv2df;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTDIV4SF:
> + icode = CODE_FOR_avx512vl_scatterdiv4sf;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTSIV2DI:
> + icode = CODE_FOR_avx512vl_scattersiv2di;
> + goto scatter_gen;
> + case IX86_BUILTIN_SCATTERALTDIV4SI:
> + icode = CODE_FOR_avx512vl_scatterdiv4si;
> + goto scatter_gen;
> case IX86_BUILTIN_GATHERPFDPS:
> icode = CODE_FOR_avx512pf_gatherpfv16sisf;
> goto vec_prefetch_gen;
> @@ -37813,6 +37878,36 @@ rdseed_step:
> emit_insn (gen (half, op3));
> op3 = half;
> break;
> + case IX86_BUILTIN_SCATTERALTSIV4DF:
> + case IX86_BUILTIN_SCATTERALTSIV4DI:
> + half = gen_reg_rtx (V4SImode);
> + if (!nonimmediate_operand (op2, V8SImode))
> + op2 = copy_to_mode_reg (V8SImode, op2);
> + emit_insn (gen_vec_extract_lo_v8si (half, op2));
> + op2 = half;
> + break;
> + case IX86_BUILTIN_SCATTERALTDIV8SF:
> + case IX86_BUILTIN_SCATTERALTDIV8SI:
> + half = gen_reg_rtx (mode3);
> + if (mode3 == V4SFmode)
> + gen = gen_vec_extract_lo_v8sf;
> + else
> + gen = gen_vec_extract_lo_v8si;
> + if (!nonimmediate_operand (op3, GET_MODE (op3)))
> + op3 = copy_to_mode_reg (GET_MODE (op3), op3);
> + emit_insn (gen (half, op3));
> + op3 = half;
> + break;
> + case IX86_BUILTIN_SCATTERALTSIV2DF:
> + case IX86_BUILTIN_SCATTERALTSIV2DI:
> + if (!nonimmediate_operand (op2, V4SImode))
> + op2 = copy_to_mode_reg (V4SImode, op2);
> + break;
> + case IX86_BUILTIN_SCATTERALTDIV4SF:
> + case IX86_BUILTIN_SCATTERALTDIV4SI:
> + if (!nonimmediate_operand (op3, GET_MODE (op3)))
> + op3 = copy_to_mode_reg (GET_MODE (op3), op3);
> + break;
> default:
> break;
> }
> @@ -38928,6 +39023,54 @@ ix86_vectorize_builtin_scatter (const_tr
> case E_V16SImode:
> code = si ? IX86_BUILTIN_SCATTERSIV16SI : IX86_BUILTIN_SCATTERALTDIV16SI;
> break;
> + case E_V4DFmode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERALTSIV4DF : IX86_BUILTIN_SCATTERDIV4DF;
> + else
> + return NULL_TREE;
> + break;
> + case E_V4DImode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERALTSIV4DI : IX86_BUILTIN_SCATTERDIV4DI;
> + else
> + return NULL_TREE;
> + break;
> + case E_V8SFmode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERSIV8SF : IX86_BUILTIN_SCATTERALTDIV8SF;
> + else
> + return NULL_TREE;
> + break;
> + case E_V8SImode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERSIV8SI : IX86_BUILTIN_SCATTERALTDIV8SI;
> + else
> + return NULL_TREE;
> + break;
> + case E_V2DFmode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERALTSIV2DF : IX86_BUILTIN_SCATTERDIV2DF;
> + else
> + return NULL_TREE;
> + break;
> + case E_V2DImode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERALTSIV2DI : IX86_BUILTIN_SCATTERDIV2DI;
> + else
> + return NULL_TREE;
> + break;
> + case E_V4SFmode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERSIV4SF : IX86_BUILTIN_SCATTERALTDIV4SF;
> + else
> + return NULL_TREE;
> + break;
> + case E_V4SImode:
> + if (TARGET_AVX512VL)
> + code = si ? IX86_BUILTIN_SCATTERSIV4SI : IX86_BUILTIN_SCATTERALTDIV4SI;
> + else
> + return NULL_TREE;
> + break;
> default:
> return NULL_TREE;
> }
> --- gcc/testsuite/gcc.target/i386/avx512f-pr88464-5.c.jj 2018-12-18 11:40:14.777856788 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512f-pr88464-5.c 2018-12-18 11:40:51.162266018 +0100
> @@ -0,0 +1,45 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512f -mprefer-vector-width=512 -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 64 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +__attribute__((noipa)) void
> +f1 (long long * __restrict__ a, const long long * __restrict__ b, const int * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (a[i] > 10)
> + a[i] = b[c[i]];
> +}
> +
> +__attribute__((noipa)) void
> +f2 (long long * __restrict__ a, const long long * __restrict__ b, const long * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (a[i] > 10)
> + a[i] = b[c[i]];
> +}
> +
> +__attribute__((noipa)) void
> +f3 (int * __restrict__ a, const int * __restrict__ b, const int * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (a[i] > 10)
> + a[i] = b[c[i]];
> +}
> +
> +__attribute__((noipa)) void
> +f4 (int * __restrict__ a, const int * __restrict__ b, const long * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (a[i] > 10)
> + a[i] = b[c[i]];
> +}
> --- gcc/testsuite/gcc.target/i386/avx512f-pr88464-6.c.jj 2018-12-18 11:41:02.175087205 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512f-pr88464-6.c 2018-12-18 11:41:58.410174126 +0100
> @@ -0,0 +1,61 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512f } } } */
> +/* { dg-options "-O3 -mavx512f -mprefer-vector-width=512 -mtune=skylake-avx512" } */
> +
> +#include "avx512f-check.h"
> +
> +#include "avx512f-pr88464-5.c"
> +
> +static void
> +avx512f_test (void)
> +{
> + long long a[1024], b[1024];
> + int c[1024], f[1024];
> + int d[1024];
> + long e[1024];
> + int i;
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + a[i] = (i % 3) != 0 ? 15 : -5;
> + b[i] = 2 * i;
> + d[i] = (i % 3) ? 1023 - i : __INT_MAX__;
> + }
> + f1 (a, b, d, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (a[i] != ((i % 3) != 0 ? (1023 - i) * 2 : -5))
> + abort ();
> + a[i] = (i % 3) != 1 ? 15 : -5;
> + b[i] = 3 * i;
> + e[i] = (i % 3) != 1 ? 1023 - i : __LONG_MAX__;
> + }
> + f2 (a, b, e, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (a[i] != ((i % 3) != 1 ? (1023 - i) * 3 : -5))
> + abort ();
> + c[i] = (i % 3) != 2 ? 15 : -5;
> + d[i] = (i % 3) != 2 ? 1023 - i : __INT_MAX__;
> + f[i] = 4 * i;
> + }
> + f3 (c, f, d, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (c[i] != ((i % 3) != 2 ? (1023 - i) * 4: -5))
> + abort ();
> + c[i] = (i % 3) != 0 ? 15 : -5;
> + e[i] = (i % 3) != 0 ? 1023 - i : __INT_MAX__;
> + f[i] = 5 * i;
> + }
> + f4 (c, f, e, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (c[i] != ((i % 3) != 0 ? (1023 - i) * 5 : -5))
> + abort ();
> + }
> +}
> --- gcc/testsuite/gcc.target/i386/avx512f-pr88464-7.c.jj 2018-12-18 11:42:21.761794974 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512f-pr88464-7.c 2018-12-18 11:42:52.647293482 +0100
> @@ -0,0 +1,45 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512f -mprefer-vector-width=512 -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 64 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +__attribute__((noipa)) void
> +f1 (long long * __restrict__ a, const long long * __restrict__ b, const int * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (b[i] > -2)
> + a[c[i]] = b[i];
> +}
> +
> +__attribute__((noipa)) void
> +f2 (long long * __restrict__ a, const long long * __restrict__ b, const long * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (b[i] > -2)
> + a[c[i]] = b[i];
> +}
> +
> +__attribute__((noipa)) void
> +f3 (int * __restrict__ a, const int * __restrict__ b, const int * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (b[i] > -2)
> + a[c[i]] = b[i];
> +}
> +
> +__attribute__((noipa)) void
> +f4 (int * __restrict__ a, const int * __restrict__ b, const long * __restrict__ c, int n)
> +{
> + int i;
> +#pragma GCC ivdep
> + for (i = 0; i < n; ++i)
> + if (b[i] > -2)
> + a[c[i]] = b[i];
> +}
> --- gcc/testsuite/gcc.target/i386/avx512f-pr88464-8.c.jj 2018-12-18 11:43:06.660065958 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512f-pr88464-8.c 2018-12-18 11:47:02.049248182 +0100
> @@ -0,0 +1,61 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512f } } } */
> +/* { dg-options "-O3 -mavx512f -mprefer-vector-width=512 -mtune=skylake-avx512" } */
> +
> +#include "avx512f-check.h"
> +
> +#include "avx512f-pr88464-7.c"
> +
> +static void
> +avx512f_test (void)
> +{
> + long long a[1024], b[1024];
> + int c[1024], f[1024];
> + int d[1024];
> + long e[1024];
> + int i;
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + a[i] = -5;
> + b[i] = (i % 3) != 0 ? 2 * i : -5;
> + d[i] = (i % 3) != 0 ? 1023 - i : __INT_MAX__;
> + }
> + f1 (a, b, d, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (a[i] != ((i % 3) != 0 ? (1023 - i) * 2 : -5))
> + abort ();
> + a[i] = -5;
> + b[i] = (i % 3) != 1 ? 3 * i : -5;
> + e[i] = (i % 3) != 1 ? 1023 - i : __LONG_MAX__;
> + }
> + f2 (a, b, e, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (a[i] != ((i % 3) != 2 ? (1023 - i) * 3 : -5))
> + abort ();
> + c[i] = -5;
> + d[i] = (i % 3) != 2 ? 1023 - i : __INT_MAX__;
> + f[i] = (i % 3) != 2 ? 4 * i : -5;
> + }
> + f3 (c, f, d, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (c[i] != ((i % 3) != 1 ? (1023 - i) * 4 : -5))
> + abort ();
> + c[i] = -5;
> + e[i] = (i % 3) != 0 ? 1023 - i : __INT_MAX__;
> + f[i] = (i % 3) != 0 ? 5 * i : -5;
> + }
> + f4 (c, f, e, 1024);
> + for (i = 0; i < 1024; i++)
> + {
> + asm volatile ("" : "+g" (i));
> + if (c[i] != ((i % 3) != 0 ? (1023 - i) * 5 : -5))
> + abort ();
> + }
> +}
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-5.c.jj 2018-12-18 11:39:36.636476084 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-5.c 2018-12-18 11:38:44.944315401 +0100
> @@ -0,0 +1,7 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=256 -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 32 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +#include "avx512f-pr88464-3.c"
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-6.c.jj 2018-12-18 11:39:39.676426727 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-6.c 2018-12-18 11:38:59.605077356 +0100
> @@ -0,0 +1,20 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512vl } } } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=256 -mtune=skylake-avx512" } */
> +
> +#define AVX512VL
> +#define AVX512F_LEN 512
> +#define AVX512F_LEN_HALF 256
> +
> +#include "avx512f-pr88464-4.c"
> +
> +static void
> +test_256 (void)
> +{
> + avx512f_test ();
> +}
> +
> +static void
> +test_128 (void)
> +{
> +}
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-7.c.jj 2018-12-18 11:39:42.568379768 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-7.c 2018-12-18 11:39:14.828830172 +0100
> @@ -0,0 +1,7 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=128 -fno-vect-cost-model -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 16 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +#include "avx512f-pr88464-3.c"
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-8.c.jj 2018-12-18 11:39:45.771327762 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-8.c 2018-12-18 11:39:30.091582351 +0100
> @@ -0,0 +1,20 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512vl } } } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=128 -fno-vect-cost-model -mtune=skylake-avx512" } */
> +
> +#define AVX512VL
> +#define AVX512F_LEN 512
> +#define AVX512F_LEN_HALF 256
> +
> +#include "avx512f-pr88464-4.c"
> +
> +static void
> +test_256 (void)
> +{
> + avx512f_test ();
> +}
> +
> +static void
> +test_128 (void)
> +{
> +}
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-9.c.jj 2018-12-18 12:00:35.984087782 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-9.c 2018-12-18 12:01:05.523608218 +0100
> @@ -0,0 +1,7 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=256 -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 32 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +#include "avx512f-pr88464-5.c"
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-10.c.jj 2018-12-18 12:00:35.991087668 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-10.c 2018-12-18 12:01:28.353237592 +0100
> @@ -0,0 +1,20 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512vl } } } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=256 -mtune=skylake-avx512" } */
> +
> +#define AVX512VL
> +#define AVX512F_LEN 512
> +#define AVX512F_LEN_HALF 256
> +
> +#include "avx512f-pr88464-6.c"
> +
> +static void
> +test_256 (void)
> +{
> + avx512f_test ();
> +}
> +
> +static void
> +test_128 (void)
> +{
> +}
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-11.c.jj 2018-12-18 12:00:35.998087554 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-11.c 2018-12-18 12:01:35.238125814 +0100
> @@ -0,0 +1,7 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=128 -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 16 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +#include "avx512f-pr88464-5.c"
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-12.c.jj 2018-12-18 12:00:36.005087441 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-12.c 2018-12-18 12:01:41.982016328 +0100
> @@ -0,0 +1,20 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512vl } } } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=128 -mtune=skylake-avx512" } */
> +
> +#define AVX512VL
> +#define AVX512F_LEN 512
> +#define AVX512F_LEN_HALF 256
> +
> +#include "avx512f-pr88464-6.c"
> +
> +static void
> +test_256 (void)
> +{
> + avx512f_test ();
> +}
> +
> +static void
> +test_128 (void)
> +{
> +}
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-13.c.jj 2018-12-18 12:00:36.012087327 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-13.c 2018-12-18 12:01:48.157916062 +0100
> @@ -0,0 +1,7 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=256 -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 32 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +#include "avx512f-pr88464-7.c"
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-14.c.jj 2018-12-18 12:00:36.019087213 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-14.c 2018-12-18 12:01:55.403798428 +0100
> @@ -0,0 +1,20 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512vl } } } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=256 -mtune=skylake-avx512" } */
> +
> +#define AVX512VL
> +#define AVX512F_LEN 512
> +#define AVX512F_LEN_HALF 256
> +
> +#include "avx512f-pr88464-8.c"
> +
> +static void
> +test_256 (void)
> +{
> + avx512f_test ();
> +}
> +
> +static void
> +test_128 (void)
> +{
> +}
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-15.c.jj 2018-12-18 12:00:36.026087099 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-15.c 2018-12-18 12:02:06.141624102 +0100
> @@ -0,0 +1,7 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do compile } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=128 -fno-vect-cost-model -mtune=skylake-avx512 -fdump-tree-vect-details" } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using 16 byte vectors" 4 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "vectorized 1 loops in function" 4 "vect" } } */
> +
> +#include "avx512f-pr88464-7.c"
> --- gcc/testsuite/gcc.target/i386/avx512vl-pr88464-16.c.jj 2018-12-18 12:00:36.033086986 +0100
> +++ gcc/testsuite/gcc.target/i386/avx512vl-pr88464-16.c 2018-12-18 12:02:12.693517738 +0100
> @@ -0,0 +1,20 @@
> +/* PR tree-optimization/88464 */
> +/* { dg-do run { target { avx512vl } } } */
> +/* { dg-options "-O3 -mavx512vl -mprefer-vector-width=128 -fno-vect-cost-model -mtune=skylake-avx512" } */
> +
> +#define AVX512VL
> +#define AVX512F_LEN 512
> +#define AVX512F_LEN_HALF 256
> +
> +#include "avx512f-pr88464-8.c"
> +
> +static void
> +test_256 (void)
> +{
> + avx512f_test ();
> +}
> +
> +static void
> +test_128 (void)
> +{
> +}
>
> Jakub