This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH i386 AVX512] [10/n] Add vector move/load/store.
- From: Uros Bizjak <ubizjak at gmail dot com>
- To: Kirill Yukhin <kirill dot yukhin at gmail dot com>
- Cc: Jakub Jelinek <jakub at redhat dot com>, Richard Henderson <rth at redhat dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 14 Aug 2014 13:45:29 +0200
- Subject: Re: [PATCH i386 AVX512] [10/n] Add vector move/load/store.
- Authentication-results: sourceware.org; auth=none
- References: <20140814113020 dot GD49937 at msticlxl57 dot ims dot intel dot com>
On Thu, Aug 14, 2014 at 1:30 PM, Kirill Yukhin <kirill.yukhin@gmail.com> wrote:
> Hello,
> This patch extends load/store insns.
> No built-ins added in this patch.
>
> Bootstrapped.
> New tests on top of patch-set all pass
> under simulator.
>
> Is it ok for trunk?
>
> gcc/
> * config/i386/i386.c
> (ix86_expand_special_args_builtin): Handle avx512vl_storev8sf_mask,
> avx512vl_storev8si_mask, avx512vl_storev4df_mask, avx512vl_storev4di_mask,
> avx512vl_storev4sf_mask, avx512vl_storev4si_mask, avx512vl_storev2df_mask,
> avx512vl_storev2di_mask, avx512vl_loadv8sf_mask, avx512vl_loadv8si_mask,
> avx512vl_loadv4df_mask, avx512vl_loadv4di_mask, avx512vl_loadv4sf_mask,
> avx512vl_loadv4si_mask, avx512vl_loadv2df_mask, avx512vl_loadv2di_mask,
> avx512bw_loadv64qi_mask, avx512vl_loadv32qi_mask, avx512vl_loadv16qi_mask,
> avx512bw_loadv32hi_mask, avx512vl_loadv16hi_mask, avx512vl_loadv8hi_mask.
> * config/i386/i386.md: Allow V32HI mode.
Please update the above entry.
> * config/i386/sse.md
> (define_mode_iterator VMOVE): Allow V4TI mode.
> (define_mode_iterator V_AVX512VL): New.
> (define_mode_iterator V): New handling for AVX512VL.
> (define_insn "avx512f_load<mode>_mask"): Delete.
> (define_insn "<avx512>_load<mode>_mask"): New.
> (define_insn "avx512f_store<mode>_mask"): Delete.
> (define_insn "<avx512>_store<mode>_mask"): New.
>
> --
> Thanks, K
>
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index 183b7be..da01ac6 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -34722,6 +34722,14 @@ ix86_expand_special_args_builtin (const struct builtin_description *d,
> case CODE_FOR_avx512f_storev16si_mask:
> case CODE_FOR_avx512f_storev8df_mask:
> case CODE_FOR_avx512f_storev8di_mask:
> + case CODE_FOR_avx512vl_storev8sf_mask:
> + case CODE_FOR_avx512vl_storev8si_mask:
> + case CODE_FOR_avx512vl_storev4df_mask:
> + case CODE_FOR_avx512vl_storev4di_mask:
> + case CODE_FOR_avx512vl_storev4sf_mask:
> + case CODE_FOR_avx512vl_storev4si_mask:
> + case CODE_FOR_avx512vl_storev2df_mask:
> + case CODE_FOR_avx512vl_storev2di_mask:
> aligned_mem = true;
> break;
> default:
> @@ -34765,6 +34773,20 @@ ix86_expand_special_args_builtin (const struct builtin_description *d,
> case CODE_FOR_avx512f_loadv16si_mask:
> case CODE_FOR_avx512f_loadv8df_mask:
> case CODE_FOR_avx512f_loadv8di_mask:
> + case CODE_FOR_avx512vl_loadv8sf_mask:
> + case CODE_FOR_avx512vl_loadv8si_mask:
> + case CODE_FOR_avx512vl_loadv4df_mask:
> + case CODE_FOR_avx512vl_loadv4di_mask:
> + case CODE_FOR_avx512vl_loadv4sf_mask:
> + case CODE_FOR_avx512vl_loadv4si_mask:
> + case CODE_FOR_avx512vl_loadv2df_mask:
> + case CODE_FOR_avx512vl_loadv2di_mask:
> + case CODE_FOR_avx512bw_loadv64qi_mask:
> + case CODE_FOR_avx512vl_loadv32qi_mask:
> + case CODE_FOR_avx512vl_loadv16qi_mask:
> + case CODE_FOR_avx512bw_loadv32hi_mask:
> + case CODE_FOR_avx512vl_loadv16hi_mask:
> + case CODE_FOR_avx512vl_loadv8hi_mask:
> aligned_mem = true;
> break;
> default:
> diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
> index a72c206..b8ce6c0 100644
> --- a/gcc/config/i386/i386.md
> +++ b/gcc/config/i386/i386.md
> @@ -1054,7 +1054,7 @@
> (V4SF "ps") (V2DF "pd")
> (V16QI "b") (V8HI "w") (V4SI "d") (V2DI "q")
> (V32QI "b") (V16HI "w") (V8SI "d") (V4DI "q")
> - (V64QI "b") (V16SI "d") (V8DI "q")])
> + (V64QI "b") (V32HI "w") (V16SI "d") (V8DI "q")])
>
> ;; SSE vector suffix for floating point modes
> (define_mode_attr ssevecmodesuffix [(SF "ps") (DF "pd")])
> diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> index 89a1842..910b29b 100644
> --- a/gcc/config/i386/sse.md
> +++ b/gcc/config/i386/sse.md
> @@ -146,10 +146,21 @@
> (V32HI "TARGET_AVX512F") (V16HI "TARGET_AVX") V8HI
> (V16SI "TARGET_AVX512F") (V8SI "TARGET_AVX") V4SI
> (V8DI "TARGET_AVX512F") (V4DI "TARGET_AVX") V2DI
> - (V2TI "TARGET_AVX") V1TI
> + (V4TI "TARGET_AVX") (V2TI "TARGET_AVX") V1TI
Are you sure TARGET_AVX is the correct condition for V4TI?
> (V16SF "TARGET_AVX512F") (V8SF "TARGET_AVX") V4SF
> (V8DF "TARGET_AVX512F") (V4DF "TARGET_AVX") V2DF])
>
> +;; All AVX512VL vector modes
> +(define_mode_iterator V_AVX512VL
> + [(V64QI "TARGET_AVX512BW") (V32QI "TARGET_AVX512VL && TARGET_AVX512BW")
> + (V16QI "TARGET_AVX512VL && TARGET_AVX512BW")
> + (V32HI "TARGET_AVX512BW") (V16HI "TARGET_AVX512VL && TARGET_AVX512BW")
Is the above && OK? So, you have to pass -m...vl and -m...bw to the
compiler to enable these modes?
Uros.