From: Richard Biener <rguenther@suse.de>
To: Jakub Jelinek <jakub@redhat.com>
Cc: gcc-patches@gcc.gnu.org, ubizjak@gmail.com
Subject: Re: [PATCH] Fix PR81175, make gather builtins pure
Date: Tue, 27 Jun 2017 10:40:00 -0000 [thread overview]
Message-ID: <alpine.LSU.2.20.1706271238030.23185@zhemvz.fhfr.qr> (raw)
In-Reply-To: <20170627100240.GQ2123@tucnak>
On Tue, 27 Jun 2017, Jakub Jelinek wrote:
> On Fri, Jun 23, 2017 at 02:54:35PM +0200, Richard Biener wrote:
> > 2017-06-23 Richard Biener <rguenther@suse.de>
> >
> > PR target/81175
> > * config/i386/i386.c (struct builtin_isa): Add pure_p member.
> > (def_builtin2): Initialize pure_p.
> > (ix86_add_new_builtins): Honor pure_p.
> > (def_builtin_pure): New function.
> > (ix86_init_mmx_sse_builtins): Use def_builtin_pure for all
> > gather builtins.
>
> > *************** ix86_init_mmx_sse_builtins (void)
> > *** 31985,32002 ****
> > IX86_BUILTIN_SCATTERALTDIV16SI);
> >
> > /* AVX512PF */
> > ! def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdpd",
> > ! VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFDPD);
> > ! def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdps",
> > ! VOID_FTYPE_HI_V16SI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFDPS);
> > ! def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqpd",
> > ! VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFQPD);
> > ! def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqps",
> > ! VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFQPS);
> > def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_scatterpfdpd",
> > VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> > IX86_BUILTIN_SCATTERPFDPD);
> > --- 32004,32021 ----
> > IX86_BUILTIN_SCATTERALTDIV16SI);
> >
> > /* AVX512PF */
> > ! def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdpd",
> > ! VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFDPD);
> > ! def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdps",
> > ! VOID_FTYPE_HI_V16SI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFDPS);
> > ! def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqpd",
> > ! VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFQPD);
> > ! def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqps",
> > ! VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> > ! IX86_BUILTIN_GATHERPFQPS);
> > def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_scatterpfdpd",
> > VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> > IX86_BUILTIN_SCATTERPFDPD);
>
> The above hunk looks wrong and regresses
> +FAIL: gcc.target/i386/avx512pf-vgatherpf0dpd-1.c scan-assembler-times vgatherpf0dpd[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf0dps-1.c scan-assembler-times vgatherpf0dps[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf0qpd-1.c scan-assembler-times vgatherpf0qpd[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf0qps-1.c scan-assembler-times vgatherpf0qps[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf1dpd-1.c scan-assembler-times vgatherpf1dpd[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf1dps-1.c scan-assembler-times vgatherpf1dps[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf1qpd-1.c scan-assembler-times vgatherpf1qpd[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> +FAIL: gcc.target/i386/avx512pf-vgatherpf1qps-1.c scan-assembler-times vgatherpf1qps[ \\\\t]+[^{\\n]*{%k[1-7]}(?:\\n|[ \\\\t]+#) 2
> The prefetch intrinsics can't be pure, after all they return void, so
> as pure are always optimized away.
>
> Ok for trunk if it passes bootstrap/regtest?
Ok.
Whoops - sorry for not noticing. Looks like "pf" means prefetch here,
so technically they are pure (do not write to memory) but we don't want to
DCE them as they have additional side-effects. We could mark them
looping-const-or-pure I guess, or "volatile" (which we can't express for
calls IIRC).
Richard.
> 2017-06-27 Jakub Jelinek <jakub@redhat.com>
>
> PR target/81175
> * config/i386/i386.c (ix86_init_mmx_sse_builtins): Use def_builtin
> rather than def_builtin_pure for __builtin_ia32_gatherpf*.
>
> --- gcc/config/i386/i386.c.jj 2017-06-27 09:16:09.000000000 +0200
> +++ gcc/config/i386/i386.c 2017-06-27 11:47:21.610627628 +0200
> @@ -32886,18 +32886,18 @@ ix86_init_mmx_sse_builtins (void)
> IX86_BUILTIN_SCATTERALTDIV16SI);
>
> /* AVX512PF */
> - def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdpd",
> - VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> - IX86_BUILTIN_GATHERPFDPD);
> - def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdps",
> - VOID_FTYPE_HI_V16SI_PCVOID_INT_INT,
> - IX86_BUILTIN_GATHERPFDPS);
> - def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqpd",
> - VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> - IX86_BUILTIN_GATHERPFQPD);
> - def_builtin_pure (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqps",
> - VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> - IX86_BUILTIN_GATHERPFQPS);
> + def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdpd",
> + VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> + IX86_BUILTIN_GATHERPFDPD);
> + def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfdps",
> + VOID_FTYPE_HI_V16SI_PCVOID_INT_INT,
> + IX86_BUILTIN_GATHERPFDPS);
> + def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqpd",
> + VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> + IX86_BUILTIN_GATHERPFQPD);
> + def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_gatherpfqps",
> + VOID_FTYPE_QI_V8DI_PCVOID_INT_INT,
> + IX86_BUILTIN_GATHERPFQPS);
> def_builtin (OPTION_MASK_ISA_AVX512PF, "__builtin_ia32_scatterpfdpd",
> VOID_FTYPE_QI_V8SI_PCVOID_INT_INT,
> IX86_BUILTIN_SCATTERPFDPD);
>
>
> Jakub
>
>
--
Richard Biener <rguenther@suse.de>
SUSE LINUX GmbH, GF: Felix Imendoerffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nuernberg)
prev parent reply other threads:[~2017-06-27 10:40 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-06-23 12:54 Richard Biener
2017-06-23 13:17 ` Marc Glisse
2017-06-23 13:22 ` Richard Biener
2017-06-23 13:26 ` Jakub Jelinek
2017-06-26 7:04 ` Uros Bizjak
2017-06-27 10:02 ` Jakub Jelinek
2017-06-27 10:09 ` Uros Bizjak
2017-06-27 10:27 ` Jakub Jelinek
2017-07-04 8:36 ` Patch ping (Re: [PATCH] Fix PR81175, make gather builtins pure) Jakub Jelinek
2017-07-04 8:38 ` Uros Bizjak
2017-07-04 8:46 ` Richard Biener
2017-06-27 10:40 ` Richard Biener [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=alpine.LSU.2.20.1706271238030.23185@zhemvz.fhfr.qr \
--to=rguenther@suse.de \
--cc=gcc-patches@gcc.gnu.org \
--cc=jakub@redhat.com \
--cc=ubizjak@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).