From: Hongtao Liu <crazylht@gmail.com>
To: "H.J. Lu" <hjl.tools@gmail.com>
Cc: liuhongt <hongtao.liu@intel.com>, GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: [PATCH] [i386] Replace ix86_gen_scratch_sse_rtx with gen_reg_rtx.
Date: Tue, 1 Mar 2022 11:01:21 +0800 [thread overview]
Message-ID: <CAMZc-by0UxRu7trwLWZPwTgFoyNSPAVf6zShvnLVet42mkkAYg@mail.gmail.com> (raw)
In-Reply-To: <CAMe9rOrnA=mYtqCcAP4OJsPctJvyJb1By1M5vMZanQO3UU9x7g@mail.gmail.com>
On Tue, Mar 1, 2022 at 10:27 AM H.J. Lu via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> On Mon, Feb 28, 2022 at 6:03 PM liuhongt <hongtao.liu@intel.com> wrote:
> >
> > .. in ix86_expand_vector_move and
> > ix86_convert_const_wide_int_to_broadcast(called by the former).
> >
> > ix86_expand_vector_move is called by emit_move_insn which is used by
> > many pre_reload passes, ix86_gen_scratch_sse_rtx will break data flow
> > when there's explict usage of xmm7/xmm15/xmm31.
> >
> > Bootstrapped and regtested on x86_64-linux-gnu{-m32,}
> > for both w/and w/o --with-cpu=native --with-arch=native.
> >
> > Ok for trunk?
> >
> > gcc/ChangeLog:
> >
> > PR target/104704
> > * config/i386/i386-expand.cc
> > (ix86_convert_const_wide_int_to_broadcast): Replace
> > ix86_gen_scratch_sse_rtx with gen_reg_rtx.
> > (ix86_expand_vector_move): Ditto.
> > * config/i386/sse.md (*vec_dupv4si): Add alternative $r and
> > corresponding splitter after it.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.target/i386/incoming-11.c: Revert r12-2665-g7f4c3943f795fd.
> > * gcc.target/i386/pr100865-11b.c: Expect vmovdqa or vmovda64.
> > * gcc.target/i386/pr100865-12b.c: Ditto.
> > * gcc.target/i386/pr100865-8b.c: Ditto.
> > * gcc.target/i386/pr100865-9b.c: Ditto.
> > * gcc.target/i386/pr82941-1.c: Expect vzeroupper for ! ia32.
> > * gcc.target/i386/pr82942-1.c: Ditto.
> > * gcc.target/i386/pr82990-1.c: Ditto.
> > * gcc.target/i386/pr82990-3.c: Ditto.
> > * gcc.target/i386/pr82990-5.c: Ditto.
> > ---
> > gcc/config/i386/i386-expand.cc | 6 +--
> > gcc/config/i386/sse.md | 41 +++++++++++++++-----
> > gcc/testsuite/gcc.target/i386/incoming-11.c | 2 +-
> > gcc/testsuite/gcc.target/i386/pr100865-11b.c | 2 +-
> > gcc/testsuite/gcc.target/i386/pr100865-12b.c | 2 +-
> > gcc/testsuite/gcc.target/i386/pr100865-8b.c | 2 +-
> > gcc/testsuite/gcc.target/i386/pr100865-9b.c | 2 +-
> > gcc/testsuite/gcc.target/i386/pr82941-1.c | 3 +-
> > gcc/testsuite/gcc.target/i386/pr82942-1.c | 3 +-
> > gcc/testsuite/gcc.target/i386/pr82990-1.c | 3 +-
> > gcc/testsuite/gcc.target/i386/pr82990-3.c | 3 +-
> > gcc/testsuite/gcc.target/i386/pr82990-5.c | 3 +-
> > 12 files changed, 45 insertions(+), 27 deletions(-)
> >
> > diff --git a/gcc/config/i386/i386-expand.cc b/gcc/config/i386/i386-expand.cc
> > index faa0191c6dd..75a28cdd89d 100644
> > --- a/gcc/config/i386/i386-expand.cc
> > +++ b/gcc/config/i386/i386-expand.cc
> > @@ -257,7 +257,7 @@ ix86_convert_const_wide_int_to_broadcast (machine_mode mode, rtx op)
> > machine_mode vector_mode;
> > if (!mode_for_vector (broadcast_mode, nunits).exists (&vector_mode))
> > gcc_unreachable ();
> > - rtx target = ix86_gen_scratch_sse_rtx (vector_mode);
> > + rtx target = gen_reg_rtx (vector_mode);
>
> I think ix86_gen_scratch_sse_rtx should check
> currently_expanding_gimple_stmt == NULL
> to return gen_reg_rtx (vector_mode) instead.
>
I'm a bit worried about continuing to use the hard register even if
only at the expand stage, if there is a recursive call to
expand_vector_move, we will still mess up the data flow.
.i.e there's emit_move_insn in ix86_expand_vector_init_duplicate.
> > bool ok = ix86_expand_vector_init_duplicate (false, vector_mode,
> > target,
> > GEN_INT (val_broadcast));
> > @@ -605,7 +605,7 @@ ix86_expand_vector_move (machine_mode mode, rtx operands[])
> > if (!register_operand (op0, mode)
> > && !register_operand (op1, mode))
> > {
> > - rtx scratch = ix86_gen_scratch_sse_rtx (mode);
> > + rtx scratch = gen_reg_rtx (mode);
> > emit_move_insn (scratch, op1);
> > op1 = scratch;
> > }
> > @@ -647,7 +647,7 @@ ix86_expand_vector_move (machine_mode mode, rtx operands[])
> > && !register_operand (op0, mode)
> > && !register_operand (op1, mode))
> > {
> > - rtx tmp = ix86_gen_scratch_sse_rtx (GET_MODE (op0));
> > + rtx tmp = gen_reg_rtx (GET_MODE (op0));
> > emit_move_insn (tmp, op1);
> > emit_move_insn (op0, tmp);
> > return;
> > diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> > index 3066ea3734a..d124545aa5d 100644
> > --- a/gcc/config/i386/sse.md
> > +++ b/gcc/config/i386/sse.md
> > @@ -25121,20 +25121,43 @@ (define_insn "vec_dupv4sf"
> > (set_attr "mode" "V4SF")])
> >
> > (define_insn "*vec_dupv4si"
> > - [(set (match_operand:V4SI 0 "register_operand" "=v,v,x")
> > + [(set (match_operand:V4SI 0 "register_operand" "=v,v,x,v")
> > (vec_duplicate:V4SI
> > - (match_operand:SI 1 "nonimmediate_operand" "Yv,m,0")))]
> > + (match_operand:SI 1 "nonimmediate_operand" "Yv,m,0,$r")))]
> > "TARGET_SSE"
> > "@
> > %vpshufd\t{$0, %1, %0|%0, %1, 0}
> > vbroadcastss\t{%1, %0|%0, %1}
> > - shufps\t{$0, %0, %0|%0, %0, 0}"
> > - [(set_attr "isa" "sse2,avx,noavx")
> > - (set_attr "type" "sselog1,ssemov,sselog1")
> > - (set_attr "length_immediate" "1,0,1")
> > - (set_attr "prefix_extra" "0,1,*")
> > - (set_attr "prefix" "maybe_vex,maybe_evex,orig")
> > - (set_attr "mode" "TI,V4SF,V4SF")])
> > + shufps\t{$0, %0, %0|%0, %0, 0}
> > + #"
> > + [(set_attr "isa" "sse2,avx,noavx,noavx512vl")
> > + (set_attr "type" "sselog1,ssemov,sselog1,sselog1")
> > + (set_attr "length_immediate" "1,0,1,1")
> > + (set_attr "prefix_extra" "0,1,*,0")
> > + (set_attr "prefix" "maybe_vex,maybe_evex,orig,maybe_vex")
> > + (set_attr "mode" "TI,V4SF,V4SF,TI")
> > + (set (attr "preferred_for_speed")
> > + (cond [(eq_attr "alternative" "3")
> > + (symbol_ref "TARGET_INTER_UNIT_MOVES_TO_VEC")
> > + ]
> > + (symbol_ref "true")))])
> > +
> > +(define_split
> > + [(set (match_operand:V4SI 0 "sse_reg_operand")
> > + (vec_duplicate:V4SI
> > + (match_operand:SI 1 "general_reg_operand")))]
> > + "TARGET_SSE && reload_completed
> > + /* Disable this splitter if avx512vl_vec_dup_gprv4si insn is
> > + available, because then we can broadcast from GPRs directly. */
> > + && !TARGET_AVX512VL"
> > + [(const_int 0)]
> > +{
> > + emit_insn (gen_vec_setv4si_0 (gen_lowpart (V4SImode, operands[0]),
> > + CONST0_RTX (V4SImode),
> > + gen_lowpart (SImode, operands[1])));
> > + emit_insn (gen_vec_duplicatev4si (operands[0], operands[0]));
> > + DONE;
> > +})
> >
> > (define_insn "*vec_dupv2di"
> > [(set (match_operand:V2DI 0 "register_operand" "=x,v,v,x")
> > diff --git a/gcc/testsuite/gcc.target/i386/incoming-11.c b/gcc/testsuite/gcc.target/i386/incoming-11.c
> > index 4b822684b88..a830c96f7d1 100644
> > --- a/gcc/testsuite/gcc.target/i386/incoming-11.c
> > +++ b/gcc/testsuite/gcc.target/i386/incoming-11.c
> > @@ -15,4 +15,4 @@ void f()
> > for (i = 0; i < 100; i++) q[i] = 1;
> > }
> >
> > -/* { dg-final { scan-assembler-not "andl\[\\t \]*\\$-16,\[\\t \]*%esp" } } */
> > +/* { dg-final { scan-assembler "andl\[\\t \]*\\$-16,\[\\t \]*%esp" } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr100865-11b.c b/gcc/testsuite/gcc.target/i386/pr100865-11b.c
> > index 7e458e85cdd..fe7736c318c 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr100865-11b.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr100865-11b.c
> > @@ -5,4 +5,4 @@
> >
> > /* { dg-final { scan-assembler-times "movabsq" 1 } } */
> > /* { dg-final { scan-assembler-times "vpbroadcastq\[\\t \]+%(?:r|e)\[^\n\]*, %xmm\[0-9\]+" 1 } } */
> > -/* { dg-final { scan-assembler-times "vmovdqa64\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > +/* { dg-final { scan-assembler-times "vmovdqa(?:64|)\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr100865-12b.c b/gcc/testsuite/gcc.target/i386/pr100865-12b.c
> > index dee0cfb016a..c9acfc7088f 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr100865-12b.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr100865-12b.c
> > @@ -5,4 +5,4 @@
> >
> > /* { dg-final { scan-assembler-times "movabsq" 1 } } */
> > /* { dg-final { scan-assembler-times "vpbroadcastq\[\\t \]+%(?:r|e)\[^\n\]*, %xmm\[0-9\]+" 1 } } */
> > -/* { dg-final { scan-assembler-times "vmovdqa64\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > +/* { dg-final { scan-assembler-times "vmovdqa(?:64|)\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr100865-8b.c b/gcc/testsuite/gcc.target/i386/pr100865-8b.c
> > index 4b7dd7cee3e..fa474c98a37 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr100865-8b.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr100865-8b.c
> > @@ -4,4 +4,4 @@
> > #include "pr100865-8a.c"
> >
> > /* { dg-final { scan-assembler-times "vpbroadcastd\[\\t \]+%(?:r|e)\[^\n\]*, %xmm\[0-9\]+" 1 } } */
> > -/* { dg-final { scan-assembler-times "vmovdqa64\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > +/* { dg-final { scan-assembler-times "vmovdqa(?:64|)\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr100865-9b.c b/gcc/testsuite/gcc.target/i386/pr100865-9b.c
> > index a315dde7c52..0714c3c9d6a 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr100865-9b.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr100865-9b.c
> > @@ -4,4 +4,4 @@
> > #include "pr100865-9a.c"
> >
> > /* { dg-final { scan-assembler-times "vpbroadcastw\[\\t \]+%(?:r|e)\[^\n\]*, %xmm\[0-9\]+" 1 } } */
> > -/* { dg-final { scan-assembler-times "vmovdqa64\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > +/* { dg-final { scan-assembler-times "vmovdqa(?:64|)\[\\t \]%xmm\[0-9\]+, " 16 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr82941-1.c b/gcc/testsuite/gcc.target/i386/pr82941-1.c
> > index c3be2f5b797..d7e530d5116 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr82941-1.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr82941-1.c
> > @@ -11,5 +11,4 @@ pr82941 ()
> > z = y;
> > }
> >
> > -/* { dg-final { scan-assembler-times "vzeroupper" 1 { target ia32 } } } */
> > -/* { dg-final { scan-assembler-not "vzeroupper" { target { ! ia32 } } } } */
> > +/* { dg-final { scan-assembler-times "vzeroupper" 1 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr82942-1.c b/gcc/testsuite/gcc.target/i386/pr82942-1.c
> > index 29ead049a67..9cdf81a9d60 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr82942-1.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr82942-1.c
> > @@ -3,5 +3,4 @@
> >
> > #include "pr82941-1.c"
> >
> > -/* { dg-final { scan-assembler-times "vzeroupper" 1 { target ia32 } } } */
> > -/* { dg-final { scan-assembler-not "vzeroupper" { target { ! ia32 } } } } */
> > +/* { dg-final { scan-assembler-times "vzeroupper" 1 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr82990-1.c b/gcc/testsuite/gcc.target/i386/pr82990-1.c
> > index bbf580fea77..ff1d6d40eb2 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr82990-1.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr82990-1.c
> > @@ -11,5 +11,4 @@ pr82941 ()
> > z = y;
> > }
> >
> > -/* { dg-final { scan-assembler-times "vzeroupper" 1 { target ia32 } } } */
> > -/* { dg-final { scan-assembler-not "vzeroupper" { target { ! ia32 } } } } */
> > +/* { dg-final { scan-assembler-times "vzeroupper" 1 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr82990-3.c b/gcc/testsuite/gcc.target/i386/pr82990-3.c
> > index 89ddb20adb3..201fa98d8d4 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr82990-3.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr82990-3.c
> > @@ -3,5 +3,4 @@
> >
> > #include "pr82941-1.c"
> >
> > -/* { dg-final { scan-assembler-times "vzeroupper" 1 { target ia32 } } } */
> > -/* { dg-final { scan-assembler-not "vzeroupper" { target { ! ia32 } } } } */
> > +/* { dg-final { scan-assembler-times "vzeroupper" 1 } } */
> > diff --git a/gcc/testsuite/gcc.target/i386/pr82990-5.c b/gcc/testsuite/gcc.target/i386/pr82990-5.c
> > index b9da0e706b1..008217af0b8 100644
> > --- a/gcc/testsuite/gcc.target/i386/pr82990-5.c
> > +++ b/gcc/testsuite/gcc.target/i386/pr82990-5.c
> > @@ -11,5 +11,4 @@ pr82941 ()
> > z = y;
> > }
> >
> > -/* { dg-final { scan-assembler-times "vzeroupper" 1 { target ia32 } } } */
> > -/* { dg-final { scan-assembler-not "vzeroupper" { target { ! ia32 } } } } */
> > +/* { dg-final { scan-assembler-times "vzeroupper" 1 } } */
> > --
> > 2.18.1
> >
>
>
> --
> H.J.
--
BR,
Hongtao
prev parent reply other threads:[~2022-03-01 2:52 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-03-01 2:03 liuhongt
2022-03-01 2:26 ` H.J. Lu
2022-03-01 2:38 ` H.J. Lu
2022-03-01 5:45 ` Hongtao Liu
2022-03-01 15:06 ` H.J. Lu
2022-03-01 22:48 ` H.J. Lu
2022-03-02 1:32 ` Hongtao Liu
2022-03-02 2:01 ` H.J. Lu
2022-03-01 3:01 ` Hongtao Liu [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=CAMZc-by0UxRu7trwLWZPwTgFoyNSPAVf6zShvnLVet42mkkAYg@mail.gmail.com \
--to=crazylht@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=hjl.tools@gmail.com \
--cc=hongtao.liu@intel.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).