From: Richard Biener <richard.guenther@gmail.com>
To: Uros Bizjak <ubizjak@gmail.com>
Cc: liuhongt <hongtao.liu@intel.com>, gcc-patches@gcc.gnu.org
Subject: Re: [PATCH] i386: Clear upper bits of XMM register for V4HFmode/V2HFmode operations [PR110762]
Date: Mon, 7 Aug 2023 13:19:49 +0200 [thread overview]
Message-ID: <CAFiYyc2VN4W0MZpQN=VErmXTtSfGEjNDNjwccWvDD+QLysWe3g@mail.gmail.com> (raw)
In-Reply-To: <CAFULd4adtebpoycts-k+q+9ndkwHyXPNnDTDntoDz6kMjv9i-g@mail.gmail.com>
On Mon, Aug 7, 2023 at 11:20 AM Uros Bizjak via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> On Mon, Aug 7, 2023 at 10:57 AM liuhongt <hongtao.liu@intel.com> wrote:
> >
> > Similar like r14-2786-gade30fad6669e5, the patch is for V4HF/V2HFmode.
> >
> > Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}.
> > Ok for trunk?
> >
> > gcc/ChangeLog:
> >
> > PR target/110762
> > * config/i386/mmx.md (<insn><mode>3): Changed from define_insn
> > to define_expand and break into ..
> > (<insn>v4hf3): .. this.
> > (divv4hf3): .. this.
> > (<insn>v2hf3): .. this.
> > (divv2hf3): .. this.
> > (movd_v2hf_to_sse): New define_expand.
> > (movq_<mode>_to_sse): Extend to V4HFmode.
> > (mmxdoublevecmode): Ditto.
> > (V2FI_V4HF): New mode iterator.
> > * config/i386/sse.md (*vec_concatv4sf): Extend to hanlde V8HF
> > by using mode iterator V4SF_V8HF, renamed to ..
> > (*vec_concat<mode>): .. this.
> > (*vec_concatv4sf_0): Extend to handle V8HF by using mode
> > iterator V4SF_V8HF, renamed to ..
> > (*vec_concat<mode>_0): .. this.
> > (*vec_concatv8hf_movss): New define_insn.
> > (V4SF_V8HF): New mode iterator.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.target/i386/pr110762-v4hf.c: New test.
>
> LGTM.
>
> Please also note the RFC patch [1] that relaxes clears for V2SFmode
> with -fno-trapping-math. The patched compiler will then emit the same
> code as clang does for -O2. Which raises another question - should gcc
> default to -fno-trapping-math?
I think we discussed this before and yes, IMHO we should default to
-fno-trapping-math at least for C/C++ to be consistent with our other
handling of the FP environment (default to -fno-rounding-math) and
lack of proper FENV access barriers for inspecting the exceptions.
Note Fortran has the -ffpe-trap= option which would then need to make
sure to also enable -ftrapping-math. Ada might have similar constraints
(it also uses -fnon-call-exceptions, but unless it enables CPU traps for
FP exceptions that would be a no-op). Note this also shows we should
possibly separate maintaining the IEEE exception state and considering
changes in the IEEE exception states to cause CPU traps (that's also
a source of common confusion on the user side).
Richard.
> [1] https://gcc.gnu.org/pipermail/gcc-patches/2023-July/625795.html
>
> Thanks,
> Uros.
>
> > ---
> > gcc/config/i386/mmx.md | 109 +++++++++++++++---
> > gcc/config/i386/sse.md | 40 +++++--
> > gcc/testsuite/gcc.target/i386/pr110762-v4hf.c | 57 +++++++++
> > 3 files changed, 177 insertions(+), 29 deletions(-)
> > create mode 100644 gcc/testsuite/gcc.target/i386/pr110762-v4hf.c
> >
> > diff --git a/gcc/config/i386/mmx.md b/gcc/config/i386/mmx.md
> > index 896af76a33f..88bdf084f54 100644
> > --- a/gcc/config/i386/mmx.md
> > +++ b/gcc/config/i386/mmx.md
> > @@ -79,9 +79,7 @@ (define_mode_iterator V_16_32_64
> > ;; V2S* modes
> > (define_mode_iterator V2FI [V2SF V2SI])
> >
> > -;; 4-byte and 8-byte float16 vector modes
> > -(define_mode_iterator VHF_32_64 [V4HF V2HF])
> > -
> > +(define_mode_iterator V2FI_V4HF [V2SF V2SI V4HF])
> > ;; Mapping from integer vector mode to mnemonic suffix
> > (define_mode_attr mmxvecsize
> > [(V8QI "b") (V4QI "b") (V2QI "b")
> > @@ -108,7 +106,7 @@ (define_mode_attr mmxintvecmodelower
> >
> > ;; Mapping of vector modes to a vector mode of double size
> > (define_mode_attr mmxdoublevecmode
> > - [(V2SF "V4SF") (V2SI "V4SI")])
> > + [(V2SF "V4SF") (V2SI "V4SI") (V4HF "V8HF")])
> >
> > ;; Mapping of vector modes back to the scalar modes
> > (define_mode_attr mmxscalarmode
> > @@ -594,7 +592,7 @@ (define_insn "sse_movntq"
> > (define_expand "movq_<mode>_to_sse"
> > [(set (match_operand:<mmxdoublevecmode> 0 "register_operand")
> > (vec_concat:<mmxdoublevecmode>
> > - (match_operand:V2FI 1 "nonimmediate_operand")
> > + (match_operand:V2FI_V4HF 1 "nonimmediate_operand")
> > (match_dup 2)))]
> > "TARGET_SSE2"
> > "operands[2] = CONST0_RTX (<MODE>mode);")
> > @@ -1927,21 +1925,94 @@ (define_expand "lroundv2sfv2si2"
> > ;;
> > ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
> >
> > -(define_insn "<insn><mode>3"
> > - [(set (match_operand:VHF_32_64 0 "register_operand" "=v")
> > - (plusminusmultdiv:VHF_32_64
> > - (match_operand:VHF_32_64 1 "register_operand" "<comm>v")
> > - (match_operand:VHF_32_64 2 "register_operand" "v")))]
> > +(define_expand "<insn>v4hf3"
> > + [(set (match_operand:V4HF 0 "register_operand")
> > + (plusminusmult:V4HF
> > + (match_operand:V4HF 1 "nonimmediate_operand")
> > + (match_operand:V4HF 2 "nonimmediate_operand")))]
> > "TARGET_AVX512FP16 && TARGET_AVX512VL"
> > - "v<insn>ph\t{%2, %1, %0|%0, %1, %2}"
> > - [(set (attr "type")
> > - (cond [(match_test "<CODE> == MULT")
> > - (const_string "ssemul")
> > - (match_test "<CODE> == DIV")
> > - (const_string "ssediv")]
> > - (const_string "sseadd")))
> > - (set_attr "prefix" "evex")
> > - (set_attr "mode" "V8HF")])
> > +{
> > + rtx op2 = gen_reg_rtx (V8HFmode);
> > + rtx op1 = gen_reg_rtx (V8HFmode);
> > + rtx op0 = gen_reg_rtx (V8HFmode);
> > +
> > + emit_insn (gen_movq_v4hf_to_sse (op2, operands[2]));
> > + emit_insn (gen_movq_v4hf_to_sse (op1, operands[1]));
> > +
> > + emit_insn (gen_<insn>v8hf3 (op0, op1, op2));
> > +
> > + emit_move_insn (operands[0], lowpart_subreg (V4HFmode, op0, V8HFmode));
> > + DONE;
> > +})
> > +
> > +(define_expand "divv4hf3"
> > + [(set (match_operand:V4HF 0 "register_operand")
> > + (div:V4HF
> > + (match_operand:V4HF 1 "nonimmediate_operand")
> > + (match_operand:V4HF 2 "nonimmediate_operand")))]
> > + "TARGET_AVX512FP16 && TARGET_AVX512VL"
> > +{
> > + rtx op2 = gen_reg_rtx (V8HFmode);
> > + rtx op1 = gen_reg_rtx (V8HFmode);
> > + rtx op0 = gen_reg_rtx (V8HFmode);
> > +
> > + emit_insn (gen_movq_v4hf_to_sse (op1, operands[1]));
> > + rtx tmp = gen_rtx_VEC_CONCAT (V8HFmode, operands[2],
> > + force_reg (V4HFmode, CONST1_RTX (V4HFmode)));
> > + emit_insn (gen_rtx_SET (op2, tmp));
> > + emit_insn (gen_divv8hf3 (op0, op1, op2));
> > + emit_move_insn (operands[0], lowpart_subreg (V4HFmode, op0, V8HFmode));
> > + DONE;
> > +})
> > +
> > +(define_expand "movd_v2hf_to_sse"
> > + [(set (match_operand:V8HF 0 "register_operand")
> > + (vec_merge:V8HF
> > + (vec_duplicate:V8HF
> > + (match_operand:V2HF 1 "nonimmediate_operand"))
> > + (match_operand:V8HF 2 "reg_or_0_operand")
> > + (const_int 3)))]
> > + "TARGET_SSE")
> > +
> > +(define_expand "<insn>v2hf3"
> > + [(set (match_operand:V2HF 0 "register_operand")
> > + (plusminusmult:V2HF
> > + (match_operand:V2HF 1 "nonimmediate_operand")
> > + (match_operand:V2HF 2 "nonimmediate_operand")))]
> > + "TARGET_AVX512FP16 && TARGET_AVX512VL"
> > +{
> > + rtx op2 = gen_reg_rtx (V8HFmode);
> > + rtx op1 = gen_reg_rtx (V8HFmode);
> > + rtx op0 = gen_reg_rtx (V8HFmode);
> > +
> > + emit_insn (gen_movd_v2hf_to_sse (op2, operands[2], CONST0_RTX (V8HFmode)));
> > + emit_insn (gen_movd_v2hf_to_sse (op1, operands[1], CONST0_RTX (V8HFmode)));
> > + emit_insn (gen_<insn>v8hf3 (op0, op1, op2));
> > +
> > + emit_move_insn (operands[0], lowpart_subreg (V2HFmode, op0, V8HFmode));
> > + DONE;
> > +})
> > +
> > +(define_expand "divv2hf3"
> > + [(set (match_operand:V2HF 0 "register_operand")
> > + (div:V2HF
> > + (match_operand:V2HF 1 "nonimmediate_operand")
> > + (match_operand:V2HF 2 "nonimmediate_operand")))]
> > + "TARGET_AVX512FP16 && TARGET_AVX512VL"
> > +{
> > + rtx op2 = gen_reg_rtx (V8HFmode);
> > + rtx op1 = gen_reg_rtx (V8HFmode);
> > + rtx op0 = gen_reg_rtx (V8HFmode);
> > +
> > + emit_insn (gen_movd_v2hf_to_sse (op2, operands[2],
> > + force_reg (V8HFmode, CONST1_RTX (V8HFmode))));
> > + emit_insn (gen_movd_v2hf_to_sse (op1, operands[1], CONST0_RTX (V8HFmode)));
> > + emit_insn (gen_divv8hf3 (op0, op1, op2));
> > +
> > + emit_move_insn (operands[0], lowpart_subreg (V2HFmode, op0, V8HFmode));
> > + DONE;
> > +})
> > +
> >
> > ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
> > ;;
> > diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
> > index ab455c3e297..7383a50ee0d 100644
> > --- a/gcc/config/i386/sse.md
> > +++ b/gcc/config/i386/sse.md
> > @@ -430,6 +430,9 @@ (define_mode_iterator VF_512
> > (define_mode_iterator VFB_512
> > [V32HF V16SF V8DF])
> >
> > +(define_mode_iterator V4SF_V8HF
> > + [V4SF V8HF])
> > +
> > (define_mode_iterator VI48_AVX512VL
> > [V16SI (V8SI "TARGET_AVX512VL") (V4SI "TARGET_AVX512VL")
> > V8DI (V4DI "TARGET_AVX512VL") (V2DI "TARGET_AVX512VL")])
> > @@ -10873,11 +10876,11 @@ (define_insn "*vec_concatv2sf_sse"
> > (set_attr "type" "sselog,ssemov,mmxcvt,mmxmov")
> > (set_attr "mode" "V4SF,SF,DI,DI")])
> >
> > -(define_insn "*vec_concatv4sf"
> > - [(set (match_operand:V4SF 0 "register_operand" "=x,v,x,v")
> > - (vec_concat:V4SF
> > - (match_operand:V2SF 1 "register_operand" " 0,v,0,v")
> > - (match_operand:V2SF 2 "nonimmediate_operand" " x,v,m,m")))]
> > +(define_insn "*vec_concat<mode>"
> > + [(set (match_operand:V4SF_V8HF 0 "register_operand" "=x,v,x,v")
> > + (vec_concat:V4SF_V8HF
> > + (match_operand:<ssehalfvecmode> 1 "register_operand" " 0,v,0,v")
> > + (match_operand:<ssehalfvecmode> 2 "nonimmediate_operand" " x,v,m,m")))]
> > "TARGET_SSE"
> > "@
> > movlhps\t{%2, %0|%0, %2}
> > @@ -10889,17 +10892,34 @@ (define_insn "*vec_concatv4sf"
> > (set_attr "prefix" "orig,maybe_evex,orig,maybe_evex")
> > (set_attr "mode" "V4SF,V4SF,V2SF,V2SF")])
> >
> > -(define_insn "*vec_concatv4sf_0"
> > - [(set (match_operand:V4SF 0 "register_operand" "=v")
> > - (vec_concat:V4SF
> > - (match_operand:V2SF 1 "nonimmediate_operand" "vm")
> > - (match_operand:V2SF 2 "const0_operand")))]
> > +(define_insn "*vec_concat<mode>_0"
> > + [(set (match_operand:V4SF_V8HF 0 "register_operand" "=v")
> > + (vec_concat:V4SF_V8HF
> > + (match_operand:<ssehalfvecmode> 1 "nonimmediate_operand" "vm")
> > + (match_operand:<ssehalfvecmode> 2 "const0_operand")))]
> > "TARGET_SSE2"
> > "%vmovq\t{%1, %0|%0, %1}"
> > [(set_attr "type" "ssemov")
> > (set_attr "prefix" "maybe_vex")
> > (set_attr "mode" "DF")])
> >
> > +(define_insn "*vec_concatv8hf_movss"
> > + [(set (match_operand:V8HF 0 "register_operand" "=x,v,v")
> > + (vec_merge:V8HF
> > + (vec_duplicate:V8HF
> > + (match_operand:V2HF 2 "nonimmediate_operand" "x,m,v"))
> > + (match_operand:V8HF 1 "reg_or_0_operand" "0,C,v" )
> > + (const_int 3)))]
> > + "TARGET_SSE"
> > + "@
> > + movss\t{%2, %0|%0, %2}
> > + %vmovss\t{%2, %0|%0, %2}
> > + vmovss\t{%2, %1, %0|%0, %1, %2}"
> > + [(set_attr "isa" "noavx,*,avx")
> > + (set_attr "type" "ssemov")
> > + (set_attr "prefix" "orig,maybe_vex,maybe_vex")
> > + (set_attr "mode" "SF")])
> > +
> > ;; Avoid combining registers from different units in a single alternative,
> > ;; see comment above inline_secondary_memory_needed function in i386.cc
> > (define_insn "vec_set<mode>_0"
> > diff --git a/gcc/testsuite/gcc.target/i386/pr110762-v4hf.c b/gcc/testsuite/gcc.target/i386/pr110762-v4hf.c
> > new file mode 100644
> > index 00000000000..332784ac694
> > --- /dev/null
> > +++ b/gcc/testsuite/gcc.target/i386/pr110762-v4hf.c
> > @@ -0,0 +1,57 @@
> > +/* PR target/110762 */
> > +/* { dg-do compile { target { ! ia32 } } } */
> > +/* { dg-options "-O2 -mavx512fp16 -mavx512vl -dp" } */
> > +
> > +typedef _Float16 v4hf __attribute__((vector_size(8)));
> > +typedef _Float16 v2hf __attribute__((vector_size(4)));
> > +
> > +v4hf
> > +foo (v4hf a, v4hf b)
> > +{
> > + return a + b;
> > +}
> > +
> > +v4hf
> > +foo2 (v4hf a, v4hf b)
> > +{
> > + return a - b;
> > +}
> > +
> > +v4hf
> > +foo3 (v4hf a, v4hf b)
> > +{
> > + return a * b;
> > +}
> > +
> > +v4hf
> > +foo1 (v4hf a, v4hf b)
> > +{
> > + return a / b;
> > +}
> > +
> > +v2hf
> > +foo4 (v2hf a, v2hf b)
> > +{
> > + return a + b;
> > +}
> > +
> > +v2hf
> > +foo5 (v2hf a, v2hf b)
> > +{
> > + return a - b;
> > +}
> > +
> > +v2hf
> > +foo6 (v2hf a, v2hf b)
> > +{
> > + return a * b;
> > +}
> > +
> > +v2hf
> > +foo7 (v2hf a, v2hf b)
> > +{
> > + return a / b;
> > +}
> > +
> > +/* { dg-final { scan-assembler-times "\\*vec_concatv8hf_0" 7 } } */
> > +/* { dg-final { scan-assembler-times "\\*vec_concatv8hf_movss" 8 } } */
> > --
> > 2.31.1
> >
next prev parent reply other threads:[~2023-08-07 11:20 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-08-07 8:57 liuhongt
2023-08-07 9:18 ` Uros Bizjak
2023-08-07 11:19 ` Richard Biener [this message]
2023-08-09 9:52 ` Uros Bizjak
2023-08-07 11:53 ` Hongtao Liu
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='CAFiYyc2VN4W0MZpQN=VErmXTtSfGEjNDNjwccWvDD+QLysWe3g@mail.gmail.com' \
--to=richard.guenther@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=hongtao.liu@intel.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).