Hi, Because we switch intrinsics for avx512bf16 to the new type __bf16. Now we could use m128/256bh for vector bf16 type instead of m128/256bf16. And unified builtin for avx512bf16/avxneconvert.
Thanks,
Lingling
> -----Original Message-----
> From: Hongtao Liu <crazy...@gmail.com>
> Sent: Tuesday, October 25, 2022 1:23 PM
> To: Kong, Lingling <lingling.k...@intel.com>
> Cc: Liu, Hongtao <hongtao....@intel.com>; gcc-patches@gcc.gnu.org; Jiang,
> Haochen <haochen.ji...@intel.com>
> Subject: Re: [PATCH 4/6] Support Intel AVX-NE-CONVERT
>
> On Mon, Oct 24, 2022 at 2:20 PM Kong, Lingling <lingling.k...@intel.com>
> wrote:
> >
> > > From: Gcc-patches
> > > <gcc-patches-bounces+lingling.kong=intel....@gcc.gnu.org>
> > > On Behalf Of Hongtao Liu via Gcc-patches
> > > Sent: Monday, October 17, 2022 1:47 PM
> > > To: Jiang, Haochen <haochen.ji...@intel.com>
> > > Cc: Liu, Hongtao <hongtao....@intel.com>; gcc-patches@gcc.gnu.org
> > > Subject: Re: [PATCH 4/6] Support Intel AVX-NE-CONVERT
> > >
> > > On Fri, Oct 14, 2022 at 3:58 PM Haochen Jiang via Gcc-patches
> > > <gcc-patches@gcc.gnu.org> wrote:
> > > >
> > > > From: Kong Lingling <lingling.k...@intel.com>
> > > > +(define_insn "vbcstne<vbcstnetype>2ps_<mode>"
> > > > + [(set (match_operand:VF1_128_256 0 "register_operand" "=x")
> > > > + (vec_duplicate:VF1_128_256
> > > > + (unspec:SF
> > > > + [(match_operand:HI 1 "memory_operand" "m")]
> > > > + VBCSTNE)))]
> > > > + "TARGET_AVXNECONVERT"
> > > > + "vbcstne<vbcstnetype>2ps\t{%1, %0|%0, %1}"
> > > > + [(set_attr "prefix" "vex")
> > > > + (set_attr "mode" "<sseinsnmode>")])
> > > Since jakub has support bf16 software emulation, can we rewrite it
> > > with general rtl ir without unspec?
> > > Like (float_extend:SF (match_operand:BF "memory_operand" "m")
> > > > +
> > > > +(define_int_iterator VCVTNEBF16
> > > > + [UNSPEC_VCVTNEEBF16SF
> > > > + UNSPEC_VCVTNEOBF16SF])
> > > > +
> > > > +(define_int_attr vcvtnebf16type
> > > > + [(UNSPEC_VCVTNEEBF16SF "ebf16")
> > > > + (UNSPEC_VCVTNEOBF16SF "obf16")]) (define_insn
> > > > +"vcvtne<vcvtnebf16type>2ps_<mode>"
> > > > + [(set (match_operand:VF1_128_256 0 "register_operand" "=x")
> > > > + (unspec:VF1_128_256
> > > > + [(match_operand:<sf_bfloat16> 1 "memory_operand" "m")]
> > > > + VCVTNEBF16))]
> > > > + "TARGET_AVXNECONVERT"
> > > > + "vcvtne<vcvtnebf16type>2ps\t{%1, %0|%0, %1}"
> > > > + [(set_attr "prefix" "vex")
> > > > + (set_attr "mode" "<sseinsnmode>")])
> > > Similar for this one and all those patterns below.
> >
> > That's great! Thanks for the review!
> > Now rewrite it without unspec and use float_extend for new define_insn.
> Ok.
> >
> > Thanks
> > Lingling
> >
> >
>
>
> --
> BR,
> Hongtao
0001-Support-Intel-AVX-NE-CONVERT.patch
Description: 0001-Support-Intel-AVX-NE-CONVERT.patch
