> From: Kumar, Venkataramanan <[email protected]>
> Sent: Monday, July 20, 2026 4:19 AM
> > From: Haochen Jiang <[email protected]>
> > Sent: Thursday, July 2, 2026 12:19 PM
> > Subject: [PATCH 7/7] Support ACEv1 outer product instructions
> >
> > --- a/gcc/config/i386/i386-expand.cc
> > +++ b/gcc/config/i386/i386-expand.cc
> > @@ -14789,7 +14789,7 @@ ix86_expand_ace_builtin (const struct
> > builtin_description *d, tree exp, {
> > tree arg;
> > rtx pat, op;
> > - unsigned int i, nargs, arg_adjust = 0;
> > + unsigned int i, nargs, arg_adjust = 0, constant = 100;
>
>
> Any reason we put 100 here ??
Just to initialize it to a number that will not lead to issue. But we need
to initialize it to ensure nothing unexpected happens since it is
conditionally set.
> > diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md index
> > 9a514161b3b..f65884921bb 100644
> > --- a/gcc/config/i386/sse.md
> > +++ b/gcc/config/i386/sse.md
> > @@ -34612,3 +34626,71 @@
> > "TARGET_ACEV1"
> > "tilemov<rowcol>\t{%2, %1, %%tmm%c0|tmm%c0, %1, %2}"
> > [(set_attr "prefix" "evex")])
> > +
> > +(define_insn "top2bf16ps"
> > + [(set (reg:V32SF TMM_REGNUM)
> > + (unspec_volatile:V32SF
> > + [(match_operand:SI 0 "const_0_to_7_operand")
>
> In previous patches we are giving QI mode, should we follow the same?
Definitely. I forgot to change them to QI here since it is a last minute
change for previous patches.
Thx,
Haochen
