On Mon, Oct 24, 2022 at 12:12 AM Richard Biener
<richard.guenther@gmail.com> wrote:
>
> On Fri, Oct 21, 2022 at 6:18 PM H.J. Lu <hjl.tools@gmail.com> wrote:
> >
> > On Fri, Oct 21, 2022 at 2:33 AM Richard Biener
> > <richard.guenther@gmail.com> wrote:
> > >
> > > On Thu, Oct 20, 2022 at 6:58 PM H.J. Lu via Gcc-patches
> > > <gcc-patches@gcc.gnu.org> wrote:
> > > >
> > > > commit e034c5c895722e0092d2239cd8c2991db77d6d39
> > > > Author: Jakub Jelinek <jakub@redhat.com>
> > > > Date:   Sat Dec 2 08:54:47 2017 +0100
> > > >
> > > >         PR target/78643
> > > >         PR target/80583
> > > >         * expr.c (get_inner_reference): If DECL_MODE of a non-bitfield
> > > >         is BLKmode for vector field with vector raw mode, use TYPE_MODE
> > > >         instead of DECL_MODE.
> > > >
> > > > fixed the case where DECL_MODE of a vector field is BLKmode and its
> > > > TYPE_MODE is a vector mode because of target attribute.  Remove the
> > > > BLKmode check for the case where DECL_MODE of a vector field is a vector
> > > > mode and its TYPE_MODE is BLKmode because of target attribute.
> > > >
> > > > gcc/
> > > >
> > > >         PR target/107304
> > > >         * expr.c (get_inner_reference): Always use TYPE_MODE for vector
> > > >         field with vector raw mode.
> > > >
> > > > gcc/testsuite/
> > > >
> > > >         PR target/107304
> > > >         * gcc.target/i386/pr107304.c: New test.
> > > > ---
> > > >  gcc/expr.cc                              |  3 +-
> > > >  gcc/testsuite/gcc.target/i386/pr107304.c | 39 ++++++++++++++++++++++++
> > > >  2 files changed, 40 insertions(+), 2 deletions(-)
> > > >  create mode 100644 gcc/testsuite/gcc.target/i386/pr107304.c
> > > >
> > > > diff --git a/gcc/expr.cc b/gcc/expr.cc
> > > > index efe387e6173..9145193c2c1 100644
> > > > --- a/gcc/expr.cc
> > > > +++ b/gcc/expr.cc
> > > > @@ -7905,8 +7905,7 @@ get_inner_reference (tree exp, poly_int64_pod *pbitsize,
> > > >           /* For vector fields re-check the target flags, as DECL_MODE
> > > >              could have been set with different target flags than
> > > >              the current function has.  */
> > > > -         if (mode == BLKmode
> > > > -             && VECTOR_TYPE_P (TREE_TYPE (field))
> > > > +         if (VECTOR_TYPE_P (TREE_TYPE (field))
> > > >               && VECTOR_MODE_P (TYPE_MODE_RAW (TREE_TYPE (field))))
> > >
> > > Isn't the check on TYPE_MODE_RAW also wrong then?  Btw, the mode could
> >
> > TYPE_MODE_RAW is always set to a vector mode for a vector type:
> >
> >        /* Find an appropriate mode for the vector type.  */
> >         if (TYPE_MODE (type) == VOIDmode)
> >           SET_TYPE_MODE (type,
> >                          mode_for_vector (SCALAR_TYPE_MODE (innertype),
> >                                           nunits).else_blk ());
>
> But mode_for_vector can return a MODE_INT!

You are right.

>   /* For integers, try mapping it to a same-sized scalar mode.  */
>   if (GET_MODE_CLASS (innermode) == MODE_INT)
>     {
>       poly_uint64 nbits = nunits * GET_MODE_BITSIZE (innermode);
>       if (int_mode_for_size (nbits, 0).exists (&mode)
>           && have_regs_of_mode[mode])
>         return mode;
>
> > But TYPE_MODE returns BLKmode if the vector mode is unsupported.
> >
> > > also be an integer mode.
> >
> > For a vector field, mode is either BLK mode or the vector mode.  Jakub,
> > can you comment on it?
>
> I think that for
>
> typedef int v2si __attribute__((vector_size(8)));
>
> struct X { int i; v2si j; };
>
> v2si should get DImode with -mno-sse?
>

Currently GCC generates

(insn 31 32 33 (set (subreg:DI (reg:V2SI 105) 0)
        (reg:DI 84 [ _3 ])) "y2.c":12:11 -1
     (nil))

With my patch, v2si gets DImode directly without SUBREG.

Here is the v2 patch with the update commit message:

Remove the BLKmode check for the case where DECL_MODE
of a vector field is a vector mode and its TYPE_MODE isn't a
vector mode because of target attribute.

OK for master?

Thanks.

-- 
H.J.