On Wed, May 20, 2026 at 9:05 AM Hongtao Liu <[email protected]> wrote:
>
> On Wed, May 20, 2026 at 2:55 PM Uros Bizjak <[email protected]> wrote:
> >
> > On Fri, May 15, 2026 at 10:40 AM Uros Bizjak <[email protected]> wrote:
> > >
> > > On Fri, May 15, 2026 at 3:46 AM Liu, Hongtao <[email protected]> 
> > > wrote:
> > > >
> > > >
> > > >
> > > > > -----Original Message-----
> > > > > From: Roger Sayle <[email protected]>
> > > > > Sent: Friday, May 15, 2026 5:23 AM
> > > > > To: 'GCC Patches' <[email protected]>
> > > > > Cc: 'Hongtao Liu' <[email protected]>; Liu, Hongtao
> > > > > <[email protected]>; 'Uros Bizjak' <[email protected]>
> > > > > Subject: [PATCH] Improve vector increment/decrement on x86.
> > > > >
> > > > >
> > > > > This patch improves the code generated by the i386 backend for 
> > > > > incrementing
> > > > > (adding one to) and decrementing (subtracting one from) a vector.  
> > > > > With SSE
> > > > > materializing the vector -1 is more efficient than materializing the 
> > > > > vector +1,
> > > > > hence x + 1 (increment) is better expressed as x - (-1), and x - 1 
> > > > > (decrement) is
> > > > > better expressed as x + (-1).  Conveniently the relevant additions and
> > > > > subtractions are specified as a single pattern, using a plusminus 
> > > > > iterator, in the
> > > > > machine description.
> > > >
> > > > Can we add pre_reload define_insn_and_split for them,
> > > >
> > > > (set (reg:V16QI 100 [ _2 ])
> > > >     (minus:V16QI (reg:V16QI 107 [ x ])
> > > >         (const_vector:V16QI [
> > > >                 (const_int 1 [0x1]) repeated x16
> > > >             ])))
> > > >
> > > > Theoretically, it should be able to capture more optimization 
> > > > opportunities (if vector +/-1 is only exposed through RTL optimization)
> >
> > Another issue is with insn canonicalization [1], where
> >
> > (minus x (const_int n)) is converted to (plus x (const_int -n)).
> >
> > According to the above, x - (-1) is not a canonical form and
> > optimizers will convert it to the canonical form anyway.
>
> In recog.cc
>
>     case MINUS:
>       if (CONST_SCALAR_INT_P (XEXP (x, 1)))
>         validate_change (object, loc,
>                          simplify_gen_binary
>                          (PLUS, GET_MODE (x), XEXP (x, 0),
>                           simplify_gen_unary (NEG,
>                                               GET_MODE (x), XEXP (x, 1),
>                                               GET_MODE (x))), 1);
>       break;
>
> Canonicalization is restricted to CONST_SCALAR_INT_P, that's why it's
> not trigger for vector.

Thanks for pointing this difference out.

Uros.

Reply via email to