Re: [PATCH 55/67] target/arm: Convert FCVT* (vector, integer) scalar to decodetree

[Richard Henderson]

On 12/6/24 10:23, Peter Maydell wrote:
> On Sun, 1 Dec 2024 at 15:21, Richard Henderson
> <richard.henderson@linaro.org> wrote:
> > Arm silliness with naming, the scalar insns described
> > as part of the vector instructions, as separate from
> > the "regular" scalar insns which output to general registers.
> >
> > Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > ---
> >   target/arm/tcg/translate-a64.c | 135 ++++++++++++++-------------------
> >   target/arm/tcg/a64.decode      |  30 ++++++++
> >   2 files changed, 87 insertions(+), 78 deletions(-)
> >
> > diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c
> > index 98a42feb7d..ad245f2c26 100644
> > --- a/target/arm/tcg/translate-a64.c
> > +++ b/target/arm/tcg/translate-a64.c
> > @@ -8678,6 +8678,16 @@ static void do_fcvt_scalar(DisasContext *s, MemOp out, 
> > MemOp esz,
> >                                    tcg_shift, tcg_fpstatus);
> >               tcg_gen_extu_i32_i64(tcg_out, tcg_single);
> >               break;
> > +        case MO_16 | MO_SIGN:
> > +            gen_helper_vfp_toshh(tcg_single, tcg_single,
> > +                                 tcg_shift, tcg_fpstatus);
> > +            tcg_gen_extu_i32_i64(tcg_out, tcg_single);
> > +            break;
> > +        case MO_16:
> > +            gen_helper_vfp_touhh(tcg_single, tcg_single,
> > +                                 tcg_shift, tcg_fpstatus);
> > +            tcg_gen_extu_i32_i64(tcg_out, tcg_single);
> > +            break;
>
> This hunk adds calls to the toshh and touhh helpers,
> but as far as I can see it doesn't remove any calls to
> those helpers that were in the old decode functions or
> any calls to the handle_simd_shift_fpint_conv() function
> which was the only one that did call them. Should this
> be in a different patch?

Removing those happens in patch 61, with FCVTZ* (vector, fixed-point).
Here we're only converting the scalar path.

> (Conversely, we remove calls to gen_helper_advsimd_f16tosinth
> and gen_helper_advsimd_f16touinth but don't have those here.)

With the conversion I'm sharing more code.  So we (eventially) remove advsimd_f16to*inth 
entirely and use only vfp_to*hh.  They both call the same float16_to_*int16_scalbn 
function in the end.


r~