On Sun, 3 Aug 2025 at 02:41, Richard Henderson
<richard.hender...@linaro.org> wrote:
>
> Signed-off-by: Richard Henderson <richard.hender...@linaro.org>
> ---
> target/arm/tcg/translate-a64.c | 44 ++++++++++++++++++++++++++++++++++
> target/arm/tcg/a64.decode | 10 ++++++++
> 2 files changed, 54 insertions(+)
>
> diff --git a/target/arm/tcg/translate-a64.c b/target/arm/tcg/translate-a64.c
> index dbf47595db..b70ae5befd 100644
> --- a/target/arm/tcg/translate-a64.c
> +++ b/target/arm/tcg/translate-a64.c
> @@ -4552,6 +4552,50 @@ TRANS(SUB_i, gen_rri, a, 1, 1, tcg_gen_sub_i64)
> TRANS(ADDS_i, gen_rri, a, 0, 1, a->sf ? gen_add64_CC : gen_add32_CC)
> TRANS(SUBS_i, gen_rri, a, 0, 1, a->sf ? gen_sub64_CC : gen_sub32_CC)
>
> +/*
> + * Min/Max (immediate)
> + */
> +
> +static void gen_wrap3_i32(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m, NeonGenTwoOpFn
> fn)
> +{
> + TCGv_i32 t1 = tcg_temp_new_i32();
> + TCGv_i32 t2 = tcg_temp_new_i32();
> +
> + tcg_gen_extrl_i64_i32(t1, n);
> + tcg_gen_extrl_i64_i32(t2, m);
> + fn(t1, t1, t2);
> + tcg_gen_extu_i32_i64(d, t1);
> +}
> +
> +static void gen_smax32_i64(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m)
> +{
> + gen_wrap3_i32(d, n, m, tcg_gen_smax_i32);
> +}
> +
> +static void gen_smin32_i64(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m)
> +{
> + gen_wrap3_i32(d, n, m, tcg_gen_smin_i32);
> +}
> +
> +static void gen_umax32_i64(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m)
> +{
> + gen_wrap3_i32(d, n, m, tcg_gen_umax_i32);
> +}
> +
> +static void gen_umin32_i64(TCGv_i64 d, TCGv_i64 n, TCGv_i64 m)
> +{
> + gen_wrap3_i32(d, n, m, tcg_gen_umin_i32);
> +}
> +
> +TRANS_FEAT(SMAX_i, aa64_cssc, gen_rri, a, 0, 0,
> + a->sf ? tcg_gen_smax_i64 : gen_smax32_i64)
> +TRANS_FEAT(SMIN_i, aa64_cssc, gen_rri, a, 0, 0,
> + a->sf ? tcg_gen_smin_i64 : gen_smin32_i64)
> +TRANS_FEAT(UMAX_i, aa64_cssc, gen_rri, a, 0, 0,
> + a->sf ? tcg_gen_umax_i64 : gen_umax32_i64)
> +TRANS_FEAT(UMIN_i, aa64_cssc, gen_rri, a, 0, 0,
> + a->sf ? tcg_gen_umin_i64 : gen_umin32_i64)
We end up doing the zero-extension twice for the 32-bit case,
once in gen_wrap3_i32(), and once in gen_rri(). Does the
extra one get optimized away ?
Otherwise
Reviewed-by: Peter Maydell <peter.mayd...@linaro.org>
thanks
-- PMM
