Re: [PATCH v2 14/25] target/ppc: Avoid tcg_const_i64 in do_vcntmb

[Daniel Henrique Barboza]

On 3/7/23 15:34, Richard Henderson wrote:
> Compute both partial results separately and accumulate
> at the end, instead of accumulating in the middle.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---

Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>

> Cc: Daniel Henrique Barboza <danielhb413@gmail.com>
> Cc: Cédric Le Goater <clg@kaod.org>
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Cc: Greg Kurz <groug@kaod.org>
> Cc: qemu-ppc@nongnu.org
> ---
>   target/ppc/translate/vmx-impl.c.inc | 21 +++++++++++----------
>   1 file changed, 11 insertions(+), 10 deletions(-)
>
> diff --git a/target/ppc/translate/vmx-impl.c.inc 
> b/target/ppc/translate/vmx-impl.c.inc
> index 7af6d7217d..ca27c11d87 100644
> --- a/target/ppc/translate/vmx-impl.c.inc
> +++ b/target/ppc/translate/vmx-impl.c.inc
> @@ -2236,24 +2236,25 @@ static bool trans_MTVSRBMI(DisasContext *ctx, arg_DX_b 
> *a)
>   
>   static bool do_vcntmb(DisasContext *ctx, arg_VX_mp *a, int vece)
>   {
> -    TCGv_i64 rt, vrb, mask;
> -    rt = tcg_const_i64(0);
> -    vrb = tcg_temp_new_i64();
> +    TCGv_i64 r[2], mask;
> +
> +    r[0] = tcg_temp_new_i64();
> +    r[1] = tcg_temp_new_i64();
>       mask = tcg_constant_i64(dup_const(vece, 1ULL << ((8 << vece) - 1)));
>   
>       for (int i = 0; i < 2; i++) {
> -        get_avr64(vrb, a->vrb, i);
> +        get_avr64(r[i], a->vrb, i);
>           if (a->mp) {
> -            tcg_gen_and_i64(vrb, mask, vrb);
> +            tcg_gen_and_i64(r[i], mask, r[i]);
>           } else {
> -            tcg_gen_andc_i64(vrb, mask, vrb);
> +            tcg_gen_andc_i64(r[i], mask, r[i]);
>           }
> -        tcg_gen_ctpop_i64(vrb, vrb);
> -        tcg_gen_add_i64(rt, rt, vrb);
> +        tcg_gen_ctpop_i64(r[i], r[i]);
>       }
>   
> -    tcg_gen_shli_i64(rt, rt, TARGET_LONG_BITS - 8 + vece);
> -    tcg_gen_trunc_i64_tl(cpu_gpr[a->rt], rt);
> +    tcg_gen_add_i64(r[0], r[0], r[1]);
> +    tcg_gen_shli_i64(r[0], r[0], TARGET_LONG_BITS - 8 + vece);
> +    tcg_gen_trunc_i64_tl(cpu_gpr[a->rt], r[0]);
>       return true;
>   }
>