Re: [Qemu-devel] [PATCH 6/7] target-tricore: Add cmp.f instruction

[Richard Henderson <rth@twiddle.net>]

On 03/01/2016 08:24 AM, Bastian Koppelmann wrote:
> Signed-off-by: Bastian Koppelmann <kbastian@mail.uni-paderborn.de>
> ---
>  target-tricore/fpu_helper.c | 29 +++++++++++++++++++++++++++++
>  target-tricore/helper.h     |  1 +
>  target-tricore/translate.c  |  3 +++
>  3 files changed, 33 insertions(+)
> 
> diff --git a/target-tricore/fpu_helper.c b/target-tricore/fpu_helper.c
> index ee8b687..ceda415 100644
> --- a/target-tricore/fpu_helper.c
> +++ b/target-tricore/fpu_helper.c
> @@ -171,3 +171,32 @@ uint32_t helper_fdiv(CPUTriCoreState *env, uint32_t r1, uint32_t r2)
>      f_update_psw_flags(env, true);
>      return (uint32_t)f_result;
>  }
> +
> +uint32_t helper_fcmp(CPUTriCoreState *env, uint32_t r1, uint32_t r2)
> +{
> +    uint32_t result = 0;
> +    uint32_t lt, eq, uo;
> +    float32 arg1 = make_float32(r1);
> +    float32 arg2 = make_float32(r2);
> +
> +    set_flush_inputs_to_zero(0, &env->fp_status);
> +
> +    lt = float32_lt_quiet(arg1, arg2, &env->fp_status);
> +    eq = float32_eq_quiet(arg1, arg2, &env->fp_status);
> +    uo = float32_unordered(arg1, arg2, &env->fp_status);

Do it in one step with float32_compare.
You don't want _quiet; see below re psw_flags.

> +    env->FPU_FS = 0;
> +    if (float32_is_signaling_nan(arg1) || float32_is_signaling_nan(arg2)) {
> +        env->FPU_FI = (1 << 31);
> +        env->FPU_FS = 1;
> +    }
> +
> +    return result;

If you return flush_inputs_to_zero to 1 here, you don't have to zero it on all
other fp operations.

Why aren't you using the same f_update_psw_flags function?  The only bit that
compare can set is invalid anyway.


r~