Re: [RFC PATCH v2] target/ppc: Enable hardfloat for PPC

[Richard Henderson]

On 2/18/20 9:10 AM, BALATON Zoltan wrote:
>  void helper_reset_fpstatus(CPUPPCState *env)
>  {
> -    set_float_exception_flags(0, &env->fp_status);
> +    set_float_exception_flags(env->default_fp_excpt_flags, &env->fp_status);
>  }

What I don't like is the forced setting of inexact.  I don't mind leaving it
set if it is already set, which corresponds to the normal accumulation of
exceptions.

In addition, if the inexact exception is unmasked, I would expect a signal to
be delivered only when an inexact exception happens.  Whereas this patch would
deliver a signal for every fp operation.

It should be just as easy to do

    flags = get_float_exception_flags(status);
    flags &= env->save_fp_exception_flags;
    set_float_exception_flags(flags, status);


> +    DEFINE_PROP_BOOL("hardfloat", PowerPCCPU, hardfloat, true),

I would also prefer a different name here -- perhaps x-no-fp-fi.


r~