Re: [Qemu-devel] [PATCH 17/37] target-ppc: Introduce DFP Add

[Richard Henderson]

On 04/18/2014 07:50 AM, Tom Musta wrote:
> +__attribute__ ((unused))
> +static void dfp_run_post_processors(struct PPC_DFP *dfp,
> +                PPC_DFP_PostProc post_processors[], const size_t n)
> +{
> +    int i;
> +
> +    for (i = 0; i < n; i++) {
> +        post_processors[i](dfp);
> +    }
> +}

...

> +PPC_DFP_PostProc ADD_PPs[] = {
> +    dfp_set_FPRF_from_FRT,
> +    dfp_check_for_OX,
> +    dfp_check_for_UX,
> +    dfp_check_for_XX,
> +    dfp_check_for_VXSNAN,
> +    dfp_check_for_VXISI_add,
> +};

Is this really any better than

static void ADD_PPs(struct PPC_DFP *dfp)
{
    dfp_set_FPRF_from_FRT(dfp);
    dfp_check_for_OX(dfp);
    dfp_check_for_UX(dfp);
    dfp_check_for_XX(dfp);
    dfp_check_for_VXSNAN(dfp);
    dfp_check_for_VXISI_add(dfp);
}

The forms seem equally readable to me, with the advantage that we'll get good
branch prediction on direct subroutine calls.  And if the compiler decides that
inlining is profitable, it'll have a chance.


r~