Hi Richard,
> ...I still think we should avoid testing can_create_pseudo_p.
> Does it work with the last part replaced by:
>
> if (!DECIMAL_FLOAT_MODE_P (mode))
> {
> if (aarch64_can_const_movi_rtx_p (src, mode)
> || aarch64_float_const_representable_p (src)
> || aarch64_float_const_zero_rtx_p (src))
> return true;
> }
>
> return false;
>
> ? If so, the patch is OK with that change from my POV, but please wait
> till Thursday morning for others' opinions.
With that every FP literal load causes an internal error, so they must be
allowed.
I could change the can_create_pseudo_p to true. This generates identical code
but
it allows literal loads to be created after regalloc (which should ultimately
crash as
there is no valid alternative).
Cheers,
Wilco
