Re: [PATCH 3/3] target/ppc: Merge fsqrt and fsqrts helpers

[Richard Henderson]

On 9/5/22 17:31, Víctor Colombo wrote:
> I have a message bookmarked here ([1]), but I don't know if there is a
> previous one with a more in depth scheme.

There may have been a previous, but that's the one I was thinking of.

> Anyway, I was also analyzing recently the idea of removing all these
> reset_fpstatus() calls from instructions helpers. I think this would
> require to actually call it from the end of the (previous) instructions instead of the 
> beginning? Like adding the call to
> do_float_check_status() and float_invalid_op_*() as a focal point to
> 'hide' the calls to reset_fpstatus().

Well, there would of course be no separate call, but do_float_check_status 
would:

    int status = get_float_exception_flags(&env->fp_status);

    set_float_exception_flags(0, &env->fp_status);

straight away.  No extra call overhead, and the steady-state of softfp exception flags 
outside of an in-progress fp operation is 0.

> However there are also insns
> helpers that don't call these auxiliary functions, which I think would
> cause the refactor to not be worthy overall.

Anything that can raise a softfp exception and doesn't do something with it, either 
immediately within the same helper, or just afterward with helper_float_check_status, is 
buggy.  With those fixed, helper_reset_fpstatus may be removed entirely.



r~