Re: [PATCH 0/1] target/arm: Check NaN mode before silencing NaN

[Peter Maydell]

On Sat, 26 Jun 2021 at 00:04, Joe Komlodi <joe.komlodi@xilinx.com> wrote:
>
> Hi all,
>
> This fixes an assertion that occurs when executing FRSQRTE, FRECPE, or FRECPX
> on a signaling NaN when the CPU has default NaN mode enabled.
>
> When attempting to silence the NaN, we hit an assertion that ensures that the
> CPU FPU status does not have default_nan_mode set.
>
> To avoid this, we check if default_nan_mode is set before trying to silence 
> the
> NaN.
> What happens then is the instruction sets any flags because of the signaling
> NaN, then gets the default NaN value due to default_nan_mode being set.

Richard, Alex: what is the assertion trying to achieve ? It doesn't
seem entirely obvious to me that because we're in default-NaN mode
(which is a property of the *output* of FPU insns) that we should
blow up on calling float*_silence_nan() (which is typically an action
performed on the *input* of FPU insns).

This used to work fine, because we would have seen the assertions
when we tested the implementation of all these Arm insns...

If we do want to keep the assertion, somebody should audit the
other frontends that use float*_silence_nan() (i386, m68k, s390x)
to see if they also need updating.

thanks
-- PMM