Re: [Qemu-devel] [PATCH 08/19] target-alpha: use new float64_unordered() function

[Peter Maydell <peter.maydell@linaro.org>]

On 12 April 2011 22:59, Aurelien Jarno <aurelien@aurel32.net> wrote:
> Use float64_unordered() in helper_cmptun() instead of doing the
> the comparison manually. This also fixes the wrong behaviours with
> sNaNs.
>
> Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
> ---
>  target-alpha/op_helper.c |    5 +++--
>  1 files changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/target-alpha/op_helper.c b/target-alpha/op_helper.c
> index 6c2ae20..e07ae69 100644
> --- a/target-alpha/op_helper.c
> +++ b/target-alpha/op_helper.c
> @@ -904,10 +904,11 @@ uint64_t helper_cmptun (uint64_t a, uint64_t b)
>     fa = t_to_float64(a);
>     fb = t_to_float64(b);
>
> -    if (float64_is_quiet_nan(fa) || float64_is_quiet_nan(fb))
> +    if (float64_unordered(fa, fb, &FP_STATUS)) {
>         return 0x4000000000000000ULL;
> -    else
> +    } else {
>         return 0;
> +    }
>  }

I'm not sure this is right. The Alpha Architecture Handbook is a
bit opaque, but the Compiler Writer's Guide is clearer:
www.compaq.com/cpq-alphaserver/technology/literature/cmpwrgd.pdf
page D-4 says that CMPTUN (like CMPTEQ) should generate InvalidOp
for SNaNs but not for QNaNs. (Contrast CMPTLT, CMPTLE, which set
InvalidOp for both SNaNs and QNaNs.)

So I think you want the _quiet version here. (And helper_cmpteq
needs to use float64_eq_quiet rather than float64_eq.)

-- PMM