On Thu, Jul 09, 2026 at 12:18:51AM +0800, Max Chou wrote:
> FloatParts64.exp holds an unbiased exponent. Subtract exp_bias to store
> the unbiased equivalent of the max-normal biased exponent.
>
> Fixes: d8be49537658 ("fpu: Add conversion routines for OCP FP8 E5M2")
> Signed-off-by: Max Chou <[email protected]>
Reviewed-by: Chao Liu <[email protected]>
Thanks,
Chao
> ---
> fpu/softfloat.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index bc244a44ff..e2e53c1d68 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -2435,7 +2435,8 @@ static void parts_float_to_e5m2(FloatParts64 *a,
> float_status *s, bool saturate)
> /* Per OCP, conversion in SATURATE mode bounds Inf to MAX. */
> if (saturate) {
> a->cls = float_class_normal;
> - a->exp = float8_e5m2_params.exp_max - 1;
> + a->exp = (float8_e5m2_params.exp_max - 1 -
> + float8_e5m2_params.exp_bias);
> a->frac = MAKE_64BIT_MASK(float8_e5m2_params.frac_shift,
> float8_e5m2_params.frac_size + 1);
> }
> --
> 2.55.0
>
