[Qemu-devel] Re: MIPS FP rounding

[Thiemo Seufer]

Daniel Jacobowitz wrote:
> Thiemo, in May you added this:
> 
> -    update_fcr31();       \
> +    update_fcr31();                                           \
> +    if (GET_FP_CAUSE(env->fcr31) & FP_INVALID)                \
> +        FST2 = 0x7fbfffff;                                    \
> +    else if (GET_FP_CAUSE(env->fcr31) & FP_UNDERFLOW) {       \
> +        if ((env->fcr31 & 0x3) == 0)                          \
> +            FST2 &= 0x80000000ULL;                            \
> +    }                     \
> 
> That last bit says that if underflow is detected and the rounding mode
> is round-to-nearest, return an signed appropriate signed zero.  Why?
> The MIPS ISA docs say a rounded result is returned in this case, and
> that's what's in FST2 if that code is removed.

It fixed an internal testcase, I'll have to check what was going
on there, probably tomorrow.


Thiemo