Re: [Qemu-devel] [PATCH 135/147] softfloat: Fix uint64_to_float64

[Peter Maydell]

On 28 September 2012 02:12, Richard Henderson <address@hidden> wrote:
> The interface to normalizeRoundAndPackFloat64 requires that the
> high bit be clear.  Perform one shift-right-and-jam if needed.
>
> Signed-off-by: Richard Henderson <address@hidden>
> ---
>  fpu/softfloat.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/fpu/softfloat.c b/fpu/softfloat.c
> index b29256a..45f105a 100644
> --- a/fpu/softfloat.c
> +++ b/fpu/softfloat.c
> @@ -1273,9 +1273,14 @@ float64 int64_to_float64( int64 a STATUS_PARAM )
>
>  float64 uint64_to_float64( uint64 a STATUS_PARAM )
>  {
> -    if ( a == 0 ) return float64_zero;
> -    return normalizeRoundAndPackFloat64( 0, 0x43C, a STATUS_VAR );
> +    int exp =  0x43C;
>
> +    if ( a == 0 ) return float64_zero;
> +    if ( (int64_t)a < 0 ) {
> +        a = (a >> 1) | (a & 1);
> +        exp += 1;
> +    }

You could just use
  shift64RightJamming(a, 1, &a);
rather than calculating a like that; gcc is smart enough to emit
the same code as your longhand version.

> +    return normalizeRoundAndPackFloat64( 0, exp, a STATUS_VAR );

I worked through it and I believe this patch does the right thing
with raising or not raising the inexact flag...

-- PMM