[cairo] _cairo_color_compute_shorts fails with FPU set to single precision

[Behdad Esfahbod]

On Wed, 2006-08-30 at 09:56 +0200, Alexander Larsson wrote:
> On Wed, 2006-08-30 at 03:45 -0400, Behdad Esfahbod wrote:
> 
> > Or better yet, use what Ali suggests.  Or even without a branch:
> > 
> >         int32_t i;
> >         i = f * 65536.0;
> >         i -= i == 65536;
> 
> Are you sure this doesn't generate a branch in the assembly? It might be
> able to use a cmov to avoid it, but maybe not.

It's using the comparison result in the subtract AFAIU:

convert:
        pushl   %ebp
        xorl    %edx, %edx
        movl    %esp, %ebp
        subl    $8, %esp
        fnstcw  -2(%ebp)
        flds    .LC0
        fmull   8(%ebp)
        movzwl  -2(%ebp), %eax
        movb    $12, %ah
        movw    %ax, -4(%ebp)
        fldcw   -4(%ebp)
        fistpl  -8(%ebp)
        fldcw   -2(%ebp)
        movl    -8(%ebp), %eax
        leave
        cmpl    $65536, %eax
        sete    %dl
        subl    %edx, %eax
        ret


(gcc -O2)

int
convert (double d)
{
  int i;
  i = d * 65536.0;
  i -= i == 65536;
  return i;
}


Carl, while you are at it, it may be good to think about what to do
about NANs in color computations.  Catching NAN in all color entry
points for example if we're not doing already.

-- 
behdad
http://behdad.org/

"Commandment Three says Do Not Kill, Amendment Two says Blood Will Spill"
        -- Dan Bern, "New American Language"