Design issues (opengl, hdr, channels, pigment(?))

[Boudewijn Rempt]

On Friday 02 February 2007, Schleimer, Ben wrote:
> Hi,
>
> > Well, graphics cards don't provide the same
> > precision that the cpu does, so
> > the results of computations can come out
> > differently:
>
> http://www.gpgpu.org/wiki/FAQ#Where_can_I_get_information_about_GPU_floatin
>g_point_precision.3F_Is_it_IEEE-compliant.3F.
>
>   The article just points out that the cards do
> rounding on floats/doubles differently then IEEE 754.
> If you convert your uint8 channels into float, I
> seriously doubt you'll have precision problem unless
> you're trying to blend thousands of channels (24bits
> mantissa - 8bits = 16bits of space before you hit
> rounding problems)

Of course, for 8 bit rgba that's not a problem, but when doing interesting 
stuff with float lab, it may become an issue.
>
> Of course this needs to be tested and obviously, the
> above doesnt apply if there are other problems with
> GPU floating point.
>




-- 
Boudewijn Rempt 
http://www.valdyas.org/fading/index.cgi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://mail.kde.org/pipermail/kimageshop/attachments/20070202/104609c1/attachment.pgp