Re: Filtering

Hi Ola,

unfortunately the floating point textures on the platform I'm working
on don't support bilinear filtering (this is the reason why I stored
the value in an RGB-texture to begin with).

/Andreas

P.S. Got your Ph.D yet?

On Fri, Oct 1, 2010 at 8:12 PM, Ola Olsson <[email protected]> wrote:
> While we're throwing ideas around, why not try using a float texture and see
> if it produces the same error?
>
> .ola
>
> P.S.
>  Hi Andreas :)
>
> ----- Original Message -----
> From: "Nathaniel Hoffman" <[email protected]>
> To: "Game Development Algorithms" <[email protected]>
> Sent: Friday, October 01, 2010 7:59 PM
> Subject: Re: [Algorithms] Filtering
>
>
>> Didn't the newer NVIDIA GPUs fix this?
>>
>>> You guessed right. The loss of precision is in the texture units.
>>> Unfortunately, 8 bit components are filtered to 8 bit results (even
>>> though
>>> they show up as floating point values in the shader). This is true for
>>> nvidia gpus for sure and probably all other gpus.
>>>
>>> -mike
>>>   ----- Original Message -----
>>>   From: Stefan Sandberg
>>>   To: Game Development Algorithms
>>>   Sent: Friday, October 01, 2010 1:45 AM
>>>   Subject: Re: [Algorithms] Filtering
>>>
>>>
>>>   Assuming you're after precision, what's wrong with doing it manually?
>>> :)
>>>   If performance is what you're after, and you're working on textures as
>>> they were intended(ie, game textures or video or something like that,
>>> not 'data'), you could separate contrast & color separately, keeping
>>> high contrast resolution, and downsampled color, and
>>>   you'd save both bandwidth and instr.
>>>   If you simply want to know 'why', I'm guessing loss of precision in the
>>> tex units?
>>>   You've already ruled out shader precision from your own manual
>>> filtering, so doesn't leave much else, imo..
>>>   Other than manipulating the data you're working on, which is the only
>>> thing you -can- change I guess, I cant really see a solution,
>>>   but far greater minds linger here than mine, so hold on for what I
>>> assume will be a lengthy description of floating point math as
>>>   it is implemented in modern gpu's :)
>>>
>>>
>>>
>>>
>>>
>>>
>>>   On Fri, Oct 1, 2010 at 9:57 AM, Andreas Brinck
>>> <[email protected]> wrote:
>>>
>>>     Hi,
>>>
>>>     I have a texture in which I use the R, G and B channel to store a
>>>     value in the [0, 1] range with very high precision. The value is
>>>     extracted like this in the (Cg) shader:
>>>
>>>     float
>>>     extractValue(float2 pos) {
>>>        float4 temp = tex2D(buffer, pos);
>>>        return (temp.x * 16711680.0 + temp.y * 65280.0 + temp.z * 255.0) *
>>>     (1.0 / 16777215.0);
>>>     }
>>>
>>>     I now want to sample this value with bilinear filtering but when I do
>>>     this I don't get a correct result. If I do the filtering manually
>>> like
>>>     this:
>>>
>>>     float
>>>     sampleValue(float2 pos) {
>>>            float2 ipos = floor(pos);
>>>            float2 fracs = pos - ipos;
>>>            float d0 = extractValue(ipos);
>>>            float d1 = extractValue(ipos + float2(1, 0));
>>>            float d2 = extractValue(ipos + float2(0, 1));
>>>            float d3 = extractValue(ipos + float2(1, 1));
>>>            return lerp(lerp(d0, d1, fracs.x), lerp(d2, d3, fracs.x),
>>> fracs.y);
>>>     }
>>>
>>>     everything works as expected. The values in the buffer can be seen as
>>>     a linear combination of three constants:
>>>
>>>     value = (C0 * r + C1 * g + C2 * b)
>>>
>>>     If we use the built in texture filtering we should get the following
>>>     if we sample somewhere between two texels: {r0, g0, b0} and {r1, g1,
>>>     b1}. For simplicity we just look at filtering along one axis:
>>>
>>>     filtered value = lerp(r0, r1, t) * C0 + lerp(g0, g1, t) * C1 +
>>>     lerp(b0, b1, t) * C2;
>>>
>>>     Doing the filtering manually:
>>>
>>>     filtered value = lerp(r0 * C0 + b0 * C1 + g0 * C2, r1 * C0 + g1 * C1
>>> +
>>>     b1 * C2, t)  =
>>>                       = (r0 * C0 + b0 * C1 + g0 * C2) * (1 - t) + (r1 *
>>>     C0 + g1 * C1 + b1 * C2) * t =
>>>                       = (r0 * C0) * (1 - t) + (r1 * C0) * t + ... =
>>>                       = lerp(r0, r1, t) * C0 + ...
>>>
>>>     So in the world of non floating point numbers these two should be
>>>     equivalent right?
>>>
>>>     My theory is that the error is caused by an unfortunate order of
>>>     floating point operations. I've tried variations like:
>>>
>>>     (temp.x * (16711680.0 / 16777215.0) + temp.y * (65280.0/16777215.0) +
>>>     temp.z * (255.0/16777215.0))
>>>
>>>     and
>>>
>>>     (((temp.x * 256.0 + temp.y) * 256.0 + temp.z) * 255.0) * (1.0 /
>>> 16777215.0)
>>>
>>>     but all exhibit the same problem. What do you think; is it possible
>>> to
>>>     solve this problem?
>>>
>>>     Regards Andreas
>>>
>>>     ------------------------------------------------------------------------------
>>>     Start uncovering the many advantages of virtual appliances
>>>     and start using them to simplify application deployment and
>>>     accelerate your shift to cloud computing.
>>>     http://p.sf.net/sfu/novell-sfdev2dev
>>>     _______________________________________________
>>>     GDAlgorithms-list mailing list
>>>     [email protected]
>>>     https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
>>>     Archives:
>>>
>>> http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list
>>>
>>>
>>>
>>>
>>>
>>> ------------------------------------------------------------------------------
>>>
>>>
>>>   ------------------------------------------------------------------------------
>>>   Start uncovering the many advantages of virtual appliances
>>>   and start using them to simplify application deployment and
>>>   accelerate your shift to cloud computing.
>>>   http://p.sf.net/sfu/novell-sfdev2dev
>>>
>>>
>>> ------------------------------------------------------------------------------
>>>
>>>
>>>   _______________________________________________
>>>   GDAlgorithms-list mailing list
>>>   [email protected]
>>>   https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
>>>   Archives:
>>>
>>> http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list------------------------------------------------------------------------------
>>> Start uncovering the many advantages of virtual appliances
>>> and start using them to simplify application deployment and
>>> accelerate your shift to cloud computing.
>>> http://p.sf.net/sfu/novell-sfdev2dev_______________________________________________
>>> GDAlgorithms-list mailing list
>>> [email protected]
>>> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
>>> Archives:
>>> http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list
>>
>>
>>
>> ------------------------------------------------------------------------------
>> Start uncovering the many advantages of virtual appliances
>> and start using them to simplify application deployment and
>> accelerate your shift to cloud computing.
>> http://p.sf.net/sfu/novell-sfdev2dev
>> _______________________________________________
>> GDAlgorithms-list mailing list
>> [email protected]
>> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
>> Archives:
>> http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list
>
>
> ------------------------------------------------------------------------------
> Start uncovering the many advantages of virtual appliances
> and start using them to simplify application deployment and
> accelerate your shift to cloud computing.
> http://p.sf.net/sfu/novell-sfdev2dev
> _______________________________________________
> GDAlgorithms-list mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
> Archives:
> http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list
>

------------------------------------------------------------------------------
Start uncovering the many advantages of virtual appliances
and start using them to simplify application deployment and
accelerate your shift to cloud computing.
http://p.sf.net/sfu/novell-sfdev2dev
_______________________________________________
GDAlgorithms-list mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
Archives:
http://sourceforge.net/mailarchive/forum.php?forum_name=gdalgorithms-list