Re: [Openexr-devel] Shade me confused

[Drew Hess]

Rock!  :)

This will do wonders for OpenEXR sequence playback ;)

-dwh-


On Thu, 27 Mar 2003, Nafees Bin Zafar wrote:

> Wheeeee.  It works!!!
> 
> NVRM version: NVIDIA Linux x86 nvidia.o Kernel Module  1.0-4348  Mon Mar 
> 24 16:36:50 PST 2003
> 
> diff -r1.1 ImageViewFragShader.cpp
> 130c130
> <     float *ptr = _floatPixels;
> ---
>  >     half *ptr = _floatPixels;
> 150c150
> <           GL_RGBA, GL_FLOAT, _floatPixels);
> ---
>  >           GL_RGBA, GL_HALF_FLOAT_NV, _floatPixels);
> 
> 
> diff -r1.1 ImageViewFragShader.h
> 92c92
> <     Imf::Array<float>   _floatPixels;
> ---
>  >     Imf::Array<half>   _floatPixels;
> 
> 
> Thanks Drew!
> 
> -n
> 
> 
> 
> Drew Hess wrote:
> 
> >It's already there.  If you look in glext.h in the Nvidia Cg SDK, you'll 
> >find this:
> >
> >#define GL_HALF_FLOAT_NV                  0x140B
> >
> >So if you have a half framebuffer, all you really need to do is change the
> >input pixel type to GL_HALF_FLOAT_NV in the glTexImage2D call (once the
> >bug is fixed) and it should work.
> >
> >We use half to represent color channels and float for things like Z,
> >normals, etc.
> >
> >
> >-dwh-
> >
> >
> >
> > On Thu, 27 Mar 2003, Nafees Bin Zafar wrote:
> >
> >  
> >
> >>Are there plans for a GL_HALF_NV type?  I would thing it would be at 
> >>least equally natural for the main program to do the conversion to half, 
> >>and explicitly specify that it's sending in half format data.  Also it 
> >>is plausible to imagine that an application's internal format is half, 
> >>in which case there would have to be a convert to float, only for a 
> >>convert back to half to occur right away in the driver.
> >>
> >>Or am I missing something?  Is comptime's internal format half, or 
> >>float?  Nuke uses floats.
> >>
> >>-n
> >>
> >>Drew Hess wrote:
> >>
> >>    
> >>
> >>>It is float data that gets uploaded to the card, due to a bug in the 
> >>>current (as of 4290) Nvidia drivers.  Once the bug is fixed, we'll be able 
> >>>to send half data to the card instead.  It may be fixed in today's 4390 
> >>>drivers on Windows, but I'm waiting for the equivalent Linux drivers to 
> >>>find out.
> >>>
> >>>The magic conversion happens in the glTexImage2D call (the GL_FLOAT 
> >>>parameter specifies the input pixel type).
> >>>
> >>>
> >>>-dwh-
> >>>
> >>>
> >>>On Thu, 27 Mar 2003, Nafees Bin Zafar wrote:
> >>>
> >>> 
> >>>
> >>>      
> >>>
> >>>>Hi Drew.  Looking through the ImageViewFragShader class, I see that the 
> >>>>texel data passed into glTexImage2D() is just regular float data.  The 
> >>>>internal format spec is set to GL_FLOAT_RGBA16_NV.  So where does the 
> >>>>magic conversion to half type occur?
> >>>>
> >>>>I am hoping that float data doesn't get transferred to the graphics card 
> >>>>because that would transfer 16 unnecessary bits.
> >>>>
> >>>>-n
> >>>>
> >>>>
> >>>>   
> >>>>
> >>>>        
> >>>>
> >>>
> >>>
> >>> 
> >>>
> >>>      
> >>>
> >>
> >>    
> >>
> >
> >
> >
> >  
> >
> 
> 
>