Hey, I'm trying to learn OpenGL using a fully programmable pipeline, rather than any fixed function bits, and I'm trying to get straight in my head a few things about OpenGL, particularly with HDRR. 1. a. Am I right in thinking that HDRR is a rendering technique where lighting is worked out linearly, using high precision floating point values, then one uses a tone mapping algorithm of some sort to convert the calculations to 8 bit RGB integers for display on a monitor, optionally applying a bloom effect to values that are clamped to pure white? b. If so, what support is there in OpenGL for this, in the way of shaders I guess. I mean, what's the 'standard' way of doing HDR rendering? I'm guessing this probably involves frame buffers or something... c. There seems to be far fewer resources for HDR in OpenGL than in DirectX. More specifically, with DirectX everyone seems to talk about Shader Model 3.0, enabling 32-bit floating point support and all that; with GLSL, is fp32 support implied, or does one have to enable it somehow? d. Also, are there any stock shaders/libraries for doing things like global illumination, tone mapping and lighting with an arbitrary number of lights? 2. a. DirectX seems to have loads of these Shader Model 3.0/4.0 graphics engines that support HDR, like X-Ray used in STALKER, Source Engine for HL2, CryEngine 2 for Crysis... Are there any modernish (preferably opensource but not necessary) engines for OpenGL that support modern rendering techniques like HDRR, soft shadows, global illumination, etc. and have no dependency on the fixed function pipeline? b. What versions of OpenGL are roughly equivelant to particular versions of DirectX/Shader Model, if that is even comparable? c. I understand that DirectX 10 is a big step up from 9.0c, is Longs Peak/MtEvans intended to be the true competition to DirectX 10, or are the current APIs a good feature match for each other, graphics wise? Thanks :)

personally im not to great a fan of HDR ( as it doesnt really seem to add much vs its cost ), perhaps things would look better once we have higher bitdepth displays.
check out the nvidia devleoper website theres a couple of examples of HDR using opengl.
yes floating textures are supported with opengl/glsl ( u will need a capable card of doing it though )

Quote:I completely disagree. This opinion shows that you have never seen a good implementation of realtime HDR. The difference is enormeous. HDR support (with good tonemapping) is absolutely vital to obtain photorealistic 3D images

care to point to any demos

Speaking from a photographer point of view, I have to agree with Yann L about HDR being vital for photorealistic scenery.

Admittedly with higher bit-depth/contrast ratio screens we'll see more of the benefit, but HDR-style techniques have been used for hundreds of years in paintings and that's what gives paintings that deeper look in comparison to photographs.

Consider the following scenario (it's a bit simplified, but the idea holds..): You have a low intensity light and a sun, but both are displayed at maximum intensity onscreen (255,255,255) due to the other objects in the scene being far lower intensity. You have a mildly reflective object... this mildly reflective object would reflect both the sun and the light at the same intensity, despite that in reality it'd reflect the sun far more than it'd reflect the low intensity light. With HDR, you could represent the sun as being far more intense than the lightbulb, and use that in all calculations, but have the actual intensities clamped in tone mapping.

Plus gamma, non-linear sampling and all that sort of thing is quite ugly; linear is nicer.

If you look at comparison screenshots, you can see where calculations like this have an impact, for example: http://en.wikipedia.org/wiki/Image:Hl2hdrcomparison.jpg

I'm only starting out with trying HDR in programming; but with photography I've used it quite a while. Another thing that has a large impact is things like motion blurring, since higher intensity lights will leave more of a trail than lower intensity objects, and as far as I know, you can only do that sort of thing with HDR imagery.

http://www.daionet.gr.jp/~masa/rthdribl/
http://www.sync.dk/~khn/research3/index.html

[Edited by - thecwin on May 21, 2007 6:51:49 AM]

yes im aware of those links ( i have studied HDR quite a lot, in fact are looking at supporting logluv32 in my game, i have done a trail )
WRT what i said "as it doesnt really seem to add much vs its cost" it does knock performance quite a bit, its not a cheap addition to make WRT speed + memory, which may stop u from adding other (more imporant) things, eg unified lighting etc.

heres a shot from the masa demo i took
http://www.zedzeek.com/junk/LDRHDR.jpg
true the difference is quite big in the reflections (but this is like a best case senerio for HDR in a more typical ingame scene the difference is gonna be far less than this). now i ask u honestly is a 30% drop in performance worth it?

btw tonemapping can be done with LDR as well