ompf2.com

For what it's worth, both AMD and Nvidia currently use LLVM in their OpenCL toolchains (Nvidia also uses it in the Cuda toolchain). Nvidia's backend is already in the main LLVM distribution (NVPTX target), and AMD's (R600 I believe, or maybe CAL) is supposed to be incorporated in LLVM 3.3. This mean...

Basically, you just have to ray march through the fire volume for every light sample, much like you would for primary rays.

It's probably worth noting that GPUs should be able to handle function pointers more efficiently than CPUs for micro-architectural reasons (if you're interested, I can explain), so be sure to give them a try. Are you referring to the GPU's latency hiding strategies with this comment? Or something e...

A CPU will do polymorphism either through function pointers or though a big switch clause, depending on what the compiler thinks is fastest. Hence, you have to eat that cost there too. It's probably worth noting that GPUs should be able to handle function pointers more efficiently than CPUs for micr...

Just write your own. This is the one I use: __constant__ unsigned int shift1[4] = {6, 2, 13, 3}; __constant__ unsigned int shift2[4] = {13, 27, 21, 12}; __constant__ unsigned int shift3[4] = {18, 2, 7, 13}; __constant__ unsigned int offset[4] = {4294967294, 4294967288, 4294967280, 4294967168}; __sha...

PPM (and VCM) can be made to be unbiased simply by having each photon store information about the previous hit, rather than the current hit. Then when the eye path lands close to a photon, you simply shoot a ray toward the photon's source (be it directly at a light or simply the last thing the photo...

Optix now has a GPU based HLBVH builder, as of version 2.5. Has anyone tried to benchmark it yet?

There's one very practical difference between unbiased and merely consistent renderers. To see it, imagine that you have a render farm, and you'd like to use a large number of separate computers to render an image. With an unbiased renderer, you can simply have them each render the image, but with f...

All that unbiased means in practice is that, thanks to the Central Limit Theorem, an infinite sum of finite runs will converge to zero error. (it also means that each run has an expected error of zero over infinitely many realizations, but this in practical terms has no value at all - all that matt...

This is quite spectacular: http://agl.unm.edu/rpf/ Moreover, it's very likely that it can be brought near to real time on a high end GPU or two, since it looks like a very GPU friendly algorithm, I'd expect a speed increase of 100x or more, since it's heavy on local number crunching and uses some tr...