I just pushed a new "Lite" version of my spherical harmonics shader library for HLSL: github.com/TheRealMJP/S... The Lite version does not use templates, and is compatible with FXC or other compilers that don't support templates and operator overloads. - ThreadSky

mjp123.bsky.social • 17 days ago

I just pushed a new "Lite" version of my spherical harmonics shader library for HLSL: https://github.com/TheRealMJP/SHforHLSL

The Lite version does not use templates, and is compatible with FXC or other compilers that don't support templates and operator overloads.

Comments

mjp123.bsky.social•17 days ago

The downside is there's no fp16 support for the Lite version. It could be done, but would double the entire contents of that header (which is why I went for templates in the original version). There's no operator overloads for add/sub/mul/div either, but that's pretty minor.

icastano.bsky.social•17 days ago

Why not use a define for the float type so that you can at least customize it to your preference?

mjp123.bsky.social•17 days ago

So I *could* do that but IMO it’s not really useful to just globally use fp16 for SH everywhere. In my experience you can have a mix depending on what you’re doing.

ingramb.bsky.social•17 days ago

Could also add another define to mangle the names, and allow the header to be included multiple times. Makes the library code ugly, but usability is decent.

mjp123.bsky.social•17 days ago

I think I would rather dupe the functions and types again rather than make the code ugly…the template version is already too ugly for my tastes. 🙂

nolram.bsky.social•17 days ago

I’m a bit of a beginner when it comes to dealing with SH, so apologies for the question - but is there functionality in there for convoluting/encoding i.e. a cubemap into SH for (ir)radiance?

mjp123.bsky.social•16 days ago

The header has the essential ProjectOntoL1/L2 functions that you need to do that, but doesn’t have a high level function for looping over a cubemap’s texels and projecting all of them onto SH. That might be a good example to add though in the comments.

mjp123.bsky.social•16 days ago

Adding a function to do it probably wouldn’t be super useful
IMO, since if you were projecting a cubemap on the GPU you would probably want to do it in parallel access many threads. Likely 1 thread per texel, followed by N parallel sum/reduction passes. Or possibly use fp atomics if available.

nolram.bsky.social•16 days ago

Oh for sure, thank you for the answer! A sample in the comments would indeed be useful and interesting.

Thank you for this fantastic library and all your hard work!

mjp123.bsky.social•16 days ago

Sounds good, I’ll add one when I get a chance

Comments

Posting Rules

Reply