TinyBVH has been updated to 1.9.1 on the main branch. Updated: * Fully revised README.md * New BVH optimizer (bvhopt) * Bonsai and BonsaiS builders * Faster builders * 2-wide AMD layout, with hw support * CPU H-PLOC builder * Vulkan vs OpenCL benchmark And more. Link: github.com/jbikker/tiny...

Show Post

New blog post! Go read about making ReSTIR better, killing ReSTIR, improving denoising, and more! jms55.github.io/posts/2026-0...

Show Post

I think most of that stuff also made it in the dx12 example, which directly compares various layouts against native DXR. In that code the experiments take turns so clock speed changes affect all; the comparisons are thus fair.

Show Post

Well actually now that you mention it, it's a lot of small things but together perhaps substantial: * Stack in group shared * Smaller stack * Compressed triangles (Baldwin & Weber) * Better forced fma * Better unpack for 4 chars * Reduced scope for less reg pressure. See traverse_cwbvh.cl.

Show Post

A better BVH. Higher average prim-per-leaf count (ideal is 3, avg is now 2.31 (was 1.8) I believe); and on average more children per interior node. A small boost was obtained by reducing the stack (it was a bit excessive). Oh and the optimizer does a better job for all BVHs now.

Show Post