The current cutlass does not compile for me because of a typo in cutlass (NVIDIA/cutlass#1783):
tiny-cuda-nn/dependencies/cutlass/include/cutlass/matrix.h:14028:3: error: ‘struct cutlass::Matrix<Element_, 4, 4>’ has no member named ‘set_slice3x3’; did you mean ‘set_slice_3x3’? [-Wtemplate-body]
[build] 14028 | m.set_slice3x3({
[build] | ^ ~~~~~~~~~~
[build] | set_slice_3x3
This merge request in cutlass resolves it: NVIDIA/cutlass#1784, latest version works for me