Recently I had fun reimplementing an old (but still usable!) code for accelerator optics. It involved transfer matrices for a 6D phase space to second order. Most of the FORTRAN77 source code was just pages and pages of hand-differentiated 6x6x6 matrices (with quite non-trivial elements) and the plumbing to painstakingly propagate those jacobians around for fitting... all replaced with a single, magic, call to jax.grad(). Felt like cheating!
I'm also super interested in its application to modelling, e.g. projects like https://github.com/deepmodeling/jax-fem -- particularly for chaining different sorts of simulations and analysis together and getting gradients through the lot. Also quite magic!
> the thing JAX was truly meant for: a graphics renderer
I mean, just like ray-tracing, SDF (ray-marching) is neat, but basically everything useful is expensive or hard to do (collisions, meshes, texturing etc.). I mean mathy stuff is easier (rotations, unions/intersections, function composition, etc.) but 3D is usually used in either modeling software or video games, which care more about the former than they do the latter.
Recently I had fun reimplementing an old (but still usable!) code for accelerator optics. It involved transfer matrices for a 6D phase space to second order. Most of the FORTRAN77 source code was just pages and pages of hand-differentiated 6x6x6 matrices (with quite non-trivial elements) and the plumbing to painstakingly propagate those jacobians around for fitting... all replaced with a single, magic, call to jax.grad(). Felt like cheating!
I'm also super interested in its application to modelling, e.g. projects like https://github.com/deepmodeling/jax-fem -- particularly for chaining different sorts of simulations and analysis together and getting gradients through the lot. Also quite magic!
I like the concept of applying Jax to SDF sphere tracing :)
I mean, just like ray-tracing, SDF (ray-marching) is neat, but basically everything useful is expensive or hard to do (collisions, meshes, texturing etc.). I mean mathy stuff is easier (rotations, unions/intersections, function composition, etc.) but 3D is usually used in either modeling software or video games, which care more about the former than they do the latter.
Also, what's the story for JAX + WebGL when it comes to targeting hardware-accelerated ray tracing?