I tested this a tiny bit on play/torch-compile by putting a torch.compiled function in the inner loop of edge calculation. The compiler seemed to function, but the speedup was negligible (4%?). I still think this is plausibly worth investigating further, but I expect it will be more of a "read a bunch of docs, think carefully, and try a bunch of things out" project, not a "magical improvements out of the box" project. Though not bad to check the box once and a while and see if magic is inside.
I tested this a tiny bit on
play/torch-compileby putting a torch.compiled function in the inner loop of edge calculation. The compiler seemed to function, but the speedup was negligible (4%?). I still think this is plausibly worth investigating further, but I expect it will be more of a "read a bunch of docs, think carefully, and try a bunch of things out" project, not a "magical improvements out of the box" project. Though not bad to check the box once and a while and see if magic is inside.