COO intersection primitives : fusing value selection with value intersection. (#86269)
As per title. This one fuses 3 kernels into 1 with about 20-10% performance improvement.
This kernel is also useful for union-like operations.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/86269
Approved by: https://github.com/amjames, https://github.com/cpuhrsch