Does TensorFlow and PyTorch have any specialized functions for multipling matrices with special properties?
For example, consider the matrix multiplication:
C := AB where A and B are n x n
The cost of a regular matmul is approximately 2n^3 floating point operations (FLOPs). However, if A is triangular, I can perform the same multiplication for half the number of FLOPs in C using an MKL kernel for triangular matrix multiplication - TRMM. Is it possible to do something like this in TensorFlow and PyTorch?