Binary comparison

[rengolin]

We have a copy of fpcmp from LLVM's test-suite, which works fine when the outputs are expected to be equal or close enough, but when we have offset issues in code generation, it may be tricky to know the difference between FP ordering and reading from the wrong offset.

LIBXSMM has a matrix comparison function that can be used in tpp-mlir.

First, we need to change libxsmm to move that method to include/libxsmm_math.h (branch), then we make use of it in a bincmp binary in tpp-mlir (branch).

[rengolin]

Function to copy:
https://github.com/libxsmm/libxsmm/blob/main/samples/xgemm/gemm_kernel.c#L2090

How to call:
https://github.com/libxsmm/libxsmm/blob/main/samples/xgemm/gemm_kernel.c#L3703

When to exit 1
https://github.com/libxsmm/libxsmm/blob/main/samples/xgemm/gemm_kernel.c#L3851