The file matmul.cpp / matmul.F90 contains a naive implementation of
matrix-matrix multiplication, which has very inefficient cache usage. Try to
improve the performance by
- Creating a transposed copy of one of the source matrices. This will allow both right hand side matrices in the innermost loop to be traversed contiguously
- Implement cache blocking for the computational loop