Fixed-size matrix products should never malloc. It's OK if this means that they never do any cache-friendliness/parallellization. Guaranteeing no malloc is more important.
for the record only fancy matrix-matrix products involving triangular or selfadjoint views are not safe regarding this issue.
progress: triangular*vector and self-adjoint*vector are now fine too.
OK, given that it's only in triangular*matrix and selfadjoint*matrix products, and does not happen in triangular.solve(matrix), I think that we can ship with that. Must fix this in 3.1.
Summary: Bug 51: make general_matrix_matrix_triangular_product use L3-blocking helper so that general symmetric rank-updates and general-matrix-to-triangular products do not trigger dynamic memory allocation for fixed size matrices.
Summary: Bug 51: add block preallocation mechanism to selfadjoint*matrix product.