Edward Grace wrote:
Note that the components above and below the diagonal are the same. In BLAS (or indeed boost::ublas) this would be a matrix of symmetric type (or Hermitian if complex). Ah in this case, we need to look if we can infer the "shape" of the matrix from the AST and then tag the said expression with the proper shape policy. IIRC I have a diag(n) that returns a diagonal matrix with element from the vector n and it was shaped properly. So the engien is here (shaped expression) but we need to be sure at CT that the said matrix has a given property. From my humble experiences, thinking you've finished means you've only just about figured out what the problem actually is! I was half-joking , mind you ;)
-- ___________________________________________ Joel Falcou - Assistant Professor PARALL Team - LRI - Universite Paris Sud XI Tel : (+33)1 69 15 66 35