Hi all,
With all your help I was able to install 2.9.4, using Quentin's SRPMS for
ufsparse and glpk and applying a patch to the makefile as detailed in a
previous email to the list by Ismail.
I am testing it against 2.1.72. both versions were configured with the
same
options and use the same blas/atlas/fftw and compilers.
I have a some code which constructs a 201^2x201^2 sparse matrix and
inverts
it. While the inversion speed is comparable for both versions, The part
which constructs the I,J,S vectors which I later turn into the sparse
matrix by s=sparse(I,J,S) takes almost twice as much time in 2.9.4 (12
seconds compared to 6 seconds). This part has a lot of loops, vector
concatenations etc...
Does this sound familiar to anyone, or should I investigate further?