 |
 |
Van Snyder wrote: >What are your favorite programs to measure the performance of linear >algebra procedures, especially vector and matrix inner product >procedures? I'm interested to compare Fortran intrinsic DOT_PRODUCT >and MATMUL to xDOT and xGEMM, respectively. To answer your first question first :-), self written. I have written a set of routines which I use to evaluate MATMUL vs LINPACK vs self written code. My compiler is now starting to get the better with MATMUL. Just getting round to doing DOT_PRODUCT and even simple vector arithmetic. One reason that I like to use self-written code is that I know I am just looking at the effect of compiled MATMUL, not at included other code which could cloud what I am trying to measure. Doesn't this really depend on your compiler? And whether you have optimised BLAS available? The x... normally rely very heavily on BLAS. As you know, their generality could slug programs which always use a stride of one. My compiler is Compaq Fortran on OpenVMS. I always have the latest version because I beta-test. I would imagine that they use the same (improving) techniques on their UNIX platform and within CVF. Any "slug" I see on any vector notation, I write to my compiler vendor, and very soon see a difference in performance of the next beta of their compiler. Regards, Paddy Paddy O'Brien, Transmission Development, TransGrid, PO Box A1000, Sydney South, NSW 2000, Australia Tel: +61 2 9284-3063 Fax: +61 2 9284-3050 Email: [log in to unmask] Either "\'" or "\s" (to escape the apostrophe) seems to work for most people, but that little whizz-bang apostrophe gives me little spam. %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%