LAPACK Archives

[Lapack] NaN propagation in BLAS

And James Demmel writes:
How to do this of course depends on how we decide to treat the cases
alpha=0, beta=0.

Behavior here is fixed by decades of code depending on current behavior.  I 
know I've used beta=0 to multiply into uninitialized memory.  GEMM documents 
that case specifically.  I have a sneaking suspicion people have used alpha=0 
to scale or initialize C (likely some submatrix of a larger matrix).

<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or