(j3.2006) Question from a colleague
Robert Corbett
Robert.Corbett
Tue Apr 7 15:36:07 EDT 2009
Van Snyder wrote:
> A colleague asked me whether there is a performance benefit if the
> compiler is informed that it is not necessary for a procedure to be
> thread safe. This actually originated in a discussion of iterators in
> python, CLU and alphard, where the activation record has to be preserved
> between suspension and resumption, as for a coroutine. My colleague did
> ask, however, whether our code would run faster if we could tell the
> compiler it didn't need to be thread safe. We use 5 million cpu hours
> per year, so a little improvement makes a big difference.
At least with Sun Fortran, thread-safety is not an issue for the
compiler. The run-time libraries are where thread-safety is an
issue. The main areas affected by thread-safety are storage
allocation and deallocation and I/O. I have seen programs that
are I/O intensive spend a surprisingly large percentage of their
time manipulating locks.
Bob Corbett
More information about the J3
mailing list