Hi David,
I think you've got a coding error, not a compiler error. Make sure
you're not using any optimizations (i.e. no -01,2 or 3 switches) as
these switches can often "magnify" errors such as the one you're
getting. Obviously it's not a solution to just leave optimization
off, but it might help you track down where the problem is.
I haven't actually compiled up your program, but immediately alarm
bells are ringing when I see things like
a = *(vec + i)
Are you 100% sure this isn't going to wander into unallocated
memory? Why not use std::vector instead of arrays? If you did then
you could call vector.size() and be safe. I think you're getting an
overrun somewhere because of your use of arrays - just check each
block of code for that. What is "a" set to just before ret gets set to zero?
Never put "using namespace std" in header files - it's very bad
practice. I can't find any use of "const" in your code - this
prevents many of the kind of errors you're seeing - apply it to all
variables and functions as appropriate.
You're not using Numerical Recipies for ideas are you? Although this
book is a classic it is full of horrid programming. Take a look at
the C++ book recommendations on www.accu.org for ideas for good books
(unfortunately in the field of numerical computing, there isn't much
to rival Press et. al).
HTH,
David.
At 17:25 31/12/2006, you wrote:
Upon receiving some advice, I have attached here the entire piece of code I'm
trying to run. It's a Newton solver; LSMbdnew is the generic solver, and
bulkfunc contains the function (f) and Jacobian (jac). The whole
thing may look
a little ugly to the experienced programmers on this list -- sorry.
Anyway, this will compile and run, but it won't do what it's supposed to. I
found out with gdb that the reason for this is that the function
LSMbdnew::norm(double *vec1, long num) is not executing the
add/assign properly.
For some reason, it sets the collecting variable _ret_ to zero
after one or two
iterations. Why? I am stumped.
Thanks,
-David
