CoCoA » CoCoALib

I found the problem with an ad hoc program to search for 6x6 matrices with large ratio hadamard/det.
In the main loop I inserted a check for the CPU time consumed. The linux/shell time command indicated that around 20% of resources were "system"; when I modified the code to check every 100 iters, then the "system" time used dropped almost to 0.

I deduce that calling CpuTime is quite costly, so should not be done every iteration inside "cheap loops".

Here is a brief comment about the design: I shall suppose that the class is called CpuTimeLimit

(1) simple use example:

  CpuTimeLimit CheckTimeLimit(3.0); // limits is 3.0 seconds
  for (int i=0; i < 1000000; ++i)
  {
     CheckTimeLimit();  // will throw when limit is exceeded
     ...BODY OF LOOP...
  }

(2) Use as fn arg:

  vector<RingElem> GBasisTimeout(const vector<RingElem>& gens, const CpuTimeLimit& CheckTimeLimit)
  {
    // Call CheckTimeLimit() inside any potentially long loops
  }

  // to call the fn do
  GB = GBasisTimeout(gens, CpuTimeLimit(3.0));

NOTE: if we do this then we could even use a better name for the fn (such as GBasis)