Cross sections are zero if compiled with nagfor -O2

[sschmidt]

W2 only gives zeros as cross sections if compiled with nagfor and FCLAGS="-O2"

| Integrating process 'eedd':
|=============================================================================|
| It      Calls  Integral[fb]  Error[fb]   Err[%]    Acc  Eff[%]   Chi2 N[It] |
|=============================================================================|
|=============================================================================|
   1          0  0.0000000E+00  0.00E+00    0.00    0.00    0.00
|=============================================================================|
n_events =  1
?rebuild_events = true
$sample = "shower"
| Initializing simulation for processes eedd:
| Simulation mode = unweighted, event_normalization = '1', unpolarized_events
******************************************************************************
*** ERROR: Simulation: sum of process integrals must be positive; skipping
******************************************************************************

I'm using nagfor 5.2(724) on 32bit Linux.

The problem first appeared somewhere between r2465 and r2475, right now trying to spot the guilty one.

This probably is a compiler bug, but should be mentioned somwhere.

[sschmidt]

As a small update: r2465 works normally, r2469 gives only zeros, the revisions in between don't compile.

[Juergen Reuter]

This is critical. WK, this happened (if SS is correct) through your splitting of command/integration/simulation modules.

[kilian]

was unrelated to optimization: in integrate_init_iterations a flag was not initialized properly, just happened to be .false. in the -O2 case [2531].