Debug vs. Release question

[dondilworth]

I have a general question: My code gives slightly different answers in debug and release mode. After a lengthy calculation, those differences diverge more and more. What can cause this? There are lots of math calls in this rather large program, and if they come back different in the last digit, that might do it.

Any ideas of what to look for or settings to change? Here are my command lines:

/nologo /debug:full /OD /QaxAVX /Qparallel /Qip /I"Debug/" /extend_source:132 /pad_source /warn:none /debug-parameters:used /Qguide-file:"Debug\\SYNOPSYS200_lib_.gap" /Qsave /align:commons /assume:byterecl /Qzero /Qfp_port /fpconstant /iface:cvf /module:"Debug/" /object:"Debug/" /Fd"Debug\\vc100.pdb" /traceback /check:all /libs:dll /threads /winapp /c


/nologo /debug:parallel /Oy- /Qparallel /I"Release/" /extend_source:132 /Qsave /assume:byterecl /iface:cvf /module:"Release/" /object:"Release/" /Fd"Release\\vc100.pdb" /traceback /libs:static /threads /winapp /c

[dondilworth]

It gets more interesting. Different compiler optimizations give different results. In debug mode, I do a lengthy calculation and get these answers:

No optimization: 0.0809
Optimization max speed: 0.01563
Optimization minimum size: 0.00767

I would expect this to be deterministic, but it looks more like chaos. Any reason why?

[Anonymous66 · Accepted solution]

Without seeing your code, it is impossible to know exactly what is happening, but optimizations canaffectthe results. Optimizations affect how some things are calculated andcan change the order of operations. It sounds like your code is very sensitive to small changes. If you want the same result at all optimization levels, I would suggest adding the options /fp:precise /fp:source.

For more information on the factors that can affect the consistency of floating-point calculations see http://software.intel.com/en-us/articles/consistency-of-floating-point-results-using-the-intel-compiler/