Unexpected access violation

[Roman1]

I am using the latest: Fortran Compiler XE 14.0.3.202

The program I am working on has started unexpectedly crashing when compiled in optimized release mode.  However, when compiled in debug mode, there are no errors or warnings.  I have spent a day debugging, and I was able to isolate the portion of the code where the problem occurs.  This code is very unstable, and doing a trivial change can make the problem go away.

Attached is the source code, project, and output BuildLog files.  When compiled in release mode, the program crashes with:
forrtl: severe (157): Program Exception - access violation

Roman

[Steven_L_Intel1]

I'm looking at this now. Seems that the optimizer has moved some code that references an omitted argument outside the if(present) test for that argument. Not nice.

[Steven_L_Intel1]

Escalated as issue DPD200359189. /Qopenmp is required to trigger the error. I will let you know here of any progress on the bug.

[Roman1]

Thanks for looking at this.  Initially, I was concerned that there was a bug in my fortran code.  Until the compiler is fixed, do you know of a safe way of getting around this problem?  I don't want to disable /Qopenmp since there is parallelization in some of the subroutines, and I am using pardiso from MKL.  While debugging, I noticed that if in fwd_model.f90, I declare the 'is' variable with a SAVE attribute, it doesn't crash.

integer,save:: is

Do you think this would be safe for now, or would it move the problem somewhere else?

Roman

[Steven_L_Intel1]

I found that all sorts of small modifications made the error go away. A simple one would be to reduce the optimization level on fwd_model.f90 without changing the code. I would be hesitant about adding SAVE declarations on code that is executed in parallel regions.

[John_Campbell]

I looked at the attachment in Quote #1 and found:

beta_range.f90    call fwd_model( )

fwd_model.f90  subroutine fwd_model(dpred)

Although routine fwd_model is contained, isn't there an argument mismatch ?
I would have expected this to be wrong in Fortran 95.

John

[mecej4]

John, please look at the code again. In subroutine FWD_MODEL, the argument DPRED is declared OPTIONAL.