ICE 11.1.064 and cluster-openmp

[Tim_Gallagher]

Hello again,

I've been having a heck of a time getting the -cluster-openmp option to work with the attached code. The threads on the remote side keep segfaulting, but I'll post the details about that in the correct forum.

I thought I may be hitting stack size issues, so I also compiled with the -heap-arrays option and that's when I came across the ICE. I think it's related to the problem I'm having with the remote threads since the ICE happens while compiling the same file that the segfaults occur.

At any rate, attached is the code setup to generate the ICE. ifort -V is:

Intel Fortran Intel 64 Compiler Professional for applications running on Intel 64, Version 11.1 Build 20091130 Package ID: l_cprof_p_11.1.064
Copyright (C) 1985-2009 Intel Corporation. All rights reserved.
FOR NON-COMMERCIAL USE ONLY

The main code has a defined macro to handle the grid size, so to make this code run:

make GRID_SIZE_DEF="-DDEF_GRID_SIZE=9" wenoAMR

Thanks again,

Tim

Edit: The other post detailing the remote thread segfault is at http://software.intel.com/en-us/forums/showthread.php?t=72507 if anybody has any ideas what the problem may be, or if they are related.

[Steven_L_Intel1]

I can reproduce the problem, but I'll comment that the Cluster OpenMP feature is not supported in 11.1 and is going to go away entirely in a future release. Escalated to development as issue DPD200151938.

If you remove -heap-arrays, it compiles.

[Tim_Gallagher]

Is there a planned replacement for it or is the idea going to be dropped entirely?

It compiles without -heap-arrays but it doesn't work -- the remote threads don't have access to the data they need that's created in the local memory even though the variable is SHARABLE and SHARED which doesn't seem correct. But if it's going to be dropped, I won't spend a lot of time worrying about it.

Thanks,

Tim

[Steven_L_Intel1]

Dropped entirely. Look at using MPI or, eventually, Coarray Fortran.