Real*16 problems after recent Mac OSX update

[Phil_Livermore]

The following code inverts a simple matrix in real*16 precision using a modified version of the standard LAPACK routines. This used to work on my Mac using ifort 11.1 quite happily. However, following a recent core system update (10.6.3 and 10.6.4) the code no longer works, producing segmentation fault errors (when compiled with -O0 or -C) which are not tracable.

This code continues to work under a Linux platform.

Any thoughts as to what I can do?

Phil

[Kevin_D_Intel]

I just posted the same to your Intel Premier issue.

The failure you experienced after the OS upgrade relates to a known issue with the linker (ld) that Apple provides with the Xcode 3.2.2 release you are also using. The problematic linker version is: ld64-97.2

We have made a temporary patch to our next 11.1 update (tentatively scheduled for release in August 2010) that avoids the linker issue.

If you wish to continue using your current 11.1.067 compiler with Xcode 3.2.2, then I confirmed you can use the work-around by compiling your program with: -use-asm

The only other alternative to that work-around is to return to the Xcode 3.2.1 release which contains an earlier linker (ld) version that does not have the defect the current linker does.

(Internal tracking id: DPD200157562)

(Resolution Update on 09/24/2010): This defect is fixed in the Intel Fortran Compiler Professional Edition 11.1 Update 7 (11.1.089 Mac OS).

[Kevin_D_Intel]

The 11.1 Update 7 (11.1.089 - Mac OS) containing the mentioned change is now available from the Intel Registration Center: https://registrationcenter.intel.com