[Wien] array FJ lower bound error with ifort 12

[Peter Blaha]

So it seems Intel did it again!
We had similar problems with very early versions of ifort 12, but intermediate versions were fine.

Either try to get a different version (I'm using composerxe-2011.3.174 and this is fine), or try to
compile the corresponding routine without optimization (-O0).
But experience from early ifort 12 version showed that in several routines errors could occur, simply
because intel does "speculative calculations", ignoring if statements, which should direct the program
to jump out of a loop.

I'm wondering how long it takes for Intel to produce a stable version again.

Am 06.09.2011 16:10, schrieb Gavin Abo:
> Hi Wien2k Users,
>
> Does anyone have a fix for the runtime error below? Is one of the older ifort 12 versions more stable?
>
> Operating System: Red Hat Enterprise Linux 6.0
> Compilers: composer_xe_2011_sp1.6.233 (ifort 12.1.0 + mkl, 30 day evaluation), cc 4.4.4 20100726
> Processor: Dual Quad Core Intel Xeon E5606
> Memory: 6x2 GB 1333 MHz DDR3 ECC SDRAM
> WIEN2k_11.1 (Release 14/6/2011)
>
> Compiles without errors with the following settings.
>
> Compiler options: -FR -mp1 -w -prec_div -pc80 -pad -ip -DINTEL_VML -O1 -xHost -traceback -g -check bounds
> Linker Flags: -L/opt/intel/composer_xe_2011_sp1.6.233/compiler/lib/intel64 -openmp -lpthread
> Preprocessor flag: '-DParallel'
> R_LIB (LAPACK+BLAS): -L/opt/intel/composer_xe_2011_sp1.6.233/mkl/lib/intel64 -lmkl_lapack95_lp64 -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -openmp -lpthread
>
> Shared Memory Architecture: y
> MPI and Scalapack: n
>
> Runtime error with TiC example with 1000 k-points:
>
> cd /home/wien/wiendata/TiC; run_lapw -cc 0.0001 -NI
>
> LAPW0 END
> LAPW1 END
>
> forrtl: severe (408): fort: (3): Subscript #1 of the array FJ has value -2147483648 which is less than the lower bound of 1
>
> Image PC Routine Line Source
> lapw2 00000000005B7A6A Unknown Unknown Unknown
> lapw2 00000000005B6566 Unknown Unknown Unknown
> lapw2 0000000000566B80 Unknown Unknown Unknown
> lapw2 00000000005233EF Unknown Unknown Unknown
> lapw2 0000000000523911 Unknown Unknown Unknown
> lapw2 0000000000507CEC sphbes_ 69 sphbes.f
> lapw2 000000000048E3AD harmon_ 13 harmon.f
> lapw2 00000000004A14FA l2main_ 671 l2main_tmp_.F
> lapw2 00000000004D4845 MAIN__ 564 lapw2_tmp_.F
> lapw2 0000000000403C5C Unknown Unknown Unknown
> libc.so.6 0000003E5321EC5D Unknown Unknown Unknown
> lapw2 0000000000403B59 Unknown Unknown Unknown
>
>  > stop error
>
> Note: For 100 k-points, runs without error and converges with 12 cycles.
>
> Thanks,
>
> Gavin
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-- 

                                       P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------