topic problem when use large memory and MKL poissson lib. in Intel® oneAPI Math Kernel Library

problem when use large memory and MKL poissson lib.

Chang_Lei — Fri, 27 Aug 2010 08:22:21 GMT

The system environment is Intel XeonE5450,16GB memory,Redhat Enterprise Linux 5.2,Intel Fortran Compiler 10.1.
I thought I have use the right compile option as follows:

"ifort -mcmodel=medium -shared-intel mkl_dfti.f90 mkl_poisson.f90 ...myprogram.f90 -L/opt/intel/cmkl/9.0/lib/em64t/ -lmkl_em64t -lguide -lpthread -lm "

however, after giving vetorize information,these errors are present:

/opt/intel/cmkl/9.0/lib/em64t//libmkl_em64t.a(_d_commit_helmholtz_3d_fb.o): In function `d_commit_helmholtz_3d_':
_d_commit_helmholtz_3d_iface.c:(.text+0x4a): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_commit_helmholtz_3d_iface.c:(.text+0xcc): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_commit_helmholtz_3d_iface.c:(.text+0xe5): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_commit_helmholtz_3d_iface.c:(.text+0xfe): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_commit_helmholtz_3d_iface.c:(.text+0x10a): relocation truncated to fit: R_X86_64_PC32 against `.bss'
/opt/intel/cmkl/9.0/lib/em64t//libmkl_em64t.a(_d_helmholtz_3d_fb.o): In function `d_helmholtz_3d_':
_d_helmholtz_3d_iface.c:(.text+0x4a): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_helmholtz_3d_iface.c:(.text+0xcc): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_helmholtz_3d_iface.c:(.text+0xe5): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_helmholtz_3d_iface.c:(.text+0xfe): relocation truncated to fit: R_X86_64_PC32 against `.bss'
_d_helmholtz_3d_iface.c:(.text+0x10a): relocation truncated to fit: R_X86_64_PC32 against `.bss'
/opt/intel/cmkl/9.0/lib/em64t//libmkl_em64t.a(_d_init_helmholtz_3d_fb.o): In function `d_init_helmholtz_3d_':
_d_init_helmholtz_3d_iface.c:(.text+0x4a): additional relocation overflows omitted from the output

I know this problem is about MKL, because when not link it I can allocate memory bigger than 2G( even to 16G) with no problem.

Any suggestionis much appreciated.

problem when use large memory and MKL poissson lib.

Gennady_F_Intel — Fri, 27 Aug 2010 09:15:04 GMT

Lei,

Yes, you are compiling by the right way, but the version of MKL 9.0 did'd suppot Integer8 to adress huge data arrays.

Therefore, there no way to resolve the problem you reported.

You can evalaute the latest MKL to work with this problem.

Please see here the link to the latest MKL 10.3 beta program.

--Gennady

problem when use large memory and MKL poissson lib.

Chang_Lei — Mon, 30 Aug 2010 06:28:17 GMT

Thanks.
I note MKL 10.3 is a beta edition.
Can non-commercial edition 10.2 or other edition of MKL support Integer8 adress?
Best regards.

problem when use large memory and MKL poissson lib.

Gennady_F_Intel — Mon, 30 Aug 2010 06:45:28 GMT

yes, 10.3 beta supports exactly the same features wich are supporting in GOLD versions.

problem when use large memory and MKL poissson lib.

Chang_Lei — Fri, 03 Sep 2010 02:04:42 GMT

Hi~

Could you please give anexplanation to GOLD versions?

Can Intel Cluster MKL 10.0.010 support Integer8 address?

With best regards.

problem when use large memory and MKL poissson lib.

Gennady_F_Intel — Fri, 03 Sep 2010 14:23:30 GMT

Yes, Integer8 is supported in MKL v10.0 also. Se ILP64 libraries for that. It should be noted thatstarting with Intel MKL 10.0, we have merged these two versions of MKL ( Intel MKL for SMP systmes and Intel MKL Cluster Edition ) and therefore Scalapack and Cluster FFT functionalities were included into one package.

--Gennady