Intel® Math Kernel Library

cannot find -lblas

I'am using Ubuntu 9.04 i386. I installed Intel Fortran Compiler. I know that this issue has been disussed yet but all workarounds didn't help me. When ifort-compiling my program I always get one of the following messages:
ld: cannot find -lblas
ld: cannot find -llapack
Going to
/opt/intel/Compiler/11.0/083/mkl/interfaces/blas95/ and doing a:
sudo make PLAT=lnx32 lib I get:
mkdir -p obj
make -f makefile objects
make[1]: Betrete Verzeichnis '/opt/intel/Compiler/11.0/083/mkl/interfaces/blas95'

Memory Management with MKL

Hi there,

I have a serious problem with memory management of MKL. And it is essential for me to overcome to this problem.

In my Fortran language program (in Linux), I have one matrix with dimension of A(TRINATOM,TRINATOM) (TRINATOM=23826); that after running of the program, I have this message:

Fortcom: ERROR: myprog.for, Line 17: A common block or variable may not exceed 2147483647 bytes.

So I try to use memory management with MKL. First of all I introduced the matrix 'A' by POINTER instead of DIMENSION as bellow:

Over- and underflows in Intel ODE solver

I am currently considering using the library to solve the ODEs that arise in conjunction with modelchecking of markov chains. The ODEs that describe these systems can be arbitrarily stiff yet maximum precision is crucial.
I would like to know how the library handles internal overflows. Can the solver deal with such situations and correct the problem or is there some kind of flag which indicates precision related problems so that I would be able to judge the credibility of the results.

Thanks in advance
Falko Dulat

MKL support for Windows CE


I know this question has been asked before, but is there any update (or plans) for MKL supporting Windows Embedded CE? We were evaluating several third party math libraries that we could potentially use in our Windows CE solution (We do use x86 hardware - Vortex86DX) but most of them rely on MKL. This could create huge opportunity for Intel - based embedded or mobile systems that currently cannot use MKL.


Direct Sparse Solver for Time Dependent Problems


I have a time-dependent problem that involves solving N quasi 9-diagonal matrices (top two rows full, with 9 nonzero diagonals) at each time step. I am currently using the direct sparse solver and I was wondering what the best method for calling DSS might be. I am storing all the nonzero elements for all N matrices in the same vector, then looping through the vector:

DO time=1,endtime
DO j=1,L

error = dss_factor_real(handle,MKL_DSS_DEFAULTS,values(((j-1)*(11*N-20)+1):(j*(11*N-20))) )
error = dss_solve_real(handle,MKL_DSS_DEFAULTS,rhs,nRhs,A(:,j))

Undefined reference

We are running into some undefined references errors when compiling:
mpif90 -o pxa-lnxem64.x sizes.o COMS.o CALL_FORWARD_C.o chord.o dersig.o dertau.o drdnor.o evecl.o evec2.o intsr1.o INVERSE_C.o IO_COEF.o lancos3e.o matr.o max1d2.o mist.o mpi_parameter.o multsend.o pwk.o REC_SORT.o scvd.o sig1.o STRING.o zolgrid.o ITER_GL1CAR_C.o ITER_GL1CAR_C_IN_FR.o ITER_GL1CAR_MC.o ITER_GL1CAR_M.o MAIN.o max2_5ce1.o -lmkl_solver -lmkl_lapack -lmkl_intel_thread -lmkl_em64t -lguide -lpthread -lmpi

dgbsv() and "double* ab" argument, banded matrix format in C++

Hello, I am now playing with the banded matrix form for the Ax=B system anddgbsv() driver function.
I have a 5 band matrix ( 2 upper + diagonal + 2lower )

I coulnd't find in the MKL manual the format expected, in C/C++, for banded matrices,
I couldn't even find the signature explicitely in the manual, but i found it in ...
the double* ab points to the banded matrix.

double ab[] = {
0., 0., 5., 6. , 7. , 8. , 9. , 10., 11., // most upper diagonal
0., 5., 6., 7. , 8. , 9. , 10., 11., 12., // upper diagonal

Non-determinism in calling Pardiso from C#


When I hooked up my C# app to MKL version 9 (.3 maybe) I had problems getting deterministic results from Pardiso. The answers were very close but different each time which is a bit of a pain in some situations. I fixed it by aligning my C# double arrays to 16 byte boundaries and that was fine. I've just upgraded to 10.1 and used the samples provided in this forum to expose the MKL functions I want in a Dll and call them in the recommended way but I can't make it become deterministic again, even when aligning to 16 byte boundaries. Is this expected?

2D DFTF using real_2d_ccs_double_ex2.f90


I want to use the "real_2d_ccs_double_ex2.f90" example, suitably modified to evaluate the FFT
of a 100x100 matrix. I have modified the N_MAX and M_MAX values to 100 and also read the 100X100
matrix into the X_IN_2D matrix. I do not modify any other value and allow the example to run.

The problem i face is that the results for the forward and backward FFT which are seen in "real_2d_ccs_double_ex2.res"
show only 200 elements and writen in a 100x3 (=300) format. I am not able to change the format into a more suitable one.

Intel® Math Kernel Library abonnieren