The Intel® Math Kernel Library (Intel® MKL) LAPACK domain contains a huge variety of routines. Now, a new tool is provided with a faster method of finding appropriate LAPACK functions in Intel® Math Kernel Library Developer Reference document. This tool would be very useful for Intel® MKL newbies and for users not familiar with LAPACK function naming conventions. By using this tool, users can specify functionality as parameters in drop down lists, descriptions of all functions satisfying the requirements will be shown through this tool. 

Please follow this link:
Intel® Math Kernel Library LAPACK Function Finding Advisor

I am installing a third party application using FFTw3 on a HPC cluster. I would like to try out the FFTW interface that MKL provides, to see if I get some performance benefits.

Anyhow I am currently stuck with the compilation that fails with a missing symbol at linking time:

../channelflow/libchflow.so: undefined reference to `fftw_mpi_execute_r2r'

I generated the wrapper library following the instructions here:

• http://software.intel.com/en-us/mkl-developer-reference-c-building-your-own-wrapper-library

with the command:

make libintel64 compiler=intel mpi=intelmpi

and looking at the symbols inside it I can find `fftw_mpi_execute_dft_r2r` but not `fftw_mpi_execute_r2r`. The documentation of FFTW:

•  http://www.fftw.org/fftw3_doc/Using-MPI-Plans.html

seems to have an inconsistent naming for the APIs, so I was wondering if this is a "bug" in `mkl/interfaces/fftw3x_cdft/wrappers/execute.c` where line 65 should be changed from:

FFTW_EXTERN void FFTW_MPI_MANGLE(execute_dft_r2r)

to:

FFTW_EXTERN void FFTW_MPI_MANGLE(execute_r2r)

I'll be waiting for your reply, thanks. Let me know if you need further details.

Dear Intel MKL User,

I try to see into a matter of MKL for solving systems of second order with symmetrical and unsymmetrical matrixes. Therefore I tried to analyze the examples and got stuck with part of code fgmres_full_funct with preconditioner calling. I didn't understand the code below (especially for tmp(ipar(23)+0,1,..4) assignment.

I wonder if anyone could explain it:

! If RCI_REQUEST=3, then apply the preconditioner on the vector
! TMP(IPAR(22)) and put the result in vector TMP(IPAR(23))
!---------------------------------------------------------------------------
      IF (RCI_REQUEST.EQ.3) THEN
          IF (IPAR(4).EQ.3) THEN
              TMP(IPAR(23)+0)=-2.0D0
              TMP(IPAR(23)+1)= 0.08519601586107672D0
              TMP(IPAR(23)+2)=-1.1590871369607090D0
              TMP(IPAR(23)+3)=-0.65791939687456980D0
              TMP(IPAR(23)+4)= 0.75660051476696133D0
          ELSE
              IF(IPAR(4).EQ.4) THEN
                  TMP(IPAR(23)+0)= 0.0D0
                  TMP(IPAR(23)+1)= 0.0D0
                  TMP(IPAR(23)+2)= 0.0D0
                  TMP(IPAR(23)+3)= 1.0D0
                  TMP(IPAR(23)+4)=-1.0D0
              ELSE
              DO I=0,N-1
                  TMP(IPAR(23)+I)=I*TMP(IPAR(22)+I)
              ENDDO
              ENDIF
          ENDIF

          GOTO 1
      ENDIF

Thanks

Ivan

I am working with the Spack package manager which builds and installs science software and development tools from source. Spack is /not/ a GUI application and I use it from scripts in a "headless" automated build farm with OS X, *BSD, and Linux hosts. In the ancient past, there was the ability to do silent / scriptable installs under OS X for Intel software tools (including MKL), but since 16.* this ability seems to have been removed:

myhost:MacOS $ sudo ./install.sh --cli-mode --silent /tmp/silent.cfg
./install.sh: line 639: /Volumes/m_mkl_2017.3.181/m_mkl_2017.3.181.app/Contents/MacOS/./../MacOS/install: No such file or directory

Attempts at running install_gui from the terminal with the same option set is semi-successful. Unfortunately, it appears that PSET_INSTALL_DIR is partially ignored as files are installed into / rather than /opt/intel. The end of the install.log reads:

1504830718:672 : user_is_root: user is root
1504830718:672 : base_user_dir: result is /opt/intel/
1504830718:672 : save_silent_config: save params ACCEPT_EULA:CONTINUE_WITH_OPTIONAL_ERROR:PSET_INSTALL_DIR:CONTINUE_WITH_INSTALLDIR_OVERWRITE:COMPONENTS:PSET_MODE:NONRPM_DB_DIR:XCODE_DIR:XCODE_INTEGRATION_NEEDED in file /opt/intel/ism/db/history/mkl.full_suite_family_2017
1504830718:672 : save_silent_config: get value of the ACCEPT_EULA
1504830718:672 : save_silent_config: get value of the CONTINUE_WITH_OPTIONAL_ERROR
1504830718:672 : save_silent_config: get value of the PSET_INSTALL_DIR
1504830718:672 : getPsetCoreSubdir: started
1504830718:672 : getPsetCoreSubdir: suite core subdir is /parallel_studio_xe_2017.4.049
1504830718:672 : save_silent_config: get value of the CONTINUE_WITH_INSTALLDIR_OVERWRITE
1504830718:672 : save_silent_config: get value of the COMPONENTS
1504830718:672 : save_silent_config: get value of the PSET_MODE
1504830718:672 : save_silent_config: get value of the XCODE_DIR
1504830718:672 : save_silent_config: get value of the XCODE_INTEGRATION_NEEDED
1504830718:672 : save_silent_config: finished
1504830718:672 : mkdtemp: result /li_plugin_c.YwHsPb
1504830718:672 : env_set_string_internal: key "PWD" with value "/pset_user_stat" (length 15)
1504830718:672 : plugin_run: library is "/pset_user_stat/user_stat", func is "send_user_stat"
1504830718 - ERROR : plugin_run: dlopen(/pset_user_stat/user_stat, 9): image not found
1504830718:672 : main: error during running of pset_user_stat layer
1504830718 - WARN : main: temporary folder is not defined
1504830718:672 : main: removing marker file /tmp/foo/intel.pset.root.running.marker
1504830718:673 : dump_debug_info: , , ,

1504830718:673 : Exit code is 0
1504830718:673 : Logging finished at Thu Sep  7 19:31:58 2017

It would be very useful to have a method to reliably install MKL from the command-line in an automated fashion under OS X or a scriptable work-around in the meantime. For bonus, it would be nice to be able to do so without root privileges.

Hello all,

I have a question regaarding the Compiler Option /4I8:

I am dealing with very very larg Matrix and I have to use Cluster_sparse_solver_64 (some components of the ia exceeds the 2^31-1). So I must use /4I8 when I compile the program (using mpiifort).

When I use /4I8, I recieve a warning that

parallel01.f90(77): warning #6075: The data type of the actual argument does not match the definition.   [IERR]
        CALL MPI_INIT( ierr )
-----------------------^
parallel01.f90(79): warning #6075: The data type of the actual argument does not match the definition.   [MKL_COMM]
        CALL MPI_COMM_RANK( MKL_COMM, myid, ierr )
----------------------------^
parallel01.f90(79): warning #6075: The data type of the actual argument does not match the definition.   [MYID]
        CALL MPI_COMM_RANK( MKL_COMM, myid, ierr )
--------------------------------------^
parallel01.f90(79): warning #6075: The data type of the actual argument does not match the definition.   [IERR]
        CALL MPI_COMM_RANK( MKL_COMM, myid, ierr )
--------------------------------------------^

and when I Change the data type of myid, ierr and mkl_comm from integer(4) to integer(8), i get another error that:

parallel01.f90(283): error #6285: There is no matching specific subroutine for this generic subroutine call.   [CLUSTER_SPARSE_SOLVER_64]
        CALL CLUSTER_SPARSE_SOLVER_64 (CPT,CMAXFCT, CMNUM, CMTYPE, CPHASE,CN,M_SPARSE,cum_row_index,col_index8,CPERM,CNRHS,CIPARM,CMSGLVL,RHS,X,MKL_COMM,CERROR)
-------------^

Also, when I do not use /4I8, every Thing is fine, but when running the program I get a runtime error.

Can somebody help me in this regard or give me an example for using Cluster_sparse_solver_64 (this is not provided i Intel document).

Stay cool

Mehdi

I have tried to compile TensorFlow with the full MKL but this failed. I logged an issue on the TensorFlow GitHub but since the MKL integration is provided by Intel they are unable to help there. Could you please look at the issue: https://github.com/tensorflow/tensorflow/issues/12849

Steps to reproduce:

git clone https://github.com/tensorflow/tensorflow.git test
cd test
git checkout r1.3
yes "" | TF_NEED_CUDA=0 TF_NEED_MKL=1 TF_DOWNLOAD_MKL=0 MKL_INSTALL_PATH=<path>/l_mkl_2017.3.196/inst/mkl ./configure
bazel build --config=mkl -c opt --verbose_failures //tensorflow/tools/pip_package:build_pip_package

OS version: Ubuntu Linux 14.04
Bazel version: 0.5.3

Error message:

ERROR: missing input file '//third_party/mkl:libmklml_intel.so'
ERROR: <path>/tensorflow/test/third_party/mkl/BUILD:16:1: //third_party/mkl:intel_binary_blob: missing input file '//third_party/mkl:libmklml_intel.so'
Target //tensorflow/tools/pip_package:build_pip_package failed to build
ERROR: <path>/tensorflow/test/third_party/mkl/BUILD:16:1 1 input file(s) do not exist

The configure script is creating symlinks in third_party/mkl/ for libmkl_rt.so (see here), which is fine, but not for libmklml_intel.so (see here), which doesn't exist in the full MKL distribution. However third_party/mkl/BUILD references libmklml_intel.so. Is this a bug or is use of the full MKL library not supported in TensorFlow 1.3?

Dear all,

I have issues when using Intel MKL to provide Blas+Lapack+ScaLapack. Specifically, i use it with OpenMPI+GCC and link against libmkl_scalapack_lp64, libmkl_blacs_openmpi_lp64, libmkl_intel_lp64, libmkl_core, libmkl_sequential. The issues is with calling pdpotri after Cholesky factorization:

pdpotri_ (&uplo,&n_columns, A_loc, &submatrix_row, &submatrix_column, descriptor,&info);

where

void pdpotri_(const char *UPLO,
                const int *N,
                double *A, const int *IA, const int *JA, const int *DESCA,
                int *INFO);

I don't have any issues with my code on both Ubuntu and macOS when using Netlib-Scalapack 2.0.2 + Openblas 0.2.20. For those cases, calculation of Cholesky factorization followed by pdpotri produce correct results and agree with serial Lapack. Also note that other small test programs I have (i.e. calculate L1 norm or do Cholesky factorization) run ok with Intel-MKL. This suggest that something is wrong in Intel-MKL implementation of pdpotri.

A bit more on the issue itself: this happens when run with 4 MPI cores (2x2 grid) and a small test program; the program runs ok with matrix 64x64 with 32 blocks but fails for 120x120 with 32 blocks. Specifically, i see a floating point exception from process rank 2. Not sure I can debug it further on my side.

p.s. That's Intel-MKL 2017.3.196.

Regards, Denis

Intel® Math Kernel Library (Intel® MKL) is a highly optimized, extensively threaded, and thread-safe library of mathematical functions for engineering, scientific, and financial applications that require maximum performance.

Intel MKL 2018 packages are now ready for download. Intel MKL is available as part of the Intel® Parallel Studio XE and Intel® System Studio. Please visit the Intel® Math Kernel Library Product Page.

Please see What's new in Intel MKL 2018 follow this link - https://software.intel.com/en-us/articles/intel-math-kernel-library-release-notes-and-new-features

Dear all,

I am a graduate student in China. Recently I am using the Pardiso solver for solving linear equations of FEM (Finite Element Method). When I increased the size of the coefficient matrix, Pardiso prompted that the memory was not enough (error -2), but I noticed the only 47% of the memory was occupied.  I don't know why and decided to use OOC model to overcome this problem. But after a few days of trying, I still don't know how to use OOC. So could you please help me with this problem?

I set iparm(60)=2 and build a "pardiso_ooc.cfg" file in the working directory whose content is,

MKL_PARDISO_OOC_PATH=./tmp

MKL_PARDISO_OOC_MAX_CORE_SIZE=60000

MKL_PARDISO_OOC_MAX_SWAP_SIZE=100000

MKL_PARDISO_OOC_KEEP_FILE=0

I'm not sure if my operation is correct. After this operation, when I use Pardiso to solve a small-size problem (which can also be solved by in-core model), 6 new files (tmp.ind, tmp.jal, tmp.lnz, tmp.lup, tmp.sin, tmp.sle) are produced automatically in the working dictionary. However, if applied to a big-size problem, Pardiso cannot work again with error -2 and no new files are produced.

The computing platform I use has 12 processors (Intel(R) Xeon(R) CPU E5-2620 v3 @ 2.4GHz) and memory of 63.0 GB, and the OS is  Linux (kernel version) 2.6.32-358.el.x86_64. The compiler is composer_xe_2013.3.163 and the ifort version is 13.1.1. The compiler is installed by seniors (the MKL is automatically installed), and I don't know how to get the version of MKL. The command I use is ''source /etc/profile" followed by "ifort xx.f -mcmodel=medium -mkl".

I am a green hand in the Linux OS and troubled by this problem for a long time. I will be appreciate if some of you can help me with it or give some advice!

Thanks for your attention!

Xudong Li

If I could do this, I would be able to avoid conversion between layers. I have done some experiment, if no conversion, my application will be faster 10%

hi there, 

We are using MKL on a RedHat Linux Network Server with Xeon Processor which has 32 Physical Core (64 Logical Core). The Application uses a thread pool to handle network requests in parallel. Each request is handled independently. The performance improves with more threads:

4 threads   : 45 seconds 
8 threads   : 23 seconds
16 threads : 15 seconds
24 threads : 14 seconds
32 threads : 15 seconds

However, the performance always caps at 16 threads, and drops a little bit with 32 threads. I replace the mkl cblas_sgemm function with atlas, then the performance keeps improving from 1 thread to 32 threads linearly. 

And limit the mkl thread count by calling mkl_set_num_threads(1) at the beginning of main function or set environment variable to 1, also doesn't work and get the same result. The multiprocess solution also have the same problem(??). Another experiment which sleeps a small amount of time before calling mkl cblas_sgemm shows linear but not ideal result. It looks like there are some resource contention inside the MKL cblas_sgemm implementation? Or do we miss anything here?

Any comment or suggestion is highly appreciated! And thanks much in advance!

Thanks,

Yu

p.p1 {margin: 0.0px 0.0px 0.0px 0.0px; font: 12.0px 'Helvetica Neue'; color: #454545}
span.s1 {font: 12.0px '.PingFang SC'}

Hi,

I try to run mkl scalapack in fortran code on an infinity band network using I_MPI_FABRICS=shm:dapl, however mkl scalapack does not work correctly when running on several nodes, e.g.  pzheev exits with error code 16. When switching to shm:tcp network fabrics it works. It also works with netlib scalapack reference implementation + mkl lapack/blas. I tried intel-2016 update 4 and intel-2017 update 4, both give the same errors. 

Any idea on this error ?

best,

marius

I was wondering if a SkyLake (AVX512) version of HPCG is already available with MKL. 

Can one build an optimized version of HPCG relying solely on optimized sparse vector matrix functions available with MKL?

Thanks...

Michael

Hello,

I understand that the MKL library can be linked with objects compiled via gcc, g++ and gfortran. In terms of performance and stability though are there any side effects when we use MKL with GNU vs Intel compilers? Should we expect any performance impact to codes not using Intel compilers?

Thanks!

Michael

Hi,

It's the first time using the MKL.

I wonder is there a example code to show how to use MKL to do 2D FFT on image?

Does MKL support pthread?

Thanks

Any hint will be appreciated!

hi,

The intel manual says iparm[2] can only be 0, 2 or 3. What happens if this has been set to 1? (It does not throw any error)

Thanks

Dinesh

I installed Fortran Intel Parallel Studio XE 2018. I am trying to compile a fortran program on my macbook using and but I am getting undefined symbols for blacs package. I read that blacs are already installed as part of the Intel Math Kernel library. Can someone help me with some pointers?

source /opt/intel/bin/compilervars.sh -arch intel64 -platform mac

user@system:~/Downloads/test$ ifort test.f 
Undefined symbols for architecture x86_64:
  "_blacs_barrier_", referenced from:
      _solve_ in ifortB6fnER.o
  "_blacs_exit_", referenced from:
      _MAIN__ in ifortB6fnER.o
  "_blacs_get_", referenced from:
      _MAIN__ in ifortB6fnER.o
  "_blacs_gridexit_", referenced from:
      _MAIN__ in ifortB6fnER.o
      _solve_ in ifortB6fnER.o
      _factr_ in ifortB6fnER.o
      _factrs_ in ifortB6fnER.o
      _facgf_ in ifortB6fnER.o
  "_blacs_gridinfo_", referenced from:
      _MAIN__ in ifortB6fnER.o
      _solve_ in ifortB6fnER.o
      _detarray_ in ifortB6fnER.o
  "_blacs_gridinit_", referenced from:
      _MAIN__ in ifortB6fnER.o
  "_blacs_pinfo_", referenced from:
      _MAIN__ in ifortB6fnER.o
  "_descinit_", referenced from:
      _solve_ in ifortB6fnER.o
      _factr_ in ifortB6fnER.o
      _factrs_ in ifortB6fnER.o
      _facgf_ in ifortB6fnER.o
  "_numroc_", referenced from:
      _MAIN__ in ifortB6fnER.o
      _netwk_ in ifortB6fnER.o
      _solve_ in ifortB6fnER.o
      _factrs_ in ifortB6fnER.o
  "_pzgetrf_", referenced from:
      _factr_ in ifortB6fnER.o
      _factrs_ in ifortB6fnER.o
      _facgf_ in ifortB6fnER.o
  "_pzgetrs_", referenced from:
      _solve_ in ifortB6fnER.o
  "_zgebr2d_", referenced from:
      _solve_ in ifortB6fnER.o
      _detarray_ in ifortB6fnER.o
  "_zgebs2d_", referenced from:
      _solve_ in ifortB6fnER.o
      _detarray_ in ifortB6fnER.o
ld: symbol(s) not found for architecture x86_64

Why do not you remove the , in

enum {
    MKL_BLACS_CUSTOM = 0,
    MKL_BLACS_MSMPI = 1,
    MKL_BLACS_INTELMPI = 2,
    MKL_BLACS_MPICH2 = 3,
    MKL_BLACS_LASTMPI = 4,
};

so we can get rid of

C:\Program Files (X86)\IntelSWTools\compilers_and_libraries_2018.0.124\windows\mkl\include\mkl_service.h(279): warning #271: trailing comma is nonstandard

      MKL_BLACS_LASTMPI = 4,

I have tried to compile TensorFlow with the full MKL but this failed. I logged an issue on the TensorFlow GitHub but since the MKL integration is provided by Intel they are unable to help there. Could you please look at the issue: https://github.com/tensorflow/tensorflow/issues/12849

Steps to reproduce:

git clone https://github.com/tensorflow/tensorflow.git test
cd test
git checkout r1.3
yes "" | TF_NEED_CUDA=0 TF_NEED_MKL=1 TF_DOWNLOAD_MKL=0 MKL_INSTALL_PATH=<path>/l_mkl_2017.3.196/inst/mkl ./configure
bazel build --config=mkl -c opt --verbose_failures //tensorflow/tools/pip_package:build_pip_package

OS version: Ubuntu Linux 14.04
Bazel version: 0.5.3

Error message:

ERROR: missing input file '//third_party/mkl:libmklml_intel.so'
ERROR: <path>/tensorflow/test/third_party/mkl/BUILD:16:1: //third_party/mkl:intel_binary_blob: missing input file '//third_party/mkl:libmklml_intel.so'
Target //tensorflow/tools/pip_package:build_pip_package failed to build
ERROR: <path>/tensorflow/test/third_party/mkl/BUILD:16:1 1 input file(s) do not exist

The configure script is creating symlinks in third_party/mkl/ for libmkl_rt.so (see here), which is fine, but not for libmklml_intel.so (see here), which doesn't exist in the full MKL distribution. However third_party/mkl/BUILD references libmklml_intel.so. Is this a bug or is use of the full MKL library not supported in TensorFlow 1.3?

Hi,

I am not a expert of Makefile and need help.

I am trying to use MKL FFTW3.x to do image 2dFFT.

1. I followed the following instruction to build a FFTW3 wrapper:

https://software.intel.com/en-us/mkl-developer-reference-c-building-your...

cd interfaces/fftw3xc
make libintel64 compiler=gnu INSTALL_DIR=/my/path

2. In my code I add the header files according to the https://software.intel.com/en-us/mkl-developer-reference-c-building-an-a....

#include "fftw3.h"

#include "fftw3_mkl.h"
#include "mkl_service.h"

3, I give the path for the header files and the library in the Makefile. 

Below is the content:

////////////////////////////////////////////////////////////////////////////////////////////////////////

fti2 := $(shell echo ${HOME})/bin/MKLfftw3/include
fftl := $(shell echo ${HOME})/bin/MKLfftw3
all:
        g++ `Magick++-config --cxxflags --cppflags` -O2 -o main main.cpp \
        `Magick++-config --ldflags --libs` -lm -lfftw3xc_gnu -I${ffti2} -L${fftl} -lfftw3xc_gnu  -g -Wall

run:
        ./main ../inputs/

clean:
        rm -f -r main

////////////////////////////////////////////////////////////////////////////////////////////////////////

Below is the result of make all.

I found the "'MKL_free" is in the  "mkl_service.h" file.

Though I add its path in the makefile, compiler still cannot find "MKL_free".

Did I miss something?

Any hint will be appreciated!

Below is the result of make all.

////////////////////////////////////////////////////////////////////////////////////////////////////////

make all

g++ `Magick++-config --cxxflags --cppflags` -O2 -o main main.cpp \
        `Magick++-config --ldflags --libs` -lm -lfftw3xc_gnu -I/home/bin/MKLfftw3/include  -L/home/bin/MKLfftw3 -lfftw3xc_gnu  -g -Wall
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_free.o): In function `fftwf_free':
fftwf_free.c:(.text+0x14): undefined reference to `MKL_free'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_malloc.o): In function `fftwf_malloc':
fftwf_malloc.c:(.text+0x1c): undefined reference to `MKL_malloc'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_plan_guru64_dft.o): In function `fftwf_plan_guru64_dft':
fftwf_plan_guru64_dft.c:(.text+0x288): undefined reference to `DftiCreateDescriptor_s_1d'
fftwf_plan_guru64_dft.c:(.text+0x2b3): undefined reference to `DftiCreateDescriptor_s_md'
fftwf_plan_guru64_dft.c:(.text+0x2e5): undefined reference to `DftiCreateDescriptor_s_1d'
fftwf_plan_guru64_dft.c:(.text+0x310): undefined reference to `DftiCreateDescriptor_s_md'
fftwf_plan_guru64_dft.c:(.text+0x335): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x369): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x38e): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x3ba): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x3df): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x582): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x5a7): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x5d0): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x5f5): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x6aa): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x6cf): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x6fb): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x720): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x74c): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x771): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x7a9): undefined reference to `DftiSetValue'
fftwf_plan_guru64_dft.c:(.text+0x7ce): undefined reference to `DftiErrorClass'
fftwf_plan_guru64_dft.c:(.text+0x7e5): undefined reference to `DftiCommitDescriptor'
fftwf_plan_guru64_dft.c:(.text+0x80a): undefined reference to `DftiErrorClass'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_plan_guru64_dft.o): In function `execute_fi':
fftwf_plan_guru64_dft.c:(.text+0x8a9): undefined reference to `DftiComputeForward'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_plan_guru64_dft.o): In function `execute_bi':
fftwf_plan_guru64_dft.c:(.text+0x8d7): undefined reference to `DftiComputeBackward'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_plan_guru64_dft.o): In function `execute_fo':
fftwf_plan_guru64_dft.c:(.text+0x90d): undefined reference to `DftiComputeForward'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftwf_plan_guru64_dft.o): In function `execute_bo':
fftwf_plan_guru64_dft.c:(.text+0x943): undefined reference to `DftiComputeBackward'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftw_version.o): In function `delete_plan':
fftw_version.c:(.text+0x7e): undefined reference to `DftiFreeDescriptor'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftw_free.o): In function `fftw_free':
fftw_free.c:(.text+0x14): undefined reference to `MKL_free'
/home/bin/MKLfftw3/libfftw3xc_gnu.a(fftw_malloc.o): In function `fftw_malloc':
fftw_malloc.c:(.text+0x1c): undefined reference to `MKL_malloc'
collect2: error: ld returned 1 exit status
Makefile:4: recipe for target 'all' failed
make: *** [all] Error 1