Please be aware that in future releases (forthcoming Intel® MKL 11.0), we will modify the way we deliver the Intel® MKL Reference Manual in uncompressed help format by delivering it on the web instead of in the product package, and we will no longer deliver man pages or help content for integration with Eclipse* IDE. We are making this change in order to reduce the amount of duplicated content, decrease the size of the product package, and decrease the time it takes to install the product.

Note that the versions of reference manual for integration into Microsoft Visual Studio* 2008 and 2010 will continue to be delivered in the product package.  Additionally, the Intel MKL Reference manual in uncompressed help HTML format and pdf format will be made available on the Intel® MKL web page for download and off-line use.

The Intel MKL web page also include other MKL documentation resources: Intel MKL Release Notes, User's Guide, LAPACK User's Guide, Intel® MKL Link Line Advisor,etc.

Forthcoming Intel® MKL 11.0 (Expected date of release: Autumn 2012)

•  Intel® MKL 11.0 will have backward incompatibility with Intel® MKL 10.2 update 3

•  Open MP static library on Microsoft* Windows

•  GMP* Arithmetic functions

•  Intel® MKL Reference Manual will be removed from product package

•  Intel® Pentium III will no longer be supported

•  Red Hat* EL4 and PGI* Fortran 10.x support will be dropped

•  PGI* Fortran 77 support will be removed

Who is affected: Users who use Convolution/Correlation routines from Intel® MKL

OS: Microsoft* Windows only

Intel® MKL version: Intel® MKL 11.0 (Expected date of release: Autumn 2012).

Problem definition: Intel® MKL 11.0 will have backward incompatibility with Intel® MKL 10.2 update 3

Recommended solution: Recompile your application with MKL version 10.2 update 4 or higher

In Intel® MKL version 10.2 update 4, Convolution/correlation kernels were moved from VSL/VML core dlls to standard Intel® MKL core dll.  We kept old kernels in VSL/VML dlls for backward compatibility between Intel® MKL 10.2 update 3 and Intel® MKL 10.2.4 and higher. We are going to remove old Convolution/Correlation functions from VSL/VML dlls in Intel® MKL 11.0.  As a result you can face backward incompatibility between Intel® MKL 11.0 and Intel® MKL 10.2 update 3 but MKL 11.0 has backward compatibility with Intel® MKL 10.2.4 and higher. We recommend recompile your application with any version of Intel® MKL higher than Intel® MKL 10.2.3. You can find summary table below.

Summary for backward compatibility of Convolution/Correlation component of Intel® MKL

Microsoft* Windows

Intel® MKL 10.2.4 and higher have backward compatibility with Intel® MKL 10.2.3

Intel® MKL 10.3 has backward compatibility with Intel® MKL 10.2.3

Intel® MKL 11.0 has backward compatibility with Intel® MKL 10.2.4 and higher but Intel® MKL 11.0 does not have backward compatibility with Intel® MKL 10.2.3

Linux

Intel® MKL 10.2.4 and higher have backward compatibility with Intel® MKL 10.2.3

Intel® MKL 10.3 has backward compatibility with Intel® MKL 10.2.3

Intel® MKL 11.0 has backward compatibility with Intel® MKL 10.2.3

Mac

Intel® MKL 10.2.4 and higher have backward compatibility with Intel® MKL 10.2.3

Intel® MKL 10.3 has backward compatibility with Intel® MKL 10.2.3

Intel® MKL 11.0 has backward compatibility with Intel® MKL 10.2.3

Please contact us if you have any questions/comments

• a powerful N-dimensional array object
• sophisticated (broadcasting) functions
• tools for integrating C/C++ and Fortran code
• useful linear algebra, Fourier transform, and random number capabilities.

Besides its obvious scientific uses, NumPy can also be used as an efficient multi-dimensional container of generic data.

For more information on NumPy, please visit http://NumPy.scipy.org/

Version Information

This application note was created to help NumPy users to make use of the latest versions of Intel MKL on Linux platforms.

The instructions given in this articles apply to Intel MKL 10.3 and above and Intel Compiler 11.0 and above.

 
Step 2 - Downloading NumPy Source Code

The NumPy source code can be downloaded from:

http://NumPy.scipy.org/

Prerequisites

Intel MKL can be obtained from the following options:

Download a FREE evaluation version of the Intel MKL product.
Download the FREE non-commercial* version of the Intel MKL product.

All of these can be obtained at: Intel® Math Kernel Library product web page.

Intel® MKL is also bundled with the following products

Intel® Parallel Studio XE 2011
Intel Parallel Composer XE 2011
Intel Cluster Studio 2011

Step 3 - Configuration

Use the following commands to extract the NumPy tar files from the downloaded NumPy-x.x.x.tar.gz.

1.	$gunzip numpy-x.x.x.tar.gz 
2.	$tar -xvf numpy-x.x.x.tar 

 
The above will create a directory named numpy-x.x.x

Make sure that C++ and FORTRAN compilers are installed and they are in PATH. Also set LD_LIBRARY_PATH to your compiler (C++ and FORTRAN), and MKL libraries.

Step 4 - Building NumPy

Change directory to numpy-x.x.x
Create a site.cfg from the existing one
 
Edit site.cfg as follows:

Add the following lines to site.cfg in your top level NumPy directory to use Intel® MKL, if you are building on Intel 64 platform:

[mkl]
library_dirs = /opt/intel/composer_xe_2011_sp1.6.233/mkl/lib/intel64
include_dirs = /opt/intel/composer_xe_2011.sp1.6.233/mkl/include
mkl_libs = mkl_rt
lapack_libs =

If you are building NumPy for 32 bit, please add as the following

[mkl]
library_dirs = /opt/intel/composer_xe_2011_sp1.6.233/mkl/lib/ia32
include_dirs = /opt/intel/composer_xe_2011_sp1.6.233/mkl/include
mkl_libs = mkl_rt
lapack_libs = 

Modify cc_exe in numpy/distutils/intelccompiler.py to be something like:

self.cc_exe = 'icc -O3 -g -fPIC -fp-model strict -fomit-frame-pointer -openmp -xhost'

Here we use, -O3, optimizations for speed and enables more aggressive loop transformations such as Fusion, Block-Unroll-and-Jam, and collapsing IF statements, -openmp for OpenMP threading and -xhost option tells the compiler to generate instructions for the highest instruction set available on the compilation host processor. If you are using the ILP64 interface, please add -DMKL_ILP64 compiler flag.

Run icc --help for more information on processor-specific options, and refer Intel Compiler documentation for more details on the various compiler flags.

Compile and install NumPy with the Intel compiler: (on 64-bit platforms replace "intel" with "intelem")

python setup.py config --compiler=intel build_clib --compiler=intel build_ext --compiler=intel install

If you build NumPY for Intel64 bit platforms:

$export LD_LIBRARY_PATH=/opt/intel/composer_xe_2011_sp1.6.233/mkl/lib/intel64:/opt/intel/composer_xe_2011_sp1.6.233/compiler/lib/intel64:$LD_LIBRARY_PATH

If you build NumPY for ia32 bit platforms:

$export LD_LIBRARY_PATH=/opt/intel/composer_xe_2011_sp1.6.233/mkl/lib/ia32:/opt/intel/composer_xe_2011_sp1.6.233/compiler/lib/ia32:$LD_LIBRARY_PATH

It is possible that LD_LIBRARY_PATH causes a problem, if you have installed MKL and Composer XE in other directories than the standard ones. The only solution I've found that always works is to build Python, NumPy and SciPy inside an environment where you've set the LD_RUN_PATH variable, e.g:

$export LD_RUN_PATH=~/opt/lib:~/intel/composer_xe_2011_sp1.6.233/compiler/lib:~/intel/composer_xe_2011_sp1.6.233/mkl/lib/ia32

Note: We recommend users to use arrays with 'C' ordering style which is row-major, which is default than Fortran Style which is column-major, and this is because NumPy uses CBLAS and also to get better performance.

Appendex A: Example:

Please see below an example Python script for matrix multiplication that you can use Numply installed with Intel MKL which has been provided for illustration purpose.

import numpy as np
import time

N = 6000
M = 10000

k_list = [64, 80, 96, 104, 112, 120, 128, 144, 160, 176, 192, 200, 208, 224, 240, 256, 384]

def get_gflops(M, N, K):
	return M*N*(2.0*K-1.0) / 1000**3

np.show_config()

for K in k_list:
	a = np.array(np.random.random((M, N)), dtype=np.double, order='C', copy=False)
	b = np.array(np.random.random((N, K)), dtype=np.double, order='C', copy=False)
	A = np.matrix(a, dtype=np.double, copy=False)
	B = np.matrix(b, dtype=np.double, copy=False)

	C = A*B

	start = time.time()

	C = A*B
	C = A*B
	C = A*B
	C = A*B
	C = A*B

	end = time.time()

	tm = (end-start) / 5.0

	print "{0:4}, {1:9.7}, {2:9.7}".format(K, tm, get_gflops(M, N, K) / tm)

Appendix B: Performance Comparison

Please click Examples.py to download the examples for LU, Cholesky and SVD.



 

Please follow the link, you can find the online documentations and C LAPACK examples.

The release of PARDISO 4.0.0 from the University of Basel is not backward compatible and thus introduces some incompatibilities with the implementation of PARDISO that is available with the Intel® Math Kernel Library. This article outlines those places where the two interfaces have diverged.

The names of routines have the following structure:
vsl[datatype]{Conv|Corr}<base name> for C-interface
The field [datatype] is optional. If present, the symbol specifies the type of the input and
output data and can be
s (for single precision real type),
d (for double precision real type),
c(for single precision complex type), or
z (for double precision complex type).

The current implementation provides:

· Fourier algorithms for one-dimensional single and double precision real and complex data
· Fourier algorithms for multi-dimensional single and double precision real and complex data
· Direct algorithms for one-dimensional single and double precision real and complex data
•  Direct algorithms for multi-dimensional single and double precision real and complex data

Sample Code

    printf("EXAMPLE executing a convolution task\n");
    rank = 1;
    mode = VSL_CONV_MODE_DIRECT;
    vslcConvNewTask(&task,mode,rank,&xshape,&yshape,&zshape);
    status = vslcConvExec(task,x,&xstride,y,&ystride,z,&zstride);

For simple sample c or fortran code, please see
${MKL Install Dir}\examples\vslc\source\vslcconvexec.c
${MKL Install Dir}\examples\vslf\source\vslcconvexec.f

Compatiblity with IBM* ESSL library
One-dimensional algorithms cover the following functions from the IBM* ESSL library:
SCONF, SCORF
SCOND, SCORD
SDCON, SDCOR
DDCON, DDCOR
SDDCON, SDDCOR.
Special wrappers are designed to simulate these ESSL functions. The wrappers are provided
as sample sources for FORTRAN and C. To reuse them, use the following directories:
${MKL install Dir}/examples/vslc/essl/vsl_wrappers
${MKL install Dir}/examples/vslf/essl/vsl_wrappers

Performance issue
One issue is reported in MKL forum Speed of vslsconv in 10.1. The speed of vslsconv for 1000x1000 and 10x10 is far slower than FFT evaluated.

MKL really seems to have problem with choosing the optimal algorithm in this situation. It erroneously favors the overlap-add method and ends up performing a series of small 2D FFTs.

This issue will be fixed in one of our future releases.

About the Author

Naveen Gv is a Technical Consulting Engineer (TCE) in the performance library team. At Intel he has specialized on Multi-core programming, Intel Performance Primitives and Intel Math Kernel Library. His professional interests are teaching Multi core Programming Methodology to software community and implementing Digital Signal Processing algorithms on x86 platform. Naveen has worked with several universities across Asia Pacific to implement Multi-Core programming in academia. His e-mail address is naveen.gv at intel.com

Software Requirement:
1. Intel® Visual FORTRAN Compiler Professional for Windows.  
The compiler was required to be installed and properly integrated to Microsoft VC 2005 or VC 2008.

2. Intel® MKL for windows
You can install the Stand alone MKL separately or the integrated MKL which comes along with Intel Fortran Compiler Professional version. 

3. Microsoft* Visual Studio 2005 or 2008. (MSVC*)
For who'd like to build x64 bit application, please install the MSVC* package which supports X64 development.

Step1 : Which kind of application: ia32, em64t or ia64 application you need to build?
Which target machine you need your application run on?
ia32: 32 bit application. 
EM64t: Intel® 64 bit application
IA64:  Intel® Itanium 64bit application.

Some more information what ia32 and intel64 Intel® architecture mean,  please see this article.

When build ia32 or em64t application on Intel®  XEON machine with windows*, in MSVC* environment, please check the project configuration manager=> active platform.  They should be  "ia32" or "X64" (which is corresponding to em64t).

Step 2 : Which MKL library you need link against?
2.1) please check the on-line KB article Intel® Math Kernel Library Link Line Advisor After choose it, you can get which libraries you need to list. 

For example,
"mkl_intel_c.lib mkl_intel_thread.lib mkl_core.lib libiomp5md.lib" is enough for a ia32 windows application generally.
"mkl_intel_lp64.lib mkl_intel_thread.lib mkl_core.lib libiomp5md.lib" is ok for a X64 windows application

Step 3: How to link MKL in your project manually?
3.1) Please follow the steps in how to use MKL in MSVC to add the include path, library path and mkl library as the article show. 

For example, add include path manually



please take care of about library path,  you need enter right directory:
<MKL install dir>\ia32,  for ia32 application
<MKL install dir>\em64t, for X64 application
<MKL install dir>\ia64, for IA64 application
they can't be mixed. 


Step 4: How to link MKL in your project automatically - One Button Support  (optional)
The recent MKL version have provided one "build" menu in MSVC environment. Click one button, all of environment setting and required libraries are ready in your project.
* This step can replace the step3 in most of cases.  
Or if the button is not show, please try step3, set the paths and library manually.

4.1) See below screenshot in MSVC 2005, click the version you need

Then the required library will be added in your project automatically.


4.2) In Visual Fortran compiler, it is a little different, to integrate MKL, please click the compiler option as below, change the "No" to value like /Qmkl:parallel or /Qmkl:sequential

Please note: some interface libraries (i.e. mkl_lapack95.lib mkl_blas95.lib fftw3xc_ms) are not MKL standard library, so the "one button" integration don't support it.  You may add them manually if you need to use them.

Step 5: About Compaq*Visual Fortran support and porting from 32bit to 64bit (Optional for who have used CVF)
CVF and Intel Fortran compiler use different call convention by default:
CVF: stdcall
Intel Fortran, C Compiler: cdecl (default interface of the Microsoft Visual C* application)

5.1) MKL support both of them on ia32 platform, but if you are porting from CVF to intel Fortran, you may take care of the required library: mkl_intel_s or mkl_intel_c.lib

• Using the CVF compiler
  The CVF compiler will link with mkl_intel_s[_dll].lib if routines are compiled with the default interface. However, if you compile with the option /iface=(cref,nomixed_str_len_arg), the compiler will link with mkl_intel_c[_dll].lib .
• Using the Intel® Fortran compiler
  The Intel Fortran compiler will link with mkl_intel_c[_dll].lib by default. But if the /Gm option is used, call mkl_intel_s  [_dll].lib (/Gm enables CVF and Powerstation calling convention compatibility, so does /iface:cvf).

5.2) MKL don't provide em64t and IA64 CVF interface support.  So if you are porting CVF 32bit application to X64 or IA64 platform with Intel Fortran compiler and MKL. Please note, you must to link "mkl_intel_lp64|ilp64, mkl_intel_thread, mkl_core.lib" and remove the compiler option /Gm or /iface:cvf.

Here is a MSVC 2005 project (Intel® Fortran 11.0.0.074 IA32 application) for your reference. DFT_VF_sample.zip

Troubleshooting
When build the project,
1. fatal error LNK1104: cannot open file 'mkl_xxx.lib'
please make sure the library is in the library path and the path you marked in step 3 or step 4 is right

2. "error #7002: Error opening the compiled module file.  Check INCLUDE paths. [MKL_DFTI]."
Please add the header file mkl_dfti.f90 in your code, for example,

! Include to build module MKL_DFTI
INCLUDE 'mkl_dfti.f90'
before  the code line
USE MKL_DFTI

Or copy the mkl_dfti.f90 file to the source directory of your project and include it in your Project.