topic Hi Chao, in Intel® oneAPI Math Kernel Library

Deal with huge matrix

chao_n_ — Tue, 26 Jul 2016 08:38:15 GMT

I have written some codes with MKL library, but recently these codes do not work. They can be complied well and do not report any error when I run it. However, they did not work for some huge matrix operations. They keep running, but no results return. Here is an example code. When the variable nID exceed 31, the program will keep running forever.

#include <stdio.h>

#include <mkl.h>

int main(){

int nID=31;

int nSNP=33;

double *mMatrix = (double*) calloc(nID*nSNP,sizeof(double));

double *AGmatrix = (double*) calloc(nID*nID,sizeof(double));

const CBLAS_LAYOUT Layout=CblasRowMajor;

double alpha=1;

double beta=0;

cblas_dgemm (Layout, CblasTrans, CblasNoTrans, nID, nID, nSNP, alpha, mMatrix,

nID, mMatrix, nID, beta, AGmatrix, nID);

}

I suppose you had in mind

TimP — Tue, 26 Jul 2016 13:50:25 GMT

I suppose you had in mind some other compiler, but I tried to make a VS gui project with a recent Intel C++ for Intel64, hoping to run under debug. I had several issues with source code format and line terminators. I wasn't able to get the mkl include path working. At command line, I was able to build, but I got immediate return to command prompt when running.

It doesn't appear that you should be approaching default limits for stack or (if in 32-bit mode) heap space, but you leave us guessing.

Hi Chao,

Ying_H_Intel — Wed, 27 Jul 2016 02:06:06 GMT

Hi Chao,

Could you please tell some details like the OS, 32bit or 64bit, how do you compile the mkl program?

for example, we run it in Red Hat Enterprise Linux Server release 6.3 on machine with16 core . with command line

$ source /opt/intel/mkl/bin/mklvars.sh intel64

$ g++ main.cpp -lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lm
$ ./a.out

it just run fine.

Best Regards,

Ying

Hi Chao,

Ying_H_Intel — Wed, 27 Jul 2016 03:20:53 GMT

Hi Chao,

You can

>export MKL_VERBOSE=1

>a.out

and let us know the result.

Best Regards,
Ying

Quote:Ying H. (Intel) wrote:

chao_n_ — Wed, 27 Jul 2016 03:30:23 GMT

Ying H. (Intel) wrote:

Hi Chao,

You can

>export MKL_VERBOSE=1

>a.out

and let us know the result.

Best Regards,
Ying

I'm sorry. I still keep running.

[ningc@localhost sig]$ gcc -lmkl_intel_lp64 -lmkl_intel_thread -liomp5 -lmkl_core\

> -lpthread -lm test.c

[ningc@localhost sig]$ export MKL_VERBOSE=1

[ningc@localhost sig]$ ./a.out

Quote:chao n. wrote:

chao_n_ — Wed, 27 Jul 2016 07:21:41 GMT

chao n. wrote:

Quote:

Ying H. (Intel) wrote:

Hi Chao,

You can

>export MKL_VERBOSE=1

>a.out

and let us know the result.

Best Regards,
Ying

I'm sorry. It still keep running.

[ningc@localhost sig]$ gcc -lmkl_intel_lp64 -lmkl_intel_thread -liomp5 -lmkl_core\

> -lpthread -lm test.c

[ningc@localhost sig]$ export MKL_VERBOSE=1

[ningc@localhost sig]$ ./a.out

When I set nID=31

[ningc@localhost sig]$ ./a.out

MKL_VERBOSE Intel(R) MKL 11.3 Update 3 Product build 20160413 for Intel(R) 64 architecture Intel(R) Streaming SIMD Extensions 4.2 (Intel(R) SSE4.2) enabled processors, Lnx 2.00GHz lp64 intel_thread NMICDev:0

MKL_VERBOSE DGEMM(N,T,31,31,33,0x7fffdf46b9d0,0x20ed010,31,0x20ed010,31,0x7fffdf46b9d8,0x20ef010,31) 80.88ms CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:32 WDiv:HOST:+0.000

Hi

Ying_H_Intel — Thu, 28 Jul 2016 02:57:05 GMT

When set nID=32,32,33, my output is like below

[yhu5@snb04 mkl_issue]$ ./a.out

Error getting SCIF driver version
MKL_VERBOSE Intel(R) MKL 11.3 Update 3 Product build 20160413 for Intel(R) 64 architecture Intel(R) Streaming SIMD Extensions 4.2 (Intel(R) SSE4.2) enabled processors, Lnx 2.70GHz lp64 intel_thread NMICDev:0
MKL_VERBOSE DGEMM(N,T,32,32,33,0x7fffa3280650,0x165e010,32,0x165e010,32,0x7fffa3280658,0x1660120,32) 3.61ms CNR:SSE4_2 Dyn:1 FastMM:1 TID:0 NThr:16 WDiv:HOST:+0.000

The notable places is maybe the machine type be " Intel(R) Streaming SIMD Extensions 4.2 (Intel(R) SSE4.2) enabled processors. and thread number".

Do you have other machine to test the code? and for thread number , how about if you

>export MKL_NUM_THREADS=1

>a.out?

Best Regards,
Ying

Hi

chao_n_ — Fri, 29 Jul 2016 08:32:35 GMT

When set export MKL_NUM_THREADS=1, it work well.

However, I can not set more than one threads. How can I solve the problem?

Ying H. (Intel) wrote:

Hi

When set nID=32,32,33, my output is like below

[yhu5@snb04 mkl_issue]$ ./a.out

Error getting SCIF driver version
MKL_VERBOSE Intel(R) MKL 11.3 Update 3 Product build 20160413 for Intel(R) 64 architecture Intel(R) Streaming SIMD Extensions 4.2 (Intel(R) SSE4.2) enabled processors, Lnx 2.70GHz lp64 intel_thread NMICDev:0
MKL_VERBOSE DGEMM(N,T,32,32,33,0x7fffa3280650,0x165e010,32,0x165e010,32,0x7fffa3280658,0x1660120,32) 3.61ms CNR:SSE4_2 Dyn:1 FastMM:1 TID:0 NThr:16 WDiv:HOST:+0.000

The notable places is maybe the machine type be " Intel(R) Streaming SIMD Extensions 4.2 (Intel(R) SSE4.2) enabled processors. and thread number".

Do you have other machine to test the code? and for thread number , how about if you

>export MKL_NUM_THREADS=1

>a.out?

Best Regards,
Ying