0x00007fffe006cbba in __vsldBRngWH () from /usr/lib64/Revo-7.3/R-3.1.1/lib64/R/lib/libmkl_vml_mc3.so

[Prabhat_T_]

STACK TRACE

> x <- rxExec(rnorm, 100, 5, RNGkind="WH")

Program received signal SIGILL, Illegal instruction.
0x00007fffe006cbba in __vsldBRngWH () from /usr/lib64/Revo-7.3/R-3.1.1/lib64/R/lib/libmkl_vml_mc3.so

Problem Description:

I am using R on CENTOS 6.7, and while executing a simple R script to generate random numbers via x <- rxExec(rnorm, 100, 5, RNGkind="WH"), the script causes SIGKILL in Math Kernel Library when RNGkind="WH", it works fine for any other Random number algorithm like MCG31, R250, MCG59, MT19937, SOBOL, NIEDERR

Any suggestion...

[Andrey_N_Intel]

Hello,

Can you provide additional details on CPU, OS, and version of Intel MKL you see the illegal instruction issue with? Do you observe the same behavior when running C example that calls Intel MKL WH based random number generation in the same environment (HW/OS/etc)?

Thanks, Andrey

[Siddharth_C_]

Hi Audrey

here is below information we could find out

[1] "Intel(R) Math Kernel Library Version 11.1.1 Product Build 20131010 for Intel(R) 64 architecture applications"  

2.6.32-573.18.1.el6.x86_64 #1 SMP Tue Feb 9 22:46:17 UTC 2016 x86_64 x86_64 x86_64 GNU/Linux

vendor_id       : AuthenticAMD
cpu family      : 16
model           : 8
model name      : AMD Opteron(tm) Processor 4171 HE

[Siddharth_C_]

Hi Andrey

Do you have C example to run WH random number generator, and steps to execute them which I can use.

Siddharth

[Gennady_F_Intel]

- the similar issues were fixed into 11.1 update 4 and latest versions of MKL ( 11.2 ). Could you check the update 4 first. You may also try to evaluate the latest 11.3 update 2 available now and let us know the result. 

- wrt to example - You may try vslgetstreamstatebrng.c example from <mklroot>\examples\vslc\source directory.

[Andrey_N_Intel]

Additionally, you may want to run the example vdrnggaussian.c for generation of Gaussian random variates which is available in the folder Gennady suggested. In the example, please replace the line #define BRNG    VSL_BRNG_MCG31 with #define BRNG    VSL_BRNG_WH.  Intel MKL supports three methods for generation of Gaussian numbers: Box-Muller, Box-Muller2, and ICDF, To check all of them please modify the line #define METHOD  VSL_RNG_METHOD_GAUSSIAN_ICDF in the example using VSL_RNG_METHOD_GAUSSIAN_BOXMULLER and VSL_RNG_METHOD_GAUSSIAN_BOXMULLER2. In case you do not have the example I attach it. To build the example, please the service of Intel MKL Link Advisor available at https://software.intel.com/en-us/articles/intel-mkl-link-line-advisor.

Please, let us know how it works for you.

Thanks,

Andrey