topic Hi MiauCat, in Intel® oneAPI Math Kernel Library

mkl_get_max_threads() return value unclear

MiauCat — Fri, 14 Jul 2017 09:22:07 GMT

I'm a bit confused about the method

mkl_get_max_threads()

According to the help documentation, it should return the number of physical cores (provided that dynamic adjustment is enabled).

However, when I run the code below on my 4-core machine, I get back "1" for "nT". How so?

			if ( 1 == mkl_get_dynamic( ) )
			{
				long nT = mkl_get_max_threads();
				mkl_set_num_threads( nT );
			}

Hi,

Ying_H_Intel — Wed, 19 Jul 2017 04:02:48 GMT

Hi,

Right, generally, it returns physical cores. What is you CPU type or do you set OMP_NUM_THREAD in environment?

Could you please add the cpu check code as below and share your output?

or what exact MKL function are you calling? you may

export MKL_VERBOSE=1

run your exe.

and share the output

Best Regards,

Ying

MKLVersion Version;
mkl_get_version(&Version);
printf("Major version: %d\n",Version.MajorVersion);
printf("Minor version: %d\n",Version.MinorVersion);
printf("Update version: %d\n",Version.UpdateVersion);
printf("Product status: %s\n",Version.ProductStatus);
printf("Build: %s\n",Version.Build);
printf("Platform: %s\n",Version.Platform);
printf("Processor optimization: %s\n",Version.Processor);
printf("================================================================\n");
printf("\n");

Thanks Ying.

MiauCat — Wed, 19 Jul 2017 06:41:47 GMT

Thanks Ying.

I've placed your code (plus a check for the environmental variables) in a separate piece of code and run it, and it seems to do the correct thing. However, the piece of code I've submitted earlier did return 1 for sure.

Unfortunately, that piece of code is scraped by now. I need to redo it for testing. It is not unlikely that the application (of which the code is only a small part of) set anything OMP related beforehand - but that (in theory) shouldn't influence the outcome, or should it?

Major version: 2017
Minor version: 0
Update version: 3
Product status: Product
Build: 20170413
Platform: Intel(R) 64 architecture
Processor optimization: Intel(R) Streaming SIMD Extensions 2 (Intel(R) SSE2) enabled processors
================================================================

mkl_get_dynamic( ) returned: 1
mkl_get_max_threads( ) returned: 4
================================================================

OMP_NUM_THREAD: -NOT SET-

Hi Ying,

MiauCat — Wed, 19 Jul 2017 07:16:53 GMT

Hi Ying,

did some more testing and can reproduce my problem.

The reason I want to use mkl_get_max_threads( ) in a first place is to "reset" the threads to the normal situation after having it forced to use 1 thread only previosly. However, the moment I've set the mkl threads to 1, the according commands can not recover my physical cores anymore:

mkl_set_num_threads( 1 );
<...some code...>

if ( 1 == mkl_get_dynamic( ) )
{
	<...debug output code...>

	long nT = mkl_get_max_threads( );
	mkl_set_num_threads( nT );	
}

I do not reset the MKL threads because nT is still 1. The debug output is:

Major version: 2017
Minor version: 0
Update version: 3
Product status: Product
Build: 20170413
Platform: Intel(R) 64 architecture
Processor optimization: Intel(R) Streaming SIMD Extensions 2 (Intel(R) SSE2) enabled processors
================================================================

	 mkl_get_dynamic( ) returned: 1
	 mkl_get_max_threads( ) returned: 1
================================================================

	 OMP_NUM_THREAD:

Hi MiauCat,

Ying_H_Intel — Thu, 20 Jul 2017 06:50:38 GMT

Hi MiauCat,

Right, that should be the root cause. So you may reset the matrix

long nT = mkl_get_max_threads( );

mkl_set_num_threads( 1 );
<...some code...>

mkl_set_num_threads( nT );

The function is based on OpenMP function. so it detected the current OpenMP thread number. The documentation seems need to rewording. We will check this and fix them.

The maximum number of threads for Intel MKL functions to
use in internal parallel regions.

Thank you!

Ying

Hi Ying,

MiauCat — Thu, 20 Jul 2017 07:17:57 GMT

Hi Ying,

thanks for clearing this up. Yes I also think that the documention needs a bit of update in this case - maybe with a nice example?

In particular the Description section was confusing in https://software.intel.com/en-us/mkl-developer-reference-fortran-mkl-get-max-threads where it currently just reads:

If the dynamic adjustment is enabled, the function returns the number of physical cores on your system.

But I also think there should be an example of "how to reset" properly in https://software.intel.com/en-us/mkl-developer-reference-fortran-mkl-set-num-threads.

A related question: In case the mkl_set_num_threads(); is called at higher level and outside "my" control, what to do?
Is it a valid or a stupid idea to use:

mkl_set_num_threads( really_high_value );

Hi MiauCat,

Ying_H_Intel — Tue, 25 Jul 2017 03:19:42 GMT

Hi MiauCat,

thanks for the suggestion. The problem was escalated to our doc team.

About the mkl_set_num_threads( really_high_value ); is called at higher level and outside "my" control,

you can use mkl_get_max_threads() to get the really_high_value

or use mkl_set_num_threads( wanted number) to set the number.

If you want to get the number of physical cores of system, maybe use OpenMP function, OMP_GET_NUM_PROCS (OP).which should be ok in mkl application. ( for your reference https://software.intel.com/en-us/forums/intel-visual-fortran-compiler-for-windows/topic/302866

Best Regards,
Ying