Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
7056 Discussions

MKL in OneAPI 2024.1

EMagee
Beginner
‎06-05-2024 03:28 PM
1,294 Views

I have run into an issue with the MKL library from OneAPI 2024.1.0 on a couple Linux platforms. I am getting incorrect results from one of the function (vdTGamma, specifically). If I compare the results from this function between using OneAPI 2023.2.0 and 2024.1.0 I get different results. I have verified the results are correct from 2023.2.0. I have attached an CPP file (with a CMake.txt) to demonstrate. To build the example:

 

> export MKLROOT=<location of onapi 2023.2>/mkl/latest
> export COMPILERLIBDIR=<location of onapi 2023.2>/compiler/latest
> mkdir build
> cd build
> cmake ..
> make
> ./vdTGamma
vdTGamma test/example program

Argument  vdTGamma   expected
==================================
-1.3499   +2.9314    +2.9314
+3.0349   +2.0660    +2.0660
+0.7254   +1.2596    +1.2596
-0.0631   16.4914    16.4914
+0.7147   +1.2754    +1.2754
-0.2050   5.7068     5.7068
-0.1241   8.7741     8.7741
+1.4897   +0.8859    +0.8859
+1.4090   +0.8868    +0.8868
+1.4172   +0.8865    +0.8865
Error! Maximum error is -0.0000

Now using 2024:

> export MKLROOT=<location of onapi 2024.1>/mkl/latest
> export COMPILERLIBDIR=<location of onapi 2024.1>/compiler/latest
> rm -rf *
> cmake ..
> make
> ./vdTGamma
vdTGamma test/example program
Argument  vdTGamma   expected
==============================
-1.3499   +2.6115    +2.9314
+3.0349   +2.0660    +2.0660
+0.7254   +1.2596    +1.2596
-0.0631   inf        16.4914
+0.7147   +1.2754    +1.2754
-0.2050   inf        5.7068
-0.1241   inf        8.7741
+1.4897   +0.8859    +0.8859
+1.4090   +0.8868    +0.8868
+1.4172   +0.8865    +0.8865
Error! Maximum error is   -inf

Has anyone else seen this?  My caode, and CMake file are attrached.

vdTGamma.tar.gz
0 Kudos

Link Copied

6 Replies
Gennady_F_Intel
Moderator
‎06-06-2024 01:34 AM
1,245 Views

What kind of CPU you are running your code here?

I see no problem here while building against oneMKL 2024.1 ( the current one) and 2023.2 and running on Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz.

I slightly updated your example by adding mkl_get_version(*.*) routine to print info about version of MKL used:

Compiling: icpx -std=c++11 -qmkl vdTGammaTest.cp

Results obtained:

MKL 2023 u2:

Argument vdTGamma expected
========================
-1.3499 +2.9314 +2.9314
+3.0349 +2.0660 +2.0660
+0.7254 +1.2596 +1.2596
-0.0631 -16.4914 -16.4914
+0.7147 +1.2754 +1.2754
-0.2050 -5.7068 -5.7068
-0.1241 -8.7741 -8.7741
+1.4897 +0.8859 +0.8859
+1.4090 +0.8868 +0.8868
+1.4172 +0.8865 +0.8865

Error! Maximum error is -0.0000

************************************
MKL version: 2023.0.2, CPU: Intel(R) Advanced Vector Extensions 512 (Intel(R) AVX-512) enabled processors
************************************

MKL 2024 u1:

Argument vdTGamma expected
===============================================================================
-1.3499 +2.9314 +2.9314
+3.0349 +2.0660 +2.0660
+0.7254 +1.2596 +1.2596
-0.0631 -16.4914 -16.4914
+0.7147 +1.2754 +1.2754
-0.2050 -5.7068 -5.7068
-0.1241 -8.7741 -8.7741
+1.4897 +0.8859 +0.8859
+1.4090 +0.8868 +0.8868
+1.4172 +0.8865 +0.8865

Error! Maximum error is -0.0000

****************************************************************
MKL version: 2024.0.1, CPU: Intel(R) Advanced Vector Extensions 512 (Intel(R) AVX-512) enabled processors
****************************************************************

--Gennady

0 Kudos
Copy link
EMagee
Beginner
‎06-06-2024 06:51 AM
1,214 Views
In response to Gennady_F_Intel

Here are the outputs from two different systems.  Let me know if you need additional information.  System 1 is an HPE Cray EX with the AMD 7H12 Rome processor  and the SLES OS.  System 2 is a Penguin Computing TrueHPC system with the AMD 7713 Milan  and the RHEL OS.  NOTE: I had to add pthread, m, and dl to the link line for system 2.

System 1 with 2023, output from CMake:

 

-- The C compiler identification is GNU 10.3.0
-- The CXX compiler identification is GNU 10.3.0
-- Check for working C compiler: /opt/cray/pe/craype/2.7.19/bin/cc
-- Check for working C compiler: /opt/cray/pe/craype/2.7.19/bin/cc - works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /opt/cray/pe/craype/2.7.19/bin/CC
-- Check for working CXX compiler: /opt/cray/pe/craype/2.7.19/bin/CC - works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
vdTGamma: Using MKL in /opt/intel/oneapi_2023.2.0.49397/mkl/latest with ...
   MKL_LIB_DIR: /opt/intel/oneapi_2023.2.0.49397/mkl/latest/lib/intel64 ...
   COMPILER_LIB_DIR: /opt/intel/oneapi_2023.2.0.49397/compiler/latest/linux/compiler/lib/intel64_lin ...
vdTGamma test/example program

Major version:          2023.0.2
Build:                  20230613
Platform:               Intel(R) 64 architecture
Processor optimization: Intel(R) Architecture processors

Argument   vdTGamma     expected
======================================
-1.3499   +2.9314     +2.9314
+3.0349   +2.0660     +2.0660
+0.7254   +1.2596     +1.2596
-0.0631   -16.4914     -16.4914
+0.7147   +1.2754     +1.2754
-0.2050   -5.7068     -5.7068
-0.1241   -8.7741     -8.7741
+1.4897   +0.8859     +0.8859
+1.4090   +0.8868     +0.8868
+1.4172   +0.8865     +0.8865

Error! Maximum error is -0.0000

 

System 1 with 2024, output from CMake:

 

-- The C compiler identification is GNU 10.3.0
-- The CXX compiler identification is GNU 10.3.0
-- Check for working C compiler: /opt/cray/pe/craype/2.7.19/bin/cc
-- Check for working C compiler: /opt/cray/pe/craype/2.7.19/bin/cc - works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /opt/cray/pe/craype/2.7.19/bin/CC
-- Check for working CXX compiler: /opt/cray/pe/craype/2.7.19/bin/CC - works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
vdTGamma: Using MKL in /opt/intel/oneapi_2024.1.0.596/mkl/latest with ...
   MKL_LIB_DIR: /opt/intel/oneapi_2024.1.0.596/mkl/latest/lib/intel64 ...
   COMPILER_LIB_DIR: /opt/intel/oneapi_2024.1.0.596/compiler/latest/lib ...
vdTGamma test/example program

Major version:          2024.0.1
Build:                  20240215
Platform:               Intel(R) 64 architecture
Processor optimization: Intel(R) Architecture processors

Argument   vdTGamma     expected
======================================
-1.3499   +2.6115     +2.9314
+3.0349   +2.0660     +2.0660
+0.7254   +1.2596     +1.2596
-0.0631     -inf     -16.4914
+0.7147   +1.2754     +1.2754
-0.2050     -inf     -5.7068
-0.1241     -inf     -8.7741
+1.4897   +0.8859     +0.8859
+1.4090   +0.8868     +0.8868
+1.4172   +0.8865     +0.8865

Error! Maximum error is   -inf

 

System 2 with 2023, CMake output:

 

-- The C compiler identification is GNU 8.5.0
-- The CXX compiler identification is GNU 8.5.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc - works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ - works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
vdTGamma: Using MKL in /p/app/compilers/intel/oneapi-2023.1.0/mkl/latest with ...
   MKL_LIB_DIR: /p/app/compilers/intel/oneapi-2023.1.0/mkl/latest/lib/intel64 ...
   COMPILER_LIB_DIR: /p/app/compilers/intel/oneapi-2023.1.0/compiler/latest/linux/compiler/lib/intel64_lin ...
vdTGamma test/example program

Major version:          2023.0.1
Build:                  20230303
Platform:               Intel(R) 64 architecture
Processor optimization: Intel(R) Architecture processors

Argument   vdTGamma     expected
========================================
-1.3499   +2.9314     +2.9314
+3.0349   +2.0660     +2.0660
+0.7254   +1.2596     +1.2596
-0.0631   -16.4914     -16.4914
+0.7147   +1.2754     +1.2754
-0.2050   -5.7068     -5.7068
-0.1241   -8.7741     -8.7741
+1.4897   +0.8859     +0.8859
+1.4090   +0.8868     +0.8868
+1.4172   +0.8865     +0.8865

Error! Maximum error is -0.0000

 


System 2 with 2024, CMake output:

 

-- The C compiler identification is GNU 8.5.0
-- The CXX compiler identification is GNU 8.5.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc - works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ - works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
vdTGamma: Using MKL in /p/app/compilers/intel/oneapi-2024.1.0/mkl/latest with ...
   MKL_LIB_DIR: /p/app/compilers/intel/oneapi-2024.1.0/mkl/latest/lib/intel64 ...
   COMPILER_LIB_DIR: /p/app/compilers/intel/oneapi-2024.1.0/compiler/latest/lib ...
vdTGamma test/example program

Major version:          2024.0.1
Build:                  20240215
Platform:               Intel(R) 64 architecture
Processor optimization: Intel(R) Architecture processors

Argument   vdTGamma     expected
======================================
-1.3499   +2.6115     +2.9314
+3.0349   +2.0660     +2.0660
+0.7254   +1.2596     +1.2596
-0.0631     -inf     -16.4914
+0.7147   +1.2754     +1.2754
-0.2050     -inf     -5.7068
-0.1241     -inf     -8.7741
+1.4897   +0.8859     +0.8859
+1.4090   +0.8868     +0.8868
+1.4172   +0.8865     +0.8865

Error! Maximum error is   -inf

 

 

In response to Gennady_F_Intel
0 Kudos
Copy link
EMagee
Beginner
‎06-06-2024 07:35 AM
1,205 Views
In response to EMagee

Side note: just out of curiosity, I tried vsTGamma with 2024 and it worked fine.

In response to EMagee
0 Kudos
Copy link
EMagee
Beginner
‎06-06-2024 01:11 PM
1,161 Views
In response to EMagee

Another side note.  I was able to test this on an additional machine using 2024 and it works fine.  System 3 is an HPE Cray EX4000 with the AMD 9654 Genoa processor  and the SLES 15 OS.  Here is the output:

vdTGamma test/example program

Major version:          2024.0.0
Build:                  20231011
Platform:               Intel(R) 64 architecture
Processor optimization: Intel(R) Architecture processors

Argument  vdTGamma   std::tgamma
==================================
-1.3499   +2.9314     +2.9314
+3.0349   +2.0660     +2.0660
+0.7254   +1.2596     +1.2596
-0.0631   -16.4914     -16.4914
+0.7147   +1.2754     +1.2754
-0.2050   -5.7068     -5.7068
-0.1241   -8.7741     -8.7741
+1.4897   +0.8859     +0.8859
+1.4090   +0.8868     +0.8868
+1.4172   +0.8865     +0.8865

Error! Maximum error is 0.0000

u

 

In response to EMagee
0 Kudos
Copy link
Gennady_F_Intel
Moderator
‎06-07-2024 12:42 AM
1,124 Views

yes, I have to reproduced the behavior while running this case on Genoa CPU. We will root-caused the problem and keep this threads updated.

--Gennady

0 Kudos
Copy link
Gennady_F_Intel
Moderator
‎06-26-2024 01:11 AM
643 Views

Erik,

yesterday the last version of oneMKL ( 2024.2) has been released and available for download. The problem You reported has been fixed.

You could take 2024.2 and evaluate this fix.

here is what I see running this example on Genoa CPU.

vdTGamma test/example program

Argument vdTGamma expected
===============================================================================
-1.3499 +2.9314 +2.9314
+3.0349 +2.0660 +2.0660
+0.7254 +1.2596 +1.2596
-0.0631 -16.4914 -16.4914
+0.7147 +1.2754 +1.2754
-0.2050 -5.7068 -5.7068
-0.1241 -8.7741 -8.7741
+1.4897 +0.8859 +0.8859
+1.4090 +0.8868 +0.8868
+1.4172 +0.8865 +0.8865

Error! Maximum error is -0.0000

****************************************************************
MKL version: 2024.0.2, CPU: Intel(R) Architecture processors
****************************************************************

I also attached the all needed details about CPU, OS, linking and obtained results - see _2024u2_Genoa.txt . 

6 KB
0 Kudos
Copy link
Reply

Community support is provided Monday to Friday. Other contact methods are available here.

Intel does not verify all solutions, including but not limited to any file transfers that may appear in this community. Accordingly, Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.

For more complete information about compiler optimizations, see our Optimization Notice.