Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
公告
FPGA community forums and blogs on community.intel.com are migrating to the new Altera Community and are read-only. For urgent support needs during this transition, please visit the FPGA Design Resources page or contact an Altera Authorized Distributor.
7234 讨论

Pardiso bunch-kaufman pivoting option 3 different in MKL 2025.0.1

StefanV1
初学者
1,149 次查看
When using PARDISO with iparm[20]=3, behavior has changed between MKL 2024.2 and 2025.0. It seems that it does not perform the solve step anymore. With 2024.2, the log said ``` Times: ====== Time spent in direct solver at solve step (solve) : 0.000027 s Time spent in additional calculations : 0.000005 s Total time spent : 0.000032 s ``` Now the line "Time spent in direct solver at solve step (solve)" disappeared and the returned solution is different. (https://github.com/coin-or/Ipopt/issues/799) Is this a know issue? Is this a feature or bug?
0 项奖励
1 解答
Ruqiu_C_Intel
主持人
1,003 次查看

Thanks to Stefan for letting us know that this is not a serious issue.

We will fix it in a future release.


在原帖中查看解决方案

0 项奖励
4 回复数
Ruqiu_C_Intel
主持人
1,095 次查看

Hi Stefan,


Thank you for posting the issue.

We can reproduce it on our side. If you set iparm[20]=0 or 1, the disappeared message will reappear.

May I ask what is the typical value of max_iterref_steps in your application? Also could you tell us how important is this issue to your project? Does this issue hinder your work? 


Best regards,

Ruqiu




0 项奖励
StefanV1
初学者
1,069 次查看

iparm[7] (max_iterref_steps) is currently set to 1 by default. Having iparm[20]=3 and iparm[7]=2 (the default) or iparm[7]=10 does not work. But setting iparm[20]=3 and iparm[7]=0 seems to work.

 

Since we have workarounds (iparm[20]=0 or 1, iparm[7]=0, or using an older MKL), this issue is not a serious blocker.

0 项奖励
Ruqiu_C_Intel
主持人
1,004 次查看

Thanks to Stefan for letting us know that this is not a serious issue.

We will fix it in a future release.


0 项奖励
Ruqiu_C_Intel
主持人
593 次查看

The fixed is available in oneMKL 2025.1.0 new release. 

0 项奖励
回复