LAPACKE_mkl_cgetrfnpi. return value

[marcsolal]

I have an issue using LAPACKE_mkl_cgetrfnpi. The function is returning me a nonzero value. I am getting a return value (info) of -4. According to documentation, I understand this means that the parameter 4 does not have a legal value. I am using 1581 for m and n and 1580 for nfact (4th parameter). It looks to me this are correct values since I have nfact<=n and nfact<=m.   Do I miss something?

Thanks for helping. 

[Gennady_F_Intel]

We had a similar issue a couple of versions ago. Which version of MKL do you use?

[marcsolal]

This is the one I am using:

Intel(R) Math Kernel Library Version 2020.0.2 Product Build 20200624 for Intel(R) 64 architecture applications

I have the same result with 2019.

[Gennady_F_Intel]

Marco,

Could you build us the short reproducer?

[marcsolal]

What do you mean? This is inside a large programme I cannot share. I can try to get the same thing in a smaller one. I do not know if I will get the same error. 

[Gennady_F_Intel]

We don't need to know the whole program, but we need to see how you call this routine and the input date you pass to this routine. And also if you use some specific libraries and / or compiler option.

The reproducer could help us to check and investigate the problem on our end faster.

and another cause we typically asked to prepare the reproducer - in 50% of all cases when the user starts to prepare the reproducer, the user identifies the problem on his side.

[mecej4]

Marcsolal: A common error in situations of this type is that integer arguments have byte-sizes (4 or 8 )  that do not match the target model being used (LP64 or ILP64). It helps to know the declared types of the argument variables and details of how the linking was performed.

[Gennady_F_Intel]

yes, I agree, it might be the cause of the problem.

[marcsolal]

It should not be the case. I am using 64 bits int in my code and when I call this function. I am using many MKL functions in this  program and they are working.

I will try to write a smaller version of my code on Monday. Hopefully, I will get the same error in the small code.  

Thanks.

[marcsolal]

I still need to reproduce the error in a smaller program.  I did not succeed on my first attempt. The code was working fine.  So I am not sure what was wrong. I think it was not an issue with the integers but it has something to do with the matrix choice.  

[Gennady_F_Intel]

Mark, are there any updates? Were you able to build the same small reproducer?

[marcsolal]

I tried to reproduce the error on a smaller case, but it was working fine. So, I think for now, this is in hold. If I find the same error again, I will post it.