dtrnlspbc_get() returns 0xbaadf00d in "iter" argument

[sim176]

I'm using MKL 11 (tried on 11.04 and 11.0.5) and I'm getting a bad return value upon the return of this function. The "iter" argument gets filled with the magic number value 0xbaadf00d, which means unitialized allocated heap memory in the Microsoft environment (according to http://en.wikipedia.org/wiki/Magic_number_(programming)).The problem is systematic in my test case.

This function has been in use in our code for more than a year without any problem. As we recently upgraded to MKL 11 from 10.x, and obviously I'm suspecting it.

Any ideas?

Thank you

[sim176]

I think the upgrade indeed changed something, and I think that we are using the function in a way that seems now illegal in MKL 11. We were calling the dtrnlspbc_get() function even if the _solve() function had not returned a negative RCI_Request (indicating the regression has finished).

What seems weird, (a probably a bug?) is the behavior I described in my first post. The _get function even returned TR_SUCCESS.

My problem seems fixed, but a confimation of my assumptions would be appreciated.

Thank you

[Zhang_Z_Intel]

sim176 wrote:

My problem seems fixed, but a confimation of my assumptions would be appreciated.

It seems to be prohibited now to call _get() before _solve() returns TR_SUCCESS. I'll double-check and let you know.

[mecej4]

The forum software has now expanded its repertoire -- typing a "quote=" tag makes it appear in different languages at different times! In Zhang's post above, it is in Russian. I had the same thing happening to my posts elsewhere.

[Zhang_Z_Intel]

Hi,

I'm not able to reproduce the issue using MKL 11.0.5. Please see my test code attached. I simply modifies  an existing code sample in the MKL package. The *_get() function is called when *_solve() does not return a negative RCI_Request, and it does return a valid "iter" value each time. Would you please post your test code and clarify more about the problems you encountered? Thanks.