PGI User Forum

Mat,

I have not had time to test out all possible variations and permutations. I just notice that if I run the same Fortran program twice in a row, the first time takes 0.5 seconds longer --yes re ...

Mat,

There still is an outstanding issue about why allocation of device memory produces enormous delays. pgcudainit does not solve this problem on our Redhat Linux system. In fact it has no effec ...

According to the release notes, the shift problem has been solved in version 11.3; see http://www.pgroup.com/support/release_tprs_2011.htm (TPR 17689).

Mat,

Your latest version of the program indeed no longer produces these wild differences in behavior of C and FORTRAN. Whether I run pgcudainit or not makes no difference. Indeed, it should not, ...

Mat,

Thanks for pointing out those differences. I changed those variables to integer*4 in the FORTRAN version and reran the short runs. Did you get a 15% increase in speed on the long ones? For m ...