Bug in tsmttsm w/ CUDA+OpenMP for some odd problem size

Create issue
Issue #329 resolved
Jonas Thies created an issue

V^TW gives wrong results in the phist tests for nrows=163, ncols=30, nprocs=12, where P1 gets a GPU. Row distribution is equal, 13 per proc and 20 for the last p11.

to reproduce: mpirun -np 12 ./phist-1.7.4-kernels-test-[Release|Debug] --gtest_filter=163_30

Comments (2)

  1. Log in to comment