Maybe you can shed some light on this. I’m not too familiar with the OpenCL codebase, and I am trying to fix some failing tests in the monorepo. I have the following failing test in
AgradRevMatrix, mat_cholesky_1st_deriv_large_gradients. Single stepping through the code, the problem arises inside
tri_inverse.hpp, when calling
WORK_PER_THREAD somehow has a value of
0, and when calculating block sizes later on causes a divide by zero error. Also, a
0-valued THREAD_BLOCK_SIZE entry is also created.
I have taken a screenshot of my debugger: https://imgur.com/a/UnQAFlU. In it, you will see that the options map has two duplicate definitions of “WORK_PER_THREAD”. I don’t even know how that is possible. Is this possibly a Google Test threading issue / have you ever seen it before?