Numba / JIT produces invalid results

0 views
Skip to first unread message

Peter Würtz

unread,
May 12, 2017, 1:32:11 PM5/12/17
to Numba Public Discussion - Public
I encountered a situation where Numba produces invalid results. The weird thing is that the compiled function jumps from being valid to invalid when changing seemingly irrelevant details within the code. I posted an example at issue

https://github.com/numba/numba/issues/2365

In the meantime I discovered the `inspect_types` function and observed that numba deduces wrong local variable types, seemingly at random. In this example there are float64 types popping up in a line of code where there are just integer bitshift operations.
Reply all
Reply to author
Forward
0 new messages