The unit tests did not catch the race condition fixed by r4c63c1569c because
they do not consider sufficiently large input sizes. We should add at least a
few test sizes between 1M to 10M for each of the major primitives (sort, scan,
reduce, transform). Basically, anything implemented with a CUDA kernel should
have an associated test.
Original issue reported on code.google.com by wnbell on 10 Jun 2010 at 8:32
Original issue reported on code.google.com by
wnbell
on 10 Jun 2010 at 8:32