Lion Krischer <lion.krisc...@gmail.com> wrote: > I added a slightly more comprehensive benchmark to the PR. Please have a > look. It tests the total time for 100 FFTs with and without cache. It is > over 30 percent faster with cache which it totally worth it in my > opinion as repeated FFTs of the same size are a very common use case.
All the calls to trancendental functions are stored in the cache. Without a cache, we get excessive calls to sin(x) and cos(x) whenever FFTs of the same size are repeated. This can indeed matter at lot. Sturla _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org https://mail.scipy.org/mailman/listinfo/numpy-discussion
