prutschman-iv
prutschman-iv
It looks like the kernel is defined as `((x > 0) - (x < 0))` numpy uses the expression `in1 > 0 ? 1 : (in1 < 0 ? -1...
I'll attempt to prepare a pull request with appropriate tests. For the tests, it looks like one approach would be adding something like this to test_misc.py. Does that seem right?...
I must be misunderstanding some aspect of the test suite. I tried adding the `test_sign_inf_nan` and running the test suite with `python setup.py test` without fixing the bug, but I...
I was successful in running the test suite for v12, thank you. From there I also observed a divergence between cupy and numpy for complex numbers, so I've fixed that...
I'm not trying to pester, but I wanted to point out that I've removed the draft status from my PR, just in case that notification isn't automatic.
I've been looking more at the PTX code generated. I think the original method of `(in0.imag() > 0) - (in0.imag() < 0)` may be superior for the complex case, and...
I have verified the changes with the main branch. I believe this is ready for merge.
@kmaehashi I added the flag. It did fix the warnings, but the error still occurred. @leofang I applied the patch. This allowed the build to succeed.
*EDIT*: about 160/second, not 16/second. So, my once/second theory doesn't explain the load. I tried profiling an idle worker and found it spends a fair fraction of its running time...
4-6% across all CPUs, not of a single thread. (It's not literally pegging a single thread, it's spread across all of them.)