Siu Kwan Lam
Siu Kwan Lam
I did a quick look at speed of handwritten dot product loop and I didn't find any performance difference between the two versions. The problem may be only occurring to...
I am able to reproduce the problem. Here's my script based on https://github.com/numba/numba/issues/8398#issuecomment-1228819198: bench.ipy ```python from numpy import * from numba import jit nRates_vf = 13 nPayment = 19 @jit(nopython=True)...
Relates to https://github.com/numba/numba/issues/6547#issuecomment-766693730 and https://github.com/numba/numba/issues/8172#issuecomment-1160474583
The problem here is that the return type request is not wired up: https://github.com/numba/numba/blob/c1e63c032899935d142aa6ead5d52b8730af53b8/numba/cuda/decorators.py#L91 `restype` is not passed to `disp.compile`.
RE: https://github.com/numba/numba/issues/8400#issuecomment-1229502785 @cstyl The behavior of `i32 + i32` promoting to `i64 + i64` is due to Numba's lack of bigint support and we opted for preventing overflow. This is...
Discussed in triage meeting to prefer returning instead of printing out. and make old behavior deprecated in the next release.
I cannot replicate the problem on mainline or 0.54.1 on OSX.
ping @Hardcode84 Dispatchers are not leaking but the MCJIT is leaking the loaded object code. It's unlikely that this problem will affect your usecase since you are not using the...
TODO: look at exception keeping Dispatcher alive
Thanks for the contribution. I'll schedule it for review.