Fix cast method not changing dtype #55
Conversation
There was a problem hiding this comment.
Looks good to me, thanks. Do you think the default None option for dtype is useful anywhere? If no, we could remove it and use a different default value (eg. "complex128") or not have a default at all and remove the if dtype is None check. This would simplify the code a bit.
|
I'm not sure, |
There was a problem hiding this comment.
I'm not sure,
self.castwithoutdtypeis called everywhere inbackend.py, but I'm not sure if we can simply replace it withDTYPECPX.
I checked and all cases where self.cast is called without dtype in backend.py are on the state vector, so in principle DTYPECPX would work. The only issue is that I think this is not accessible from this class, only from the global qibojit backend class. Given that dtype=None is also the default for the np.array() call it should be fine to leave it as it is.
Other than that this looks good to me.
|
Can we merge this? |
|
Yes, I believe so. |
Closes #47. While working to fix #47, I noticed that PR qiboteam/qibo#528 do not resolve the problem with qibojit.
In fact, the cast method of CuPy doesn't change the dtype if the input array is already a CuPy array. I fixed it, and I also make some additional changes to improve the cast for both CuPy and Numba, because some times the dtype is ignored. Let me know your opinion.