Hamming weight phasing with configurable number of ancilla

[dobbse42]

#645: Added HammingWeightPhasingWithConfigurableAncilla and a temp test file, still debugging. TODO: add unit tests to hamming_weight_phasing_test.py

Currently having issues joining the slices of the phased register back together with
x = bb.join(x_parts, dtype=QUInt(self.bitsize.bit_length()))
at the end of build_composite_bloq() (line 272).

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[mpharrigan]

Thanks for opening a PR! Please ping me or @tanujkhattar when this is ready for review or if you have any specific questions

[mpharrigan]

try bb.join(np.array(x_parts), ...)

Elsewhere, we tried to be careful about accepting either lists or numpy arrays but seem to have missed this one.
#1470

[dobbse42]

@mpharrigan This should be in a functional state at least. There are a few TODO comments which indicate things I couldn't immediately tell how to do but aren't essential, so suggestions for those would be appreciated. I also have a few questions:

• Do we want to perform input validation (e.g. ancillasize < bitsize-1)? If so, what is the preferred method?
• Should I write gatecount assertions to get gatecounts of subcomponents, or just hardcode them? e.g. should I just write "4*num_toffolis" or is there some way to get the T-count of a specified toffoli decomposition in tests?
• Is there any particular logic to the choice of parameter values to test, or just common sense?
• Are we testing against what we logically expect or what the paper says should happen? i.e. we can often get more precise gatecounts by just thinking about the situation (in which case we might assert that counts are exactly equal to what we expect), whereas papers are often just concerned with asymptotics or upper bounds (in which case we would assert that gatecounts are <= what the paper reports). Further, there are situations where computing the precise gatecount for every input is tedious and not very helpful for testing compared to just saying "less than" or "approx equal to".
• When is a build_call_graph override necessary? For example, HWphasegradient doesn't have one but I got errors when I tried to test HWConfigurableancilla without one.

All feedback is welcome, especially on the tests as I do not have much experience with pytest. I expect to have to make quite a few modifications to this.