Fix dtypes in IntVector and PlusEqualsProduct

[anurudhp]

(part of #1142)

[tanujkhattar]

I'm surprised this does not raise an error because the call site in the decomposition of QFTPhaseGradient is passing in a register of type QFxp and would need to be updated to include a cast; at the very least.

It's maybe best to open an issue to track the changes needed at the call site and merge this change so we don't go down the rabbit hole. This is a positive improvement so LGTM

[anurudhp]

I'm surprised this does not raise an error because the call site in the decomposition of QFTPhaseGradient is passing in a register of type QFxp and would need to be updated to include a cast; at the very least.

I think it's because it uses a cirq style decomp which may not be respecting the dtypes?

Qualtran/qualtran/bloqs/qft/qft_phase_gradient.py

Lines 82 to 97 in 174b549

	def decompose_from_registers(
	self, *, context: cirq.DecompositionContext, **quregs: NDArray[cirq.Qid] # type: ignore[type-var]
	) -> Iterator[cirq.OP_TREE]:
	if self.bitsize == 1:
	yield cirq.H(*quregs['q'])
	return
	q, phase_grad = quregs['q'], quregs['phase_grad']
	a, b = q[: self.bitsize // 2], q[self.bitsize // 2 :]
	yield QFTPhaseGradient(len(a), False).on_registers(q=a, phase_grad=phase_grad[: len(a)])
	yield PlusEqualProduct(len(a), len(b), len(phase_grad)).on_registers(
	a=a[::-1], b=b, result=phase_grad
	)
	yield QFTPhaseGradient(len(b), False).on_registers(q=b, phase_grad=phase_grad[: len(b)])
	if self.with_reverse:
	for i in range(self.bitsize // 2):
	yield cirq.SWAP(q[i], q[-i - 1])

[tanujkhattar]

I think it's because it uses a cirq style decomp which may not be respecting the dtypes?

I'd be very surprised if that's the root cause.
Update:
I think what's happening is that the
assert_valid_bloq_decomposition(qft_bloq) is testing in the test is wrongly testing the decomposition of TestQFTWithPhaseGradient instead of QFTPhaseGradient.

Qualtran/qualtran/bloqs/qft/qft_phase_gradient_test.py

Line 59 in 174b549

assert_valid_bloq_decomposition(qft_bloq)

Can you add a test for assert_valid_bloq_decomposition that tests the QFTPhaseGradient bloq instead? I think that test should fail here.

---

cc @fdmalone Are data type verification checks turned on by default? Do you know what may be going on here?

[fdmalone]

@tanujkhattar The type checking will only raise an error if things are grossly wrong. By default it's set to LOOSE (which allows a limited set of conversions without a cast from wholly fractional / wholly integer fxps to uints, is that the case here?)
https://github.com/quantumlib/Qualtran/blob/main/qualtran/_infra/data_types.py#L728