Using 2D grid in cudaKronprod for state initialization

[sacpis]

Using 2D grid in cudaKronprod to parallelize over both state dimensions.

Fixes QA bug 5961696.

The bug is in kronprod which calls cudaKronprod  as a 1D grid. The kernel uses blockIdx.y / gridDim.y to go over the user state (for n = 30 qubits, it will be 1 billion elments), which is tsize2 parameter. Now, when gridDim.y = 1, only the first block ran. It went over all 2^n elements sequentially. Other blocks just launched and exited.

The 2D grid shows that there are 65535 blocks each handling 4 chunks in parallel.

Used the code below

import cudaq
import numpy as np

cudaq.set_target("nvidia")

@cudaq.kernel
def kernel(vec : cudaq.State):
    p = cudaq.qubit()
    q = cudaq.qvector(vec)

    mz(p)
    mz(q)

n = 30
v = np.zeros(2**n, dtype=cudaq.complex())
v[-1] = 1.
state = cudaq.State.from_data(v)

print(cudaq.sample(kernel, state))

This change now executes the kernel within 1 minute.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.

[github-actions]

CUDA Quantum Docs Bot: A preview of the documentation can be found here.