Performance Regression for Kernel Launching 13.0->14.0

[mawolf2023]

Required prerequisites

• Consult the security policy. If reporting a security vulnerability, do not report the bug using this form. Use the process described in the policy to report the issue.
• Make sure you've read the documentation. Your issue may be addressed there.
• Search the issue tracker to verify that this hasn't already been reported. +1 or comment there if it has.
• If possible, make a PR with a failing test to give us a starting point to work on!

Describe the bug

There seems to be a slowdown in kernel launches between CUDA-Q v13.0 and 14.0(2).

The code below, produced the following timings.

Steps to reproduce the bug

import time
import cudaq
from cudaq import spin

cudaq.set_target("nvidia", option="fp64")

@cudaq.kernel
def trivial(n: int):
    q = cudaq.qvector(n)
    for i in range(n):
        h(q[i])

H = spin.z(0)  # the simplest possible observable

# ---- Warm-up: trigger JIT for both code paths ----
for _ in range(5):
    cudaq.get_state(trivial, 4)
    cudaq.observe(trivial, H, 4).expectation()

N = 1000

# ---- Measure get_state ----
t0 = time.time()
for _ in range(N):
    cudaq.get_state(trivial, 4)
t_state_total = time.time() - t0
t_state_per   = t_state_total / N

# ---- Measure observe ----
t0 = time.time()
for _ in range(N):
    cudaq.observe(trivial, H, 4).expectation()
t_obs_total = time.time() - t0
t_obs_per   = t_obs_total / N

print(f"cudaq:    {cudaq.__version__.split()[2]}")
print(f"target:   nvidia fp64")
print(f"kernel:   trivial 4-qubit Hadamard")
print(f"calls:    {N} per op")
print()
print(f"  get_state  total: {t_state_total:7.2f} s   per call: {t_state_per*1000:7.3f} ms")
print(f"  observe    total: {t_obs_total:7.2f} s   per call: {t_obs_per*1000:7.3f} ms")

Expected behavior

Faster times

Is this a regression? If it is, put the last known working version (or commit) here.

13.0

Environment

• CUDA-Q version: 13.0 compared to 14.2 (and 14.0)
• Python version: 3.11
• Operating system: WSL

Suggestions

No response