Performance of get/set field data

[FedericoNegri]

Before submitting the issue

• I have checked for Compatibility issues
• I have searched among the existing issues
• I am using a Python virtual environment

Description of the bug

Some recent changes (f7be394?) seem to have slowed down manipulating field data (by orders of magnitude!).

Steps To Reproduce

This is a minimal script to reproduce the issue. Using uv I could easily run it with different "versions" of PyDPF-Core and compare timings.

Using the latest released version (0.13.6) it takes ~20ms to run. With current main it instead takes >5s.

Note: I'm aware the operations tested here could be done in a different way, by setting the field data and scoping outside of the loop. But still, I feel like this could indicate some bigger problem.

# /// script
# requires-python = "==3.12.9"
# dependencies = [
#     "numpy==2.2.4",
#     "ansys-dpf-core==0.13.6", #fast
#     #"ansys-dpf-core @ git+https://[email protected]/ansys/pydpf-core.git@e2a1cf0caa01d3376e956cbea026590a40ccbf48", #fast
#     #"ansys-dpf-core @ git+https://[email protected]/ansys/pydpf-core.git@f7be394515266766b6208ef4f341a5fdfec558db", #slow
# ]
# ///
"""
Script to measure performance of creating a DPF field.

How to:

1. Install uv if not already installed:
    pip install uv
   See https://docs.astral.sh/uv/getting-started/installation/#standalone-installer

2. Save this code as, say, `create_field.py` and run it:
    uv run create_field.py

3. Change the ansys-dpf-core version in the script header and re-run the script:
    uv run create_field.py
"""

import time
import numpy as np
from ansys.dpf.core import fields_factory

from ansys.dpf import core as dpf

server = dpf.start_local_server(
    context=dpf.AvailableServerContexts.premium,
    config=dpf.AvailableServerConfigs.InProcessServer,
    as_global=True,
)

num_entities = int(1e+6)
field = fields_factory.create_scalar_field(
    num_entities=num_entities, location=dpf.locations.elemental, server=server
)
field.name = "my_field"
field.data = np.zeros(num_entities, dtype=np.int32)
field.scoping.ids = np.zeros(num_entities, dtype=np.int32)

all_indices = np.arange(num_entities)
chunks = np.array_split(all_indices, 200)

start_time = time.perf_counter()
for index, chunk in enumerate(chunks):

    field.data[chunk] = int(index)
    field.scoping.ids[chunk] = chunk

print(
    f"Filling DPF field done in {(time.perf_counter() - start_time):.3f} s"
)

Which Operating System causes the issue?

Windows

Which DPF/Ansys version are you using?

DPF Server 2025.2.pre0

Which Python version causes the issue?

3.12

Installed packages

See script above

[FedericoNegri]

Adding an extra step to that script

for i in range(10):
    op = dpf.operators.math.scale(
        field=field,
        weights=1.5,
    )
    op.run()

    start_time = time.perf_counter()

    _ = op.outputs.field().data

    print(
        f"Getting the data in {(time.perf_counter() - start_time)*1000:.3f} ms"
    )

shows an average time to get the field data of 80ms, compared to ~3ms before!!

@cbellot000 @PProfizi @rafacanton can you please take a look?