Basic benchmark script for a very simple kernel loop

benchmark_kernelloop.py

@@ -0,0 +1,37 @@
+import numpy as np
+import parcels
+
+runtime = np.timedelta64(10, "D")
+dt = np.timedelta64(30, "m")
+
+parcelsv4 = True
+try:
+    from parcels._datasets.structured.generic import datasets as datasets_structured
+except ImportError:
+    from tests.utils import create_fieldset_zeros_unit_mesh
+    parcelsv4 = False
+
+# create fieldset
+if parcelsv4:
+    ds = datasets_structured["ds_2d_left"]
+    grid = parcels.xgrid.XGrid(parcels.xgcm.Grid(ds))
+    U = parcels.Field("U", ds["U (A grid)"], grid, mesh_type="flat")
+    V = parcels.Field("V", ds["V (A grid)"], grid, mesh_type="flat")
+    fieldset = parcels.FieldSet([U, V])
+else:
+    fieldset = create_fieldset_zeros_unit_mesh()
+
+pclass = parcels.Particle if parcelsv4 else parcels.JITParticle
+
+# use a kernel that doesn't involve field access/interpolation
+def DoNothing(particle, fieldset, time):
+    pass
+
+for npart in [1, 10, 100, 1000, 10000]:
+    lon = np.linspace(-10, 10, npart)
+    lat = np.linspace(-30, -20, npart)
+
+    pset = parcels.ParticleSet(fieldset=fieldset, pclass=pclass, lon=lon, lat=lat)
+
+    print(f"Running {len(lon)} particles with parcels v{4 if parcelsv4 else 3}")
+    pset.execute(DoNothing, runtime=runtime, dt=dt)