def test_hycom_advection():
print("Loading Data...")
field = generate_field.hycom_surface(months=list(range(1, 13)))
field.x = np.linspace(min(field.x), max(field.x), len(field.x))
land = np.isnan(field.U[0])
field.U[:, land] = 0
field.V[:, land] = 0
field.time = (
field.time - np.datetime64('1970-01-01T00:00:00')) / np.timedelta64(
1, 's') # convert from numpy datetime to seconds since epoch
# initialize particles
[X, Y] = np.meshgrid(field.x, field.y)
ocean_points = np.array([X[~land.T], Y[~land.T]]).T
num_particles = 5000
p0 = ocean_points[np.random.choice(np.arange(len(ocean_points)),
size=num_particles,
replace=False)]
# initialize advection parameters
t_start = field.time[0]
t_end = field.time[-1]
num_timesteps = 365 * 24
time = np.linspace(t_start, t_end, num_timesteps)
dt = time[1] - time[0]
print(f'dt: {dt/3600: .1f} hours')
save_every = 24
device_index = 2 # amd
P, buf_time, kernel_time = openCL_advect(field,
p0,
t_start,
num_timesteps,
save_every,
dt,
device_index,
verbose=True,
kernel='lat_lon')
P = np.concatenate([p0[:, np.newaxis], P], axis=1)
out_time = time[0] + dt * np.arange(num_timesteps + 1, step=save_every)
# make an animation
proj = ccrs.PlateCarree()
fig = plt.figure(figsize=[14, 8])
ax = plt.axes(projection=proj)
ax.coastlines()
dot, = ax.plot([], [], '.')
def update(i):
dot.set_xdata(P[:, i, 0])
dot.set_ydata(P[:, i, 1])
ax.set_title(datetime.utcfromtimestamp(out_time[i]))
ax.set_ylim(-90, 90)
return dot,
ani = FuncAnimation(fig, update, frames=len(out_time))
#plt.show()
print('saving animation...')
ani.save('plots/hycom_2015.mp4', fps=30)
def test_ocean():
field = generate_field.multiple_timestep_ocean()
# initialize particles
[X, Y] = np.meshgrid(np.linspace(-180, 180, 47), np.linspace(-85, 85, 29))
p0 = np.array([X.flatten(), Y.flatten()]).T
# initialize advection parameters
num_timesteps = 100
save_every = 1
dt = 3600 * 1 # 1 hrs
device_index = 2 # amd
t0 = 0
P, buf_time, kernel_time = openCL_advect(field,
p0,
t0,
num_timesteps,
save_every,
dt,
device_index,
verbose=True,
kernel='lat_lon')
plot_advection(P, dt * np.arange(num_timesteps, step=save_every), field)
return P
def advect_circles(dt, field):
# advect particles
p0 = np.array([[.4, 0]])
num_timesteps = int(10 / dt)
save_every = 1
P, buffer_seconds, kernel_seconds = openCL_advect(
field=field,
p0=p0,
t0=0,
num_timesteps=num_timesteps,
save_every=save_every,
dt=dt,
device_index=0,
verbose=False)
return P[0, :, 0], P[0, :, 1]
def test_hycom():
field = generate_field.hycom_surface(months=list(range(1, 13)) * 2)
land = np.isnan(field.U[0])
field.U[:, land] = 0
field.V[:, land] = 0
field.time = (
field.time - np.datetime64('1970-01-01T00:00:00')) / np.timedelta64(
1, 's') # convert from numpy datetime to seconds since epoch
# initialize particles
[X, Y] = np.meshgrid(field.x, field.y)
ocean_points = np.array([X[~land.T], Y[~land.T]]).T
num_particles = 5000
p0 = ocean_points[np.random.choice(np.arange(len(ocean_points)),
size=num_particles,
replace=False)]
# initialize advection parameters
t_start = field.time[0]
t_end = field.time[-1]
num_timesteps = 2 * 365 * 10
time = np.linspace(t_start, t_end, num_timesteps)
dt = time[1] - time[0]
print(f'dt: {dt/3600: .1f} hours')
save_every = 20
device_index = 2 # amd
P, buf_time, kernel_time = openCL_advect(field,
p0,
t_start,
num_timesteps,
save_every,
dt,
device_index,
verbose=True,
kernel='lat_lon')
P = np.concatenate([p0[:, np.newaxis], P], axis=1)
field.x = np.linspace(min(field.x), max(field.x), len(field.x))
plot_ocean_advection(
P,
np.linspace(t_start, t_end + (t_end - t_start),
num_timesteps // save_every))
return P
def run_opencl(num_particles,
num_timesteps,
save_every=1,
device='amd',
verbose=False):
field = generate_field.converge()
p0 = np.random.rand(num_particles, 2) * [
field.x.max() - field.x.min(),
field.y.max() - field.y.min()
] + [field.x.min(), field.y.min()]
t0 = 0
dt = 1
device_index = {'cpu': 0, 'iris': 1, 'amd': 2}[device]
P, buffer_seconds, kernel_seconds = openCL_advect(field, p0, t0,
num_timesteps,
save_every, dt,
device_index, verbose)
return P, buffer_seconds, kernel_seconds
def test_cartesian_electric_dipole():
field1 = generate_field.electric_dipole(nx=1000)
field2 = generate_field.electric_dipole(nx=1000)
field = Field2D(x=field1.x,
y=field1.y,
time=np.array([0, .5]),
U=np.concatenate([field1.U, -field2.U], axis=0),
V=np.concatenate([field1.V, -field2.V], axis=0))
# advect particles
np.random.seed(2)
p0 = np.random.rand(1000, 2) * [
field.x.max() - field.x.min(),
field.y.max() - field.y.min()
] + [field.x.min(), field.y.min()]
num_timesteps = 600
save_every = 10
dt = 1e-3
P, buffer_seconds, kernel_seconds = openCL_advect(
field=field,
p0=p0,
t0=0,
num_timesteps=num_timesteps,
save_every=save_every,
dt=dt,
device_index=2,
verbose=True)
fig, ax = plt.subplots(1, 1)
ax.set_aspect(1)
dot, = ax.plot([], [], '.', markersize=5, color='C0')
charge1 = ax.add_artist(Circle((-1, 0), 0.05, color='blue'))
charge2 = ax.add_artist(Circle((1, 0), 0.05, color='red'))
ax.streamplot(field.x,
field.y,
field.U[0].T,
field.V[0].T,
linewidth=1,
density=2,
arrowsize=0,
color='gray')
ax.set_xlim([min(field.x), max(field.x)])
ax.set_ylim([min(field.y), max(field.y)])
ax.axes.xaxis.set_visible(False)
ax.axes.yaxis.set_visible(False)
time = dt * np.arange(num_timesteps, step=save_every)
def update(i):
if i == 0:
charge1.set_color('blue')
charge2.set_color('red')
elif i == 24:
charge1.set_color('red')
charge2.set_color('blue')
dot.set_xdata(P[:, i, 0])
dot.set_ydata(P[:, i, 1])
ax.set_title(f't={i}')
ax.set_xlim([min(field.x), max(field.x)])
ax.set_ylim([min(field.y), max(field.y)])
return dot, charge1, charge2
ani = FuncAnimation(fig, update, frames=len(time))
plt.show()
import generate_field import numpy as np from openCL_driver import openCL_advect from plot_advection import plot_advection field = generate_field.converge_diverge() p0 = np.random.rand(500, 2) * [field.x.max() - field.x.min(), field.y.max() - field.y.min()] + [field.x.min(), field.y.min()] dt = 1 device_index = 0 num_timesteps = 100 save_every = 1 verbose = True t0 = field.time[0] P, buffer_seconds, kernel_seconds = openCL_advect(field, p0, t0, num_timesteps, save_every, dt, device_index, verbose) plot_advection(P, np.arange(num_timesteps), field)