def reduce(self, dst):
dst.tmp_comp[0] = serial_reduce_array(dst.psi >= 0.0, 'sum')
dst.tmp_comp[1] = serial_reduce_array(dst.psi**2, 'sum')
dst.get_carray('tmp_comp').set_data(
parallel_reduce_array(dst.tmp_comp, 'sum'))
epsilon = sqrt(dst.tmp_comp[1] / dst.tmp_comp[0])
if epsilon <= 1e-3:
self.eqn_has_converged = 1
def reduce(self, dst, t, dt):
dst.tmp_comp[0] = serial_reduce_array(dst.compression > 0.0, 'sum')
dst.tmp_comp[1] = serial_reduce_array(dst.compression, 'sum')
dst.tmp_comp[:] = parallel_reduce_array(dst.tmp_comp, 'sum')
if dst.tmp_comp[0] > 0:
avg_rho = dst.tmp_comp[1] / dst.tmp_comp[0]
else:
avg_rho = self.rho0
self.compression = fabs(avg_rho - self.rho0) / self.rho0
def reduce(self, dst):
dst.tmp_comp[0] = serial_reduce_array(dst.array.compression > 0.0, 'sum')
dst.tmp_comp[1] = serial_reduce_array(dst.array.compression, 'sum')
dst.tmp_comp.set_data(parallel_reduce_array(dst.tmp_comp, 'sum'))
if dst.tmp_comp[0] > 0:
comp = dst.tmp_comp[1]/dst.tmp_comp[0]/self.rho0
else:
comp = 0.0
self.compression = comp
def reduce(self, dst):
n = len(dst.x)
tmp_sum_logrho = serial_reduce_array(dst.logrho, 'sum')
sum_logrho = parallel_reduce_array(tmp_sum_logrho, 'sum')
g = exp(sum_logrho / n)
lamda = self.k * numpy.power(g / dst.rho, self.eps)
dst.h[:] = lamda * dst.h0
def reduce(self, d_rho, d_h, d_h0, dst):
n = declare('int')
k = declare('int')
n = len(dst.x)
tmp_sum_logrho = serial_reduce_array(dst.array.logrho, 'sum')
sum_logrho = parallel_reduce_array(tmp_sum_logrho, 'sum')
g = exp(sum_logrho / n)
for k in range(n):
lamda = self.k * pow(g / d_rho[k], self.eps)
d_h[k] = lamda * d_h0[k]
def main():
comm = mpi.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
n = 5
data = np.ones(n) * (rank + 1)
full_data = []
for i in range(size):
full_data = np.concatenate([full_data, np.ones(n) * (i + 1)])
for op in ('sum', 'prod', 'min', 'max'):
serial_data = serial_reduce_array(data, op)
result = mpi_reduce_array(serial_data, op)
expect = getattr(np, op)(full_data)
msg = "For op %s: Expected %s, got %s" % (op, expect, result)
assert expect == result, msg
def reduce(self, dst):
m = serial_reduce_array(dst.m, op='sum')
dst.total_mass[0] = parallel_reduce_array(m, op='sum')
def reduce(self, dst, t, dt):
dst.total_mass[0] = serial_reduce_array(dst.m, op='sum')
if dst.gpu is not None:
dst.gpu.push('total_mass')
def reduce(self, dst):
dst.total_mass[0] = serial_reduce_array(dst.array.m, op='sum')
def py_initialize(self, dst, t, dt):
from numpy import sqrt
vmag = sqrt(dst.u**2 + dst.v**2 + dst.w**2)
dst.vmax[0] = serial_reduce_array(vmag, 'max')
dst.vmax[:] = parallel_reduce_array(dst.vmax, 'max')
"""Test if the mpi_reduce_array function works correctly.
"""
import mpi4py.MPI as mpi
import numpy as np
from pysph.base.reduce_array import serial_reduce_array, mpi_reduce_array
comm = mpi.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
n = 5
data = np.ones(n)*(rank + 1)
full_data = []
for i in range(size):
full_data = np.concatenate([full_data, np.ones(n)*(i+1)])
for op in ('sum', 'prod', 'min', 'max'):
serial_data = serial_reduce_array(data, op)
result = mpi_reduce_array(serial_data, op)
expect = getattr(np, op)(full_data)
msg = "For op %s: Expected %s, got %s"%(op, expect, result)
assert expect == result, msg
def test_reduce_sum_works(self):
x = np.linspace(0, 10, 100)
expect = np.sum(x)
result = serial_reduce_array(x, 'sum')
self.assertAlmostEqual(result, expect)
def test_reduce_prod_works(self):
x = np.linspace(0, 10, 100)
expect = np.prod(x)
result = serial_reduce_array(x, 'prod')
self.assertAlmostEqual(result, expect)
