-
Notifications
You must be signed in to change notification settings - Fork 6
/
MPIderivHelperFuncs.py
112 lines (89 loc) · 3.77 KB
/
MPIderivHelperFuncs.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
import numpy as np
import FFTHelperFuncs
def MPIderiv2(comm,var,dim):
"""Returns first derivative (2-point central finite difference)
of a 3-dimensional, real space, uniform grid (with L = 1) variable.
Assumes that the field is split on axis 0 between processes.
Args:
comm -- MPI world communicator
var -- input field
dim -- axis along the derivative should be taken
"""
rank = comm.Get_rank()
size = comm.Get_size()
sl_m1 = slice(None,-2,None)
sl_p1 = slice(2,None,None)
sl_c = slice(1,-1,None)
ds = 2.0/float(var.shape[2]) # assumes axis 2 covers entire grid with L = 1
N = np.array(FFTHelperFuncs.FFT.global_shape(), dtype=int)
loc_slc = FFTHelperFuncs.local_shape
n_proc = N // loc_slc
if (dim == 0):
next_proc = (rank + n_proc[1]) % (n_proc[0] * n_proc[1])
prev_proc = (rank - n_proc[1]) % (n_proc[0] * n_proc[1])
# send right slice of local proc as left slab to follow proc
leftSlice = None
leftSlice = comm.sendrecv(sendobj=var[-1:,:,:],dest=next_proc,source=prev_proc)
# send left slice of local proc as right slab to follow proc
rightSlice = None
rightSlice = comm.sendrecv(sendobj=var[:1,:,:],dest=prev_proc,source=next_proc)
tmp = np.concatenate((leftSlice,var,rightSlice),axis=0)
p1 = tmp[sl_p1,:,:]
m1 = tmp[sl_m1,:,:]
elif (dim == 1):
next_proc = (rank + 1) % n_proc[1] + (rank // n_proc[1]) * n_proc[1]
prev_proc = (rank - 1) % n_proc[1] + (rank // n_proc[1]) * n_proc[1]
# send right slice of local proc as left slab to follow proc
leftSlice = None
leftSlice = comm.sendrecv(sendobj=var[:,-1:,:],dest=next_proc,source=prev_proc)
# send left slice of local proc as right slab to follow proc
rightSlice = None
rightSlice = comm.sendrecv(sendobj=var[:,:1,:],dest=prev_proc,source=next_proc)
tmp = np.concatenate((leftSlice,var,rightSlice),axis=1)
p1 = tmp[:,sl_p1,:]
m1 = tmp[:,sl_m1,:]
elif (dim == 2):
# nothing special required here as we do pencil decomp in x-y
tmp = np.concatenate((var[:,:,-1:],var,var[:,:,:1]),axis=2)
p1 = tmp[:,:,sl_p1]
m1 = tmp[:,:,sl_m1]
else:
print("watch out for dimension!")
del tmp
return np.array((p1 - m1)/ds)
def MPIXdotGradYScalar(comm,X,Y):
""" returns (X . grad) Y
"""
return X[0] * MPIderiv2(comm,Y,0) + X[1] * MPIderiv2(comm,Y,1) + X[2] * MPIderiv2(comm,Y,2)
def MPIXdotGradY(comm,X,Y):
""" returns (X . grad) Y
"""
res = np.zeros_like(X)
for i in range(3):
res[i] = (X[0] * MPIderiv2(comm,Y[i],0) + X[1] * MPIderiv2(comm,Y[i],1) + X[2] * MPIderiv2(comm,Y[i],2))
return res
def MPIdivX(comm,X):
""" returns div X = x_dx + y_dy + z_dz
"""
return MPIderiv2(comm,X[0],0) + MPIderiv2(comm,X[1],1) + MPIderiv2(comm,X[2],2)
def MPIdivXY(comm,X,Y):
""" returns pd_j X_j Y_i
"""
res = np.zeros_like(Y)
for i in range(3):
res[i] = MPIderiv2(comm,X[0]*Y[i],0) + MPIderiv2(comm,X[1]*Y[i],1) + MPIderiv2(comm,X[2]*Y[i],2)
return res
def MPIgradX(comm,X):
""" returns grad X = [ x_dx, y_dy, z_dz ]
"""
return np.array([MPIderiv2(comm,X,0),
MPIderiv2(comm,X,1),
MPIderiv2(comm,X,2),
])
def MPIrotX(comm,X):
""" returns curl X = [ z_dy - y_dz, x_dz - z_dx, y_dx - x_dy ]
"""
return np.array([MPIderiv2(comm,X[2],1) - MPIderiv2(comm,X[1],2),
MPIderiv2(comm,X[0],2) - MPIderiv2(comm,X[2],0),
MPIderiv2(comm,X[1],0) - MPIderiv2(comm,X[0],1),
])