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
|
#
# Demonstrates writing joined array with ADIOS.
# Joined Array is a Global Array, where in one dimension we don't need to
# calculate starting offsets for each processes, nor the global dimension.
# Instead, the per-process blocks will be virtually joined together in the
# specified dimension, in the order of the blocks in the output file.
# User should not care of the order of blocks, but it will always be consistent
# ordering across multiple joined array variables as long as every process
# always writes the same number of blocks into each variable.
#
from mpi4py import MPI
import numpy
import adios2
import random
# MPI
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
NSTEPS = 5
NCOLS = 4
with adios2.Stream("joinedArray-py.bp", "w", comm) as fh:
# Either use begin_step/end_step explicitely within a loop, or
# use stepping ADIOS stream, then use fh.current_step() for the step value
vtable = fh._io.define_variable(
"table", numpy.zeros([1], dtype=numpy.double), [adios2.JoinedDim, NCOLS], [], [1, NCOLS]
)
for _ in fh.steps(NSTEPS):
# random size per process, 5..10 each
nrows = random.randint(0, 5) + 5
mytable = numpy.zeros([nrows, NCOLS], dtype=numpy.double)
for row in range(0, nrows):
for col in range(0, NCOLS):
mytable[row, col] = rank * 1.0 + row * 0.1 + col * 0.01
vtable.set_selection([[], [nrows, NCOLS]])
fh.write("table", mytable, [adios2.JoinedDim, NCOLS], [], [nrows, NCOLS])
# Read content:
# bpls -l joinedArray-py.bp/ -D -t
# bpls -l joinedArray-py.bp/ -Dd -n 4 -f "%5.2f"
|
