def parallel_transform(comm, image_data):
rank = comm.Get_rank()
size = comm.Get_size()
print "transforming for rank i = ", rank
# We assume number of processes is a power of 2
# start and end are row indices
start = rank * int(IMAGE_SIZE / size)
end = ((rank+1) * int(IMAGE_SIZE / size))
# Load partial data
b = data_transformer(SAMPLE_SIZE, IMAGE_SIZE)
data = np.zeros((IMAGE_SIZE, IMAGE_SIZE))
print "about to transform for rank ",rank
for i in xrange(start, end):
print "working on row = ",i
data += b.transform(image_data[i-start], -np.pi/2048*i)
print "transformed image for rank ",rank
if not (rank == 0):
print "sending transformed data for rank ",rank
comm.Send([data, MPI.FLOAT], dest=0)
print "sent transformed data for rank ",rank
return data
def parallel_transform(comm, image_data):
rank = comm.Get_rank()
size = comm.Get_size()
print "transforming for rank i = ", rank
# We assume number of processes is a power of 2
# start and end are row indices
start = rank * int(IMAGE_SIZE / size)
end = ((rank + 1) * int(IMAGE_SIZE / size))
# Load partial data
b = data_transformer(SAMPLE_SIZE, IMAGE_SIZE)
data = np.zeros((IMAGE_SIZE, IMAGE_SIZE))
print "about to transform for rank ", rank
for i in xrange(start, end):
print "working on row = ", i
data += b.transform(image_data[i - start], -np.pi / 2048 * i)
print "transformed image for rank ", rank
if not (rank == 0):
print "sending transformed data for rank ", rank
comm.Send([data, MPI.FLOAT], dest=0)
print "sent transformed data for rank ", rank
return data
# Compute partial transformation
for i in xrange(start, end):
print 'transforming for row ', i
print 'process = ', rank
data += data_trans.transform(image_data[i - start],
-np.pi / IMAGE_SIZE * i)
# Reduce to root process
result = comm.reduce(data, op=MPI.SUM, root=p_root)
return result
if __name__ == '__main__':
dt = np.dtype('d')
data_trans = data_transformer(SAMPLE_SIZE, IMAGE_SIZE)
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
im_data = []
result = np.zeros((IMAGE_SIZE, IMAGE_SIZE))
start_end = {}
for i in xrange(size):
start = i * int(IMAGE_SIZE / size)
end = ((i + 1) * int(IMAGE_SIZE / size))
start_end[i] = (start, end)
print start_end
# Compute partial transformation
for i in xrange(start, end):
print 'transforming for row ',i
print 'process = ',rank
data += data_trans.transform(image_data[i-start], -np.pi/IMAGE_SIZE*i)
# Reduce to root process
result = comm.reduce(data, op=MPI.SUM, root=p_root)
return result
if __name__ == '__main__':
dt = np.dtype('d')
data_trans = data_transformer(SAMPLE_SIZE, IMAGE_SIZE)
comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()
im_data = []
result = np.zeros((IMAGE_SIZE, IMAGE_SIZE))
start_end = {}
for i in xrange(size):
start = i * int(IMAGE_SIZE / size)
end = ((i+1) * int(IMAGE_SIZE / size))
start_end[i] = (start, end)
print start_end