def _saveAutocorrelation(self, autocorrelation_function, number_modes, x_coordinates, y_coordinates, filename):
new_coordinates_x = self.determineNewCoordinates(x_coordinates, self._hard_x_min, self._hard_x_max)
new_coordinates_y = self.determineNewCoordinates(y_coordinates, self._hard_y_min, self._hard_y_max)
twoform = autocorrelation_function.Twoform()
eigenvalues = twoform.eigenvalues().copy()
distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=number_modes, n_columns=len(new_coordinates_x)*len(new_coordinates_y))
if distribution_plan.myRank() == 0:
twoform_vectors = TwoformVectorsWavefronts(new_coordinates_x, new_coordinates_y, filename)
diagonal_elements, tde = self._getDiagonalElements(eigenvalues, twoform_vectors, distribution_plan)
new_twoform = Twoform(new_coordinates_x,
new_coordinates_y,
diagonal_elements,
eigenvalues,
twoform_vectors)
autocorrelation_function._setTwoform(new_twoform)
autocorrelation_function.info().set("propagation_seperator", self.seperator())
autocorrelation_function.info().set("propagation", self.log())
autocorrelation_function.save(filename)
return autocorrelation_function
def _saveAutocorrelation(self, autocorrelation_function, number_modes,
x_coordinates, y_coordinates, filename):
new_coordinates_x = self.determineNewCoordinates(
x_coordinates, self._hard_x_min, self._hard_x_max)
new_coordinates_y = self.determineNewCoordinates(
y_coordinates, self._hard_y_min, self._hard_y_max)
twoform = autocorrelation_function.Twoform()
eigenvalues = twoform.eigenvalues().copy()
distribution_plan = DistributionPlan(mpi.COMM_WORLD,
n_rows=number_modes,
n_columns=len(new_coordinates_x) *
len(new_coordinates_y))
if distribution_plan.myRank() == 0:
twoform_vectors = TwoformVectorsWavefronts(new_coordinates_x,
new_coordinates_y,
filename)
diagonal_elements, tde = self._getDiagonalElements(
eigenvalues, twoform_vectors, distribution_plan)
new_twoform = Twoform(new_coordinates_x, new_coordinates_y,
diagonal_elements, eigenvalues,
twoform_vectors)
autocorrelation_function._setTwoform(new_twoform)
autocorrelation_function.info().set("propagation_seperator",
self.seperator())
autocorrelation_function.info().set("propagation", self.log())
autocorrelation_function.save(filename)
return autocorrelation_function
def _createIterationMatrices(self, H, Q, A, number_eigenvectors):
if Q is None or H is None:
start_index = 1
m = number_eigenvectors * 2 + 1
q = self.randomVector(A.totalShape()[0])
q /= norm(q)
distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=m, n_columns=A.totalShape()[0])
Q = ParallelMatrix(distribution_plan)
H = np.zeros((m, m), dtype=np.complex128)
if 0 in Q.localRows():
Q.setRow(0, q)
else:
start_index = Q.totalShape()[0]
m = start_index + number_eigenvectors + 1
new_distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=m, n_columns=A.totalShape()[0])
# if new_distribution_plan.totalShape()[0] <= Q.distributionPlan().totalShape()[0]:
# new_distribution_plan = Q.distributionPlan()
Q = Q.enlargeTo(new_distribution_plan)
H = self.resizeCopy(H, m, m)
return H, Q, start_index, m
def testIndices(self):
plan = DistributionPlan(mpi.COMM_WORLD, 10, 10)
global_index = 5
local_index = plan.globalToLocalIndex(global_index)
result = plan.localToGlobalIndex(local_index)
self.assertEqual(global_index, result)
def __init__(self, N_e, sigma_matrix, weighted_fields, x_coordinates, y_coordinates, k, number_modes):
log("Setting up autocorrelation operator")
self._action = 0
self._builder = AutocorrelationBuilder(N_e, sigma_matrix, weighted_fields, x_coordinates, y_coordinates, k, strategy=BuilderStrategyPython)
self._distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD, n_columns=self.dimensionSize(), n_rows=self.dimensionSize())
log("Setting up PETSc interface")
self._petSc_operator = PetScOperator(self)
self._number_modes = number_modes
mpi.COMM_WORLD.barrier()
def evaluateAllR_2_Fredholm_parallel_direct(self, v_in, v_out):
if not self._has_phases:
self._setUpPhases()
v = v_in.fullData().copy()
H = np.zeros_like(self._field)
tmp2 = np.zeros((self._field.shape[0] * self._field.shape[1]), dtype=np.complex128)
result = np.zeros_like(self._field)
tmp = np.zeros( (self._field_y_coordinates.shape[0], self._field.shape[0] * self._field.shape[1]), dtype=np.complex128)
e_1 = np.zeros_like(self._field)
e_2 = np.zeros_like(self._field)
distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD, n_rows=len(self._field_x_coordinates), n_columns=len(self._field_y_coordinates))
local_rows = distribution_plan.localRows()
range_y_coordinates = tuple(range(len(self._field_y_coordinates)))
for i_field in range(self.numberFields()):
self._setActiveField(i_field)
for i_r_x in local_rows:
i_x_1 = self._coordinate_map_x[i_r_x]
i_x_minus_1 = self._coordinate_map_minus_x[i_r_x]
for i_r_y in range_y_coordinates:
i_y_1 = self._coordinate_map_y[i_r_y]
i_y_minus_1 = self._coordinate_map_minus_y[i_r_y]
self.relativeShiftedCopy(i_x_1, i_y_1, self._field_conj, e_1)
tmp[i_r_y, :] = e_1.ravel()
tmp[i_r_y, :] *= self._rho_phase_exp_y[i_y_minus_1, :, :].ravel()
tmp2[:] = v * self._rho_phase_exp_x[i_x_minus_1, :, :].ravel()
H[i_r_x, :] = tmp.dot(tmp2)
v_out.sumFullData(H.ravel())
field_product = self._rho.flatten() * v_out.fullData()
for i_r_x in local_rows:
i_x_2 = self._coordinate_map_x[i_r_x]
for i_r_y in range_y_coordinates:
i_y_2 = self._coordinate_map_y[i_r_y]
self.relativeShiftedCopy(i_x_2, i_y_2, self._field, e_2)
tmp[i_r_y, :] = e_2.ravel()
tmp[i_r_y, :] *= self._rho_phase_exp_y[i_y_2, :, :].ravel()
tmp2[:] = field_product * self._rho_phase_exp_x[i_x_2, :, :].ravel()
result[i_r_x, :] += tmp.dot(tmp2)
# it is only dV to first power because the solver normalizes for one integration
result *= self._grid_area
v_out.sumFullData(result.ravel())
def evaluateAllR_2_Fredholm_parallel_convolution(self, v_in, v_out):
f = v_in.fullData().reshape(self._x_coordinates.shape[0], self._y_coordinates.shape[0])
distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD, n_rows=self.numberFields(), n_columns=self.numberFields())
local_rows = distribution_plan.localRows()
res = np.zeros_like(self._field)
for i_field in local_rows:
self._setActiveField(i_field)
scal_prod_action = self._convolution.convolve2D(f, self._field_reverse_conj)
self._field_tmp[:, :] = scal_prod_action * self._rho
res += self._convolution.convolve2D(self._field, self._field_tmp)
res *= self._grid_area
v_out.sumFullData(res.ravel())
def crateParallelMatrixFromLocal(local_matrix):
n_rows = local_matrix.shape[1]
n_columns = local_matrix.shape[0]
plan = DistributionPlan(mpi.COMM_WORLD, n_columns=n_columns, n_rows=n_rows)
parallel_matrix = ParallelMatrix(plan)
parallel_matrix.broadcast(local_matrix.transpose(), root=0)
return parallel_matrix
def __init__(self, twoform):
self._parent = twoform
vector = self._parent.vector(0)
self._n_size = vector.size
self._petsc_matrix = PETSc.Mat().create()
self._petsc_matrix.setSizes([self._n_size, self._n_size])
self._petsc_matrix.setUp()
self._parallel_matrix = ParallelMatrixPETSc(self._petsc_matrix)
plan = self._parallel_matrix.distributionPlan()
self._parent._distribution_plan = plan
self._vector_in = ParallelVector(plan)
self._vector_out = ParallelVector(plan)
self._distribution_plan = DistributionPlan(
communicator=mpi.COMM_WORLD,
n_columns=self.dimensionSize(),
n_rows=self.dimensionSize())
def eigenfunctions(self, matrix, number_modes):
import sys, slepc4py
slepc4py.init(sys.argv)
from petsc4py import PETSc
from slepc4py import SLEPc
E = SLEPc.EPS()
E.create()
E.setOperators(matrix.petScMatrix())
E.setProblemType(SLEPc.EPS.ProblemType.HEP)
#E.setType(SLEPc.EPS.Type.ARNOLDI)
E.setFromOptions()
E.setTolerances(tol=1e-9, max_it=200)
E.setDimensions(nev=number_modes)
E.solve()
Print = PETSc.Sys.Print
iterations = E.getIterationNumber()
self.log("Number of iterations of the method: %d" % iterations)
eps_type = E.getType()
self.log("Solution method: %s" % eps_type)
nev, ncv, mpd = E.getDimensions()
self.log("Number of requested eigenvalues: %d" % nev)
tol, maxit = E.getTolerances()
self.log("Stopping condition: tol=%.4g, maxit=%d" % (tol, maxit))
nconv = E.getConverged()
self.log("Number of converged eigenpairs %d" % nconv)
eigenvalues = np.zeros(nconv, dtype=np.complex128)
result_vector = ParallelVector(matrix.distributionPlan())
plan = DistributionPlan(mpi.COMM_WORLD,
n_columns=matrix.totalShape()[1],
n_rows=nconv)
eigenvectors_parallel = ParallelMatrix(plan)
# Create the results vectors
vr, wr = matrix.petScMatrix().getVecs()
vi, wi = matrix.petScMatrix().getVecs()
#
for i in range(nconv):
k = E.getEigenpair(i, vr, vi)
result_vector.setCollective(vr.getArray())
eigenvalues[i] = k
if i in eigenvectors_parallel.localRows():
eigenvectors_parallel.setRow(i, result_vector.fullData())
return eigenvalues, eigenvectors_parallel
def _createParallelMatrix(self, f_gamma):
log("Building matrix")
return self._createParallelMatrixPETSc(f_gamma)
product_coordinates=self.productCoordinates()
n_coordinates = product_coordinates.shape[0]
distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD,
n_rows=product_coordinates.shape[0],
n_columns=product_coordinates.shape[0])
matrix = ParallelMatrix(distribution_plan=distribution_plan)
if self._mode_element_wise:
for i_row in distribution_plan.localRows():
self._printProgress(n_coordinates, i_row)
r_i = product_coordinates[i_row, :]
for i_column in range(n_coordinates):
r_j = product_coordinates[i_column, :]
value = f_gamma(r_i, r_j)
# TODO
raise NotImplementedError("Can only handle entire rows")
# matrix.setElement(i_row, i_column, value)
else:
for i_row in distribution_plan.localRows():
self._printProgress(len(distribution_plan.localRows()), i_row)
r_i = product_coordinates[i_row, :]
value = f_gamma(r_i)
value = value.reshape(value.size)
matrix.setRow(global_index=i_row,
content=value)
if distribution_plan.communicator().Get_rank() == 0:
log("done")
return matrix
def evaluateAllR_2_Fredholm_parallel_convolution(self, v_in, v_out):
f = v_in.fullData().reshape(self._x_coordinates.shape[0],
self._y_coordinates.shape[0])
distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD,
n_rows=self.numberFields(),
n_columns=self.numberFields())
local_rows = distribution_plan.localRows()
res = np.zeros_like(self._field)
for i_field in local_rows:
self._setActiveField(i_field)
scal_prod_action = self._convolution.convolve2D(
f, self._field_reverse_conj)
self._field_tmp[:, :] = scal_prod_action * self._rho
res += self._convolution.convolve2D(self._field, self._field_tmp)
res *= self._grid_area
v_out.sumFullData(res.ravel())
def __init__(self, communicator, petsc_object):
blocks = petsc_object.getOwnershipRanges()
t = []
for i_block in range(len(blocks) - 1):
t.append(np.arange(blocks[i_block], blocks[i_block + 1]))
self._rows_for_rank = np.array(t)
if "Vec" in str(type(petsc_object)):
n_rows = petsc_object.getSizes()[1]
n_columns = petsc_object.getSizes()[1]
else:
n_rows = petsc_object.getSizes()[0][1]
n_columns = petsc_object.getSizes()[1][1]
DistributionPlan.__init__(self,
communicator=communicator,
n_rows=n_rows,
n_columns=n_columns)
def __init__(self, twoform):
self._parent = twoform
vector = self._parent.vector(0)
self._n_size = vector.size
self._petsc_matrix = PETSc.Mat().create()
self._petsc_matrix.setSizes([self._n_size, self._n_size])
self._petsc_matrix.setUp()
self._parallel_matrix = ParallelMatrixPETSc(self._petsc_matrix)
plan = self._parallel_matrix.distributionPlan()
self._parent._distribution_plan = plan
self._vector_in = ParallelVector(plan)
self._vector_out = ParallelVector(plan)
self._distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD, n_columns=self.dimensionSize(), n_rows=self.dimensionSize())
def __init__(self, x_coordinates, y_coordinates, intensity,
eigenvalues_spatial, eigenvectors_parallel,
phase_space_density, method):
self._method = method
if self._method not in ["accurate", "quick"]:
raise Exception("Unknown divergence action %s" % self._method)
communicator = mpi.COMM_WORLD
n_vectors = eigenvalues_spatial.size
self._intensity = intensity
eigenvalues = eigenvalues_spatial
self._number_modes = n_vectors
self._x_coordinates = x_coordinates
self._y_coordinates = y_coordinates
self._petSc_operator = PetScOperatorDivergence(self)
self._my_distribution_plan = DistributionPlan(
communicator=communicator,
n_rows=n_vectors,
n_columns=self.dimensionSize())
self._prepareEigenvectors(communicator, eigenvectors_parallel)
self._my_eigenvalues = eigenvalues[
self._my_distribution_plan.localRows()]
self._my_eigenvectors_conjugated = self._my_eigenvectors.conj()
self._my_eigenvectors_times_eigenvalues = self._my_eigenvectors
self._my_eigenvectors = None
for i_e, e in enumerate(self._my_eigenvalues):
self._my_eigenvectors_times_eigenvalues[i_e, :, :] *= e
self._phase_space_density = phase_space_density
self._sigma_p_x = phase_space_density.divergencePartSigmaX()
self._sigma_p_y = phase_space_density.divergencePartSigmaY()
self._prefactor = phase_space_density.normalizationConstant()
log("Divergence action sigma x/y: %e %e" %
(self._sigma_p_x, self._sigma_p_y))
x_coordinates_weights = x_coordinates[
(x_coordinates > -5 * self._sigma_p_x)
& (x_coordinates < 5 * self._sigma_p_x)]
y_coordinates_weights = y_coordinates[
(y_coordinates > -5 * self._sigma_p_y)
& (y_coordinates < 5 * self._sigma_p_y)]
log("Calculating phase space density xy")
weight_function = np.zeros(
(x_coordinates_weights.shape[0], y_coordinates_weights.shape[0]),
dtype=np.complex128)
for i_x, x in enumerate(x_coordinates_weights):
for i_y, y in enumerate(y_coordinates_weights):
weight_function[i_x, i_y] = phase_space_density.staticPart(
np.array([x, y]))
weight_function_horizontal = np.zeros((x_coordinates.shape[0]),
dtype=np.complex128)
weight_function_vertical = np.zeros((y_coordinates.shape[0]),
dtype=np.complex128)
log("Calculating phase space density x")
for i_x, x in enumerate(x_coordinates_weights):
weight_function_horizontal[i_x] = phase_space_density.staticPart(
np.array([x, 0.0]))
log("Calculating phase space density y")
for i_y, y in enumerate(y_coordinates_weights):
weight_function_vertical[i_y] = phase_space_density.staticPart(
np.array([0.0, y]))
#plot(x_coordinates, weight_function_horizontal)
#plot(y_coordinates, weight_function_vertical)
self._weight_function = weight_function
self._weight_function_horizontal = weight_function_horizontal
self._weight_function_vertical = weight_function_vertical
self._i_action = 0
self._convolution = Convolution()
def arnoldi(self, A, n = 25, accuracy=1e-8, accuracy_projection=None):
n = min(A.totalShape()[0], n)
# H: Hessenbergmatrix
# Q: Schurbasis
H = None
Q = None
my_rank = A.distributionPlan().myRank()
for i in range(5):
H, Q = self.arnoldi_iteration(H, Q, A, n)
r = np.linalg.eigh(H)
eig_val = r[0][::-1]
eig_vec = r[1].transpose()[::-1, :]
eig_vec = eig_vec.transpose()
schur_vec = np.zeros((A.totalShape()[0], n), dtype=np.complex128)
n = min(H.shape[0], n)
q_distribution_plan = Q.distributionPlan()
qt_distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=Q.totalShape()[1], n_columns=Q.totalShape()[0])
parallel_vector_in = ParallelVector(q_distribution_plan)
parallel_vector_out = ParallelVector(qt_distribution_plan)
for i in range(n):
t = eig_vec[:,i]
full_data = np.append(eig_vec[:,i], np.zeros(Q.totalShape()[0]-eig_vec[:,i].shape[0]))
parallel_vector_in._full_data[:] = full_data
Q.dotForTransposed(parallel_vector_in, parallel_vector_out)
schur_vec[:, i] = parallel_vector_out.fullData()
parallel_vector_out._full_data[:] = schur_vec[:, n-1]
A.dot(parallel_vector_out)
acc = scipy.linalg.norm( parallel_vector_out.fullData()/eig_val.max() - (eig_val[n-1]/eig_val.max()) * schur_vec[:, n-1])
acc2 = np.abs(H[-1, -2] / eig_val[n-2])
if my_rank == 0:
print("Accuracy last Schur/ritz vector for normalized matrix: %e"% acc)
print("Accuracy projection vs smallest eigenvalue: %e"% acc2)
if accuracy_projection is not None:
if acc2 <= accuracy_projection and acc <= accuracy:
if my_rank == 0:
print("Converged")
sys.stdout.flush()
break
else:
if acc <= accuracy:
if my_rank == 0:
print("Converged")
sys.stdout.flush()
break
return eig_val[0:n], schur_vec[:,0:n]
def testIndicesUndergo(self):
plan = DistributionPlan(mpi.COMM_WORLD, 1, 1)
print(plan.localRows())
class AutocorrelationOperator(object):
def __init__(self, N_e, sigma_matrix, weighted_fields, x_coordinates, y_coordinates, k, number_modes):
log("Setting up autocorrelation operator")
self._action = 0
self._builder = AutocorrelationBuilder(N_e, sigma_matrix, weighted_fields, x_coordinates, y_coordinates, k, strategy=BuilderStrategyPython)
self._distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD, n_columns=self.dimensionSize(), n_rows=self.dimensionSize())
log("Setting up PETSc interface")
self._petSc_operator = PetScOperator(self)
self._number_modes = number_modes
mpi.COMM_WORLD.barrier()
def numberModes(self):
return self._number_modes
def action(self, v):
return self._builder._strategy.evaluateAllR_2_Fredholm(v)
#return self._builder.evaluateAllR_2(v)
def xCoordinates(self):
return self._builder.xCoordinates()
def yCoordinates(self):
return self._builder.yCoordinates()
def parrallelLinearOperator(self):
return self
def communicator(self):
return self._distribution_plan.communicator()
def parallelDot(self, v):
v_in = ParallelVector(self._distribution_plan)
v_in.broadcast(v, root=0)
self.dot(v_in, v_in)
return v_in.fullData()
def dot(self, v_in, v_out=None):
self._action += 1
logProgress(self._number_modes, self._action, "Fredholm operator")
return self._builder._strategy.evaluateAllR_2_Fredholm_parallel(v_in, v_out)
def dimensionSize(self):
dimension_size = self._builder._x_coordinates.shape[0] * self._builder._y_coordinates.shape[0]
return dimension_size
def totalShape(self):
dimension_size = self.dimensionSize()
shape = (dimension_size, dimension_size)
return shape
def distributionPlan(self):
return self._distribution_plan
def __rmul__(self, scalar):
return self
def trace(self):
return self._builder.calculateIntensity()
def releaseMemory(self):
pass
def petScMatrix(self):
context = self._petSc_operator
A = PETSc.Mat().createPython([self.dimensionSize(),self.dimensionSize()], context)
A.setUp()
return A
def propagate(self, autocorrelation_function, filename, method='SRW', python_to_be_used="python"):
source_filename = autocorrelation_function._io.fromFile()
try:
source_uid = autocorrelation_function.info().uid()
except:
source_uid = "None"
autocorrelation_function.info().logStart()
logAll("Propagating %s (%s)" % (source_filename, source_uid))
if self._maximum_mode is None:
number_modes = autocorrelation_function.numberModes()
else:
number_modes = self._maximum_mode
if isMaster():
if not os.path.exists("tmp"):
os.mkdir("tmp")
distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=number_modes, n_columns=1)
n_rank = mpi.COMM_WORLD.Get_rank()
x_coordinates = []
y_coordinates = []
for i_mode in distribution_plan.localRows():
for i in range(1):
logAll("%i doing mode index: %i/%i (max mode index: %i)" % (n_rank, i_mode, max(distribution_plan.localRows()), number_modes-1))
if n_rank == 0:
sys.stdout.flush()
wavefront = autocorrelation_function.coherentModeAsWavefront(i_mode)
#wavefront._e_field[np.abs(wavefront._e_field)<0.000001]=0.0
if method == 'SRW':
# CHANGE THIS FOR WOFRY
srw_wavefront = propagateWavefront(self.__srw_beamline,
wavefront,
autocorrelation_function.SRWWavefrontRx(),
autocorrelation_function.SRWWavefrontDRx(),
autocorrelation_function.SRWWavefrontRy(),
autocorrelation_function.SRWWavefrontDRy(), 1.0, 1.0, i_mode,
python_to_be_used=python_to_be_used)
elif method == 'WOFRY':
srw_wavefront = propagateWavefrontWofry(self.__srw_beamline,wavefront,i_mode,python_to_be_used=python_to_be_used)
else:
raise Exception("Method not known: %s"%method)
# norm_mode = trapez2D( np.abs(srw_wavefront.E_field_as_numpy()[0,:,:,0])**2, 1, 1)**0.5
# if norm_mode > 1e2 or np.isnan(norm_mode):
# print("TRY %i AFTER PROPAGATION:" % i, i_mode,norm_mode)
# sys.stdout.flush()
#else:
# break
#if i==19:
# exit()
# if np.any(norm_srw_wavefront > 10):
# exit()
#
# if np.any(norm_wavefront > 10):
# exit()
adjusted_wavefront = self._adjustWavefrontSize(srw_wavefront)
# norm_mode = trapez2D( np.abs(adjusted_wavefront.E_field_as_numpy()[0,:,:,0])**2, 1, 1)**0.5
# if norm_mode > 1e2 or np.isnan(norm_mode):
# print("TRY %i AFTER ADJUSTMENT:" % i, i_mode,norm_mode)
# sys.stdout.flush()
# exit()
# writes a file for every wavefront
TwoformVectorsWavefronts.pushWavefront(filename, adjusted_wavefront, index=i_mode)
#print("Saving wavefront %i" % i_mode)
x_coordinates.append(adjusted_wavefront.absolute_x_coordinates().copy())
y_coordinates.append(adjusted_wavefront.absolute_y_coordinates().copy())
mpi.COMM_WORLD.barrier()
# replace the wavefronts bu the propagated ones
af = self._saveAutocorrelation(autocorrelation_function, number_modes, x_coordinates, y_coordinates, filename)
# convert from one file per wavefront to one big array
af.Twoform().convertToTwoformVectorsEigenvectors()
af.info().setEndTime()
filelist = glob.glob(filename+"*")
for f in filelist:
os.remove(f)
return af
def createDistributionPlan(rows=10, columns=10):
return DistributionPlan(mpi.COMM_WORLD, n_rows=rows, n_columns=columns)
def propagate(self,
autocorrelation_function,
filename,
method='SRW',
python_to_be_used="python"):
source_filename = autocorrelation_function._io.fromFile()
try:
source_uid = autocorrelation_function.info().uid()
except:
source_uid = "None"
autocorrelation_function.info().logStart()
logAll("Propagating %s (%s)" % (source_filename, source_uid))
if self._maximum_mode is None:
number_modes = autocorrelation_function.numberModes()
else:
number_modes = self._maximum_mode
if isMaster():
if not os.path.exists("tmp"):
os.mkdir("tmp")
distribution_plan = DistributionPlan(mpi.COMM_WORLD,
n_rows=number_modes,
n_columns=1)
n_rank = mpi.COMM_WORLD.Get_rank()
x_coordinates = []
y_coordinates = []
for i_mode in distribution_plan.localRows():
for i in range(1):
logAll("%i doing mode index: %i/%i (max mode index: %i)" %
(n_rank, i_mode, max(
distribution_plan.localRows()), number_modes - 1))
if n_rank == 0:
sys.stdout.flush()
wavefront = autocorrelation_function.coherentModeAsWavefront(
i_mode)
#wavefront._e_field[np.abs(wavefront._e_field)<0.000001]=0.0
if method == 'SRW':
# CHANGE THIS FOR WOFRY
srw_wavefront = propagateWavefront(
self.__srw_beamline,
wavefront,
autocorrelation_function.SRWWavefrontRx(),
autocorrelation_function.SRWWavefrontDRx(),
autocorrelation_function.SRWWavefrontRy(),
autocorrelation_function.SRWWavefrontDRy(),
1.0,
1.0,
i_mode,
python_to_be_used=python_to_be_used)
elif method == 'WOFRY':
srw_wavefront = propagateWavefrontWofry(
self.__srw_beamline,
wavefront,
i_mode,
python_to_be_used=python_to_be_used)
else:
raise Exception("Method not known: %s" % method)
# norm_mode = trapez2D( np.abs(srw_wavefront.E_field_as_numpy()[0,:,:,0])**2, 1, 1)**0.5
# if norm_mode > 1e2 or np.isnan(norm_mode):
# print("TRY %i AFTER PROPAGATION:" % i, i_mode,norm_mode)
# sys.stdout.flush()
#else:
# break
#if i==19:
# exit()
# if np.any(norm_srw_wavefront > 10):
# exit()
#
# if np.any(norm_wavefront > 10):
# exit()
adjusted_wavefront = self._adjustWavefrontSize(srw_wavefront)
# norm_mode = trapez2D( np.abs(adjusted_wavefront.E_field_as_numpy()[0,:,:,0])**2, 1, 1)**0.5
# if norm_mode > 1e2 or np.isnan(norm_mode):
# print("TRY %i AFTER ADJUSTMENT:" % i, i_mode,norm_mode)
# sys.stdout.flush()
# exit()
# writes a file for every wavefront
TwoformVectorsWavefronts.pushWavefront(filename,
adjusted_wavefront,
index=i_mode)
#print("Saving wavefront %i" % i_mode)
x_coordinates.append(
adjusted_wavefront.absolute_x_coordinates().copy())
y_coordinates.append(
adjusted_wavefront.absolute_y_coordinates().copy())
mpi.COMM_WORLD.barrier()
# replace the wavefronts bu the propagated ones
af = self._saveAutocorrelation(autocorrelation_function, number_modes,
x_coordinates, y_coordinates, filename)
# convert from one file per wavefront to one big array
af.Twoform().convertToTwoformVectorsEigenvectors()
af.info().setEndTime()
filelist = glob.glob(filename + "*")
for f in filelist:
os.remove(f)
return af
class DivergenceAction(object):
def __init__(self, x_coordinates, y_coordinates, intensity,
eigenvalues_spatial, eigenvectors_parallel,
phase_space_density, method):
self._method = method
if self._method not in ["accurate", "quick"]:
raise Exception("Unknown divergence action %s" % self._method)
communicator = mpi.COMM_WORLD
n_vectors = eigenvalues_spatial.size
self._intensity = intensity
eigenvalues = eigenvalues_spatial
self._number_modes = n_vectors
self._x_coordinates = x_coordinates
self._y_coordinates = y_coordinates
self._petSc_operator = PetScOperatorDivergence(self)
self._my_distribution_plan = DistributionPlan(
communicator=communicator,
n_rows=n_vectors,
n_columns=self.dimensionSize())
self._prepareEigenvectors(communicator, eigenvectors_parallel)
self._my_eigenvalues = eigenvalues[
self._my_distribution_plan.localRows()]
self._my_eigenvectors_conjugated = self._my_eigenvectors.conj()
self._my_eigenvectors_times_eigenvalues = self._my_eigenvectors
self._my_eigenvectors = None
for i_e, e in enumerate(self._my_eigenvalues):
self._my_eigenvectors_times_eigenvalues[i_e, :, :] *= e
self._phase_space_density = phase_space_density
self._sigma_p_x = phase_space_density.divergencePartSigmaX()
self._sigma_p_y = phase_space_density.divergencePartSigmaY()
self._prefactor = phase_space_density.normalizationConstant()
log("Divergence action sigma x/y: %e %e" %
(self._sigma_p_x, self._sigma_p_y))
x_coordinates_weights = x_coordinates[
(x_coordinates > -5 * self._sigma_p_x)
& (x_coordinates < 5 * self._sigma_p_x)]
y_coordinates_weights = y_coordinates[
(y_coordinates > -5 * self._sigma_p_y)
& (y_coordinates < 5 * self._sigma_p_y)]
log("Calculating phase space density xy")
weight_function = np.zeros(
(x_coordinates_weights.shape[0], y_coordinates_weights.shape[0]),
dtype=np.complex128)
for i_x, x in enumerate(x_coordinates_weights):
for i_y, y in enumerate(y_coordinates_weights):
weight_function[i_x, i_y] = phase_space_density.staticPart(
np.array([x, y]))
weight_function_horizontal = np.zeros((x_coordinates.shape[0]),
dtype=np.complex128)
weight_function_vertical = np.zeros((y_coordinates.shape[0]),
dtype=np.complex128)
log("Calculating phase space density x")
for i_x, x in enumerate(x_coordinates_weights):
weight_function_horizontal[i_x] = phase_space_density.staticPart(
np.array([x, 0.0]))
log("Calculating phase space density y")
for i_y, y in enumerate(y_coordinates_weights):
weight_function_vertical[i_y] = phase_space_density.staticPart(
np.array([0.0, y]))
#plot(x_coordinates, weight_function_horizontal)
#plot(y_coordinates, weight_function_vertical)
self._weight_function = weight_function
self._weight_function_horizontal = weight_function_horizontal
self._weight_function_vertical = weight_function_vertical
self._i_action = 0
self._convolution = Convolution()
def _prepareEigenvectors(self, communicator, parallel_eigenvectors):
distribution_plan = self._my_distribution_plan
self._my_eigenvectors = parallel_eigenvectors.localMatrix().reshape(
len(distribution_plan.localRows()), len(self._x_coordinates),
len(self._y_coordinates))
def communicator(self):
return self._distribution_plan.communicator()
def dimension_size(self):
return self._weight_function.size
def parallelDot(self, v):
v_in = ParallelVector(self._distribution_plan)
v_in.broadcast(v, root=0)
self.dot(v_in, v_in)
return v_in.fullData()
def parrallelLinearOperator(self):
return self
def dot_accurate(self, v_in, v_out=None):
if v_out is None:
v_out = v_in
self._i_action += 1
eigenvalues = self._my_eigenvalues
eigenvectors_times_eigenvalues = self._my_eigenvectors_times_eigenvalues
c_eigenvectors = self._my_eigenvectors_conjugated
v_r = v_in.fullData().reshape((self._x_coordinates.shape[0],
self._y_coordinates.shape[0])).copy()
res = np.zeros_like(v_r)
tmp = np.zeros_like(v_r)
logProgress(self._number_modes, self._i_action,
"Divergence action[accurate]")
for i in range(len(eigenvalues)):
tmp[:, :] = c_eigenvectors[i, :, :] * v_r
c = self._convolution.convolve2D(tmp, self._weight_function)
tmp[:, :] = eigenvectors_times_eigenvalues[i, :, :] * c
res[:, :] += tmp
v_out.sumFullData(res.ravel())
def dot_quick(self, v_in, v_out=None):
if v_out is None:
v_out = v_in
self._i_action += 1
eigenvalues = self._my_eigenvalues
eigenvectors_times_eigenvalues = self._my_eigenvectors_times_eigenvalues
c_eigenvectors = self._my_eigenvectors_conjugated
v_r = v_in.fullData().reshape((self._x_coordinates.shape[0],
self._y_coordinates.shape[0])).copy()
res = np.zeros_like(v_r)
tmp = np.zeros_like(v_r)
sigmas = [
self._sigma_p_x /
(self._x_coordinates[1] - self._x_coordinates[0]),
self._sigma_p_y / (self._y_coordinates[1] - self._y_coordinates[0])
]
logProgress(self._number_modes, self._i_action,
"Divergence action[quick]")
for i in range(len(eigenvalues)):
tmp[:, :] = c_eigenvectors[i, :, :] * v_r
t_i = tmp.imag.copy()
tmp[:, :] = gaussian_filter(tmp.real, sigmas)
tmp[:, :] += 1j * gaussian_filter(t_i, sigmas)
tmp[:, :] *= eigenvectors_times_eigenvalues[i, :, :]
res[:, :] += tmp
# no dV because solver normalizes for one integration
normalization = 2 * np.pi * sigmas[0] * sigmas[1] * self._prefactor
res *= normalization
v_out.sumFullData(res.ravel())
def dot(self, v_in, v_out=None):
if self._method == "accurate":
return self.dot_accurate(v_in, v_out)
if self._method == "quick":
return self.dot_quick(v_in, v_out)
raise Exception("No suite able divergence method.")
def apply(self, number_modes=None):
eigenmoder = Eigenmoder(self._x_coordinates, self._y_coordinates)
if number_modes is None:
number_modes = self._number_modes - 2
twoform = eigenmoder.eigenmodes(self, number_modes)
return twoform
def trace(self):
return self._intensity
def dimensionSize(self):
return len(self._x_coordinates) * len(self._y_coordinates)
def totalShape(self):
dimension_size = self.dimensionSize()
shape = (dimension_size, dimension_size)
return shape
def distributionPlan(self):
return self._distribution_plan
def releaseMemory(self):
pass
def petScMatrix(self):
context = self._petSc_operator
A = PETSc.Mat().createPython(
[self.dimensionSize(), self.dimensionSize()], context)
A.setUp()
return A
def evaluateAllR_2_Fredholm_parallel_direct(self, v_in, v_out):
if not self._has_phases:
self._setUpPhases()
v = v_in.fullData().copy()
H = np.zeros_like(self._field)
tmp2 = np.zeros((self._field.shape[0] * self._field.shape[1]),
dtype=np.complex128)
result = np.zeros_like(self._field)
tmp = np.zeros((self._field_y_coordinates.shape[0],
self._field.shape[0] * self._field.shape[1]),
dtype=np.complex128)
e_1 = np.zeros_like(self._field)
e_2 = np.zeros_like(self._field)
distribution_plan = DistributionPlan(
communicator=mpi.COMM_WORLD,
n_rows=len(self._field_x_coordinates),
n_columns=len(self._field_y_coordinates))
local_rows = distribution_plan.localRows()
range_y_coordinates = tuple(range(len(self._field_y_coordinates)))
for i_field in range(self.numberFields()):
self._setActiveField(i_field)
for i_r_x in local_rows:
i_x_1 = self._coordinate_map_x[i_r_x]
i_x_minus_1 = self._coordinate_map_minus_x[i_r_x]
for i_r_y in range_y_coordinates:
i_y_1 = self._coordinate_map_y[i_r_y]
i_y_minus_1 = self._coordinate_map_minus_y[i_r_y]
self.relativeShiftedCopy(i_x_1, i_y_1, self._field_conj,
e_1)
tmp[i_r_y, :] = e_1.ravel()
tmp[i_r_y, :] *= self._rho_phase_exp_y[
i_y_minus_1, :, :].ravel()
tmp2[:] = v * self._rho_phase_exp_x[i_x_minus_1, :, :].ravel()
H[i_r_x, :] = tmp.dot(tmp2)
v_out.sumFullData(H.ravel())
field_product = self._rho.flatten() * v_out.fullData()
for i_r_x in local_rows:
i_x_2 = self._coordinate_map_x[i_r_x]
for i_r_y in range_y_coordinates:
i_y_2 = self._coordinate_map_y[i_r_y]
self.relativeShiftedCopy(i_x_2, i_y_2, self._field, e_2)
tmp[i_r_y, :] = e_2.ravel()
tmp[i_r_y, :] *= self._rho_phase_exp_y[i_y_2, :, :].ravel()
tmp2[:] = field_product * self._rho_phase_exp_x[
i_x_2, :, :].ravel()
result[i_r_x, :] += tmp.dot(tmp2)
# it is only dV to first power because the solver normalizes for one integration
result *= self._grid_area
v_out.sumFullData(result.ravel())
def arnoldi_iteration(self, H, Q, A, number_eigenvectors):
H, Q, start_index, m = self._createIterationMatrices(H, Q, A, number_eigenvectors)
qt_distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=A.totalShape()[0], n_columns=m)
q_distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=m, n_columns=A.totalShape()[0])
parallel_vector = ParallelVector(qt_distribution_plan)
parallel_vector._full_data[:] = Q.globalRow(start_index-1)
for k in range(start_index, m):
A.dot(parallel_vector, parallel_vector)
if k in Q.localRows():
Q.setRow(k, parallel_vector.fullData())
q_k = Q.globalRow(k)
if k == m or True:
for j in range(k):
q_j = Q.cachedGlobalRow(j)
H[j, k-1] = np.vdot(q_j, q_k)
q_k -= H[j, k-1] * q_j
# else:
# pv = ParallelVector(qt_distribution_plan)
# pv2 = ParallelVector(q_distribution_plan)
# pv._full_data[:] = q_k
# Q.dot(pv,pv2,complex_conjugate=True)
# H[:, k-1] = pv2.fullData()[:]
#
# p=H[:, k-1]
# p[k:]=0
# pv2._full_data[:] = p
# Q.dotForTransposed(pv2, pv)
#
# q_k -= pv.fullData()
# H[k, k-1] = norm(q_k)
Q.resetCacheGlobalRow()
if k in Q.localRows():
Q.setRow(k, q_k)
row_data = Q.globalRow(k)
H[k, k-1] = norm(row_data)
norm_row_data = row_data / H[k, k-1]
if k%100==0 and Q.distributionPlan().myRank()==0:
print("Arnoldi iteration: %i/%i"% (k, m-1))
sys.stdout.flush()
# Is invariant null space
if np.abs(H[k, k-1]) < 1e-100:
break
if k in Q.localRows():
Q.setRow(k, norm_row_data)
parallel_vector._full_data[:]= norm_row_data
new_distribution_plan = DistributionPlan(mpi.COMM_WORLD, n_rows=k, n_columns=A.totalShape()[1])
Q = Q.shrinkTo(new_distribution_plan)
return H[0:k,0:k], Q
class TwoformPETSc(object):
def __init__(self, twoform):
self._parent = twoform
vector = self._parent.vector(0)
self._n_size = vector.size
self._petsc_matrix = PETSc.Mat().create()
self._petsc_matrix.setSizes([self._n_size, self._n_size])
self._petsc_matrix.setUp()
self._parallel_matrix = ParallelMatrixPETSc(self._petsc_matrix)
plan = self._parallel_matrix.distributionPlan()
self._parent._distribution_plan = plan
self._vector_in = ParallelVector(plan)
self._vector_out = ParallelVector(plan)
self._distribution_plan = DistributionPlan(communicator=mpi.COMM_WORLD, n_columns=self.dimensionSize(), n_rows=self.dimensionSize())
def mult(self, A, x, y):
xx = x.getArray(readonly=1)
yy = y.getArray(readonly=0)
yy[:] = self._1d_dot(xx)
def _1d_dot(self, xx):
x_2d = self.from1dTo2d(xx)
result_2d = self._parent.dot(x_2d)
return self.from2dTo1d(result_2d)
def from1dTo2d(self, x_1d):
x_2d = x_1d.reshape((len(self.xCoordinates()),
len(self.yCoordinates())))
return x_2d
def from2dTo1d(self, x_2d):
x_1d = x_2d.reshape(len(self.xCoordinates()) * len(self.yCoordinates()))
return x_1d
def xCoordinates(self):
return self._parent.xCoordinates()
def yCoordinates(self):
return self._parent.yCoordinates()
def dimensionSize(self):
return self._n_size
def totalShape(self):
return (self._n_size, self._n_size)
def trace(self):
return self._parent.intensity()
def petScMatrix(self):
context = self
A = PETSc.Mat().createPython([self.dimensionSize(),self.dimensionSize()], context)
A.setUp()
return A
def communicator(self):
return self._distribution_plan.communicator()
def distributionPlan(self):
return self._distribution_plan
def dot(self, v_in, v_out=None):
if v_out is None:
v_out = v_in
v_out.broadcast(self._1d_dot(v_in.fullData()), root=0)
def releaseMemory(self):
pass
def getVecs(self):
return self._petsc_matrix.getVecs()
def diagonalize(self, target_number_modes=50):
print("Doing diagonalization")
new_twoform = Eigenmoder(self.xCoordinates(), self.yCoordinates()).eigenmodes(self, target_number_modes)
new_twoform._eigenvalues /= np.diff(self.xCoordinates())[0] * np.diff(self.yCoordinates())[0]
return new_twoform
from comsyl.utils.Logger import logAll, log
from comsyl.parallel.utils import isMaster, barrier
from socket import gethostname
from comsyl.parallel.DistributionPlan import DistributionPlan
import os
if isMaster():
print("Hello from master")
if isMaster():
if not os.path.exists("tmp"):
os.mkdir("tmp")
logAll("Using LogAll")
log("Using Log")
s_id = str(mpi.COMM_WORLD.Get_rank()) + "_" + gethostname()
print("s_id: ", s_id)
print("str(mpi.COMM_WORLD.Get_rank()): ", str(mpi.COMM_WORLD.Get_rank()))
number_modes = 1000
distribution_plan = DistributionPlan(mpi.COMM_WORLD,
n_rows=number_modes,
n_columns=1)
print(distribution_plan)
f = open("./tmp/TMP%s_in" % s_id, 'w')
f.write(">>>>>>>>>")
f.close
class TwoformPETSc(object):
def __init__(self, twoform):
self._parent = twoform
vector = self._parent.vector(0)
self._n_size = vector.size
self._petsc_matrix = PETSc.Mat().create()
self._petsc_matrix.setSizes([self._n_size, self._n_size])
self._petsc_matrix.setUp()
self._parallel_matrix = ParallelMatrixPETSc(self._petsc_matrix)
plan = self._parallel_matrix.distributionPlan()
self._parent._distribution_plan = plan
self._vector_in = ParallelVector(plan)
self._vector_out = ParallelVector(plan)
self._distribution_plan = DistributionPlan(
communicator=mpi.COMM_WORLD,
n_columns=self.dimensionSize(),
n_rows=self.dimensionSize())
def mult(self, A, x, y):
xx = x.getArray(readonly=1)
yy = y.getArray(readonly=0)
yy[:] = self._1d_dot(xx)
def _1d_dot(self, xx):
x_2d = self.from1dTo2d(xx)
result_2d = self._parent.dot(x_2d)
return self.from2dTo1d(result_2d)
def from1dTo2d(self, x_1d):
x_2d = x_1d.reshape(
(len(self.xCoordinates()), len(self.yCoordinates())))
return x_2d
def from2dTo1d(self, x_2d):
x_1d = x_2d.reshape(
len(self.xCoordinates()) * len(self.yCoordinates()))
return x_1d
def xCoordinates(self):
return self._parent.xCoordinates()
def yCoordinates(self):
return self._parent.yCoordinates()
def dimensionSize(self):
return self._n_size
def totalShape(self):
return (self._n_size, self._n_size)
def trace(self):
return self._parent.intensity()
def petScMatrix(self):
context = self
A = PETSc.Mat().createPython(
[self.dimensionSize(), self.dimensionSize()], context)
A.setUp()
return A
def communicator(self):
return self._distribution_plan.communicator()
def distributionPlan(self):
return self._distribution_plan
def dot(self, v_in, v_out=None):
if v_out is None:
v_out = v_in
v_out.broadcast(self._1d_dot(v_in.fullData()), root=0)
def releaseMemory(self):
pass
def getVecs(self):
return self._petsc_matrix.getVecs()
def diagonalize(self, target_number_modes=50):
print("Doing diagonalization")
new_twoform = Eigenmoder(self.xCoordinates(),
self.yCoordinates()).eigenmodes(
self, target_number_modes)
new_twoform._eigenvalues /= np.diff(self.xCoordinates())[0] * np.diff(
self.yCoordinates())[0]
return new_twoform
