Vectorization of python ...getSVD version
Former-commit-id: 47b491c768dd30c86055cbb36ab0f7748a8a5202
This commit is contained in:
Marek Nečada
parent
a662c885d5
commit
4b614d8e0b
|
|
@ -477,9 +477,52 @@ def hexlattice_zsym_getSVD(lMax, TMatrices_om, epsilon_b, hexside, maxlayer, ome
|
||||||
minsvTMlist = np.full((klist.shape[0], onlyNmin),np.nan)
|
minsvTMlist = np.full((klist.shape[0], onlyNmin),np.nan)
|
||||||
|
|
||||||
leftmatrixlist = np.full((klist.shape[0],2,2,nelem,2,2,nelem),np.nan,dtype=complex)
|
leftmatrixlist = np.full((klist.shape[0],2,2,nelem,2,2,nelem),np.nan,dtype=complex)
|
||||||
isNaNlist = np.zeros((klist.shape[0]), dtype=bool)
|
#isNaNlist = np.zeros((klist.shape[0]), dtype=bool)
|
||||||
|
isNaNlist = (k_0*k_0 - klist[:,0]**2 - klist[:,1]**2 < 0)
|
||||||
|
nnlist = np.logical_not(isNaNlist)
|
||||||
|
|
||||||
sbtime = _time_b(verbose, step='Initialization of matrices for SVD for a given list of k\'s')
|
sbtime = _time_b(verbose, step='Initialization of matrices for SVD for a given list of k\'s')
|
||||||
|
|
||||||
|
|
||||||
|
#ki = np.arange(klist.shape[0])[k_0*k_0 - klist[:,0]**2 - klist[:,1]**2 >= 0]
|
||||||
|
k = klist[nnlist]
|
||||||
|
|
||||||
|
phases_self = np.exp(1j*np.tensordot(k,unitcell_translations,axes=(-1,-1)))
|
||||||
|
phases_u2d = np.exp(1j*np.tensordot(k,u2d_translations,axes=(-1,-1)))
|
||||||
|
phases_d2u = np.exp(1j*np.tensordot(k,d2u_translations,axes=(-1,-1)))
|
||||||
|
if gaussianSigma:
|
||||||
|
phases_self *= unitcell_envelope
|
||||||
|
phases_u2d *= u2d_envelope
|
||||||
|
phases_d2u *= d2u_envelope
|
||||||
|
leftmatrix = np.zeros((len(ki),2,2,nelem, 2,2,nelem), dtype=complex)
|
||||||
|
# 0:[u,E<--u,E ]
|
||||||
|
# 1:[d,M<--d,M ]
|
||||||
|
leftmatrix[:,0,0,:,0,0,:] = np.tensordot(a_self,phases_self, axes=(0,-1)) # u2u, E2E
|
||||||
|
leftmatrix[:,1,0,:,1,0,:] = leftmatrix[:,0,0,:,0,0,:] # d2d, E2E
|
||||||
|
leftmatrix[:,0,1,:,0,1,:] = leftmatrix[:,0,0,:,0,0,:] # u2u, M2M
|
||||||
|
leftmatrix[:,1,1,:,1,1,:] = leftmatrix[:,0,0,:,0,0,:] # d2d, M2M
|
||||||
|
leftmatrix[:,0,0,:,0,1,:] = np.tensordot(b_self,phases_self, axes=(0,-1)) # u2u, M2E
|
||||||
|
leftmatrix[:,0,1,:,0,0,:] = leftmatrix[:,0,0,:,0,1,:] # u2u, E2M
|
||||||
|
leftmatrix[:,1,1,:,1,0,:] = leftmatrix[:,0,0,:,0,1,:] # d2d, E2M
|
||||||
|
leftmatrix[:,1,0,:,1,1,:] = leftmatrix[:,0,0,:,0,1,:] # d2d, M2E
|
||||||
|
leftmatrix[:,0,0,:,1,0,:] = np.tensordot(a_d2u, phases_d2u,axes=(0,-1)) #d2u,E2E
|
||||||
|
leftmatrix[:,0,1,:,1,1,:] = leftmatrix[:,0,0,:,1,0,:] #d2u, M2M
|
||||||
|
leftmatrix[:,1,0,:,0,0,:] = np.tensordot(a_u2d, phases_u2d,axes=(0,-1)) #u2d,E2E
|
||||||
|
leftmatrix[:,1,1,:,0,1,:] = leftmatrix[:,1,0,:,0,0,:] #u2d, M2M
|
||||||
|
leftmatrix[:,0,0,:,1,1,:] = np.tensordot(b_d2u, phases_d2u,axes=(0,-1)) #d2u,M2E
|
||||||
|
leftmatrix[:,0,1,:,1,0,:] = leftmatrix[:,0,0,:,1,1,:] #d2u, E2M
|
||||||
|
leftmatrix[:,1,0,:,0,1,:] = np.tensordot(b_u2d, phases_u2d,axes=(0,-1)) #u2d,M2E
|
||||||
|
leftmatrix[:,1,1,:,0,0,:] = leftmatrix[:,1,0,:,0,1,:] #u2d, E2M
|
||||||
|
#leftmatrix is now the translation matrix T
|
||||||
|
for j in range(2):
|
||||||
|
leftmatrix[:,j] = -np.tensordot(TMatrices_om[j], leftmatrix[:,j], axes=([-2,-1],[1,2]))
|
||||||
|
# at this point, jth row of leftmatrix is that of -MT
|
||||||
|
leftmatrix[:,j,:,:,j,:,:] += n2id
|
||||||
|
#now we are done, 1-MT
|
||||||
|
|
||||||
|
leftmatrixlist[nnlist] = leftmatrix
|
||||||
|
|
||||||
|
'''
|
||||||
# sem nějaká rozumná smyčka
|
# sem nějaká rozumná smyčka
|
||||||
for ki in range(klist.shape[0]):
|
for ki in range(klist.shape[0]):
|
||||||
k = klist[ki]
|
k = klist[ki]
|
||||||
|
|
@ -521,8 +564,8 @@ def hexlattice_zsym_getSVD(lMax, TMatrices_om, epsilon_b, hexside, maxlayer, ome
|
||||||
#now we are done, 1-MT
|
#now we are done, 1-MT
|
||||||
|
|
||||||
leftmatrixlist[ki] = leftmatrix
|
leftmatrixlist[ki] = leftmatrix
|
||||||
|
'''
|
||||||
|
|
||||||
nnlist = np.logical_not(isNaNlist)
|
|
||||||
leftmatrixlist_s = np.reshape(leftmatrixlist,(klist.shape[0], 2*2*nelem,2*2*nelem))[nnlist]
|
leftmatrixlist_s = np.reshape(leftmatrixlist,(klist.shape[0], 2*2*nelem,2*2*nelem))[nnlist]
|
||||||
TEč, TMč = symz_indexarrays(lMax, 2)
|
TEč, TMč = symz_indexarrays(lMax, 2)
|
||||||
leftmatrixlist_TE = leftmatrixlist_s[np.ix_(np.arange(leftmatrixlist_s.shape[0]),TEč,TEč)]
|
leftmatrixlist_TE = leftmatrixlist_s[np.ix_(np.arange(leftmatrixlist_s.shape[0]),TEč,TEč)]
|
||||||
|
|
|
||||||
|
|
@ -0,0 +1,130 @@
|
||||||
|
import math
|
||||||
|
import numpy as np
|
||||||
|
cimport numpy as np
|
||||||
|
nx = None
|
||||||
|
|
||||||
|
cdef double _s3 = math.sqrt(3)
|
||||||
|
|
||||||
|
from scipy.constants import c
|
||||||
|
from .timetrack import _time_b, _time_e
|
||||||
|
from .qpms_p import symz_indexarrays
|
||||||
|
from .hexpoints import hexlattice_get_AB
|
||||||
|
|
||||||
|
cpdef hexlattice_zsym_getSVD(int lMax, TMatrices_om, double epsilon_b, double hexside, size_t maxlayer, double omega, klist, gaussianSigma=False, int onlyNmin=0, verbose=False):
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 2] klist_c = klist
|
||||||
|
btime = _time_b(verbose)
|
||||||
|
cdef size_t nelem = lMax * (lMax + 2)
|
||||||
|
_n2id = np.identity(2*nelem)
|
||||||
|
_n2id.shape = (2,nelem,2,nelem)
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 4] n2id = _n2id
|
||||||
|
cdef double nan = float('nan')
|
||||||
|
k_0 = omega * math.sqrt(epsilon_b) / c
|
||||||
|
tdic = hexlattice_get_AB(lMax,k_0*hexside,maxlayer)
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] a_self = tdic['a_self'][:,:nelem,:nelem]
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] b_self = tdic['b_self'][:,:nelem,:nelem]
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] a_u2d = tdic['a_u2d'][:,:nelem,:nelem]
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] b_u2d = tdic['b_u2d'][:,:nelem,:nelem]
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] a_d2u = tdic['a_d2u'][:,:nelem,:nelem]
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] b_d2u = tdic['b_d2u'][:,:nelem,:nelem]
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 2] unitcell_translations = tdic['self_tr']*hexside*_s3
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 2] u2d_translations = tdic['u2d_tr']*hexside*_s3
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 2] d2u_translations = tdic['d2u_tr']*hexside*_s3
|
||||||
|
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 1] unitcell_envelope, u2d_envelope, d2u_envelope
|
||||||
|
if gaussianSigma:
|
||||||
|
sbtime = _time_b(verbose, step='Calculating gaussian envelope')
|
||||||
|
unitcell_envelope = np.exp(-np.sum(tdic['self_tr']**2,axis=-1)/(2*gaussianSigma**2))
|
||||||
|
u2d_envelope = np.exp(-np.sum(tdic['u2d_tr']**2,axis=-1)/(2*gaussianSigma**2))
|
||||||
|
d2u_envelope = np.exp(-np.sum(tdic['d2u_tr']**2,axis=-1)/(2*gaussianSigma**2))
|
||||||
|
_time_e(sbtime, verbose, step='Calculating gaussian envelope')
|
||||||
|
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 3] svUfullTElist, svVfullTElist, svUfullTMlist, svVfullTMlist
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 2] svSfullTElist, svSfullTMlist
|
||||||
|
cdef np.ndarray[np.npy_double, ndim = 2] minsvTElist, minsvTMlist
|
||||||
|
|
||||||
|
#TMatrices_om = TMatrices_interp(omega)
|
||||||
|
if(not onlyNmin):
|
||||||
|
svUfullTElist = np.full((klist_c.shape[0], 2*nelem, 2*nelem), np.nan, dtype=complex)
|
||||||
|
svVfullTElist = np.full((klist_c.shape[0], 2*nelem, 2*nelem), np.nan, dtype=complex)
|
||||||
|
svSfullTElist = np.full((klist_c.shape[0], 2*nelem), np.nan, dtype=complex)
|
||||||
|
svUfullTMlist = np.full((klist_c.shape[0], 2*nelem, 2*nelem), np.nan, dtype=complex)
|
||||||
|
svVfullTMlist = np.full((klist_c.shape[0], 2*nelem, 2*nelem), np.nan, dtype=complex)
|
||||||
|
svSfullTMlist = np.full((klist_c.shape[0], 2*nelem), np.nan, dtype=complex)
|
||||||
|
else:
|
||||||
|
minsvTElist = np.full((klist_c.shape[0], onlyNmin),np.nan)
|
||||||
|
minsvTMlist = np.full((klist_c.shape[0], onlyNmin),np.nan)
|
||||||
|
|
||||||
|
cdef np.ndarray[np.npy_cdouble] leftmatrixlist = np.full((klist_c.shape[0],2,2,nelem,2,2,nelem),np.nan,dtype=complex)
|
||||||
|
cdef np.ndarray[np.npy_bool, ndim=1] isNaNlist = np.zeros((klist_c.shape[0]), dtype=bool)
|
||||||
|
|
||||||
|
sbtime = _time_b(verbose, step='Initialization of matrices for SVD for a given list of k\'s')
|
||||||
|
# sem nějaká rozumná smyčka
|
||||||
|
|
||||||
|
# declarations for the ki loop:
|
||||||
|
cdef size_t ki
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 1] phases_self
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 1] phases_u2d
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim = 1] phases_d2u
|
||||||
|
cdef np.ndarray[np.npy_cdouble, ndim=6] leftmatrix
|
||||||
|
cdef np.ndarray[np.npy_double, ndim=1] k
|
||||||
|
cdef int j
|
||||||
|
|
||||||
|
for ki in range(klist_c.shape[0]):
|
||||||
|
k = klist_c[ki]
|
||||||
|
if (k_0*k_0 - k[0]*k[0] - k[1]*k[1] < 0):
|
||||||
|
isNaNlist[ki] = True
|
||||||
|
continue
|
||||||
|
|
||||||
|
phases_self = np.exp(1j*np.tensordot(k,unitcell_translations,axes=(0,-1)))
|
||||||
|
phases_u2d = np.exp(1j*np.tensordot(k,u2d_translations,axes=(0,-1)))
|
||||||
|
phases_d2u = np.exp(1j*np.tensordot(k,d2u_translations,axes=(0,-1)))
|
||||||
|
if gaussianSigma:
|
||||||
|
phases_self *= unitcell_envelope
|
||||||
|
phases_u2d *= u2d_envelope
|
||||||
|
phases_d2u *= d2u_envelope
|
||||||
|
leftmatrix = np.zeros((2,2,nelem, 2,2,nelem), dtype=complex)
|
||||||
|
# 0:[u,E<--u,E ]
|
||||||
|
# 1:[d,M<--d,M ]
|
||||||
|
leftmatrix[0,0,:,0,0,:] = np.tensordot(a_self,phases_self, axes=(0,-1)) # u2u, E2E
|
||||||
|
leftmatrix[1,0,:,1,0,:] = leftmatrix[0,0,:,0,0,:] # d2d, E2E
|
||||||
|
leftmatrix[0,1,:,0,1,:] = leftmatrix[0,0,:,0,0,:] # u2u, M2M
|
||||||
|
leftmatrix[1,1,:,1,1,:] = leftmatrix[0,0,:,0,0,:] # d2d, M2M
|
||||||
|
leftmatrix[0,0,:,0,1,:] = np.tensordot(b_self,phases_self, axes=(0,-1)) # u2u, M2E
|
||||||
|
leftmatrix[0,1,:,0,0,:] = leftmatrix[0,0,:,0,1,:] # u2u, E2M
|
||||||
|
leftmatrix[1,1,:,1,0,:] = leftmatrix[0,0,:,0,1,:] # d2d, E2M
|
||||||
|
leftmatrix[1,0,:,1,1,:] = leftmatrix[0,0,:,0,1,:] # d2d, M2E
|
||||||
|
leftmatrix[0,0,:,1,0,:] = np.tensordot(a_d2u, phases_d2u,axes=(0,-1)) #d2u,E2E
|
||||||
|
leftmatrix[0,1,:,1,1,:] = leftmatrix[0,0,:,1,0,:] #d2u, M2M
|
||||||
|
leftmatrix[1,0,:,0,0,:] = np.tensordot(a_u2d, phases_u2d,axes=(0,-1)) #u2d,E2E
|
||||||
|
leftmatrix[1,1,:,0,1,:] = leftmatrix[1,0,:,0,0,:] #u2d, M2M
|
||||||
|
leftmatrix[0,0,:,1,1,:] = np.tensordot(b_d2u, phases_d2u,axes=(0,-1)) #d2u,M2E
|
||||||
|
leftmatrix[0,1,:,1,0,:] = leftmatrix[0,0,:,1,1,:] #d2u, E2M
|
||||||
|
leftmatrix[1,0,:,0,1,:] = np.tensordot(b_u2d, phases_u2d,axes=(0,-1)) #u2d,M2E
|
||||||
|
leftmatrix[1,1,:,0,0,:] = leftmatrix[1,0,:,0,1,:] #u2d, E2M
|
||||||
|
#leftmatrix is now the translation matrix T
|
||||||
|
for j in range(2):
|
||||||
|
leftmatrix[j] = -np.tensordot(TMatrices_om[j], leftmatrix[j], axes=([-2,-1],[0,1]))
|
||||||
|
# at this point, jth row of leftmatrix is that of -MT
|
||||||
|
leftmatrix[j,:,:,j,:,:] += n2id
|
||||||
|
#now we are done, 1-MT
|
||||||
|
|
||||||
|
leftmatrixlist[ki] = leftmatrix
|
||||||
|
|
||||||
|
nnlist = np.logical_not(isNaNlist)
|
||||||
|
leftmatrixlist_s = np.reshape(leftmatrixlist,(klist_c.shape[0], 2*2*nelem,2*2*nelem))[nnlist]
|
||||||
|
TEc, TMc = symz_indexarrays(lMax, 2)
|
||||||
|
leftmatrixlist_TE = leftmatrixlist_s[np.ix_(np.arange(leftmatrixlist_s.shape[0]),TEc,TEc)]
|
||||||
|
leftmatrixlist_TM = leftmatrixlist_s[np.ix_(np.arange(leftmatrixlist_s.shape[0]),TMc,TMc)]
|
||||||
|
_time_e(sbtime, verbose, step='Initializing matrices for SVD for a given list of k\'s')
|
||||||
|
|
||||||
|
sbtime = _time_b(verbose, step='Calculating SVDs for a given list of k\'s.')
|
||||||
|
if(not onlyNmin):
|
||||||
|
svUfullTElist[nnlist], svSfullTElist[nnlist], svVfullTElist[nnlist] = np.linalg.svd(leftmatrixlist_TE, compute_uv=True)
|
||||||
|
svUfullTMlist[nnlist], svSfullTMlist[nnlist], svVfullTMlist[nnlist] = np.linalg.svd(leftmatrixlist_TM, compute_uv=True)
|
||||||
|
_time_e(sbtime, verbose, step='Calculating SVDs for a given list of k\'s.')
|
||||||
|
return ((svUfullTElist, svSfullTElist, svVfullTElist), (svUfullTMlist, svSfullTMlist, svVfullTMlist))
|
||||||
|
else:
|
||||||
|
minsvTElist[nnlist] = np.linalg.svd(leftmatrixlist_TE, compute_uv=False)[...,-onlyNmin:]
|
||||||
|
minsvTMlist[nnlist] = np.linalg.svd(leftmatrixlist_TM, compute_uv=False)[...,-onlyNmin:]
|
||||||
|
_time_e(sbtime, verbose, step='Calculating SVDs for a given list of k\'s.')
|
||||||
|
return (minsvTElist, minsvTMlist)
|
||||||
5
setup.py
5
setup.py
|
|
@ -18,7 +18,8 @@ if("LD_LIBRARY_PATH" in os.environ):
|
||||||
print(os.environ['LD_LIBRARY_PATH'].split(':'))
|
print(os.environ['LD_LIBRARY_PATH'].split(':'))
|
||||||
|
|
||||||
qpms_c = Extension('qpms_c',
|
qpms_c = Extension('qpms_c',
|
||||||
sources = ['qpms/qpms_c.pyx','qpms/gaunt.c',#'qpms/gaunt.h','qpms/vectors.h','qpms/translations.h',
|
sources = ['qpms/qpms_c.pyx', #'qpms/hexpoints_c.pyx',
|
||||||
|
'qpms/gaunt.c',#'qpms/gaunt.h','qpms/vectors.h','qpms/translations.h',
|
||||||
# FIXME http://stackoverflow.com/questions/4259170/python-setup-script-extensions-how-do-you-include-a-h-file
|
# FIXME http://stackoverflow.com/questions/4259170/python-setup-script-extensions-how-do-you-include-a-h-file
|
||||||
'qpms/translations.c'],
|
'qpms/translations.c'],
|
||||||
extra_compile_args=['-std=c99','-ggdb','-O3',
|
extra_compile_args=['-std=c99','-ggdb','-O3',
|
||||||
|
|
@ -33,7 +34,7 @@ qpms_c = Extension('qpms_c',
|
||||||
)
|
)
|
||||||
|
|
||||||
setup(name='qpms',
|
setup(name='qpms',
|
||||||
version = "0.2.9",
|
version = "0.2.10",
|
||||||
packages=['qpms'],
|
packages=['qpms'],
|
||||||
# setup_requires=['setuptools_cython'],
|
# setup_requires=['setuptools_cython'],
|
||||||
install_requires=['cython>=0.21','quaternion','spherical_functions','py_gmm'],
|
install_requires=['cython>=0.21','quaternion','spherical_functions','py_gmm'],
|
||||||
|
|
|
||||||
Loading…
Reference in New Issue