QPMS
/
qpms
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Revert scatsystem cython parallelisation. 

It didn't work correctly on my laptop.
This must be reviewed later, as the waiting is annoying.
This commit is contained in:
Marek Nečada 2021-12-08 14:18:32 +02:00
parent 1e5418e7d0
commit 788a990a29
2 changed files with 21 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

40
qpms/qpms_c.pyx
View File

@ -959,8 +959,8 @@ cdef class ScatteringSystem:
cdef ccart3_t res cdef ccart3_t res
cdef cart3_t pos cdef cart3_t pos
cdef Py_ssize_t i cdef Py_ssize_t i
with nogil, parallel(), boundscheck(False), wraparound(False): with nogil:
for i in prange(evalpos_a.shape[0]): for i in range(evalpos_a.shape[0]):
pos.x = evalpos_a[i,0] pos.x = evalpos_a[i,0]
pos.y = evalpos_a[i,1] pos.y = evalpos_a[i,1]
pos.z = evalpos_a[i,2] pos.z = evalpos_a[i,2]
@ -1000,6 +1000,7 @@ def empty_lattice_modes_xy(EpsMu epsmu, reciprocal_basis, wavevector, double max
free(omegas_c) free(omegas_c)
return omegas return omegas
cdef class _ScatteringSystemAtOmegaK: cdef class _ScatteringSystemAtOmegaK:
''' '''
Wrapper over the C qpms_scatsys_at_omega_k_t structure Wrapper over the C qpms_scatsys_at_omega_k_t structure
@ -1045,8 +1046,8 @@ cdef class _ScatteringSystemAtOmegaK:
cdef ccart3_t res cdef ccart3_t res
cdef cart3_t pos cdef cart3_t pos
cdef Py_ssize_t i cdef Py_ssize_t i
with nogil, wraparound(False), parallel(): with nogil:
for i in prange(evalpos_a.shape[0]): for i in range(evalpos_a.shape[0]):
pos.x = evalpos_a[i,0] pos.x = evalpos_a[i,0]
pos.y = evalpos_a[i,1] pos.y = evalpos_a[i,1]
pos.z = evalpos_a[i,2] pos.z = evalpos_a[i,2]
@ -1089,9 +1090,9 @@ cdef class _ScatteringSystemAtOmegaK:
cdef ccart3_t *res cdef ccart3_t *res
cdef cart3_t pos cdef cart3_t pos
cdef Py_ssize_t i, j cdef Py_ssize_t i, j
with nogil, wraparound(False), parallel(): with nogil:
res = <ccart3_t *> malloc(fecv_size*sizeof(ccart3_t)) res = <ccart3_t *> malloc(fecv_size*sizeof(ccart3_t))
for i in prange(evalpos_a.shape[0]): for i in range(evalpos_a.shape[0]):
pos.x = evalpos_a[i,0] pos.x = evalpos_a[i,0]
pos.y = evalpos_a[i,1] pos.y = evalpos_a[i,1]
pos.z = evalpos_a[i,2] pos.z = evalpos_a[i,2]
@ -1259,18 +1260,17 @@ cdef class _ScatteringSystemAtOmega:
cdef ccart3_t res cdef ccart3_t res
cdef cart3_t pos cdef cart3_t pos
cdef Py_ssize_t i cdef Py_ssize_t i
with wraparound(False), nogil, parallel(): for i in range(evalpos_a.shape[0]):
for i in prange(evalpos_a.shape[0]): pos.x = evalpos_a[i,0]
pos.x = evalpos_a[i,0] pos.y = evalpos_a[i,1]
pos.y = evalpos_a[i,1] pos.z = evalpos_a[i,2]
pos.z = evalpos_a[i,2] if alt:
if alt: res = qpms_scatsysw_scattered_E__alt(self.ssw, btyp_c, &scv_view[0], pos)
res = qpms_scatsysw_scattered_E__alt(self.ssw, btyp_c, &scv_view[0], pos) else:
else: res = qpms_scatsysw_scattered_E(self.ssw, btyp_c, &scv_view[0], pos)
res = qpms_scatsysw_scattered_E(self.ssw, btyp_c, &scv_view[0], pos) results[i,0] = res.x
results[i,0] = res.x results[i,1] = res.y
results[i,1] = res.y results[i,2] = res.z
results[i,2] = res.z
return results.reshape(evalpos.shape) return results.reshape(evalpos.shape)
@boundscheck(False) @boundscheck(False)
@ -1316,9 +1316,9 @@ cdef class _ScatteringSystemAtOmega:
cdef ccart3_t *res cdef ccart3_t *res
cdef cart3_t pos cdef cart3_t pos
cdef Py_ssize_t i, j cdef Py_ssize_t i, j
with nogil, wraparound(False), parallel(): with nogil:
res = <ccart3_t *> malloc(basissize*sizeof(ccart3_t)) # thread-local res = <ccart3_t *> malloc(basissize*sizeof(ccart3_t)) # thread-local
for i in prange(evalpos_a.shape[0]): for i in range(evalpos_a.shape[0]):
pos.x = evalpos_a[i,0] pos.x = evalpos_a[i,0]
pos.y = evalpos_a[i,1] pos.y = evalpos_a[i,1]
pos.z = evalpos_a[i,2] pos.z = evalpos_a[i,2]

2
setup.py
View File

@ -150,7 +150,7 @@ qpms_c = Extension('qpms.qpms_c',
sources = ['qpms/qpms_c.pyx',], sources = ['qpms/qpms_c.pyx',],
libraries = common_libs, libraries = common_libs,
include_dirs=['amos', 'qpms', numpy_includes], include_dirs=['amos', 'qpms', numpy_includes],
extra_link_args=['-fopenmp'], extra_link_args=['-fopenmp'], # needed?
extra_compile_args=['-fopenmp'], extra_compile_args=['-fopenmp'],
) )