The array version of the AB coeff calculation still has bugs.

rewriting to C


Former-commit-id: c6582a03a8b73b46156416eb0faa3f866e2638ce
This commit is contained in:
Marek Nečada 2017-05-03 02:22:14 +03:00
parent 2d88a1ef0d
commit 40809ce90a
1 changed files with 53 additions and 54 deletions

View File

@ -554,7 +554,7 @@ cdef class trans_calculator:
# TODO CHECK (and try to cast) INPUT ARRAY TYPES (now is done)
# BIG FIXME: make skalars valid arguments, now r, theta, phi, r_ge_d have to be ndarrays
cdef:
int daxis, saxis, smallaxis, bigaxis, reslen, longest_axis, i, j, d, ax, errval
int daxis, saxis, smallaxis, bigaxis, resnd, longest_axis, i, j, d, ax, errval
np.npy_intp sstride, dstride, longi, longstride
int *local_indices
int *innerloop_shape
@ -565,36 +565,54 @@ cdef class trans_calculator:
char *a_p
char *b_p
# Process the array shapes
baseshape = np.broadcast.shape(r,theta,phi,r_ge_d)
baseshape = np.broadcast(r,theta,phi,r_ge_d).shape # nope, does not work as needed
'''
cdef int r_orignd = r.ndim if hasattr(r, "ndim") else 0
cdef int theta_orignd = theta.ndim if hasattr(theta, "ndim") else 0
cdef int phi_orignd = phi.ndim if hasattr(phi, "ndim") else 0
cdef int r_ge_d_orignd = r_ge_d.ndim if hasattr(r_ge_d, "__len__") else 0
cdef int basend = max(r_orignd, theta_orignd, phi_orignd, r_ge_d_orignd)
baseshape = list()
for d in range(basend):
baseshape.append(max(
r.shape[d+r_orignd-basend] if d+r_orignd-basend >= 0 else 1,
theta.shape[d+theta_orignd-basend] if d+theta_orignd-basend >= 0 else 1,
phi.shape[d+phi_orignd-basend] if d+phi_orignd-basend >= 0 else 1,
r_ge_d.shape[d+r_ge_d_orignd-basend] if d+r_ge_d_orignd-basend >= 0 else 1,
))
baseshape = tuple(baseshape)
'''
if not expand:
reslen = len(baseshape)
if reslen < 2:
resnd = len(baseshape)
if resnd < 2:
raise ValueError('Translation matrix arrays must have at least 2 dimensions!')
daxis = (reslen-2) if destaxis is None else destaxis
saxis = (reslen-1) if srcaxis is None else srcaxis
daxis = (resnd-2) if destaxis is None else destaxis
saxis = (resnd-1) if srcaxis is None else srcaxis
if daxis < 0:
daxis = reslen + daxis
daxis = resnd + daxis
if saxis < 0:
saxis = reslen + saxis
if daxis < 0 or saxis < 0 or daxis >= reslen or saxis >= reslen or daxis == saxis:
saxis = resnd + saxis
if daxis < 0 or saxis < 0 or daxis >= resnd or saxis >= resnd or daxis == saxis:
raise ValueError('invalid axes provided (destaxis = %d, srcaxis = %d, # of axes: %d'
% (daxis, saxis, reslen))
% (daxis, saxis, resnd))
if baseshape[daxis] != 1 or baseshape[saxis] != 1:
raise ValueError('dimension mismatch (input argument dimensions have to be 1 both at'
'destaxis (==%d) and srcaxis (==%d) but are %d and %d' %
(daxis, saxis, baseshape[daxis], baseshape[saxis]))
resultshape = list(baseshape)
else:
reslen = len(baseshape)+2
if destaxis is None:
daxis = -2
if srcaxis is None:
saxis = -1
resnd = len(baseshape)+2
daxis = (resnd-2) if destaxis is None else destaxis
saxis = (resnd-1) if srcaxis is None else srcaxis
print(daxis)
print(saxis)
if daxis < 0:
daxis = reslen + daxis
daxis = resnd + daxis
print(daxis)
if saxis < 0:
saxis = reslen + saxis
if daxis < 0 or saxis < 0 or daxis >= reslen or saxis >= reslen or daxis == saxis:
saxis = resnd + saxis
print(saxis)
if daxis < 0 or saxis < 0 or daxis >= resnd or saxis >= resnd or daxis == saxis:
raise ValueError('invalid axes provided') # TODO better error formulation
resultshape = list(baseshape)
if daxis > saxis:
@ -602,23 +620,26 @@ cdef class trans_calculator:
bigaxis = daxis
else:
smallaxis = daxis
bixagis = saxis
bigaxis = saxis
resultshape.insert(smallaxis,1)
resultshape.insert(bigaxis,1)
r = np.expand_dims(np.expand_dims(r.astype(np.float_, copy=False), smallaxis), bigaxis)
theta = np.expand_dims(np.expand_dims(theta.astype(np.float_, copy=False), smallaxis), bigaxis)
phi = np.expand_dims(np.expand_dims(phi.astype(np.float_, copy=False), smallaxis), bigaxis)
r_ge_d = np.expand_dims(np.expand_dims(r_ge_d(np.bool_, copy=False), smallaxis), bigaxis)
r_ge_d = np.expand_dims(np.expand_dims(r_ge_d.astype(np.bool_, copy=False), smallaxis), bigaxis)
print(baseshape)
print(len(baseshape), resnd,smallaxis, bigaxis)
print(r.shape, theta.shape,phi.shape,r_ge_d.shape)
longestaxis = 0
longest_axis = 0
# FIxME: the whole thing with longest_axis will fail if none is longer than 1
for i in range(reslen):
for i in range(resnd):
if resultshape[i] > resultshape[longest_axis]:
longestaxis = i
innerloop_shape = <int *> malloc(reslen * sizeof(int))
longest_axis = i
innerloop_shape = <int *> malloc(resnd * sizeof(int))
if innerloop_shape == NULL:
abort()
for i in range(reslen):
for i in range(resnd):
innerloop_shape[i] = resultshape[i]
innerloop_shape[longest_axis] = 1 # longest axis will be iterated in the outer (parallelized) loop. Therefore, longest axis, together with saxis and daxis, will not be iterated in the inner loop
resultshape[daxis] = self.c[0].nelem
@ -634,20 +655,20 @@ cdef class trans_calculator:
longstride = a.strides[longest_axis]
# TODO write this in C (as a function) and parallelize there
with nogil: #, parallel(): # FIXME rewrite this part in C
local_indices = <int *> calloc(reslen, sizeof(int))
local_indices = <int *> calloc(resnd, sizeof(int))
if local_indices == NULL: abort()
for longi in range(a.shape[longest_axis]): # outer loop (to be parallelized)
# this might be done also in the inverse order, but this is more 'c-contiguous' way of incrementing the indices
ax = reslen - 1
ax = resnd - 1
while ax >= 0:
# calculate the correct index/pointer for each array used. This can be further optimized from O(reslen * total size of the result array) to O(total size of the result array), but fick that now
# calculate the correct index/pointer for each array used. This can be further optimized from O(resnd * total size of the result array) to O(total size of the result array), but fick that now
r_p = r_c.data + r_c.strides[longest_axis] * longi
theta_p = theta_c.data + theta_c.strides[longest_axis] * longi
phi_p = phi_c.data + phi_c.strides[longest_axis] * longi
r_ge_d_p = r_ge_d_c.data + r_ge_d_c.strides[longest_axis] * longi
a_p = a.data + a.strides[longest_axis] * longi
b_p = b.data + b.strides[longest_axis] * longi
for i in range(reslen):
for i in range(resnd):
if i == longest_axis: continue
r_p += r_c.strides[i] * local_indices[i]
theta_p += theta_c.strides[i] * local_indices[i]
@ -664,14 +685,14 @@ cdef class trans_calculator:
(<double*>r_p)[0], (<double*>theta_p)[0], (<double*>phi_p)[0], <int>((<np.npy_bool*>r_ge_d_p)[0]), J)
if errval: abort()
# increment the last index 'digit' (ax is now reslen-1; we don't have do-while loop in python)
# increment the last index 'digit' (ax is now resnd-1; we don't have do-while loop in python)
local_indices[ax] += 1
while (local_indices[ax] == innerloop_shape[ax] and ax >= 0): # overflow to the next digit but stop when we reach below the last one
local_indices[ax] = 0
ax -= 1
local_indices[ax] += 1
if ax >= 0: # did not overflow, get back to the lowest index
ax = reslen - 1
ax = resnd - 1
@ -684,27 +705,5 @@ cdef class trans_calculator:
continue
free(local_indices)
free(innerloop_shape)
return a, b
# TODO make possible to access the attributes (to show normalization etc)