223031012
/
DAMASK_EICMD
mirror of https://github.com/achalhp/DAMASK_EICMD.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

improved performance for large packing values

This commit is contained in:
Martin Diehl 2012-10-26 12:59:43 +00:00
parent 4e5ae4056b
commit b9aa50c59a
1 changed files with 22 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
processing/post/averageDown.py
View File

@ -129,9 +129,12 @@ for file in files:
options.packing[2] = 1
options.shift[2] = 0
dimension[2] = min(dimension[:2]/resolution[:2]) # z spacing equal to smaller of x or y spacing
downSized = numpy.maximum(numpy.ones(3,'i'),resolution//options.packing)
packing = numpy.array(options.packing,'i')
shift = numpy.array(options.shift,'i')
downSized = numpy.maximum(numpy.ones(3,'i'),resolution//packing)
outSize = numpy.ceil(numpy.array(resolution,'d')/numpy.array(packing,'d'))
print '\t%s @ %s --> %s'%(dimension,resolution,downSized)
# ------------------------------------------ assemble header ---------------------------------------
@ -140,40 +143,29 @@ for file in files:
# ------------------------------------------ process data ---------------------------------------
table.data_rewind()
data = numpy.zeros(resolution.tolist()+[len(table.labels)]).reshape(resolution[0],\
resolution[1],\
resolution[2],\
[len(table.labels)])
for z in xrange(resolution[2]):
for y in xrange(resolution[1]):
for x in xrange(resolution[0]):
data = numpy.zeros(outSize.tolist()+[len(table.labels)])
print numpy.shape(data)
p = numpy.zeros(3,'i')
for p[2] in xrange(resolution[2]):
for p[1] in xrange(resolution[1]):
for p[0] in xrange(resolution[0]):
d = ((p-shift)%resolution)//packing
table.data_read()
data[x,y,z,:] = numpy.array(table.data_asFloat(),'d') # convert to numpy array
sum = numpy.zeros(numpy.shape(data))
data[d[0],d[1],d[2],:] += numpy.array(table.data_asFloat(),'d') # convert to numpy array
data /= packing.prod()
data = numpy.roll(data,axis=2,shift=-options.shift[2])
data = numpy.roll(data,axis=1,shift=-options.shift[1])
data = numpy.roll(data,axis=0,shift=-options.shift[0])
for axis2 in xrange(options.packing[2]):
shiftedZ = numpy.roll(data,shift=-axis2,axis=2)
for axis1 in xrange(options.packing[1]):
shiftedZY = numpy.roll(shiftedZ,shift=-axis1,axis=1)
for axis0 in xrange(options.packing[0]):
sum += numpy.roll(shiftedZY,shift=-axis0,axis=0)
averagedDown = sum[::options.packing[0],::options.packing[1],::options.packing[2],::] / options.packing.prod() # normalize data by element count
posOffset = (options.shift+[0.5,0.5,0.5])*dimension/resolution
elementSize = dimension/resolution*options.packing
posOffset = (shift+[0.5,0.5,0.5])*dimension/resolution
elementSize = dimension/resolution*packing
elem = 1
for c in xrange(downSized[2]):
for b in xrange(downSized[1]):
for a in xrange(downSized[0]):
averagedDown[a,b,c,locationCol:locationCol+3] = posOffset + [a,b,c]*elementSize
averagedDown[a,b,c,elemCol] = elem
table.data = averagedDown[a,b,c,:].tolist()
data[a,b,c,locationCol:locationCol+3] = posOffset + [a,b,c]*elementSize
data[a,b,c,elemCol] = elem
table.data = data[a,b,c,:].tolist()
table.data_write() # output processed line
elem += 1