def getGlobalMatrices(order, arch):
architecture = Arch.getArchitectureByIdentifier(arch)
stiffnessMatrices = ['kXiDivM', 'kEtaDivM', 'kZetaDivM']
groups = {'stiffnessTransposed': map(lambda x: x + 'T', stiffnessMatrices)}
if architecture.name in ['knc', 'knl']:
blockMerge = 2
configs = {
'kXiDivM': [],
'kEtaDivM': [],
'kZetaDivM': [],
'stiffnessTransposed': [],
'fP1': [],
'fP2': [],
'fP3': [],
'r1DivM': [],
'r2DivM': [],
'r3DivM': [],
'r4DivM': [],
'rT1': [],
'rT2': [],
'rT3': [],
'rT4': []
}
else:
blockMerge = 1
configs = {
'kXiDivM': [True],
'kEtaDivM': [True],
'kZetaDivM': [True],
'stiffnessTransposed': [True],
'fP1': [True],
'fP2': [True],
'fP3': [True],
'r1DivM': [True],
'r2DivM': [True],
'r3DivM': [True],
'r4DivM': [True],
'rT1': [True],
'rT2': [True],
'rT3': [True],
'rT4': [True]
}
transposedStiffnessBlocks = list()
for o in range(2, order + 1):
stoprow = Tools.numberOfBasisFunctions(o - 1)
startcol = Tools.numberOfBasisFunctions(o - 1)
stopcol = Tools.numberOfBasisFunctions(o)
transposedStiffnessBlocks.append((0, stoprow, startcol, stopcol))
for i in range(blockMerge):
if len(transposedStiffnessBlocks) >= 2:
# merge first blocks
block1 = transposedStiffnessBlocks.pop(0)
block2 = transposedStiffnessBlocks.pop(0)
transposedStiffnessBlocks.insert(0, mergeBlock(block1, block2))
stiffnessBlocks = [(block[2], block[3], block[0], block[1])
for block in transposedStiffnessBlocks]
noMemsetStiffnessBlocks = list()
for i, block in enumerate(stiffnessBlocks):
startrow = noMemsetStiffnessBlocks[i - 1][1] if i > 0 else block[0]
stoprow = architecture.getAlignedIndex(
block[1]) if i != len(stiffnessBlocks) - 1 else block[1]
noMemsetStiffnessBlocks.append((startrow, stoprow, block[2], block[3]))
for matrix in stiffnessMatrices:
configs[matrix].append(stiffnessBlocks)
configs[matrix].append(noMemsetStiffnessBlocks)
if groups.has_key('stiffnessTransposed'):
configs['stiffnessTransposed'].append(transposedStiffnessBlocks)
else:
for matrix in stiffnessMatrices:
configs[matrix + 'T'].append(transposedStiffnessBlocks)
# fP matrices
fPBlocks = list()
for o in range(1, order + 1):
start = Tools.numberOfBasisFunctions2D(o - 1)
stop = Tools.numberOfBasisFunctions2D(o)
fPBlocks.append((start, stop, start, stop))
# merge first three blocks
for i in range(blockMerge + 1):
if len(fPBlocks) >= 2:
block1 = fPBlocks.pop(0)
block2 = fPBlocks.pop(0)
fPBlocks.insert(0, mergeBlock(block1, block2))
for i in range(1, 4):
configs['fP{}'.format(i)].append(fPBlocks)
# rT matrices
rTBlocks = list()
for o in range(1, order + 1):
stoprow = Tools.numberOfBasisFunctions2D(o)
startcol = Tools.numberOfBasisFunctions(o - 1)
stopcol = Tools.numberOfBasisFunctions(o)
rTBlocks.append((0, stoprow, startcol, stopcol))
# merge first three blocks
for i in range(blockMerge + 1):
if len(rTBlocks) >= 2:
block1 = rTBlocks.pop(0)
block2 = rTBlocks.pop(0)
rTBlocks.insert(0, mergeBlock(block1, block2))
rBlocks = [(block[2], block[3], block[0], block[1]) for block in rTBlocks]
noMemsetRBlocks = list()
for i, block in enumerate(rBlocks):
startrow = noMemsetRBlocks[i - 1][1] if i > 0 else block[0]
stoprow = architecture.getAlignedIndex(
block[1]) if i != len(rBlocks) - 1 else block[1]
noMemsetRBlocks.append((startrow, stoprow, block[2], block[3]))
for i in range(1, 5):
configs['r{}DivM'.format(i)].append(rBlocks)
configs['r{}DivM'.format(i)].append(noMemsetRBlocks)
configs['rT{}'.format(i)].append(rTBlocks)
# fMrT and rT have the same sparsity pattern
for i in range(1, 5):
configs['fMrT{}'.format(i)] = copy.deepcopy(configs['rT{}'.format(i)])
return (groups, configs)
