def test_initial_version(self, backend, mat, g, x):
assert mat._version == 1
assert g._version == 1
assert x._version == 1
c = op2.Const(1, 1, name='c2', dtype=numpy.uint32)
assert c._version == 1
c.remove_from_namespace()
def test_1d_read(self, backend, set, dat):
kernel = """
void kernel_1d_read(int *x) { *x = myconstant; }
"""
constant = op2.Const(1, 100, dtype=numpy.int32, name="myconstant")
op2.par_loop(op2.Kernel(kernel, "kernel_1d_read"), set, dat(op2.WRITE))
constant.remove_from_namespace()
assert all(dat.data == constant.data)
def test_change_const_dim_matters(self, backend, iterset, diterset):
d = op2.Dat(diterset, range(nelems), numpy.uint32)
self.cache.clear()
assert len(self.cache) == 0
k = op2.Kernel("""void k(unsigned int *x) {}""", 'k')
c = op2.Const(1, 1, name='c', dtype=numpy.uint32)
op2.par_loop(k, iterset, d(op2.WRITE))
op2.base._trace.evaluate(set([d]), set())
assert len(self.cache) == 1
c.remove_from_namespace()
c = op2.Const(2, (1, 1), name='c', dtype=numpy.uint32)
op2.par_loop(k, iterset, d(op2.WRITE))
op2.base._trace.evaluate(set([d]), set())
assert len(self.cache) == 2
c.remove_from_namespace()
def test_change_constant_works(self, backend, set, dat):
k = """
void k(int *x) { *x = myconstant; }
"""
constant = op2.Const(1, 10, dtype=numpy.int32, name="myconstant")
op2.par_loop(op2.Kernel(k, 'k'), set, dat(op2.WRITE))
assert all(dat.data == constant.data)
constant.data == 11
op2.par_loop(op2.Kernel(k, 'k'), set, dat(op2.WRITE))
constant.remove_from_namespace()
assert all(dat.data == constant.data)
def test_change_constant_doesnt_require_parloop_regen(
self, backend, set, dat):
k = """
void k(int *x) { *x = myconstant; }
"""
cache = op2.base.JITModule._cache
cache.clear()
constant = op2.Const(1, 10, dtype=numpy.int32, name="myconstant")
op2.par_loop(op2.Kernel(k, 'k'), set, dat(op2.WRITE))
assert all(dat.data == constant.data)
assert len(cache) == 1
constant.data == 11
op2.par_loop(op2.Kernel(k, 'k'), set, dat(op2.WRITE))
constant.remove_from_namespace()
assert all(dat.data == constant.data)
assert len(cache) == 1
pp[2 * e] = n
pp[2 * e + 1] = i2 - 1 + (j2 - 1) * (NN - 1)
A[e] = 0.25
e += 1
nodes = op2.Set(nnode, "nodes")
edges = op2.Set(nedge, "edges")
ppedge = op2.Map(edges, nodes, 2, pp, "ppedge")
p_A = op2.Dat(edges, data=A, name="p_A")
p_r = op2.Dat(nodes, data=r, name="p_r")
p_u = op2.Dat(nodes, data=u, name="p_u")
p_du = op2.Dat(nodes, data=du, name="p_du")
alpha = op2.Const(1, data=1.0, name="alpha", dtype=fp_type)
beta = op2.Global(1, data=1.0, name="beta", dtype=fp_type)
res = op2.Kernel(
"""void res(%(t)s *A, %(t)s *u, %(t)s *du, const %(t)s *beta){
*du += (*beta)*(*A)*(*u);
}""" % {'t': "double" if fp_type == np.float64 else "float"}, "res")
update = op2.Kernel(
"""
void update(%(t)s *r, %(t)s *du, %(t)s *u, %(t)s *u_sum, %(t)s *u_max) {
*u += *du + alpha * (*r);
*du = %(z)s;
*u_sum += (*u)*(*u);
*u_max = *u_max > *u ? *u_max : *u;
def main(opt):
from aero_kernels import dirichlet, dotPV, dotR, init_cg, res_calc, spMV, \
update, updateP, updateUR
try:
with h5py.File(opt['mesh'], 'r') as f:
# sets
nodes = op2.Set.fromhdf5(f, 'nodes')
bnodes = op2.Set.fromhdf5(f, 'bedges')
cells = op2.Set.fromhdf5(f, 'cells')
# maps
pbnodes = op2.Map.fromhdf5(bnodes, nodes, f, 'pbedge')
pcell = op2.Map.fromhdf5(cells, nodes, f, 'pcell')
pvcell = op2.Map.fromhdf5(cells, nodes, f, 'pcell')
# dats
p_xm = op2.Dat.fromhdf5(nodes ** 2, f, 'p_x')
p_phim = op2.Dat.fromhdf5(nodes, f, 'p_phim')
p_resm = op2.Dat.fromhdf5(nodes, f, 'p_resm')
p_K = op2.Dat.fromhdf5(cells ** 16, f, 'p_K')
p_V = op2.Dat.fromhdf5(nodes, f, 'p_V')
p_P = op2.Dat.fromhdf5(nodes, f, 'p_P')
p_U = op2.Dat.fromhdf5(nodes, f, 'p_U')
except IOError:
import sys
print "Failed reading mesh: Could not read from %s\n" % opt['mesh']
sys.exit(1)
# Constants
gam = 1.4
gm1 = op2.Const(1, gam - 1.0, 'gm1', dtype=np.double)
op2.Const(1, 1.0 / gm1.data, 'gm1i', dtype=np.double)
op2.Const(2, [0.5, 0.5], 'wtg1', dtype=np.double)
op2.Const(2, [0.211324865405187, 0.788675134594813], 'xi1',
dtype=np.double)
op2.Const(4, [0.788675134594813, 0.211324865405187,
0.211324865405187, 0.788675134594813],
'Ng1', dtype=np.double)
op2.Const(4, [-1, -1, 1, 1], 'Ng1_xi', dtype=np.double)
op2.Const(4, [0.25] * 4, 'wtg2', dtype=np.double)
op2.Const(16, [0.622008467928146, 0.166666666666667,
0.166666666666667, 0.044658198738520,
0.166666666666667, 0.622008467928146,
0.044658198738520, 0.166666666666667,
0.166666666666667, 0.044658198738520,
0.622008467928146, 0.166666666666667,
0.044658198738520, 0.166666666666667,
0.166666666666667, 0.622008467928146],
'Ng2', dtype=np.double)
op2.Const(32, [-0.788675134594813, 0.788675134594813,
-0.211324865405187, 0.211324865405187,
-0.788675134594813, 0.788675134594813,
-0.211324865405187, 0.211324865405187,
-0.211324865405187, 0.211324865405187,
-0.788675134594813, 0.788675134594813,
-0.211324865405187, 0.211324865405187,
-0.788675134594813, 0.788675134594813,
-0.788675134594813, -0.211324865405187,
0.788675134594813, 0.211324865405187,
-0.211324865405187, -0.788675134594813,
0.211324865405187, 0.788675134594813,
-0.788675134594813, -0.211324865405187,
0.788675134594813, 0.211324865405187,
-0.211324865405187, -0.788675134594813,
0.211324865405187, 0.788675134594813],
'Ng2_xi', dtype=np.double)
minf = op2.Const(1, 0.1, 'minf', dtype=np.double)
op2.Const(1, minf.data ** 2, 'm2', dtype=np.double)
op2.Const(1, 1, 'freq', dtype=np.double)
op2.Const(1, 1, 'kappa', dtype=np.double)
op2.Const(1, 0, 'nmode', dtype=np.double)
op2.Const(1, 1.0, 'mfan', dtype=np.double)
niter = 20
for i in xrange(1, niter + 1):
op2.par_loop(res_calc, cells,
p_xm(op2.READ, pvcell),
p_phim(op2.READ, pcell),
p_K(op2.WRITE),
p_resm(op2.INC, pcell))
op2.par_loop(dirichlet, bnodes,
p_resm(op2.WRITE, pbnodes[0]))
c1 = op2.Global(1, data=0.0, name='c1')
c2 = op2.Global(1, data=0.0, name='c2')
c3 = op2.Global(1, data=0.0, name='c3')
# c1 = R' * R
op2.par_loop(init_cg, nodes,
p_resm(op2.READ),
c1(op2.INC),
p_U(op2.WRITE),
p_V(op2.WRITE),
p_P(op2.WRITE))
# Set stopping criteria
res0 = sqrt(c1.data)
res = res0
res0 *= 0.1
it = 0
maxiter = 200
while res > res0 and it < maxiter:
# V = Stiffness * P
op2.par_loop(spMV, cells,
p_V(op2.INC, pcell),
p_K(op2.READ),
p_P(op2.READ, pcell))
op2.par_loop(dirichlet, bnodes,
p_V(op2.WRITE, pbnodes[0]))
c2.data = 0.0
# c2 = P' * V
op2.par_loop(dotPV, nodes,
p_P(op2.READ),
p_V(op2.READ),
c2(op2.INC))
alpha = op2.Global(1, data=c1.data / c2.data, name='alpha')
# U = U + alpha * P
# resm = resm - alpha * V
op2.par_loop(updateUR, nodes,
p_U(op2.INC),
p_resm(op2.INC),
p_P(op2.READ),
p_V(op2.RW),
alpha(op2.READ))
c3.data = 0.0
# c3 = resm' * resm
op2.par_loop(dotR, nodes,
p_resm(op2.READ),
c3(op2.INC))
beta = op2.Global(1, data=c3.data / c1.data, name="beta")
# P = beta * P + resm
op2.par_loop(updateP, nodes,
p_resm(op2.READ),
p_P(op2.RW),
beta(op2.READ))
c1.data = c3.data
res = sqrt(c1.data)
it += 1
rms = op2.Global(1, data=0.0, name='rms')
# phim = phim - Stiffness \ Load
op2.par_loop(update, nodes,
p_phim(op2.RW),
p_resm(op2.WRITE),
p_U(op2.READ),
rms(op2.INC))
print "rms = %10.5e iter: %d" % (sqrt(rms.data) / sqrt(nodes.size), it)