def split(first, split_grid):
assert len(first) > 0
lattices = first
gcoor = gpt.coordinates((split_grid, lattices[0].checkerboard()))
lcoor = gpt.coordinates((split_grid, lattices[0].checkerboard()))
assert len(lattices) % split_grid.sranks == 0
return split_lattices(lattices, lcoor, gcoor, split_grid,
len(lattices) // split_grid.sranks)
def split_by_rank(first):
if type(first) != list:
return split_by_rank([first])[0]
assert len(first) > 0
# TODO: split types
lattices = first
grid = lattices[0].grid
mpi_split = [1, 1, 1, 1]
fdimensions = [grid.fdimensions[i] // grid.mpi[i] for i in range(grid.nd)]
split_grid = grid.split(mpi_split, fdimensions)
gcoor = gpt.coordinates(lattices[0])
lcoor = gpt.coordinates((split_grid, lattices[0].checkerboard()))
return split_lattices(lattices, lcoor, gcoor, split_grid, len(lattices))
def __init__(self, grid, local_coordinates):
assert grid.cb.n == 1
self.grid = grid
self.local_coordinates = local_coordinates
# create a minimally embedding lattice geometry
n = len(local_coordinates)
N = self.grid.Nprocessors
l = np.zeros(N, dtype=np.uint64)
l[self.grid.processor] = 2**int(np.ceil(np.log(n) / np.log(2)))
l = grid.globalsum(l)
self.L = [int(np.max(l)) * self.grid.mpi[0]] + self.grid.mpi[1:]
cb_simd_only_first_dimension = gpt.general(1, [0] * grid.nd,
[1] + [0] * (grid.nd - 1))
# create grid as subcommunicator so that sparse domains play nice with split grid
self.embedding_grid = gpt.grid(
self.L,
grid.precision,
cb_simd_only_first_dimension,
None,
self.grid.mpi,
grid,
)
self.embedded_coordinates = np.ascontiguousarray(
gpt.coordinates(self.embedding_grid)[0:n])
self.embedded_cache = {}
self.local_cache = {}
self.coordinate_lattices_cache = None
self.weight_cache = None
self.one_mask_cache = None
def __init__(self, n, precision):
self.n = n
self.fdimensions = [2**n]
self.grid = g.grid(self.fdimensions, precision)
self.verbose = g.default.is_verbose("qis_map")
self.zero_coordinate = (0, ) # |00000 ... 0> state
t = g.timer("map_init")
t("coordinates")
# TODO: need to split over multiple dimensions, single dimension can hold at most 32 bits
self.coordinates = g.coordinates(self.grid)
self.not_coordinates = [
np.bitwise_xor(self.coordinates, 2**i) for i in range(n)
]
for i in range(n):
self.not_coordinates[i].flags["WRITEABLE"] = False
t("masks")
self.one_mask = []
self.zero_mask = []
for i in range(n):
proj = np.bitwise_and(self.coordinates, 2**i)
mask = g.complex(self.grid)
g.coordinate_mask(mask, proj != 0)
self.one_mask.append(mask)
mask = g.complex(self.grid)
g.coordinate_mask(mask, proj == 0)
self.zero_mask.append(mask)
t()
if self.verbose:
g.message(t)
def open_view(self, xk, iview, write, mpi, fdimensions, g_cb, l_cb):
cv = gpt.cartesian_view(iview if iview is not None else -1, mpi,
fdimensions, g_cb, l_cb)
dn, fn = get_local_name(self.root, cv)
loc_desc = cv.describe() + "/" + ("Write" if write else "Read")
tag = "%d-%s" % (xk, str(iview))
tag_pos = "%s-%s-%s-%s" % (tag, str(fdimensions), str(g_cb), str(l_cb))
if loc_desc != self.loc_desc:
self.close_views()
self.loc_desc = loc_desc
if self.verbose:
gpt.message("Switching view to %s" % self.loc_desc)
if tag not in self.loc:
if write and dn is not None:
os.makedirs(dn, exist_ok=True)
self.loc[tag] = gpt.FILE(
fn, "a+b" if write else "rb") if fn is not None else None
if tag_pos not in self.pos:
self.pos[tag_pos] = gpt.coordinates(cv)
return self.loc[tag], self.pos[tag_pos]
def merge_indices(dst, src, st, cache=default_merge_indices_cache):
pos = gpt.coordinates(dst)
assert st is not None
result_otype = st[-1]()
if result_otype is None:
dst @= src
return
ndim = dst.otype.shape[st[0]]
rank = len(st) - 1
islice = [slice(None, None, None) for i in range(len(dst.otype.shape))]
ivec = [0] * rank
cache_key = f"merge_indices_{dst.describe()}_{result_otype.__name__}_{dst.grid.obj}"
tidx = []
src_i = []
for i in range(ndim**rank):
idx = i
for j in range(rank):
c = idx % ndim
islice[st[j]] = c
ivec[j] = c
idx //= ndim
src_i.append(src[tuple(ivec)])
tidx.append(tuple(islice))
if cache_key not in cache:
plan = gpt.copy_plan(dst, src_i)
for i in range(ndim**rank):
plan.destination += dst.view[(pos, ) + tidx[i]]
plan.source += src_i[i].view[:]
cache[cache_key] = plan()
cache[cache_key](dst, src_i)
def coordinate_mask(field, mask):
assert type(mask == numpy.ndarray)
assert field.otype == gpt.ot_singlet
x = gpt.coordinates(field)
field[x] = mask.astype(field.grid.precision.complex_dtype).reshape(
(len(mask), 1))
def perform(self, root):
global basis_size, T, current_config
if current_config is not None and current_config.conf_file != self.conf_file:
current_config = None
if current_config is None:
current_config = config(self.conf_file)
c = None
vcj = [
g.vcolor(current_config.l_exact.U_grid) for jr in range(basis_size)
]
for vcjj in vcj:
vcjj[:] = 0
for tprime in range(T):
basis_evec, basis_evals = g.load(self.basis_fmt %
(self.conf, tprime))
plan = g.copy_plan(vcj[0],
basis_evec[0],
embed_in_communicator=vcj[0].grid)
c = g.coordinates(basis_evec[0])
plan.destination += vcj[0].view[np.hstack(
(c, np.ones((len(c), 1), dtype=np.int32) * tprime))]
plan.source += basis_evec[0].view[c]
plan = plan()
for l in range(basis_size):
plan(vcj[l], basis_evec[l])
for l in range(basis_size):
g.message("Check norm:", l, g.norm2(vcj[l]))
g.save(f"{root}/{self.name}/basis", vcj)
def split(self, mpi_split):
split_grid = self.U_grid.split(mpi_split, self.U_grid.fdimensions)
U_split = [gpt.lattice(split_grid, x.otype) for x in self.U]
pos_split = gpt.coordinates(U_split[0])
for i, x in enumerate(U_split):
x[pos_split] = self.U[i][pos_split]
return self.updated(U_split)
def __init__(self, grid, base, dimensions=None):
self.base = base
if dimensions is None:
dimensions = list(range(grid.nd))
self.fft = g.fft(dimensions)
# create FA mask
cache = {}
self.weight = g.complex(grid)
self.weight[:] = 0
coor = g.coordinates(self.weight)
for mu in dimensions:
c_mu = coor[:, mu].astype(np.complex128)
c_mu_l = g.complex(grid)
c_mu_l[coor, cache] = c_mu
c_mu_l @= g.component.sin(c_mu_l * (np.pi / grid.gdimensions[mu]))
c_mu_l @= c_mu_l * c_mu_l * complex(4.0)
self.weight += c_mu_l
# special consideration for zero
self.weight[0, 0, 0, 0] = (2.0 * np.pi)**2.0 / np.prod(
[grid.gdimensions[mu]
for mu in dimensions])**(2.0 / len(dimensions))
# invert
self.weight @= g.component.inv(self.weight) * complex(
4.0 * len(dimensions))
self.weight = [self.weight]
def sample(self, t, p):
if type(t) == list:
for x in t:
self.sample(x, p)
elif t is None:
return cgpt.random_sample(self.obj, t, p)
elif type(t) == gpt.lattice:
if "pos" in p:
pos = p["pos"]
else:
pos = gpt.coordinates(t)
t0 = gpt.time()
mv = cgpt.random_sample(
self.obj,
pos,
{
**p,
**{
"shape": list(t.otype.shape),
"grid": t.grid.obj,
"precision": t.grid.precision,
},
},
)
t1 = gpt.time()
t[pos] = mv
if self.verbose:
szGB = mv.size * mv.itemsize / 1024.0**3.0
gpt.message("Generated %g GB of random data at %g GB/s" %
(szGB, szGB / (t1 - t0)))
return t
else:
assert 0
def apply_exp_ixp(dst, src, p):
# TODO: add sparse field support (x.internal_coordinates(), x.coordinates())
x = gpt.coordinates(src)
# create phase field
phase = gpt.complex(src.grid)
phase.checkerboard(src.checkerboard())
phase[x] = cgpt.coordinates_momentum_phase(x, p, src.grid.precision)
dst @= phase * src
def split(first,
split_grid,
cache=None,
group_policy=split_group_policy.separate):
assert len(first) > 0
lattices = first
gcoor = gpt.coordinates((split_grid, lattices[0].checkerboard()))
lcoor = gpt.coordinates((split_grid, lattices[0].checkerboard()))
assert len(lattices) % split_grid.sranks == 0
return split_lattices(
lattices,
lcoor,
gcoor,
split_grid,
len(lattices) // split_grid.sranks,
cache,
group_policy,
)
def perform(self, root):
global current_config, current_light_quark
if current_config is not None and current_config.conf_file != self.conf_file:
current_config = None
if current_config is None:
current_config = config(self.conf_file)
if (current_light_quark is not None
and current_light_quark.evec_dir != self.evec_dir):
current_light_quark = None
if current_light_quark is None:
current_light_quark = light_quark(current_config, self.evec_dir)
prop_l = {
"sloppy": current_light_quark.prop_l_sloppy,
"exact": current_light_quark.prop_l_exact,
}[self.solver]
vcj = g.load(f"{root}/{self.conf}/pm_basis/basis")
c = g.coordinates(vcj[0])
c = c[c[:, 3] == self.t]
g.message(
f"t = {self.t}, ilist = {self.ilist}, basis size = {len(vcj)}, solver = {self.solver}"
)
root_job = f"{root}/{self.name}"
output = g.gpt_io.writer(f"{root_job}/propagators")
# create sources
srcD = [
g.vspincolor(current_config.l_exact.U_grid) for spin in range(4)
]
for i in self.ilist:
for spin in range(4):
srcD[spin][:] = 0
srcD[spin][c, spin, :] = vcj[i][c]
g.message("Norm of source:", g.norm2(srcD[spin]))
if i == 0:
g.message("Source at origin:", srcD[spin][0, 0, 0, 0])
g.message("Source at time-origin:", srcD[spin][0, 0, 0,
self.t])
prop = g.eval(prop_l * srcD)
g.mem_report(details=False)
for spin in range(4):
output.write(
{f"t{self.t}s{spin}c{i}_{self.solver}": prop[spin]})
output.flush()
def __init__(self, grid, margin, cb=None):
super().__init__()
if cb is None:
cb = g.none
self.grid = grid
self.cb = cb
dim = grid.nd
self.local_grid = grid.split([1] * dim, [
grid.fdimensions[i] // grid.mpi[i] + 2 * margin[i]
for i in range(dim)
])
self.gcoor = g.coordinates((grid, cb), margin=margin)
self.lcoor = g.coordinates((self.local_grid, cb))
top = np.array(margin, dtype=np.int32)
bottom = np.array(self.local_grid.fdimensions, dtype=np.int32) - top
self.bcoor = np.sum(np.logical_and(self.lcoor >= top,
self.lcoor < bottom),
axis=1) == len(top)
def perform(self, root):
global basis_size, sloppy_per_job, T, current_config, compress_ratio
if current_config is not None and current_config.conf_file != self.conf_file:
current_config = None
if current_config is None:
current_config = config(self.conf_file)
U = current_config.U
reduced_mpi = [x for x in U[0].grid.mpi]
for i in range(len(reduced_mpi)):
if reduced_mpi[i] % 2 == 0:
reduced_mpi[i] //= 2
# create random selection of points with same spatial sites on each sink time slice
# use different spatial sites for each source time-slice
# this should be optimal for the local operator insertions
rng = g.random(f"sparse2_{self.conf}_{self.t}")
grid = U[0].grid
t0 = grid.ldimensions[3] * grid.processor_coor[3]
t1 = t0 + grid.ldimensions[3]
spatial_sites = int(compress_ratio * np.prod(grid.ldimensions[0:3]))
spatial_coordinates = rng.choice(g.coordinates(U[0]), spatial_sites)
local_coordinates = np.repeat(spatial_coordinates, t1 - t0, axis=0)
for t in range(t0, t1):
local_coordinates[t - t0::t1 - t0, 3] = t
sdomain = g.domain.sparse(current_config.l_exact.U_grid,
local_coordinates)
half_peramb = {"sparse_domain": sdomain}
for i0 in range(0, basis_size, sloppy_per_job):
for l in g.load(
f"{root}/{self.conf}/pm_{self.solver}_t{self.t}_i{i0}/propagators"
):
for x in l:
S = sdomain.lattice(l[x].otype)
sdomain.project(S, l[x])
half_peramb[x] = S
g.message(x)
g.save(
f"{root}/{self.name}/propagators",
half_peramb,
g.format.gpt({"mpi": reduced_mpi}),
)
def __call__(self):
plan = g.copy_plan(self.destinations, self.sources)
buffer_descriptions = []
for i in range(len(self.sources)):
src = self.sources[i]
src_cb = src.checkerboard()
coordinates = g.coordinates(src)
L = src.grid.fdimensions
for x in self.displacements[i]:
buffer_descriptions.append((src.grid, src.otype, src_cb))
plan.destination += self.destinations[self.indices[i]
[x]].view[:]
plan.source += src.view[cgpt.coordinates_shift(
coordinates, x, L)]
return cshift_executer(buffer_descriptions, plan())
def apply_exp_ixp(dst, src, p, origin, cache):
cache_key = f"{src.grid}_{src.checkerboard().__name__}_{origin}_{p}"
if cache_key not in cache:
x = gpt.coordinates(src)
phase = gpt.complex(src.grid)
phase.checkerboard(src.checkerboard())
x_relative = x
if origin is not None:
x_relative = relative_coordinates(x, origin, src.grid.fdimensions)
phase[x] = cgpt.coordinates_momentum_phase(x_relative, p,
src.grid.precision)
cache[cache_key] = phase
dst @= cache[cache_key] * src
def merge_indices(dst, src, st):
pos = gpt.coordinates(dst)
assert st is not None
result_otype = st[-1]()
if result_otype is None:
dst @= src
return
ndim = dst.otype.shape[st[0]]
rank = len(st) - 1
islice = [slice(None, None, None) for i in range(len(dst.otype.shape))]
ivec = [0] * rank
for i in range(ndim ** rank):
idx = i
for j in range(rank):
c = idx % ndim
islice[st[j]] = c
ivec[j] = c
idx //= ndim
dst[(pos,) + tuple(islice)] = src[tuple(ivec)][:]
def separate_indices(x, st, cache=default_merge_indices_cache):
pos = gpt.coordinates(x)
cb = x.checkerboard()
assert st is not None
result_otype = st[-1]()
if result_otype is None:
return x
ndim = x.otype.shape[st[0]]
rank = len(st) - 1
islice = [slice(None, None, None) for i in range(len(x.otype.shape))]
ivec = [0] * rank
result = {}
keys = []
tidx = []
dst = []
for i in range(ndim**rank):
idx = i
for j in range(rank):
c = idx % ndim
islice[st[j]] = c
ivec[j] = c
idx //= ndim
keys.append(tuple(ivec))
tidx.append(tuple(islice))
for i in keys:
v = gpt.lattice(x.grid, result_otype)
v.checkerboard(cb)
result[i] = v
dst.append(v)
cache_key = f"separate_indices_{cb.__name__}_{result_otype.__name__}_{x.otype.__name__}_{x.grid.describe()}_{x.grid.obj}"
if cache_key not in cache:
plan = gpt.copy_plan(dst, x)
for i in range(len(tidx)):
plan.destination += result[keys[i]].view[pos]
plan.source += x.view[(pos, ) + tidx[i]]
cache[cache_key] = plan()
cache[cache_key](dst, x)
return result
def distribute_cartesian_file(fdimensions, grid, cb):
ldimensions = [x for x in fdimensions]
dimdiv = len(ldimensions) - 1
primes = [7, 5, 3, 2]
nreader = 1
found = True
while found:
found = False
for p in primes:
if ldimensions[dimdiv] % p == 0 and nreader * p <= grid.Nprocessors:
nreader *= p
ldimensions[dimdiv] //= p
if ldimensions[dimdiv] == 1 and dimdiv > 0:
dimdiv -= 1
found = True
cv_desc = [a // b for a, b in zip(fdimensions, ldimensions)]
cv = gpt.cartesian_view(grid.processor, cv_desc, fdimensions, grid.cb, cb)
return gpt.coordinates(cv), nreader
def map_pos(grid, cb, key):
# if list, convert to numpy array
if type(key) == list:
key = numpy.array(key, dtype=numpy.int32)
# if key is numpy array, no further processing needed
if isinstance(key, numpy.ndarray):
return key
# if not, we expect a tuple of slices
assert type(key) == tuple
# slices without specified start/stop corresponds to memory view limitation for this rank
if all([k == slice(None, None, None) for k in key]):
# go through gpt.coordinates to use its caching feature
return gpt.coordinates((grid, cb), order="lexicographic")
nd = grid.nd
key = tuple([k if type(k) == slice else slice(k, k + 1) for k in key])
assert all([k.step is None for k in key])
top = [
grid.fdimensions[i] // grid.mpi[i] *
grid.processor_coor[i] if k.start is None else k.start
for i, k in enumerate(key)
]
bottom = [
grid.fdimensions[i] // grid.mpi[i] *
(1 + grid.processor_coor[i]) if k.stop is None else k.stop
for i, k in enumerate(key)
]
assert all([
0 <= top[i] and top[i] <= bottom[i]
and bottom[i] <= grid.fdimensions[i] for i in range(nd)
])
return cgpt.coordinates_from_cartesian_view(top, bottom, grid.cb.cb_mask,
cb.tag, "lexicographic")
def mk_qlat_gpt_copy_plan(ctype, total_site, multiplicity, tag):
geo = q.Geometry(total_site, multiplicity)
f_gpt = mk_gpt_field(ctype, geo)
f_qlat = q.Field(ctype, geo)
lexicographic_coordinates = g.coordinates(f_gpt)
buf = f_qlat.mview()
if tag == "qlat_from_gpt":
qlat_from_gpt = g.copy_plan(buf, f_gpt)
qlat_from_gpt.destination += g.global_memory_view(
f_gpt.grid, [[f_gpt.grid.processor, buf, 0, buf.nbytes]])
qlat_from_gpt.source += f_gpt.view[lexicographic_coordinates]
qlat_from_gpt = qlat_from_gpt(local_only=True)
return qlat_from_gpt
elif tag == "gpt_from_qlat":
gpt_from_qlat = g.copy_plan(f_gpt, buf)
gpt_from_qlat.source += g.global_memory_view(
f_gpt.grid, [[f_gpt.grid.processor, buf, 0, buf.nbytes]])
gpt_from_qlat.destination += f_gpt.view[lexicographic_coordinates]
gpt_from_qlat = gpt_from_qlat(local_only=True)
return gpt_from_qlat
else:
q.displayln_info(tag)
raise Exception("mk_qlat_gpt_copy_plan")
def separate_indices(x, st):
pos = gpt.coordinates(x)
cb = x.checkerboard()
assert st is not None
result_otype = st[-1]()
if result_otype is None:
return x
ndim = x.otype.shape[st[0]]
rank = len(st) - 1
islice = [slice(None, None, None) for i in range(len(x.otype.shape))]
ivec = [0] * rank
result = {}
for i in range(ndim ** rank):
idx = i
for j in range(rank):
c = idx % ndim
islice[st[j]] = c
ivec[j] = c
idx //= ndim
v = gpt.lattice(x.grid, result_otype)
v.checkerboard(cb)
v[pos] = x[(pos,) + tuple(islice)]
result[tuple(ivec)] = v
return result
else:
work_dir = "."
# load configuration
# U = g.load("/hpcgpfs01/work/clehner/configs/32IDfine/ckpoint_lat.200")
# assert abs(g.qcd.gauge.plaquette(U) - float(U[0].metadata["PLAQUETTE"])) < 1e-9
# Show metadata of field
# g.message("Metadata", U[0].metadata)
rng = g.random("test")
U = g.qcd.gauge.random(g.grid([8, 8, 8, 16], g.double), rng)
# create a sparse sub-domain and a sparse lattice S with 1% of points
sdomain = g.domain.sparse(
U[0].grid,
rng.choice(g.coordinates(U[0]), int(0.01 * U[0].grid.gsites / U[0].grid.Nprocessors)),
)
# test sparse domain
S = sdomain.lattice(U[0].otype)
sdomain.project(S, U[0])
U0prime = g.lattice(U[0])
U0prime[:] = 0
sdomain.promote(U0prime, S)
assert np.linalg.norm(U0prime[sdomain.local_coordinates] - U[0][sdomain.local_coordinates]) < 1e-14
s_slice = sdomain.slice(S, 3)
# save in default gpt format
g.save(
f"{work_dir}/out",
{
rng = g.random("test")
l_dp = rng.cnormal(g.vcolor(grid_dp))
l_sp = g.convert(l_dp, g.single)
# and convert precision
l_dp_prime = g.convert(l_sp, g.double)
eps2 = g.norm2(l_dp - l_dp_prime) / g.norm2(l_dp)
assert eps2 < 1e-14
eps2 = g.norm2(l_dp[0, 0, 0, 0] - l_sp[0, 0, 0, 0])
assert eps2 < 1e-14
################################################################################
# Test mview
################################################################################
c = g.coordinates(l_dp)
x = l_dp[c]
mv = g.mview(x)
assert mv.itemsize == 1 and mv.shape[0] == len(mv)
assert sys.getrefcount(x) == 3
del mv
assert sys.getrefcount(x) == 2
################################################################################
# Test pinning
################################################################################
l_v = g.complex(grid_sp)
pin = g.pin(l_v, g.accelerator)
del l_v
del pin
def create_links(A, fmat, basis, params):
# NOTE: we expect the blocks in the basis vectors
# to already be orthogonalized!
# parameters
make_hermitian = params["make_hermitian"]
save_links = params["save_links"]
assert not (make_hermitian and not save_links)
# verbosity
verbose = gpt.default.is_verbose("coarsen")
# setup timings
t = gpt.timer("coarsen")
t("setup")
# get grids
f_grid = basis[0].grid
c_grid = A[0].grid
# directions/displacements we coarsen for
dirs = [1, 2, 3, 4] if f_grid.nd == 5 else [0, 1, 2, 3]
disp = +1
dirdisps_full = list(zip(dirs * 2, [+1] * 4 + [-1] * 4))
dirdisps_forward = list(zip(dirs, [disp] * 4))
nhops = len(dirdisps_full)
selflink = nhops
# setup fields
Mvr = [gpt.lattice(basis[0]) for i in range(nhops)]
tmp = gpt.lattice(basis[0])
oproj = gpt.vcomplex(c_grid, len(basis))
selfproj = gpt.vcomplex(c_grid, len(basis))
# setup masks
onemask, blockevenmask, blockoddmask = (
gpt.complex(f_grid),
gpt.complex(f_grid),
gpt.complex(f_grid),
)
dirmasks = [gpt.complex(f_grid) for p in range(nhops)]
# auxilliary stuff needed for masks
t("masks")
onemask[:] = 1.0
coor = gpt.coordinates(blockevenmask)
block = numpy.array(f_grid.ldimensions) / numpy.array(c_grid.ldimensions)
block_cb = coor[:, :] // block[:]
# fill masks for sites within even/odd blocks
gpt.coordinate_mask(blockevenmask, numpy.sum(block_cb, axis=1) % 2 == 0)
blockoddmask @= onemask - blockevenmask
# fill masks for sites on borders of blocks
dirmasks_forward_np = coor[:, :] % block[:] == block[:] - 1
dirmasks_backward_np = coor[:, :] % block[:] == 0
for mu in dirs:
gpt.coordinate_mask(dirmasks[mu], dirmasks_forward_np[:, mu])
gpt.coordinate_mask(dirmasks[mu + 4], dirmasks_backward_np[:, mu])
# save applications of matrix and coarsening if possible
dirdisps = dirdisps_forward if save_links else dirdisps_full
# create block maps
t("blockmap")
dirbms = [
gpt.block.map(c_grid, basis, dirmasks[p])
for p, (mu, fb) in enumerate(dirdisps)
]
fullbm = gpt.block.map(c_grid, basis)
for i, vr in enumerate(basis):
# apply directional hopping terms
# this triggers len(dirdisps) comms -> TODO expose DhopdirAll from Grid
# BUT problem with vector<Lattice<...>> in rhs
t("apply_hop")
[fmat.Mdir(*dirdisp)(Mvr[p], vr) for p, dirdisp in enumerate(dirdisps)]
# coarsen directional terms + write to link
for p, (mu, fb) in enumerate(dirdisps):
t("coarsen_hop")
dirbms[p].project(oproj, Mvr[p])
t("copy_hop")
A[p][:, :, :, :, :, i] = oproj[:]
# fast diagonal term: apply full matrix to both block cbs separately and discard hops into other cb
t("apply_self")
tmp @= (blockevenmask * fmat * vr * blockevenmask +
blockoddmask * fmat * vr * blockoddmask)
# coarsen diagonal term
t("coarsen_self")
fullbm.project(selfproj, tmp)
# write to self link
t("copy_self")
A[selflink][:, :, :, :, :, i] = selfproj[:]
if verbose:
gpt.message("coarsen: done with vector %d" % i)
# communicate opposite links
if save_links:
t("comm")
communicate_links(A, dirdisps_forward, make_hermitian)
t()
if verbose:
gpt.message(t)
# Show metadata of field
g.message("Metadata", U[0].metadata)
# to single precision
#U = g.convert(U, g.single)
# save in default gpt format
g.save(
"out",
{
"va\nl": [
0, 1, 3, "tes\n\0t", 3.123456789123456789, 1.123456789123456789e-7,
1 + 3.1231251251234123413j
], # fundamental data types
"np":
g.coordinates(U[0].grid), # write numpy array from root node
"U":
U # write list of lattices
})
# save in custom gpt format with different mpi distribution of local views
g.save(
"out2",
{
"val": [
0, 1, 3, "test", 3.123456789123456789, 1.123456789123456789e-7,
1 + 3.1231251251234123413j
], # fundamental data types
"np":
g.coordinates(U[0].grid), # write numpy array from root node
"U":
import numpy as np
import sys, cgpt
# grid
L = [16, 16, 16, 32]
grid_dp = g.grid(L, g.double)
grid_sp = g.grid(L, g.single)
# test fields
l_dp = g.random("test").cnormal(g.vcolor(grid_dp))
l_sp = g.convert(l_dp, g.single)
################################################################################
# Test mview
################################################################################
c = g.coordinates(l_dp)
x = l_dp[c]
mv = g.mview(x)
assert mv.itemsize == 1 and mv.shape[0] == len(mv)
assert sys.getrefcount(x) == 3
del mv
assert sys.getrefcount(x) == 2
################################################################################
# Test assignments
################################################################################
pos = l_dp.mview_coordinates()
lhs = g.lattice(l_dp)
def assign_copy():
def coordinates(src, position, spacing):
coor = gpt.coordinates(src)
return coor[np.sum(np.mod(coor - position, spacing), axis=1) == 0]
