Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Apply coadd to only non-zero pixels #787

Merged
merged 1 commit into from
Sep 24, 2024
Merged

Apply coadd to only non-zero pixels #787

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 6 additions & 1 deletion src/toast/_libtoast/map_cov.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -378,7 +378,12 @@ void init_map_cov(py::module & m) {
if (nv != nb) {
auto log = toast::Logger::get();
std::ostringstream o;
o << "Buffer sizes are not consistent.";
o << "Buffer sizes are not consistent. "
<< "npix_matrix = " << nb
<< ", npix_map = " << nv
<< ", matrix_size = " << info_mat.size
<< ", block = " << block
<< ", nnz = " << nnz;
log.error(o.str().c_str());
throw std::runtime_error(o.str().c_str());
}
Expand Down
274 changes: 148 additions & 126 deletions src/toast/scripts/toast_healpix_coadd.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,14 +22,7 @@
from toast._libtoast import cov_apply_diag, cov_eigendecompose_diag
from toast.covariance import covariance_apply, covariance_invert
from toast.mpi import MPI, Comm, get_world
from toast.pixels_io_healpix import (
filename_is_fits,
filename_is_hdf5,
read_healpix,
write_healpix,
write_healpix_fits,
write_healpix_hdf5,
)
from toast.pixels_io_healpix import read_healpix, write_healpix
from toast.timing import Timer
from toast.utils import Environment, Logger

Expand Down Expand Up @@ -118,6 +111,7 @@ def main():
noiseweighted_sum = None
invcov_sum = None
nnz, nnz2, npix = None, None, None
hit_pixels = None
if len(args.inmap) == 1:
# Only one file provided, try interpreting it as a text file with a list
try:
Expand All @@ -139,10 +133,11 @@ def main():
if ifile % ntask != rank:
continue
log.info(f"{prefix}Loading file {ifile + 1} / {nfile} : {infile_map}")
inmap = read_healpix(infile_map, None, nest=True, dtype=float, verbose=False)
log.info_rank(f"{prefix}Loaded {infile_map} in", timer=timer1, comm=None)
inmap = read_healpix(infile_map, None, nest=True, dtype=float)
log.info_rank(f"{prefix}Loaded {infile_map} {inmap.shape} in", timer=timer1, comm=None)
if nnz is None:
nnz, npix = inmap.shape
hit_pixels = np.zeros(npix, dtype=bool)
else:
nnz_test, npix_test = inmap.shape
if nnz != nnz_test:
Expand All @@ -154,22 +149,42 @@ def main():
noiseweighted = "noiseweighted" in infile_map

# Determine the name of the covariance matrix file
if "unfiltered_map" in infile_map:
mapstring = "unfiltered_map"
elif "filtered_map" in infile_map:
mapstring = "filtered_map"
if "unfiltered" in infile_map:
mapstring = "unfiltered_"
elif "filtered" in infile_map:
mapstring = "filtered_"
else:
mapstring = "map"
mapstring = ""
if "binmap" in infile_map:
mapstring += "binmap"
else:
mapstring += "map"
if noiseweighted:
mapstring = f"noiseweighted_{mapstring}"

infile_invcov = infile_map.replace(f"_{mapstring}.", "_invcov.")
if infile_invcov == infile_map:
raise RuntimeError(
f"Failed to derive name of a covariance matrix file from {infile_map}."
)

if os.path.isfile(infile_invcov):
log.info(f"{prefix}Loading {infile_invcov}")
invcov = read_healpix(
infile_invcov, None, nest=True, dtype=float, verbose=False
infile_invcov, None, nest=True, dtype=float
)
if nnz2 is None:
nnz2, npix_test = invcov.shape
good = invcov[0] != 0
hit_pixels[good] = True
ngood = np.sum(good)
fsky = ngood / good.size
log.info_rank(
f"{prefix}Loaded {infile_invcov} {invcov.shape}, fsky = {fsky:.4f} in",
timer=timer1,
comm=None,
)
log.info_rank(f"{prefix}Loaded {infile_invcov} in", timer=timer1, comm=None)
invcov = invcov[:, good].copy()
else:
# Inverse covariance does not exist. Load and invert the
# covariance matrix
Expand All @@ -182,20 +197,32 @@ def main():
)
raise RuntimeError(msg)
log.info(f"{prefix}Loading {infile_cov}")
cov = read_healpix(infile_cov, None, nest=True, dtype=float, verbose=False)
log.info_rank(f"{prefix}Loaded {infile_cov} in", timer=timer1, comm=None)
nsubmap = npix
npix_submap = 1
rcond = np.zeros(npix, dtype=float)
cov = read_healpix(infile_cov, None, nest=True, dtype=float)
if nnz2 is None:
nnz2, npix_test = cov.shape
cov_shape = cov.shape
good = cov[0] != 0
ngood = np.sum(good)
hit_pixels[good] = True
fsky = ngood / good.size
log.info_rank(
f"{prefix}Loaded {infile_cov} {cov_shape}, fsky = {fsky:.4f} in",
timer=timer1,
comm=None,
)
rcond = np.zeros(ngood, dtype=float)
log.info(f"{prefix}Inverting matrix")
cov = cov.T.ravel().astype(float).copy()
cov_eigendecompose_diag(
nsubmap, npix_submap, nnz, cov, rcond, args.rcond_limit, True
cov = cov[:, good].T.ravel().copy()
cov_eigendecompose_diag(ngood, 1, nnz, cov, rcond, args.rcond_limit, True)
invcov = cov.reshape(ngood, -1).T.copy()
log.info_rank(
f"{prefix}Inverted matrix {invcov.shape} in", timer=timer1, comm=None
)
invcov = cov.reshape(npix, -1).T.copy()
log.info_rank(f"{prefix}Inverted matrix in", timer=timer1, comm=None)
del cov

# Trim off empty pixels
inmap = inmap[:, good].copy()

# Optionally scale the maps
if args.scale is not None:
if noiseweighted:
Expand All @@ -210,23 +237,18 @@ def main():
log.info(f"{prefix}Applying inverse matrix")
invcov = invcov.T.ravel().astype(float).copy()
inmap = inmap.T.ravel().astype(float).copy()
nsubmap = npix
npix_submap = 1
cov_apply_diag(nsubmap, npix_submap, nnz, invcov.data, inmap.data)
inmap = inmap.reshape(npix, -1).T.copy()
invcov = invcov.reshape(npix, -1).T.copy()
cov_apply_diag(ngood, 1, nnz, invcov.data, inmap.data)
inmap = inmap.reshape(ngood, -1).T.copy()
invcov = invcov.reshape(ngood, -1).T.copy()
log.info_rank(f"{prefix}Applied inverse matrix in", timer=timer1, comm=None)

if nnz2 is None:
nnz2, npix_test = invcov.shape

if noiseweighted_sum is None:
noiseweighted_sum = inmap
invcov_sum = invcov
else:
noiseweighted_sum += inmap
invcov_sum += invcov
log.info_rank(f"{prefix}Co-added maps in", timer=timer1, comm=None)
noiseweighted_sum = np.zeros([nnz, npix], dtype=float)
invcov_sum = np.zeros([nnz2, npix], dtype=float)
noiseweighted_sum[:, good] += inmap
invcov_sum[:, good] += invcov
log.info_rank(f"{prefix}Co-added maps in", timer=timer1, comm=None)

del invcov
del inmap

Expand All @@ -236,105 +258,105 @@ def main():
nnz = comm.bcast(nnz)
nnz2 = comm.bcast(nnz2)
npix = comm.bcast(npix)
if hit_pixels is None:
hit_pixels = np.zeros(npix, dtype=bool)
comm.Allreduce(MPI.IN_PLACE, hit_pixels, op=MPI.LOR)
good = hit_pixels
ngood = np.sum(hit_pixels)
fsky = ngood / npix
if noiseweighted_sum is None:
noiseweighted_sum = np.zeros([nnz, npix], dtype=float)
invcov_sum = np.zeros([nnz2, npix], dtype=float)
noiseweighted_sum = np.zeros([nnz, ngood], dtype=float)
invcov_sum = np.zeros([nnz2, ngood], dtype=float)
else:
noiseweighted_sum = noiseweighted_sum[:, good].copy()
invcov_sum = invcov_sum[:, good].copy()
comm.Allreduce(MPI.IN_PLACE, noiseweighted_sum, op=MPI.SUM)
comm.Allreduce(MPI.IN_PLACE, invcov_sum, op=MPI.SUM)
log.info_rank(f"Reduced inputs in", timer=timer, comm=comm)
else:
good = hit_pixels
ngood = np.sum(hit_pixels)
fsky = ngood / npix
log.info_rank(f"fsky = {fsky:.4f}", comm=comm)

if args.invcov is not None:
log.info_rank(f"Writing {args.invcov}", comm=comm)
if rank == 0:
full_invcov = np.zeros([nnz2, npix])
full_invcov[:, good] = invcov_sum
write_healpix(
args.invcov, invcov_sum, nest=True, overwrite=True, dtype=dtype
args.invcov, full_invcov, nest=True, overwrite=True, dtype=dtype
)
del full_invcov
log.info_rank(f"Wrote {args.invcov} in", timer=timer, comm=comm)

# Assign submaps, invert and apply local portions of the matrix

npix_submap = 12 * args.nside_submap**2
nsubmap = npix // npix_submap
local_submaps = [submap for submap in range(nsubmap) if submap % ntask == rank]
dist = PixelDistribution(
n_pix=npix, n_submap=nsubmap, local_submaps=local_submaps, comm=comm
)
dist_map = PixelData(dist, float, n_value=nnz)
dist_cov = PixelData(dist, float, n_value=nnz2)
for local_submap, global_submap in enumerate(local_submaps):
pix_start = global_submap * npix_submap
pix_stop = pix_start + npix_submap
dist_map.data[local_submap] = noiseweighted_sum[:, pix_start:pix_stop].T
dist_cov.data[local_submap] = invcov_sum[:, pix_start:pix_stop].T
del noiseweighted_sum
del invcov_sum

log.info_rank("Inverting matrix", comm=comm)
dist_rcond = PixelData(dist, float, n_value=1)
covariance_invert(dist_cov, args.rcond_limit, rcond=dist_rcond, use_alltoallv=True)
log.info_rank(f"Inverted matrix in", timer=timer, comm=comm)

if args.rcond is not None:
log.info_rank(f"Writing {args.rcond}", comm=comm)
if filename_is_fits(args.rcond):
write_healpix_fits(
dist_rcond,
args.rcond,
nest=True,
single_precision=not args.double_precision,
)
else:
write_healpix_hdf5(
dist_rcond,
args.rcond,
nest=True,
single_precision=not args.double_precision,
force_serial=True,
)
log.info_rank(f"Wrote {args.rcond}", timer=timer, comm=comm)
del dist_rcond

log.info_rank("Applying matrix", comm=comm)
covariance_apply(dist_cov, dist_map, use_alltoallv=True)
log.info_rank(f"Applied matrix in", timer=timer, comm=comm)

if args.cov is not None:
log.info_rank(f"Writing {args.cov}", comm=comm)
if filename_is_fits(args.cov):
write_healpix_fits(
dist_cov,
args.cov,
nest=True,
single_precision=not args.double_precision,
)
else:
write_healpix_hdf5(
dist_cov,
args.cov,
nest=True,
single_precision=not args.double_precision,
force_serial=True,
)
log.info_rank(f"Wrote {args.cov}", timer=timer, comm=comm)
del dist_cov

log.info_rank(f"Writing {args.outmap}", comm=comm)
if filename_is_fits(args.outmap):
write_healpix_fits(
dist_map,
args.outmap,
nest=True,
single_precision=not args.double_precision,
# Each task processes a segment of hit pixels

npix_task = ngood // ntask + 1
first_pix = rank * npix_task
last_pix = min(first_pix + npix_task, ngood)
if first_pix < last_pix:
my_npix = last_pix - first_pix
ind = slice(first_pix, last_pix)
my_map = noiseweighted_sum[:, ind].T.ravel().copy()
my_cov = invcov_sum[:, ind].T.ravel().copy()
my_rcond = np.zeros(my_npix, dtype=float)
log.debug(f"{prefix}Inverting {my_npix} pixels")
cov_eigendecompose_diag(
my_npix, 1, nnz, my_cov, my_rcond, args.rcond_limit, True
)
log.debug(f"{prefix}Multiplying {my_npix} pixels")
cov_apply_diag(my_npix, 1, nnz, my_cov.data, my_map.data)
my_map = my_map.reshape(my_npix, -1).T.copy()
my_cov = my_cov.reshape(my_npix, -1).T.copy()
else:
write_healpix_hdf5(
dist_map,
args.outmap,
nest=True,
single_precision=not args.double_precision,
force_serial=True,
)
log.info_rank(f"Wrote {args.outmap}", timer=timer, comm=comm)
my_map = np.zeros([nnz, 0], dtype=float)
my_cov = np.zeros([nnz2, 0], dtype=float)
my_rcond = np.zeros([0], dtype=float)

log.info_rank(f"Inverted and applied covariance in", timer=timer, comm=comm)

# Gather to root process and write

if comm is None:
total_map = [my_map]
total_cov = [my_cov]
total_rcond = [my_rcond]
else:
total_map = comm.gather(my_map, root=0)
total_cov = comm.gather(my_cov, root=0)
total_rcond = comm.gather(my_rcond, root=0)
log.info_rank(f"Gathered map and covariance in", timer=timer, comm=comm)

if rank == 0:
if args.double_precision:
dtype = np.float64
else:
dtype = np.float32
if args.rcond is not None:
log.info(f"Writing {args.rcond}")
total_rcond = np.hstack(total_rcond)
full_rcond = np.zeros(npix, dtype=dtype)
full_rcond[good] = total_rcond
write_healpix(args.rcond, full_rcond, nest=True, dtype=dtype)
del full_rcond
del total_rcond
log.info_rank(f"Wrote {args.rcond}", timer=timer, comm=None)
if args.cov is not None:
log.info(f"Writing {args.cov}")
total_cov = np.hstack(total_cov)
full_cov = np.zeros([nnz, npix])
full_cov[:, good] = total_cov
write_healpix(args.cov, full_cov, nest=True, dtype=dtype)
del full_cov
del total_cov
log.info_rank(f"Wrote {args.cov}", timer=timer, comm=None)
log.info(f"Writing {args.outmap}")
total_map = np.hstack(total_map)
full_map = np.zeros([nnz, npix])
full_map[:, good] = total_map
write_healpix(args.outmap, full_map, nest=True, dtype=dtype)
log.info_rank(f"Wrote {args.outmap}", timer=timer, comm=None)

log.info_rank(f"Co-add done in", timer=timer0, comm=comm)

Expand Down