|
mpi_initialize_alm_tools
This subroutine initializes the mpi_alm_tools module, and must be
run prior to any of the advanced interface working routines by all
processors in the MPI communicator. The root processor must supply all arguments,
while it is optional for the slaves. However, the information is disregarded
if they do.
A major advantage of MPI parallelization is large quantities
of memory, allowing for pre-computation of the Legendre
polynomials even with high
Nside and
lmax, since each processor only needs a fraction
(1/Nprocs) of the complete table. This feature is
controlled by the ``precompute_plms'' parameter. In general, the CPU
time can be expected to decrease by roughly 50% using pre-computed
Legendre polynomials for temperature calculations, and by about 30%
for polarization calculations.
Location in HEALPix directory tree:
src/f90/mod/mpi_alm_tools.f90
FORMAT call mpi_initialize_alm_tools(
comm, [nsmax], [nlmax], [nmmax], [zbounds], [polarization], [precompute_plms], [w8ring]
)
ARGUMENTS
| name & dimensionality |
kind |
in/out |
description |
|---|
| |
|
|
|
|
comm |
I4B |
IN |
MPI communicator. |
| nsmax |
I4B |
IN |
the Nside value of the HEALPix map. (OPTIONAL) |
| nlmax |
I4B |
IN |
the maximum l value used for the alm. (OPTIONAL) |
| nmmax |
I4B |
IN |
the maximum m value used for the alm. (OPTIONAL) |
| zbounds(1:2) |
DP |
IN |
section of the map on which to perform the alm
analysis, expressed in terms of
 . If zbounds(1)<zbounds(2), the analysis is
performed on the strip zbounds(1)<z<zbounds(2); if not,
it is performed outside of the strip
zbounds(2)<z<zbounds(1). (OPTIONAL) |
| polarization |
LGT |
IN |
if polarization is required, this should be
set to true, else it should be set to false. (OPTIONAL) |
| precompute_plms |
I4B |
IN |
0 = do not pre-compute any Pl
m's; 1 = pre-compute
PlmT; 2 = pre-compute
PlmT and
PlmP. (OPTIONAL) |
| w8ring_TQU(1:2*nsmax, 1:p) |
DP |
IN |
ring weights for quadrature corrections. If ring weights are not used, this array should be 1 everywhere. p is 1 for a temperature analysis and 3 for (T,Q,U). (OPTIONAL) |
EXAMPLE:
call mpi_comm_rank(comm, myid, ierr)
if (myid == root) then
call mpi_initialize_alm_tools(comm, nsmax, nlmax, nmmax,
zbounds,polarization, precompute_plms)
call mpi_map2alm(map, alms)
else
call mpi_initialize_alm_tools(comm)
call mpi_map2alm_slave
end
call mpi_cleanup_alm_tools
This example 1) initializes the mpi_alm_tools module (i.e.,
allocates internal arrays and defines required parameters), 2)
executes a parallel map2alm operation, and 3) frees the previously
allocated memory.
RELATED ROUTINESThis section lists the routines related to mpi_initialize_alm_tools
-
mpi_cleanup_alm_tools
- Frees memory that is allocated by the current routine.
-
mpi_alm2map
- Routine for executing a parallel inverse spherical harmonics transform (root processor interface)
-
mpi_alm2map_slave
- Routine for executing a parallel inverse spherical harmonics transform (slave processor interface)
-
mpi_map2alm
- Routine for executing a parallel spherical harmonics transform (root processor interface)
-
mpi_map2alm_slave
- Routine for executing a parallel spherical harmonics transform (slave processor interface)
-
mpi_alm2map_simple
- One-line interface to the parallel inverse spherical harmonics transform
-
mpi_map2alm_simple
- One-line interface to the parallel spherical harmonics transform
Eric Hivon
2010-06-18
|
|
