From 23ed2f4afc4d8ecaff1bcfc46b2f3cf53923188b Mon Sep 17 00:00:00 2001
From: Cristian C Lalescu <Cristian.Lalescu@ds.mpg.de>
Date: Tue, 25 Sep 2018 11:30:48 +0200
Subject: [PATCH] split inner_computer into hpp and cpp
---
bfps/cpp/field.cpp | 7 +
bfps/cpp/full_code/NSVE.cpp | 1 +
.../particles/particles_inner_computer.cpp | 155 ++++++++++++++++++
.../particles/particles_inner_computer.hpp | 121 ++------------
setup.py | 8 +-
5 files changed, 185 insertions(+), 107 deletions(-)
create mode 100644 bfps/cpp/particles/particles_inner_computer.cpp
diff --git a/bfps/cpp/field.cpp b/bfps/cpp/field.cpp
index d5bc78a5..87e18b67 100644
--- a/bfps/cpp/field.cpp
+++ b/bfps/cpp/field.cpp
@@ -79,6 +79,7 @@ field<rnumber, be, fc>::field(
hsize_t tmp_local_size;
ptrdiff_t local_n0, local_0_start;
ptrdiff_t local_n1, local_1_start;
+ variable_used_only_in_assert(tmp_local_size);
tmp_local_size = fftw_interface<rnumber>::mpi_local_size_many_transposed(
3, nfftw, ncomp(fc),
FFTW_MPI_DEFAULT_BLOCK, FFTW_MPI_DEFAULT_BLOCK, this->comm,
@@ -227,6 +228,7 @@ int field<rnumber, be, fc>::io(
H5Tequal(dset_type, H5T_IEEE_F64LE) ||
H5Tequal(dset_type, H5T_INTEL_F64) ||
H5Tequal(dset_type, H5T_NATIVE_DOUBLE));
+ variable_used_only_in_assert(io_for_real);
H5Tclose(dset_type);
assert(this->real_space_representation == io_for_real);
}
@@ -307,6 +309,7 @@ int field<rnumber, be, fc>::io(
/* check file space */
int ndims_fspace = H5Sget_simple_extent_dims(fspace, dims, NULL);
+ variable_used_only_in_assert(ndims_fspace);
assert(((unsigned int)(ndims_fspace)) == ndim(fc));
if (this->real_space_representation)
{
@@ -417,6 +420,7 @@ int field<rnumber, be, fc>::io_database(
H5Tequal(dset_type, H5T_IEEE_F64LE) ||
H5Tequal(dset_type, H5T_INTEL_F64) ||
H5Tequal(dset_type, H5T_NATIVE_DOUBLE));
+ variable_used_only_in_assert(io_for_real);
H5Tclose(dset_type);
assert(this->real_space_representation == io_for_real);
}
@@ -493,6 +497,7 @@ int field<rnumber, be, fc>::io_database(
/* check file space */
int ndims_fspace = H5Sget_simple_extent_dims(fspace, dims, NULL);
+ variable_used_only_in_assert(ndims_fspace);
assert(ndims_fspace == int(ndim(fc) + 1));
offset[0] = toffset;
if (this->real_space_representation)
@@ -714,6 +719,7 @@ int field<rnumber, be, fc>::write_filtered(
}
/* check file space */
int ndims_fspace = H5Sget_simple_extent_dims(fspace, fdims, NULL);
+ variable_used_only_in_assert(ndims_fspace);
assert(((unsigned int)(ndims_fspace)) == ndim(fc));
for (unsigned int i=0; i<ndim(fc); i++)
{
@@ -1567,6 +1573,7 @@ int joint_rspace_PDF(
hid_t dset, wspace;
hsize_t dims[5];
int ndims;
+ variable_used_only_in_assert(ndims);
if (fc == THREE)
{
dset = H5Dopen(
diff --git a/bfps/cpp/full_code/NSVE.cpp b/bfps/cpp/full_code/NSVE.cpp
index ecec7db3..efd82fb3 100644
--- a/bfps/cpp/full_code/NSVE.cpp
+++ b/bfps/cpp/full_code/NSVE.cpp
@@ -18,6 +18,7 @@ int NSVE<rnumber>::initialize(void)
// set caching parameters
hid_t fapl = H5Pcreate(H5P_FILE_ACCESS);
herr_t cache_err = H5Pset_cache(fapl, 0, 521, 134217728, 1.0);
+ variable_used_only_in_assert(cache_err);
DEBUG_MSG("when setting stat_file cache I got %d\n", cache_err);
this->stat_file = H5Fopen(
(this->simname + ".h5").c_str(),
diff --git a/bfps/cpp/particles/particles_inner_computer.cpp b/bfps/cpp/particles/particles_inner_computer.cpp
new file mode 100644
index 00000000..85286190
--- /dev/null
+++ b/bfps/cpp/particles/particles_inner_computer.cpp
@@ -0,0 +1,155 @@
+#include "base.hpp"
+#include "particles_utils.hpp"
+#include "particles_inner_computer.hpp"
+
+#include <cmath>
+
+template <class real_number, class partsize_t>
+template <int size_particle_positions, int size_particle_rhs>
+void particles_inner_computer<real_number, partsize_t>::compute_interaction(
+ const partsize_t nb_particles,
+ const real_number pos_part[],
+ real_number rhs_part[]) const{
+ static_assert(size_particle_positions == 6, "This kernel works only with 6 values for one particle's position");
+ static_assert(size_particle_rhs == 6, "This kernel works only with 6 values per particle's rhs");
+
+ #pragma omp parallel for
+ for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
+ // Add attr × V0 to the field interpolation
+ rhs_part[idx_part*size_particle_rhs + IDX_X] += pos_part[idx_part*size_particle_positions + 3+IDX_X]*v0;
+ rhs_part[idx_part*size_particle_rhs + IDX_Y] += pos_part[idx_part*size_particle_positions + 3+IDX_Y]*v0;
+ rhs_part[idx_part*size_particle_rhs + IDX_Z] += pos_part[idx_part*size_particle_positions + 3+IDX_Z]*v0;
+ }
+}
+
+ // for given orientation and right-hand-side, recompute right-hand-side such
+ // that it is perpendicular to the current orientation.
+ // this is the job of the Lagrange multiplier terms, hence the
+ // "add_Lagrange_multipliers" name of the method.
+template <class real_number, class partsize_t>
+template <int size_particle_positions, int size_particle_rhs>
+void particles_inner_computer<real_number, partsize_t>::add_Lagrange_multipliers(
+ const partsize_t nb_particles,
+ const real_number pos_part[],
+ real_number rhs_part[]) const{
+ static_assert(size_particle_positions == 6, "This kernel works only with 6 values for one particle's position");
+ static_assert(size_particle_rhs == 6, "This kernel works only with 6 values per particle's rhs");
+
+ #pragma omp parallel for
+ for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
+ const partsize_t idx0 = idx_part*size_particle_positions + 3;
+ const partsize_t idx1 = idx_part*size_particle_rhs + 3;
+ // check that orientation is unit vector:
+ real_number orientation_size = sqrt(
+ pos_part[idx0+IDX_X]*pos_part[idx0+IDX_X] +
+ pos_part[idx0+IDX_Y]*pos_part[idx0+IDX_Y] +
+ pos_part[idx0+IDX_Z]*pos_part[idx0+IDX_Z]);
+ variable_used_only_in_assert(orientation_size);
+ assert(orientation_size > 0.99);
+ assert(orientation_size < 1.01);
+ // I call "rotation" to be the right hand side of the orientation part of the ODE
+ // project rotation on orientation:
+ real_number projection = (
+ pos_part[idx0+IDX_X]*rhs_part[idx1+IDX_X] +
+ pos_part[idx0+IDX_Y]*rhs_part[idx1+IDX_Y] +
+ pos_part[idx0+IDX_Z]*rhs_part[idx1+IDX_Z]);
+
+ // now remove parallel bit.
+ rhs_part[idx1+IDX_X] -= pos_part[idx0+IDX_X]*projection;
+ rhs_part[idx1+IDX_Y] -= pos_part[idx0+IDX_Y]*projection;
+ rhs_part[idx1+IDX_Z] -= pos_part[idx0+IDX_Z]*projection;
+
+ // DEBUG
+ // sanity check, for debugging purposes
+ // compute dot product between orientation and orientation change
+ //real_number dotproduct = (
+ // rhs_part[idx1 + IDX_X]*pos_part[idx0 + IDX_X] +
+ // rhs_part[idx1 + IDX_Y]*pos_part[idx0 + IDX_Y] +
+ // rhs_part[idx1 + IDX_Z]*pos_part[idx0 + IDX_Z]);
+ //if (dotproduct > 0.1)
+ //{
+ // DEBUG_MSG("dotproduct = %g, projection = %g\n"
+ // "pos_part[%d] = %g, pos_part[%d] = %g, pos_part[%d] = %g\n"
+ // "rhs_part[%d] = %g, rhs_part[%d] = %g, rhs_part[%d] = %g\n",
+ // dotproduct,
+ // projection,
+ // IDX_X, pos_part[idx0 + IDX_X],
+ // IDX_Y, pos_part[idx0 + IDX_Y],
+ // IDX_Z, pos_part[idx0 + IDX_Z],
+ // IDX_X, rhs_part[idx1 + IDX_X],
+ // IDX_Y, rhs_part[idx1 + IDX_Y],
+ // IDX_Z, rhs_part[idx1 + IDX_Z]);
+ // assert(false);
+ //}
+ //assert(dotproduct <= 0.1);
+ }
+ }
+
+template <class real_number, class partsize_t>
+template <int size_particle_positions, int size_particle_rhs, int size_particle_rhs_extra>
+void particles_inner_computer<real_number, partsize_t>::compute_interaction_with_extra(
+ const partsize_t nb_particles,
+ const real_number pos_part[],
+ real_number rhs_part[],
+ const real_number rhs_part_extra[]) const{
+ static_assert(size_particle_rhs_extra == 3, "This kernel works only with 3 values for one particle's rhs extra");
+
+ // call plain compute_interaction first
+ compute_interaction<size_particle_positions, size_particle_rhs>(nb_particles, pos_part, rhs_part);
+
+ // now add vorticity term
+ #pragma omp parallel for
+ for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
+ // Cross product vorticity/orientation
+ rhs_part[idx_part*size_particle_rhs + 3+IDX_X] += 0.5*(rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Y]*pos_part[idx_part*size_particle_positions + 3+IDX_Z] -
+ rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Z]*pos_part[idx_part*size_particle_positions + 3+IDX_Y]);
+ rhs_part[idx_part*size_particle_rhs + 3+IDX_Y] += 0.5*(rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Z]*pos_part[idx_part*size_particle_positions + 3+IDX_X] -
+ rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_X]*pos_part[idx_part*size_particle_positions + 3+IDX_Z]);
+ rhs_part[idx_part*size_particle_rhs + 3+IDX_Z] += 0.5*(rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_X]*pos_part[idx_part*size_particle_positions + 3+IDX_Y] -
+ rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Y]*pos_part[idx_part*size_particle_positions + 3+IDX_X]);
+ }
+}
+
+// meant to be called AFTER executing the time-stepping operation.
+// once the particles have been moved, ensure that the orientation is a unit vector.
+template <class real_number, class partsize_t>
+template <int size_particle_positions>
+void particles_inner_computer<real_number, partsize_t>::enforce_unit_orientation(
+ const partsize_t nb_particles,
+ real_number pos_part[]) const{
+ static_assert(size_particle_positions == 6, "This kernel works only with 6 values for one particle's position");
+
+ #pragma omp parallel for
+ for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
+ const partsize_t idx0 = idx_part*size_particle_positions + 3;
+ // compute orientation size:
+ real_number orientation_size = sqrt(
+ pos_part[idx0+IDX_X]*pos_part[idx0+IDX_X] +
+ pos_part[idx0+IDX_Y]*pos_part[idx0+IDX_Y] +
+ pos_part[idx0+IDX_Z]*pos_part[idx0+IDX_Z]);
+ // now renormalize
+ pos_part[idx0 + IDX_X] /= orientation_size;
+ pos_part[idx0 + IDX_Y] /= orientation_size;
+ pos_part[idx0 + IDX_Z] /= orientation_size;
+ }
+}
+
+template class particles_inner_computer<double, long long>;
+
+template void particles_inner_computer<double, long long>::compute_interaction<6, 6>(
+ const long long nb_particles,
+ const double pos_part[],
+ double rhs_part[]) const;
+template void particles_inner_computer<double, long long>::add_Lagrange_multipliers<6, 6>(
+ const long long nb_particles,
+ const double pos_part[],
+ double rhs_part[]) const;
+template void particles_inner_computer<double, long long>::compute_interaction_with_extra <6, 6, 3>(
+ const long long nb_particles,
+ const double pos_part[],
+ double rhs_part[],
+ const double rhs_part_extra[]) const;
+template void particles_inner_computer<double, long long>:: enforce_unit_orientation<6>(
+ const long long nb_particles,
+ double pos_part[]) const;
+
diff --git a/bfps/cpp/particles/particles_inner_computer.hpp b/bfps/cpp/particles/particles_inner_computer.hpp
index b2eb95dd..3e233370 100644
--- a/bfps/cpp/particles/particles_inner_computer.hpp
+++ b/bfps/cpp/particles/particles_inner_computer.hpp
@@ -15,120 +15,31 @@ public:
}
template <int size_particle_positions, int size_particle_rhs>
- void compute_interaction(const partsize_t nb_particles, const real_number pos_part[], real_number rhs_part[]) const{
- static_assert(size_particle_positions == 6, "This kernel works only with 6 values for one particle's position");
- static_assert(size_particle_rhs == 6, "This kernel works only with 6 values per particle's rhs");
-
- #pragma omp parallel for
- for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
- // Add attr × V0 to the field interpolation
- rhs_part[idx_part*size_particle_rhs + IDX_X] += pos_part[idx_part*size_particle_positions + 3+IDX_X]*v0;
- rhs_part[idx_part*size_particle_rhs + IDX_Y] += pos_part[idx_part*size_particle_positions + 3+IDX_Y]*v0;
- rhs_part[idx_part*size_particle_rhs + IDX_Z] += pos_part[idx_part*size_particle_positions + 3+IDX_Z]*v0;
- }
- }
-
+ void compute_interaction(
+ const partsize_t nb_particles,
+ const real_number pos_part[],
+ real_number rhs_part[]) const;
// for given orientation and right-hand-side, recompute right-hand-side such
// that it is perpendicular to the current orientation.
// this is the job of the Lagrange multiplier terms, hence the
// "add_Lagrange_multipliers" name of the method.
template <int size_particle_positions, int size_particle_rhs>
- void add_Lagrange_multipliers(const partsize_t nb_particles, const real_number pos_part[], real_number rhs_part[]) const{
- static_assert(size_particle_positions == 6, "This kernel works only with 6 values for one particle's position");
- static_assert(size_particle_rhs == 6, "This kernel works only with 6 values per particle's rhs");
-
- #pragma omp parallel for
- for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
- const partsize_t idx0 = idx_part*size_particle_positions + 3;
- const partsize_t idx1 = idx_part*size_particle_rhs + 3;
- // check that orientation is unit vector:
- real_number orientation_size = sqrt(
- pos_part[idx0+IDX_X]*pos_part[idx0+IDX_X] +
- pos_part[idx0+IDX_Y]*pos_part[idx0+IDX_Y] +
- pos_part[idx0+IDX_Z]*pos_part[idx0+IDX_Z]);
- variable_used_only_in_assert(orientation_size);
- assert(orientation_size > 0.99);
- assert(orientation_size < 1.01);
- // I call "rotation" to be the right hand side of the orientation part of the ODE
- // project rotation on orientation:
- real_number projection = (
- pos_part[idx0+IDX_X]*rhs_part[idx1+IDX_X] +
- pos_part[idx0+IDX_Y]*rhs_part[idx1+IDX_Y] +
- pos_part[idx0+IDX_Z]*rhs_part[idx1+IDX_Z]);
-
- // now remove parallel bit.
- rhs_part[idx1+IDX_X] -= pos_part[idx0+IDX_X]*projection;
- rhs_part[idx1+IDX_Y] -= pos_part[idx0+IDX_Y]*projection;
- rhs_part[idx1+IDX_Z] -= pos_part[idx0+IDX_Z]*projection;
-
- // DEBUG
- // sanity check, for debugging purposes
- // compute dot product between orientation and orientation change
- //real_number dotproduct = (
- // rhs_part[idx1 + IDX_X]*pos_part[idx0 + IDX_X] +
- // rhs_part[idx1 + IDX_Y]*pos_part[idx0 + IDX_Y] +
- // rhs_part[idx1 + IDX_Z]*pos_part[idx0 + IDX_Z]);
- //if (dotproduct > 0.1)
- //{
- // DEBUG_MSG("dotproduct = %g, projection = %g\n"
- // "pos_part[%d] = %g, pos_part[%d] = %g, pos_part[%d] = %g\n"
- // "rhs_part[%d] = %g, rhs_part[%d] = %g, rhs_part[%d] = %g\n",
- // dotproduct,
- // projection,
- // IDX_X, pos_part[idx0 + IDX_X],
- // IDX_Y, pos_part[idx0 + IDX_Y],
- // IDX_Z, pos_part[idx0 + IDX_Z],
- // IDX_X, rhs_part[idx1 + IDX_X],
- // IDX_Y, rhs_part[idx1 + IDX_Y],
- // IDX_Z, rhs_part[idx1 + IDX_Z]);
- // assert(false);
- //}
- //assert(dotproduct <= 0.1);
- }
- }
-
+ void add_Lagrange_multipliers(
+ const partsize_t nb_particles,
+ const real_number pos_part[],
+ real_number rhs_part[]) const;
template <int size_particle_positions, int size_particle_rhs, int size_particle_rhs_extra>
- void compute_interaction_with_extra(const partsize_t nb_particles, const real_number pos_part[], real_number rhs_part[],
- const real_number rhs_part_extra[]) const{
- static_assert(size_particle_rhs_extra == 3, "This kernel works only with 3 values for one particle's rhs extra");
-
- // call plain compute_interaction first
- compute_interaction<size_particle_positions, size_particle_rhs>(nb_particles, pos_part, rhs_part);
-
- // now add vorticity term
- #pragma omp parallel for
- for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
- // Cross product vorticity/orientation
- rhs_part[idx_part*size_particle_rhs + 3+IDX_X] += 0.5*(rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Y]*pos_part[idx_part*size_particle_positions + 3+IDX_Z] -
- rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Z]*pos_part[idx_part*size_particle_positions + 3+IDX_Y]);
- rhs_part[idx_part*size_particle_rhs + 3+IDX_Y] += 0.5*(rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Z]*pos_part[idx_part*size_particle_positions + 3+IDX_X] -
- rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_X]*pos_part[idx_part*size_particle_positions + 3+IDX_Z]);
- rhs_part[idx_part*size_particle_rhs + 3+IDX_Z] += 0.5*(rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_X]*pos_part[idx_part*size_particle_positions + 3+IDX_Y] -
- rhs_part_extra[idx_part*size_particle_rhs_extra + IDX_Y]*pos_part[idx_part*size_particle_positions + 3+IDX_X]);
- }
- }
-
+ void compute_interaction_with_extra(
+ const partsize_t nb_particles,
+ const real_number pos_part[],
+ real_number rhs_part[],
+ const real_number rhs_part_extra[]) const;
// meant to be called AFTER executing the time-stepping operation.
// once the particles have been moved, ensure that the orientation is a unit vector.
template <int size_particle_positions>
- void enforce_unit_orientation(const partsize_t nb_particles, real_number pos_part[]) const{
- static_assert(size_particle_positions == 6, "This kernel works only with 6 values for one particle's position");
-
- #pragma omp parallel for
- for(partsize_t idx_part = 0 ; idx_part < nb_particles ; ++idx_part){
- const partsize_t idx0 = idx_part*size_particle_positions + 3;
- // compute orientation size:
- real_number orientation_size = sqrt(
- pos_part[idx0+IDX_X]*pos_part[idx0+IDX_X] +
- pos_part[idx0+IDX_Y]*pos_part[idx0+IDX_Y] +
- pos_part[idx0+IDX_Z]*pos_part[idx0+IDX_Z]);
- // now renormalize
- pos_part[idx0 + IDX_X] /= orientation_size;
- pos_part[idx0 + IDX_Y] /= orientation_size;
- pos_part[idx0 + IDX_Z] /= orientation_size;
- }
- }
-
+ void enforce_unit_orientation(
+ const partsize_t nb_particles,
+ real_number pos_part[]) const;
bool isEnable() const {
return isActive;
diff --git a/setup.py b/setup.py
index 3094c692..34f63082 100644
--- a/setup.py
+++ b/setup.py
@@ -126,7 +126,8 @@ src_file_list = [
'full_code/test_interpolation',
'full_code/NSVEparticles',
'full_code/NSVEcomplex_particles',
- 'full_code/NSVEp_extra_sampling']
+ 'full_code/NSVEp_extra_sampling',
+ 'particles/particles_inner_computer']
particle_headers = [
'cpp/particles/abstract_particles_input.hpp',
@@ -144,7 +145,7 @@ particle_headers = [
'cpp/particles/particles_field_computer.hpp',
'cpp/particles/particles_generic_interp.hpp',
'cpp/particles/particles_inner_computer_empty.hpp',
- 'cpp/particles/particles_inner_computer.hpp',
+ #'cpp/particles/particles_inner_computer.hpp',
'cpp/particles/particles_input_hdf5.hpp',
'cpp/particles/particles_output_hdf5.hpp',
'cpp/particles/particles_output_mpiio.hpp',
@@ -210,6 +211,9 @@ class CompileLibCommand(distutils.cmd.Command):
if not os.path.isdir('obj/full_code'):
os.makedirs('obj/full_code')
need_to_compile = True
+ if not os.path.isdir('obj/particles'):
+ os.makedirs('obj/particles')
+ need_to_compile = True
if not os.path.isfile('bfps/libbfps.a'):
need_to_compile = True
else:
--
GitLab