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

Simplify P3M C++ API #4969

Merged
merged 3 commits into from
Aug 2, 2024
Merged

Simplify P3M C++ API #4969

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
52b98d9
Split FFT functions
jngrad Aug 1, 2024
2f97dd8
P3M class templates
jngrad Aug 1, 2024
0263159
Split FFT operations from halo communication
jngrad Aug 1, 2024
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
14 changes: 8 additions & 6 deletions cmake/FindFFTW3.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,25 +31,27 @@ if(FFTW3_INCLUDE_DIR)
endif(FFTW3_INCLUDE_DIR)

find_path(FFTW3_INCLUDE_DIR fftw3.h)
find_library(FFTW3_LIBRARIES NAMES fftw3)
find_library(FFTW3_LIBRARIES fftw3)
find_library(FFTW3F_LIBRARIES fftw3f)
find_path(FFTW3_MPI_INCLUDE_DIR fftw3-mpi.h)
find_library(FFTW3_MPI_LIBRARIES NAMES fftw3_mpi)
find_library(FFTW3_MPI_LIBRARIES fftw3_mpi)
find_library(FFTW3F_MPI_LIBRARIES fftw3f_mpi)

# handle the QUIETLY and REQUIRED arguments and set FFTW3_FOUND to TRUE if all
# listed variables are TRUE
include(FindPackageHandleStandardArgs)
find_package_handle_standard_args(FFTW3 DEFAULT_MSG FFTW3_LIBRARIES
find_package_handle_standard_args(FFTW3 DEFAULT_MSG FFTW3_LIBRARIES FFTW3F_LIBRARIES
FFTW3_INCLUDE_DIR)
set(FPHSA_NAME_MISMATCHED 1)
find_package_handle_standard_args(FFTW3_MPI DEFAULT_MSG FFTW3_MPI_LIBRARIES
find_package_handle_standard_args(FFTW3_MPI DEFAULT_MSG FFTW3_MPI_LIBRARIES FFTW3F_MPI_LIBRARIES
FFTW3_MPI_INCLUDE_DIR)
unset(FPHSA_NAME_MISMATCHED)

mark_as_advanced(FFTW3_LIBRARIES FFTW3_INCLUDE_DIR FFTW3_MPI_LIBRARIES FFTW3_MPI_INCLUDE_DIR)
mark_as_advanced(FFTW3_LIBRARIES FFTW3F_LIBRARIES FFTW3_INCLUDE_DIR FFTW3_MPI_LIBRARIES FFTW3F_MPI_LIBRARIES FFTW3_MPI_INCLUDE_DIR)


if(FFTW3_FOUND AND NOT TARGET FFTW3::FFTW3)
add_library(FFTW3::FFTW3 INTERFACE IMPORTED)
target_include_directories(FFTW3::FFTW3 INTERFACE "${FFTW3_INCLUDE_DIR}")
target_link_libraries(FFTW3::FFTW3 INTERFACE "${FFTW3_LIBRARIES}")
target_link_libraries(FFTW3::FFTW3 INTERFACE "${FFTW3_LIBRARIES}" "${FFTW3F_LIBRARIES}")
endif()
1 change: 0 additions & 1 deletion src/core/electrostatics/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,5 @@ target_sources(
${CMAKE_CURRENT_SOURCE_DIR}/mmm1d.cpp
${CMAKE_CURRENT_SOURCE_DIR}/mmm-modpsi.cpp
${CMAKE_CURRENT_SOURCE_DIR}/p3m.cpp
${CMAKE_CURRENT_SOURCE_DIR}/p3m_gpu.cpp
${CMAKE_CURRENT_SOURCE_DIR}/scafacos_impl.cpp
${CMAKE_CURRENT_SOURCE_DIR}/specfunc.cpp)
5 changes: 0 additions & 5 deletions src/core/electrostatics/coulomb.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,11 +207,6 @@ struct LongRangeForce {
void operator()(std::shared_ptr<CoulombP3M> const &actor) const {
actor->add_long_range_forces(m_particles);
}
#ifdef CUDA
void operator()(std::shared_ptr<CoulombP3MGPU> const &actor) const {
actor->add_long_range_forces(m_particles);
}
#endif // CUDA
void
operator()(std::shared_ptr<ElectrostaticLayerCorrection> const &actor) const {
actor->add_long_range_forces(m_particles);
Expand Down
4 changes: 0 additions & 4 deletions src/core/electrostatics/coulomb.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,6 @@
#include "electrostatics/icc.hpp"
#include "electrostatics/mmm1d.hpp"
#include "electrostatics/p3m.hpp"
#include "electrostatics/p3m_gpu.hpp"
#include "electrostatics/reaction_field.hpp"
#include "electrostatics/scafacos.hpp"

Expand All @@ -49,9 +48,6 @@ using ElectrostaticsActor =
std::variant<std::shared_ptr<DebyeHueckel>,
#ifdef P3M
std::shared_ptr<CoulombP3M>,
#ifdef CUDA
std::shared_ptr<CoulombP3MGPU>,
#endif // CUDA
std::shared_ptr<ElectrostaticLayerCorrection>,
#endif // P3M
std::shared_ptr<CoulombMMM1D>,
Expand Down
70 changes: 34 additions & 36 deletions src/core/electrostatics/elc.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,6 @@

#include "electrostatics/coulomb.hpp"
#include "electrostatics/p3m.hpp"
#include "electrostatics/p3m_gpu.hpp"

#include "BoxGeometry.hpp"
#include "Particle.hpp"
Expand Down Expand Up @@ -1107,39 +1106,34 @@ ElectrostaticLayerCorrection::ElectrostaticLayerCorrection(
adapt_solver();
}

static void p3m_assign_image_charge(elc_data const &elc, CoulombP3M &p3m,
double q, Utils::Vector3d const &pos) {
if (pos[2] < elc.space_layer) {
auto const q_eff = elc.delta_mid_bot * q;
p3m.assign_charge(q_eff, {pos[0], pos[1], -pos[2]});
}
if (pos[2] > (elc.box_h - elc.space_layer)) {
auto const q_eff = elc.delta_mid_top * q;
p3m.assign_charge(q_eff, {pos[0], pos[1], 2. * elc.box_h - pos[2]});
}
}

template <ChargeProtocol protocol, typename combined_ranges>
void charge_assign(elc_data const &elc, CoulombP3M &solver,
combined_ranges const &p_q_pos_range) {
if (protocol == ChargeProtocol::BOTH or protocol == ChargeProtocol::IMAGE) {
solver.p3m.inter_weights.reset(solver.p3m.params.cao);
}
/* prepare local FFT mesh */
for (int i = 0; i < solver.p3m.local_mesh.size; i++)
solver.p3m.mesh.rs_scalar[i] = 0.;

solver.prepare_fft_mesh(protocol == ChargeProtocol::BOTH or
protocol == ChargeProtocol::IMAGE);

for (auto zipped : p_q_pos_range) {
auto const p_q = boost::get<0>(zipped);
auto const &p_pos = boost::get<1>(zipped);
if (p_q != 0.) {
// assign real charges
if (protocol == ChargeProtocol::BOTH or
protocol == ChargeProtocol::REAL) {
solver.assign_charge(p_q, p_pos, solver.p3m.inter_weights);
solver.assign_charge(p_q, p_pos, false);
}
// assign image charges
if (protocol == ChargeProtocol::BOTH or
protocol == ChargeProtocol::IMAGE) {
p3m_assign_image_charge(elc, solver, p_q, p_pos);
if (p_pos[2] < elc.space_layer) {
auto const q_eff = elc.delta_mid_bot * p_q;
solver.assign_charge(q_eff, {p_pos[0], p_pos[1], -p_pos[2]}, true);
}
if (p_pos[2] > (elc.box_h - elc.space_layer)) {
auto const q_eff = elc.delta_mid_top * p_q;
solver.assign_charge(
q_eff, {p_pos[0], p_pos[1], 2. * elc.box_h - p_pos[2]}, true);
}
}
}
}
Expand All @@ -1149,7 +1143,9 @@ template <ChargeProtocol protocol, typename combined_range>
void modify_p3m_sums(elc_data const &elc, CoulombP3M &solver,
combined_range const &p_q_pos_range) {

Utils::Vector3d node_sums{};
auto local_n = 0;
auto local_q2 = 0.0;
auto local_q = 0.0;
for (auto zipped : p_q_pos_range) {
auto const p_q = boost::get<0>(zipped);
auto const &p_pos = boost::get<1>(zipped);
Expand All @@ -1158,33 +1154,35 @@ void modify_p3m_sums(elc_data const &elc, CoulombP3M &solver,

if (protocol == ChargeProtocol::BOTH or
protocol == ChargeProtocol::REAL) {
node_sums[0] += 1.;
node_sums[1] += Utils::sqr(p_q);
node_sums[2] += p_q;
local_n++;
local_q2 += Utils::sqr(p_q);
local_q += p_q;
}

if (protocol == ChargeProtocol::BOTH or
protocol == ChargeProtocol::IMAGE) {
if (p_z < elc.space_layer) {
node_sums[0] += 1.;
node_sums[1] += Utils::sqr(elc.delta_mid_bot * p_q);
node_sums[2] += elc.delta_mid_bot * p_q;
local_n++;
local_q2 += Utils::sqr(elc.delta_mid_bot * p_q);
local_q += elc.delta_mid_bot * p_q;
}

if (p_z > (elc.box_h - elc.space_layer)) {
node_sums[0] += 1.;
node_sums[1] += Utils::sqr(elc.delta_mid_top * p_q);
node_sums[2] += elc.delta_mid_top * p_q;
local_n++;
local_q2 += Utils::sqr(elc.delta_mid_top * p_q);
local_q += elc.delta_mid_top * p_q;
}
}
}
}

auto const tot_sums =
boost::mpi::all_reduce(comm_cart, node_sums, std::plus<>());
solver.p3m.sum_qpart = static_cast<int>(tot_sums[0] + 0.1);
solver.p3m.sum_q2 = tot_sums[1];
solver.p3m.square_sum_q = Utils::sqr(tot_sums[2]);
auto global_n = 0;
auto global_q2 = 0.;
auto global_q = 0.;
boost::mpi::all_reduce(comm_cart, local_n, global_n, std::plus<>());
boost::mpi::all_reduce(comm_cart, local_q2, global_q2, std::plus<>());
boost::mpi::all_reduce(comm_cart, local_q, global_q, std::plus<>());
solver.count_charged_particles_elc(global_n, global_q2, Utils::sqr(global_q));
}

double ElectrostaticLayerCorrection::long_range_energy(
Expand Down
8 changes: 2 additions & 6 deletions src/core/electrostatics/elc.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,13 +37,13 @@
#include "actor/traits.hpp"

#include "electrostatics/p3m.hpp"
#include "electrostatics/p3m_gpu.hpp"

#include "BoxGeometry.hpp"
#include "Particle.hpp"
#include "ParticleRange.hpp"

#include <utils/Vector.hpp>
#include <utils/math/sqr.hpp>

#include <memory>
#include <optional>
Expand Down Expand Up @@ -166,11 +166,7 @@ struct elc_data {

struct ElectrostaticLayerCorrection
: public Coulomb::Actor<ElectrostaticLayerCorrection> {
using BaseSolver = std::variant<
#ifdef CUDA
std::shared_ptr<CoulombP3MGPU>,
#endif // CUDA
std::shared_ptr<CoulombP3M>>;
using BaseSolver = std::variant<std::shared_ptr<CoulombP3M>>;

elc_data elc;
BoxGeometry *m_box_geo;
Expand Down
7 changes: 5 additions & 2 deletions src/core/electrostatics/icc.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,7 @@
#include "communication.hpp"
#include "electrostatics/coulomb.hpp"
#include "electrostatics/coulomb_inline.hpp"
#include "electrostatics/p3m.hpp"
#include "errorhandling.hpp"
#include "integrators/Propagation.hpp"
#include "system/System.hpp"
Expand Down Expand Up @@ -248,8 +249,10 @@ struct SanityChecksICC {
template <typename T> void operator()(std::shared_ptr<T> const &) const {}
#ifdef P3M
#ifdef CUDA
[[noreturn]] void operator()(std::shared_ptr<CoulombP3MGPU> const &) const {
throw std::runtime_error("ICC does not work with P3MGPU");
void operator()(std::shared_ptr<CoulombP3M> const &p) const {
if (p->is_gpu()) {
throw std::runtime_error("ICC does not work with P3MGPU");
}
}
#endif // CUDA
void
Expand Down
Loading