Homec4science
Diffusion µSpectre (feat/split_tests_T2189) Browse

muSpectre/src/fft/pfft_engine.cc
656437418c7efeat/split_tests_T2189

Last Change
Raw File

pfft_engine.cc
View Options

/**
* @file pfft_engine.cc
*
* @author Lars Pastewka <lars.pastewka@imtek.uni-freiburg.de>
*
* @date 06 Mar 2017
*
* @brief implements the MPI-parallel pfft engine
*
* Copyright © 2017 Till Junge
*
* µSpectre is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation, either version 3, or (at
* your option) any later version.
*
* µSpectre is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with GNU Emacs; see the file COPYING. If not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 02111-1307, USA.
*/
#include "common/ccoord_operations.hh"
#include "fft/pfft_engine.hh"
namespace muSpectre {
template <Dim_t DimsS, Dim_t DimM>
int PFFTEngine<DimsS, DimM>::nb_engines{0};
template <Dim_t DimS, Dim_t DimM>
PFFTEngine<DimS, DimM>::PFFTEngine(Ccoord resolutions, Rcoord lengths,
Communicator comm)
:Parent{resolutions, lengths, comm}, mpi_comm{comm.get_mpi_comm()}
{
if (!this->nb_engines) pfft_init();
this->nb_engines++;
int size{comm.size()};
int dim_x{size};
int dim_y{1};
// Note: All TODOs below don't affect the function of the PFFT engine. It
// presently uses slab decompositions, the TODOs are what needs to be done
// to get stripe decomposition to work - but it does not work yet. Slab
// vs stripe decomposition may have an impact on how the code scales.
// TODO: Enable this to enable 2d process mesh. This does not pass tests.
//if (DimS > 2) {
if (false) {
dim_y = int(sqrt(size));
while ((size/dim_y)*dim_y != size) dim_y--;
dim_x = size/dim_y;
}
// TODO: Enable this to enable 2d process mesh. This does not pass tests.
//if (DimS > 2) {
if (false) {
if (pfft_create_procmesh_2d(this->comm.get_mpi_comm(), dim_x, dim_y,
&this->mpi_comm)) {
throw std::runtime_error("Failed to create 2d PFFT process mesh.");
}
}
std::array<ptrdiff_t, DimS> narr;
std::copy(this->domain_resolutions.begin(), this->domain_resolutions.end(),
narr.begin());
ptrdiff_t res[DimS], loc[DimS], fres[DimS], floc[DimS];
this->workspace_size =
pfft_local_size_many_dft_r2c(DimS, narr.data(), narr.data(), narr.data(),
Field_t::nb_components,
PFFT_DEFAULT_BLOCKS, PFFT_DEFAULT_BLOCKS,
this->mpi_comm,
PFFT_TRANSPOSED_OUT,
res, loc, fres, floc);
std::copy(res, res+DimS, this->resolutions.begin());
std::copy(loc, loc+DimS, this->locations.begin());
std::copy(fres, fres+DimS, this->fourier_resolutions.begin());
std::copy(floc, floc+DimS, this->fourier_locations.begin());
// TODO: Enable this to enable 2d process mesh. This does not pass tests.
//for (int i = 0; i < DimS-1; ++i) {
for (int i = 0; i < 1; ++i) {
std::swap(this->fourier_resolutions[i], this->fourier_resolutions[i+1]);
std::swap(this->fourier_locations[i], this->fourier_locations[i+1]);
}
for (auto & n: this->resolutions) {
if (n == 0) {
throw std::runtime_error("PFFT planning returned zero resolution. "
"You may need to run on fewer processes.");
}
}
for (auto & n: this->fourier_resolutions) {
if (n == 0) {
throw std::runtime_error("PFFT planning returned zero Fourier "
"resolution. You may need to run on fewer "
"processes.");
}
}
for (auto && pixel:
std::conditional_t<
DimS==2,
CcoordOps::Pixels<DimS, 1, 0>,
// TODO: This should be the correct order of dimension for a 2d
// process mesh, but tests don't pass.
//CcoordOps::Pixels<DimS, 1, 2, 0>
CcoordOps::Pixels<DimS, 1, 0, 2>
>(this->fourier_resolutions, this->fourier_locations)) {
this->work_space_container.add_pixel(pixel);
}
}
/* ---------------------------------------------------------------------- */
template <Dim_t DimS, Dim_t DimM>
void PFFTEngine<DimS, DimM>::initialise(FFT_PlanFlags plan_flags) {
if (this->initialised) {
throw std::runtime_error("double initialisation, will leak memory");
}
// Initialize parent after local resolutions have been determined and
// work space has been initialized
Parent::initialise(plan_flags);
this->real_workspace = pfft_alloc_real(2*this->workspace_size);
// We need to check whether the workspace provided by our field is large
// enough. PFFT may request a workspace size larger than the nominal size
// of the complex buffer.
if (long(this->work.size()*Field_t::nb_components) < this->workspace_size) {
this->work.set_pad_size(this->workspace_size -
Field_t::nb_components*this->work.size());
}
unsigned int flags;
switch (plan_flags) {
case FFT_PlanFlags::estimate: {
flags = PFFT_ESTIMATE;
break;
}
case FFT_PlanFlags::measure: {
flags = PFFT_MEASURE;
break;
}
case FFT_PlanFlags::patient: {
flags = PFFT_PATIENT;
break;
}
default:
throw std::runtime_error("unknown planner flag type");
break;
}
std::array<ptrdiff_t, DimS> narr;
std::copy(this->domain_resolutions.begin(), this->domain_resolutions.end(),
narr.begin());
Real * in{this->real_workspace};
pfft_complex * out{reinterpret_cast<pfft_complex*>(this->work.data())};
this->plan_fft = pfft_plan_many_dft_r2c(DimS, narr.data(), narr.data(),
narr.data(), Field_t::nb_components,
PFFT_DEFAULT_BLOCKS,
PFFT_DEFAULT_BLOCKS,
in, out, this->mpi_comm,
PFFT_FORWARD,
PFFT_TRANSPOSED_OUT | flags);
if (this->plan_fft == nullptr) {
throw std::runtime_error("r2c plan failed");
}
pfft_complex * i_in{reinterpret_cast<pfft_complex*>(this->work.data())};
Real * i_out{this->real_workspace};
this->plan_ifft = pfft_plan_many_dft_c2r(DimS, narr.data(), narr.data(),
narr.data(),
Field_t::nb_components,
PFFT_DEFAULT_BLOCKS,
PFFT_DEFAULT_BLOCKS,
i_in, i_out,
this->mpi_comm,
PFFT_BACKWARD,
PFFT_TRANSPOSED_IN | flags);
if (this->plan_ifft == nullptr) {
throw std::runtime_error("c2r plan failed");
}
this->initialised = true;
}
/* ---------------------------------------------------------------------- */
template <Dim_t DimS, Dim_t DimM>
PFFTEngine<DimS, DimM>::~PFFTEngine<DimS, DimM>() noexcept {
if (this->real_workspace != nullptr) pfft_free(this->real_workspace);
if (this->plan_fft != nullptr) pfft_destroy_plan(this->plan_fft);
if (this->plan_ifft != nullptr) pfft_destroy_plan(this->plan_ifft);
if (this->mpi_comm != this->comm.get_mpi_comm()) {
MPI_Comm_free(&this->mpi_comm);
}
// TODO: We cannot run fftw_mpi_cleanup since also calls fftw_cleanup
// and any running FFTWEngine will fail afterwards.
//this->nb_engines--;
//if (!this->nb_engines) pfft_cleanup();
}
/* ---------------------------------------------------------------------- */
template <Dim_t DimS, Dim_t DimM>
typename PFFTEngine<DimS, DimM>::Workspace_t &
PFFTEngine<DimS, DimM>::fft (Field_t & field) {
if (!this->plan_fft) {
throw std::runtime_error("fft plan not allocated");
}
if (field.size() != CcoordOps::get_size(this->resolutions)) {
throw std::runtime_error("size mismatch");
}
// Copy field data to workspace buffer. This is necessary because workspace
// buffer is larger than field buffer.
std::copy(field.data(), field.data()+Field_t::nb_components*field.size(),
this->real_workspace);
pfft_execute_dft_r2c(
this->plan_fft, this->real_workspace,
reinterpret_cast<pfft_complex*>(this->work.data()));
return this->work;
}
/* ---------------------------------------------------------------------- */
template <Dim_t DimS, Dim_t DimM>
void
PFFTEngine<DimS, DimM>::ifft (Field_t & field) const {
if (!this->plan_ifft) {
throw std::runtime_error("ifft plan not allocated");
}
if (field.size() != CcoordOps::get_size(this->resolutions)) {
throw std::runtime_error("size mismatch");
}
pfft_execute_dft_c2r(
this->plan_ifft, reinterpret_cast<pfft_complex*>(this->work.data()),
this->real_workspace);
std::copy(this->real_workspace,
this->real_workspace+Field_t::nb_components*field.size(),
field.data());
}
template class PFFTEngine<twoD, twoD>;
template class PFFTEngine<twoD, threeD>;
template class PFFTEngine<threeD, threeD>;
} // muSpectre
c4science · Help