Homec4science
Diffusion tamaas (features/netcdf-mpi) Browse

tamaas/src/core/cuda/cufft_engine.hh
483b0d63989bfeatures/netcdf-mpi

Last Change
Raw File

cufft_engine.hh
View Options

/*
* SPDX-License-Indentifier: AGPL-3.0-or-later
*
* Copyright (©) 2016-2021 EPFL (École Polytechnique Fédérale de Lausanne),
* Laboratory (LSMS - Laboratoire de Simulation en Mécanique des Solides)
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as published
* by the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program 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 Affero General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <https://www.gnu.org/licenses/>.
*
*/
/* -------------------------------------------------------------------------- */
#ifndef CUFFT_ENGINE_H
#define CUFFT_ENGINE_H
/* -------------------------------------------------------------------------- */
#include "fft_engine.hh"
#include <cufft.h>
/* -------------------------------------------------------------------------- */
namespace cufft {
struct plan {
cufftHandle _plan;
plan() = default;
plan(plan&& o) noexcept : _plan(o._plan) {}
plan& operator=(plan&& o) noexcept {
_plan = o._plan;
return *this;
}
/// Destroy plan
~plan() noexcept { cufftDestroy(_plan); }
/// For seamless use with fftw api
operator cufftHandle() const { return _plan; }
};
} // namespace cufft
namespace tamaas {
class CuFFTEngine : public FFTEngine {
private:
using plan_t = std::pair<cufft::plan, cufft::plan>;
/// Perform forward (R2C) transform
template <UInt dim>
void forwardImpl(const Grid<Real, dim>& real,
GridHermitian<Real, dim>& spectral);
/// Perform backward (C2R) transform
template <UInt dim>
void backwardImpl(Grid<Real, dim>& real,
const GridHermitian<Real, dim>& spectral);
/// Return the plans pair for a given transform signature
plan_t& getPlans(key_t key);
public:
/// Initialize with flags
explicit CuFFTEngine(unsigned int flags = FFTW_ESTIMATE) noexcept
: _flags(flags), plans() {}
void forward(const Grid<Real, 1>& real,
GridHermitian<Real, 1>& spectral) override {
forwardImpl(real, spectral);
}
void forward(const Grid<Real, 2>& real,
GridHermitian<Real, 2>& spectral) override {
forwardImpl(real, spectral);
}
void backward(Grid<Real, 1>& real,
GridHermitian<Real, 1>& spectral) override {
backwardImpl(real, spectral);
}
void backward(Grid<Real, 2>& real,
GridHermitian<Real, 2>& spectral) override {
backwardImpl(real, spectral);
}
unsigned int flags() const { return _flags; }
/// Cast to FFTW complex type
static auto cast(Complex* data) {
return reinterpret_cast<cufftDoubleComplex*>(data);
}
static auto cast(const Complex* data) {
return const_cast<cufftDoubleComplex*>(
reinterpret_cast<const cufftDoubleComplex*>(data));
}
protected:
unsigned int _flags; ///< FFTW flags
std::map<key_t, plan_t> plans; ///< plans corresponding to signatures
};
/* -------------------------------------------------------------------------- */
template <UInt dim>
void CuFFTEngine::forwardImpl(const Grid<Real, dim>& real,
GridHermitian<Real, dim>& spectral) {
auto& plans = getPlans(make_key(real, spectral));
if (cufftExecD2Z(plans.first, const_cast<Real*>(real.getInternalData()),
cast(spectral.getInternalData())) != CUFFT_SUCCESS)
TAMAAS_EXCEPTION("Forward transform fail");
cudaDeviceSynchronize(); ///< TODO ask the SCITAS guys
}
template <UInt dim>
void CuFFTEngine::backwardImpl(Grid<Real, dim>& real,
const GridHermitian<Real, dim>& spectral) {
auto& plans = getPlans(make_key(real, spectral));
if (cufftExecZ2D(plans.second, cast(spectral.getInternalData()),
real.getInternalData()) != CUFFT_SUCCESS)
TAMAAS_EXCEPTION("Backward transform fail");
cudaDeviceSynchronize(); ///< TODO ask the SCITAS guys
// Normalize
real *= (1. / real.getNbPoints());
}
} // namespace tamaas
#endif
c4science · Help