2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2016,2017,2018,2019,2020,2021, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
37 * \brief Implements GPU 3D FFT routines for OpenCL.
39 * \author Aleksei Iupinov <a.yupinov@gmail.com>
40 * \author Mark Abraham <mark.j.abraham@gmail.com>
46 #include "gpu_3dfft.h"
53 #include "gromacs/gpu_utils/device_context.h"
54 #include "gromacs/gpu_utils/device_stream.h"
55 #include "gromacs/gpu_utils/gmxopencl.h"
56 #include "gromacs/utility/exceptions.h"
57 #include "gromacs/utility/gmxassert.h"
58 #include "gromacs/utility/stringutil.h"
66 Impl(ivec realGridSize,
67 ivec realGridSizePadded,
68 ivec complexGridSizePadded,
69 bool useDecomposition,
70 bool performOutOfPlaceFFT,
71 const DeviceContext& context,
72 const DeviceStream& pmeStream,
73 DeviceBuffer<float> realGrid,
74 DeviceBuffer<float> complexGrid);
77 clfftPlanHandle planR2C_;
78 clfftPlanHandle planC2R_;
79 std::vector<cl_command_queue> commandStreams_;
84 //! Throws the exception on clFFT error
85 static void handleClfftError(clfftStatus status, const char* msg)
87 // Supposedly it's just a superset of standard OpenCL errors
88 if (status != CLFFT_SUCCESS)
90 GMX_THROW(InternalError(formatString("%s: %d", msg, status)));
94 Gpu3dFft::Impl::Impl(ivec realGridSize,
95 ivec realGridSizePadded,
96 ivec complexGridSizePadded,
97 const bool useDecomposition,
98 const bool performOutOfPlaceFFT,
99 const DeviceContext& context,
100 const DeviceStream& pmeStream,
101 DeviceBuffer<float> realGrid,
102 DeviceBuffer<float> complexGrid) :
103 realGrid_(realGrid), complexGrid_(complexGrid)
105 GMX_RELEASE_ASSERT(!useDecomposition, "FFT decomposition not implemented");
107 cl_context clContext = context.context();
108 commandStreams_.push_back(pmeStream.stream());
110 // clFFT expects row-major, so dimensions/strides are reversed (ZYX instead of XYZ)
111 std::array<size_t, DIM> realGridDimensions = { size_t(realGridSize[ZZ]),
112 size_t(realGridSize[YY]),
113 size_t(realGridSize[XX]) };
114 std::array<size_t, DIM> realGridStrides = {
115 1, size_t(realGridSizePadded[ZZ]), size_t(realGridSizePadded[YY] * realGridSizePadded[ZZ])
117 std::array<size_t, DIM> complexGridStrides = {
118 1, size_t(complexGridSizePadded[ZZ]), size_t(complexGridSizePadded[YY] * complexGridSizePadded[ZZ])
121 constexpr clfftDim dims = CLFFT_3D;
122 handleClfftError(clfftCreateDefaultPlan(&planR2C_, clContext, dims, realGridDimensions.data()),
123 "clFFT planning failure");
124 handleClfftError(clfftSetResultLocation(planR2C_, performOutOfPlaceFFT ? CLFFT_OUTOFPLACE : CLFFT_INPLACE),
125 "clFFT planning failure");
126 handleClfftError(clfftSetPlanPrecision(planR2C_, CLFFT_SINGLE), "clFFT planning failure");
127 constexpr cl_float scale = 1.0;
128 handleClfftError(clfftSetPlanScale(planR2C_, CLFFT_FORWARD, scale),
129 "clFFT coefficient setup failure");
130 handleClfftError(clfftSetPlanScale(planR2C_, CLFFT_BACKWARD, scale),
131 "clFFT coefficient setup failure");
133 // The only difference between 2 plans is direction
134 handleClfftError(clfftCopyPlan(&planC2R_, clContext, planR2C_), "clFFT plan copying failure");
136 handleClfftError(clfftSetLayout(planR2C_, CLFFT_REAL, CLFFT_HERMITIAN_INTERLEAVED),
137 "clFFT R2C layout failure");
138 handleClfftError(clfftSetLayout(planC2R_, CLFFT_HERMITIAN_INTERLEAVED, CLFFT_REAL),
139 "clFFT C2R layout failure");
141 handleClfftError(clfftSetPlanInStride(planR2C_, dims, realGridStrides.data()),
142 "clFFT stride setting failure");
143 handleClfftError(clfftSetPlanOutStride(planR2C_, dims, complexGridStrides.data()),
144 "clFFT stride setting failure");
146 handleClfftError(clfftSetPlanInStride(planC2R_, dims, complexGridStrides.data()),
147 "clFFT stride setting failure");
148 handleClfftError(clfftSetPlanOutStride(planC2R_, dims, realGridStrides.data()),
149 "clFFT stride setting failure");
151 handleClfftError(clfftBakePlan(planR2C_, commandStreams_.size(), commandStreams_.data(), nullptr, nullptr),
152 "clFFT precompiling failure");
153 handleClfftError(clfftBakePlan(planC2R_, commandStreams_.size(), commandStreams_.data(), nullptr, nullptr),
154 "clFFT precompiling failure");
156 // TODO: implement solve kernel as R2C FFT callback
157 // TODO: disable last transpose (clfftSetPlanTransposeResult)
160 Gpu3dFft::Impl::~Impl()
162 clfftDestroyPlan(&planR2C_);
163 clfftDestroyPlan(&planC2R_);
166 void Gpu3dFft::perform3dFft(gmx_fft_direction dir, CommandEvent* timingEvent)
168 cl_mem tempBuffer = nullptr;
169 constexpr std::array<cl_event, 0> waitEvents{ {} };
171 clfftPlanHandle plan;
172 clfftDirection direction;
173 cl_mem * inputGrids, *outputGrids;
177 case GMX_FFT_REAL_TO_COMPLEX:
178 plan = impl_->planR2C_;
179 direction = CLFFT_FORWARD;
180 inputGrids = &impl_->realGrid_;
181 outputGrids = &impl_->complexGrid_;
183 case GMX_FFT_COMPLEX_TO_REAL:
184 plan = impl_->planC2R_;
185 direction = CLFFT_BACKWARD;
186 inputGrids = &impl_->complexGrid_;
187 outputGrids = &impl_->realGrid_;
190 GMX_THROW(NotImplementedError("The chosen 3D-FFT case is not implemented on GPUs"));
192 handleClfftError(clfftEnqueueTransform(plan,
194 impl_->commandStreams_.size(),
195 impl_->commandStreams_.data(),
202 "clFFT execution failure");
205 Gpu3dFft::Gpu3dFft(ivec realGridSize,
206 ivec realGridSizePadded,
207 ivec complexGridSizePadded,
208 const bool useDecomposition,
209 const bool performOutOfPlaceFFT,
210 const DeviceContext& context,
211 const DeviceStream& pmeStream,
212 DeviceBuffer<float> realGrid,
213 DeviceBuffer<float> complexGrid) :
214 impl_(std::make_unique<Impl>(realGridSize,
216 complexGridSizePadded,
218 performOutOfPlaceFFT,
226 Gpu3dFft::~Gpu3dFft() = default;