#ifndef OPENMM_OPENCLCONTEXT_H_
#define OPENMM_OPENCLCONTEXT_H_
/* -------------------------------------------------------------------------- *
* OpenMM *
* -------------------------------------------------------------------------- *
* This is part of the OpenMM molecular simulation toolkit originating from *
* Simbios, the NIH National Center for Physics-Based Simulation of *
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
* This program is free software: you can redistribute it and/or modify *
* it under the terms of the GNU Lesser 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 Lesser General Public License for more details. *
* *
* You should have received a copy of the GNU Lesser General Public License *
* along with this program. If not, see . *
* -------------------------------------------------------------------------- */
#define __CL_ENABLE_EXCEPTIONS
#include
namespace OpenMM {
template
class OpenCLArray;
class OpenCLForceInfo;
class System;
/**
* We can't use predefined vector types like cl_float4, since different OpenCL implementations currently define
* them in incompatible ways. Hopefully that will be fixed in the future. In the mean time, we define our own
* types to represent them on the host.
*/
typedef struct {
cl_float x, y;
} mm_float2;
typedef struct {
cl_float x, y, z, w;
} mm_float4;
typedef struct {
cl_int x, y;
} mm_int2;
typedef struct {
cl_int x, y, z, w;
} mm_int4;
/**
* This class contains the information associated with a Context by the OpenCL Platform.
*/
class OpenCLContext {
public:
static const int ThreadBlockSize = 64;
static const int TileSize = 32;
OpenCLContext(int numParticles, int platformIndex, int deviceIndex);
~OpenCLContext();
/**
* This is called to initialize internal data structures after all Forces in the system
* have been initialized.
*/
void initialize(const System& system);
/**
* Add an OpenCLForce to this context.
*/
void addForce(OpenCLForceInfo* force);
/**
* Get the cl::Context associated with this object.
*/
cl::Context& getContext() {
return context;
}
/**
* Get the cl::CommandQueue associated with this object.
*/
cl::CommandQueue& getQueue() {
return queue;
}
/**
* Get the array which contains the position and charge of each atom.
*/
OpenCLArray& getPosq() {
return *posq;
}
/**
* Get the array which contains the velocity and massof each atom.
*/
OpenCLArray& getVelm() {
return *velm;
}
/**
* Get the array which contains the force on each atom.
*/
OpenCLArray& getForce() {
return *force;
}
/**
* Get the array which contains the buffers in which forces are computed.
*/
OpenCLArray& getForceBuffers() {
return *forceBuffers;
}
/**
* Get the array which contains the buffer in which energy is computed.
*/
OpenCLArray& getEnergyBuffer() {
return *energyBuffer;
}
/**
* Get the array which contains the index of each atom.
*/
OpenCLArray& getAtomIndex() {
return *atomIndex;
}
/**
* Load OpenCL source code from a file in the kernels directory.
*/
std::string loadSourceFromFile(const std::string& filename) const;
/**
* Create an OpenCL Program from source code.
*/
cl::Program createProgram(const std::string source);
/**
* Execute a kernel.
*
* @param kernel the kernel to execute
* @param workUnits the maximum number of work units that should be used
*/
void executeKernel(cl::Kernel& kernel, int workUnits);
/**
* Set all elements of an array to 0.
*/
void clearBuffer(OpenCLArray& array);
/**
* Set all elements of an array to 0.
*/
void clearBuffer(OpenCLArray& array);
/**
* Given a collection of buffers packed into an array, sum them and store
* the sum in the first buffer.
*
* @param array the array containing the buffers to reduce
* @param numBuffers the number of buffers packed into the array
*/
void reduceBuffer(OpenCLArray& array, int numBuffers);
/**
* Get the number of atoms.
*/
int getNumAtoms() const {
return numAtoms;
}
/**
* Get the number of atoms, rounded up to a multiple of TileSize. This is the actual size of
* most arrays with one element per atom.
*/
int getPaddedNumAtoms() const {
return paddedNumAtoms;
}
/**
* Get the number of blocks of TileSize atoms.
*/
int getNumAtomBlocks() const {
return numAtomBlocks;
}
/**
* Get the standard number of thread blocks to use when executing kernels.
*/
int getNumThreadBlocks() const {
return numThreadBlocks;
}
/**
* Get the total number of tiles used for nonbonded computation.
*/
int getNumTiles() const {
return numTiles;
}
/**
* Get the number of force buffers.
*/
int getNumForceBuffers() const {
return numForceBuffers;
}
private:
int numAtoms;
int paddedNumAtoms;
int numAtomBlocks;
int numTiles;
int numThreadBlocks;
int numForceBuffers;
cl::Context context;
cl::Device device;
cl::CommandQueue queue;
cl::Program utilities;
cl::Kernel clearBufferKernel;
cl::Kernel reduceFloat4Kernel;
std::vector forces;
OpenCLArray* posq;
OpenCLArray* velm;
OpenCLArray* force;
OpenCLArray* forceBuffers;
OpenCLArray* energyBuffer;
OpenCLArray* atomIndex;
};
} // namespace OpenMM
#endif /*OPENMM_OPENCLCONTEXT_H_*/