CudaKernels.cpp 36 KB
Newer Older
1
2
3
4
5
6
7
8
/* -------------------------------------------------------------------------- *
 *                                   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.               *
 *                                                                            *
9
 * Portions copyright (c) 2008-2009 Stanford University and the Authors.      *
10
11
12
 * Authors: Peter Eastman                                                     *
 * Contributors:                                                              *
 *                                                                            *
13
14
15
16
 * 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.                                        *
17
 *                                                                            *
18
19
20
21
 * 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.                        *
22
 *                                                                            *
23
24
 * You should have received a copy of the GNU Lesser General Public License   *
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.      *
25
26
27
 * -------------------------------------------------------------------------- */

#include "CudaKernels.h"
28
#include "openmm/LangevinIntegrator.h"
29
30
#include "openmm/Context.h"
#include "openmm/internal/ContextImpl.h"
31
#include "openmm/internal/NonbondedForceImpl.h"
32
#include "kernels/gputypes.h"
33
#include "kernels/cudaKernels.h"
34
35
36
37
38
39
40
#include <cmath>

extern "C" int gpuSetConstants( gpuContext gpu );

using namespace OpenMM;
using namespace std;

41
42
43
44
void CudaCalcForcesAndEnergyKernel::initialize(const System& system) {
}

void CudaCalcForcesAndEnergyKernel::beginForceComputation(ContextImpl& context) {
45
    _gpuContext* gpu = data.gpu;
46
    if (data.nonbondedMethod != NO_CUTOFF && data.computeForceCount%100 == 0)
47
        gpuReorderAtoms(gpu);
48
    data.computeForceCount++;
49
    kClearForces(gpu);
50
51
52
53
}

void CudaCalcForcesAndEnergyKernel::finishForceComputation(ContextImpl& context) {
    _gpuContext* gpu = data.gpu;
Mark Friedrichs's avatar
Mark Friedrichs committed
54
55

    if (gpu->bIncludeGBSA || gpu->bIncludeGBVI) {
56
        gpu->bRecalculateBornRadii = true;
57
58
        kCalculateCDLJObcGbsaForces1(gpu);
        kReduceObcGbsaBornForces(gpu);
Mark Friedrichs's avatar
Mark Friedrichs committed
59
60
61
62
63
        if (gpu->bIncludeGBSA ) { 
           kCalculateObcGbsaForces2(gpu);
        } else {
           kCalculateGBVIForces2(gpu);
        }
64
    }
65
    else if (data.hasNonbonded)
66
        kCalculateCDLJForces(gpu);
67
68
    if (data.hasCustomNonbonded)
        kCalculateCustomNonbondedForces(gpu, data.hasNonbonded);
69
    kCalculateLocalForces(gpu);
70
    kReduceForces(gpu);
71
72
}

73
void CudaCalcForcesAndEnergyKernel::beginEnergyComputation(ContextImpl& context) {
74
    _gpuContext* gpu = data.gpu;
75
76
77
78
    if (data.nonbondedMethod != NO_CUTOFF && data.stepCount%100 == 0)
        gpuReorderAtoms(gpu);
    data.stepCount++;
    kClearEnergy(gpu);
79
80
81
82
}

double CudaCalcForcesAndEnergyKernel::finishEnergyComputation(ContextImpl& context) {
    _gpuContext* gpu = data.gpu;
Mark Friedrichs's avatar
Mark Friedrichs committed
83
    if (gpu->bIncludeGBSA || gpu->bIncludeGBVI) {
84
85
86
        gpu->bRecalculateBornRadii = true;
        kCalculateCDLJObcGbsaForces1(gpu);
        kReduceObcGbsaBornForces(gpu);
Mark Friedrichs's avatar
Mark Friedrichs committed
87
88
89
90
91
        if (gpu->bIncludeGBSA ) {
           kCalculateObcGbsaForces2(gpu);
        } else {
           kCalculateGBVIForces2(gpu);
        }
92
    }
93
94
95
96
97
98
    else if (data.hasNonbonded)
        kCalculateCDLJForces(gpu);
    if (data.hasCustomNonbonded)
        kCalculateCustomNonbondedForces(gpu, data.hasNonbonded);
    kCalculateLocalForces(gpu);
    return kReduceEnergy(gpu)+data.ewaldSelfEnergy;
99
100
}

101
void CudaUpdateStateDataKernel::initialize(const System& system) {
102
103
}

104
double CudaUpdateStateDataKernel::getTime(const ContextImpl& context) const {
105
106
107
    return data.time;
}

108
void CudaUpdateStateDataKernel::setTime(ContextImpl& context, double time) {
109
110
111
    data.time = time;
}

112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
void CudaUpdateStateDataKernel::getPositions(ContextImpl& context, std::vector<Vec3>& positions) {
    _gpuContext* gpu = data.gpu;
    gpu->psPosq4->Download();
    int* order = gpu->psAtomIndex->_pSysData;
    int numParticles = context.getSystem().getNumParticles();
    positions.resize(numParticles);
    for (int i = 0; i < numParticles; ++i) {
        float4 pos = (*gpu->psPosq4)[i];
        int3 offset = gpu->posCellOffsets[i];
        positions[order[i]] = Vec3(pos.x-offset.x*gpu->sim.periodicBoxSizeX, pos.y-offset.y*gpu->sim.periodicBoxSizeY, pos.z-offset.z*gpu->sim.periodicBoxSizeZ);
    }
}

void CudaUpdateStateDataKernel::setPositions(ContextImpl& context, const std::vector<Vec3>& positions) {
    _gpuContext* gpu = data.gpu;
    int* order = gpu->psAtomIndex->_pSysData;
    int numParticles = context.getSystem().getNumParticles();
    for (int i = 0; i < numParticles; ++i) {
        float4& pos = (*gpu->psPosq4)[i];
        const Vec3& p = positions[order[i]];
        pos.x = p[0];
        pos.y = p[1];
        pos.z = p[2];
    }
    gpu->psPosq4->Upload();
    for (int i = 0; i < gpu->posCellOffsets.size(); i++)
        gpu->posCellOffsets[i] = make_int3(0, 0, 0);
}

void CudaUpdateStateDataKernel::getVelocities(ContextImpl& context, std::vector<Vec3>& velocities) {
    _gpuContext* gpu = data.gpu;
    gpu->psVelm4->Download();
    int* order = gpu->psAtomIndex->_pSysData;
    int numParticles = context.getSystem().getNumParticles();
    velocities.resize(numParticles);
    for (int i = 0; i < numParticles; ++i) {
        float4 vel = (*gpu->psVelm4)[i];
        velocities[order[i]] = Vec3(vel.x, vel.y, vel.z);
    }
}

void CudaUpdateStateDataKernel::setVelocities(ContextImpl& context, const std::vector<Vec3>& velocities) {
    _gpuContext* gpu = data.gpu;
    int* order = gpu->psAtomIndex->_pSysData;
    int numParticles = context.getSystem().getNumParticles();
    for (int i = 0; i < numParticles; ++i) {
        float4& vel = (*gpu->psVelm4)[i];
        const Vec3& v = velocities[order[i]];
        vel.x = v[0];
        vel.y = v[1];
        vel.z = v[2];
    }
    gpu->psVelm4->Upload();
}

void CudaUpdateStateDataKernel::getForces(ContextImpl& context, std::vector<Vec3>& forces) {
    _gpuContext* gpu = data.gpu;
    int* order = gpu->psAtomIndex->_pSysData;
    gpu->psForce4->Download();
    int numParticles = context.getSystem().getNumParticles();
    forces.resize(numParticles);
    for (int i = 0; i < numParticles; ++i) {
        float4 force = (*gpu->psForce4)[i];
        forces[order[i]] = Vec3(force.x, force.y, force.z);
    }
}

179
180
181
182
183
CudaCalcHarmonicBondForceKernel::~CudaCalcHarmonicBondForceKernel() {
}

void CudaCalcHarmonicBondForceKernel::initialize(const System& system, const HarmonicBondForce& force) {
    data.hasBonds = true;
184
    numBonds = force.getNumBonds();
Peter Eastman's avatar
Peter Eastman committed
185
186
    vector<int> particle1(numBonds);
    vector<int> particle2(numBonds);
187
188
189
190
    vector<float> length(numBonds);
    vector<float> k(numBonds);
    for (int i = 0; i < numBonds; i++) {
        double lengthValue, kValue;
Peter Eastman's avatar
Peter Eastman committed
191
        force.getBondParameters(i, particle1[i], particle2[i], lengthValue, kValue);
192
193
194
        length[i] = (float) lengthValue;
        k[i] = (float) kValue;
    }
Peter Eastman's avatar
Peter Eastman committed
195
    gpuSetBondParameters(data.gpu, particle1, particle2, length, k);
196
197
}

198
void CudaCalcHarmonicBondForceKernel::executeForces(ContextImpl& context) {
199
200
}

201
double CudaCalcHarmonicBondForceKernel::executeEnergy(ContextImpl& context) {
202
203
204
    return 0.0;
}

205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
CudaCalcCustomBondForceKernel::~CudaCalcCustomBondForceKernel() {
}

void CudaCalcCustomBondForceKernel::initialize(const System& system, const CustomBondForce& force) {
    numBonds = force.getNumBonds();
    vector<int> particle1(numBonds);
    vector<int> particle2(numBonds);
    vector<vector<double> > params(numBonds);
    for (int i = 0; i < numBonds; i++)
        force.getBondParameters(i, particle1[i], particle2[i], params[i]);
    vector<string> paramNames;
    for (int i = 0; i < force.getNumPerBondParameters(); i++)
        paramNames.push_back(force.getPerBondParameterName(i));
    globalParamNames.resize(force.getNumGlobalParameters());
    globalParamValues.resize(force.getNumGlobalParameters());
    for (int i = 0; i < force.getNumGlobalParameters(); i++) {
        globalParamNames[i] = force.getGlobalParameterName(i);
        globalParamValues[i] = (float) force.getGlobalParameterDefaultValue(i);
    }
    gpuSetCustomBondParameters(data.gpu, particle1, particle2, params, force.getEnergyFunction(), paramNames, globalParamNames);
    if (globalParamValues.size() > 0)
        SetCustomBondGlobalParams(&globalParamValues[0]);
}

void CudaCalcCustomBondForceKernel::executeForces(ContextImpl& context) {
    updateGlobalParams(context);
    kCalculateCustomBondForces(data.gpu);
}

double CudaCalcCustomBondForceKernel::executeEnergy(ContextImpl& context) {
    updateGlobalParams(context);
    kCalculateCustomBondForces(data.gpu);
    return 0.0;
}

void CudaCalcCustomBondForceKernel::updateGlobalParams(ContextImpl& context) {
    bool changed = false;
    for (int i = 0; i < (int) globalParamNames.size(); i++) {
        float value = (float) context.getParameter(globalParamNames[i]);
        if (value != globalParamValues[i])
            changed = true;
        globalParamValues[i] = value;
    }
    if (changed)
        SetCustomBondGlobalParams(&globalParamValues[0]);
}

252
253
254
255
256
CudaCalcHarmonicAngleForceKernel::~CudaCalcHarmonicAngleForceKernel() {
}

void CudaCalcHarmonicAngleForceKernel::initialize(const System& system, const HarmonicAngleForce& force) {
    data.hasAngles = true;
257
    numAngles = force.getNumAngles();
258
    const float RadiansToDegrees = (float) (180.0/3.14159265);
Peter Eastman's avatar
Peter Eastman committed
259
260
261
    vector<int> particle1(numAngles);
    vector<int> particle2(numAngles);
    vector<int> particle3(numAngles);
262
263
264
265
    vector<float> angle(numAngles);
    vector<float> k(numAngles);
    for (int i = 0; i < numAngles; i++) {
        double angleValue, kValue;
Peter Eastman's avatar
Peter Eastman committed
266
        force.getAngleParameters(i, particle1[i], particle2[i], particle3[i], angleValue, kValue);
267
268
        angle[i] = (float) (angleValue*RadiansToDegrees);
        k[i] = (float) kValue;
269
    }
Peter Eastman's avatar
Peter Eastman committed
270
    gpuSetBondAngleParameters(data.gpu, particle1, particle2, particle3, angle, k);
271
}
272

273
void CudaCalcHarmonicAngleForceKernel::executeForces(ContextImpl& context) {
274
275
}

276
double CudaCalcHarmonicAngleForceKernel::executeEnergy(ContextImpl& context) {
277
278
279
280
281
282
283
284
285
    return 0.0;
}

CudaCalcPeriodicTorsionForceKernel::~CudaCalcPeriodicTorsionForceKernel() {
}

void CudaCalcPeriodicTorsionForceKernel::initialize(const System& system, const PeriodicTorsionForce& force) {
    data.hasPeriodicTorsions = true;
    numTorsions = force.getNumTorsions();
286
    const float RadiansToDegrees = (float)(180.0/3.14159265);
Peter Eastman's avatar
Peter Eastman committed
287
288
289
290
    vector<int> particle1(numTorsions);
    vector<int> particle2(numTorsions);
    vector<int> particle3(numTorsions);
    vector<int> particle4(numTorsions);
291
292
293
294
295
    vector<float> k(numTorsions);
    vector<float> phase(numTorsions);
    vector<int> periodicity(numTorsions);
    for (int i = 0; i < numTorsions; i++) {
        double kValue, phaseValue;
Peter Eastman's avatar
Peter Eastman committed
296
        force.getTorsionParameters(i, particle1[i], particle2[i], particle3[i], particle4[i], periodicity[i], phaseValue, kValue);
297
298
        k[i] = (float) kValue;
        phase[i] = (float) (phaseValue*RadiansToDegrees);
299
    }
Peter Eastman's avatar
Peter Eastman committed
300
    gpuSetDihedralParameters(data.gpu, particle1, particle2, particle3, particle4, k, phase, periodicity);
301
302
}

303
void CudaCalcPeriodicTorsionForceKernel::executeForces(ContextImpl& context) {
304
305
}

306
double CudaCalcPeriodicTorsionForceKernel::executeEnergy(ContextImpl& context) {
307
308
309
310
311
312
313
314
315
    return 0.0;
}

CudaCalcRBTorsionForceKernel::~CudaCalcRBTorsionForceKernel() {
}

void CudaCalcRBTorsionForceKernel::initialize(const System& system, const RBTorsionForce& force) {
    data.hasRB = true;
    numTorsions = force.getNumTorsions();
Peter Eastman's avatar
Peter Eastman committed
316
317
318
319
    vector<int> particle1(numTorsions);
    vector<int> particle2(numTorsions);
    vector<int> particle3(numTorsions);
    vector<int> particle4(numTorsions);
320
321
322
323
324
325
326
327
    vector<float> c0(numTorsions);
    vector<float> c1(numTorsions);
    vector<float> c2(numTorsions);
    vector<float> c3(numTorsions);
    vector<float> c4(numTorsions);
    vector<float> c5(numTorsions);
    for (int i = 0; i < numTorsions; i++) {
        double c[6];
Peter Eastman's avatar
Peter Eastman committed
328
        force.getTorsionParameters(i, particle1[i], particle2[i], particle3[i], particle4[i], c[0], c[1], c[2], c[3], c[4], c[5]);
329
330
331
332
333
334
        c0[i] = (float) c[0];
        c1[i] = (float) c[1];
        c2[i] = (float) c[2];
        c3[i] = (float) c[3];
        c4[i] = (float) c[4];
        c5[i] = (float) c[5];
335
    }
Peter Eastman's avatar
Peter Eastman committed
336
    gpuSetRbDihedralParameters(data.gpu, particle1, particle2, particle3, particle4, c0, c1, c2, c3, c4, c5);
337
338
}

339
void CudaCalcRBTorsionForceKernel::executeForces(ContextImpl& context) {
340
341
}

342
double CudaCalcRBTorsionForceKernel::executeEnergy(ContextImpl& context) {
343
344
345
346
347
348
    return 0.0;
}

CudaCalcNonbondedForceKernel::~CudaCalcNonbondedForceKernel() {
}

349
void CudaCalcNonbondedForceKernel::initialize(const System& system, const NonbondedForce& force) {
350
    data.hasNonbonded = true;
Peter Eastman's avatar
Peter Eastman committed
351
    numParticles = force.getNumParticles();
352
    _gpuContext* gpu = data.gpu;
353
354
355
356

    // Identify which exceptions are 1-4 interactions.

    vector<pair<int, int> > exclusions;
357
    vector<int> exceptions;
358
359
360
361
362
363
    for (int i = 0; i < force.getNumExceptions(); i++) {
        int particle1, particle2;
        double chargeProd, sigma, epsilon;
        force.getExceptionParameters(i, particle1, particle2, chargeProd, sigma, epsilon);
        exclusions.push_back(pair<int, int>(particle1, particle2));
        if (chargeProd != 0.0 || epsilon != 0.0)
364
            exceptions.push_back(i);
365
366
    }

367
368
    // Initialize nonbonded interactions.
    
369
    {
Peter Eastman's avatar
Peter Eastman committed
370
371
372
373
        vector<int> particle(numParticles);
        vector<float> c6(numParticles);
        vector<float> c12(numParticles);
        vector<float> q(numParticles);
374
        vector<char> symbol;
Peter Eastman's avatar
Peter Eastman committed
375
376
        vector<vector<int> > exclusionList(numParticles);
        for (int i = 0; i < numParticles; i++) {
377
            double charge, radius, depth;
Peter Eastman's avatar
Peter Eastman committed
378
379
            force.getParticleParameters(i, charge, radius, depth);
            particle[i] = i;
380
381
382
            q[i] = (float) charge;
            c6[i] = (float) (4*depth*pow(radius, 6.0));
            c12[i] = (float) (4*depth*pow(radius, 12.0));
383
384
            exclusionList[i].push_back(i);
        }
385
        for (int i = 0; i < (int)exclusions.size(); i++) {
386
387
388
            exclusionList[exclusions[i].first].push_back(exclusions[i].second);
            exclusionList[exclusions[i].second].push_back(exclusions[i].first);
        }
Peter Eastman's avatar
Peter Eastman committed
389
        Vec3 boxVectors[3];
390
        system.getPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
Peter Eastman's avatar
Peter Eastman committed
391
        gpuSetPeriodicBoxSize(gpu, (float)boxVectors[0][0], (float)boxVectors[1][1], (float)boxVectors[2][2]);
392
393
        CudaNonbondedMethod method = NO_CUTOFF;
        if (force.getNonbondedMethod() != NonbondedForce::NoCutoff) {
394
            gpuSetNonbondedCutoff(gpu, (float)force.getCutoffDistance(), force.getReactionFieldDielectric());
395
396
397
398
399
            method = CUTOFF;
        }
        if (force.getNonbondedMethod() == NonbondedForce::CutoffPeriodic) {
            method = PERIODIC;
        }
400
401
        if (force.getNonbondedMethod() == NonbondedForce::Ewald || force.getNonbondedMethod() == NonbondedForce::PME) {
            if (force.getNonbondedMethod() == NonbondedForce::Ewald) {
402
403
404
                double alpha;
                int kmaxx, kmaxy, kmaxz;
                NonbondedForceImpl::calcEwaldParameters(system, force, alpha, kmaxx, kmaxy, kmaxz);
405
406
407
408
                gpuSetEwaldParameters(gpu, (float) alpha, kmaxx, kmaxy, kmaxz);
                method = EWALD;
            }
            else {
409
410
411
                double alpha;
                int gridSizeX, gridSizeY, gridSizeZ;
                NonbondedForceImpl::calcPMEParameters(system, force, alpha, gridSizeX, gridSizeY, gridSizeZ);
Peter Eastman's avatar
Peter Eastman committed
412
                gpuSetPMEParameters(gpu, (float) alpha, gridSizeX, gridSizeY, gridSizeZ);
413
414
                method = PARTICLE_MESH_EWALD;
            }
415
        }
416
417
        data.nonbondedMethod = method;
        gpuSetCoulombParameters(gpu, 138.935485f, particle, c6, c12, q, symbol, exclusionList, method);
418
419
420
421

        // Compute the Ewald self energy.

        data.ewaldSelfEnergy = 0.0;
422
423
424
425
426
        if (force.getNonbondedMethod() == NonbondedForce::Ewald || force.getNonbondedMethod() == NonbondedForce::PME) {
            double selfEnergyScale = gpu->sim.epsfac*gpu->sim.alphaEwald/std::sqrt(PI);
                for (int i = 0; i < numParticles; i++)
                    data.ewaldSelfEnergy -= selfEnergyScale*q[i]*q[i];
        }
427
428
429
430
    }

    // Initialize 1-4 nonbonded interactions.
    
431
    {
432
433
434
435
436
437
438
439
        int numExceptions = exceptions.size();
        vector<int> particle1(numExceptions);
        vector<int> particle2(numExceptions);
        vector<float> c6(numExceptions);
        vector<float> c12(numExceptions);
        vector<float> q1(numExceptions);
        vector<float> q2(numExceptions);
        for (int i = 0; i < numExceptions; i++) {
440
            double charge, sig, eps;
441
            force.getExceptionParameters(exceptions[i], particle1[i], particle2[i], charge, sig, eps);
442
443
            c6[i] = (float) (4*eps*pow(sig, 6.0));
            c12[i] = (float) (4*eps*pow(sig, 12.0));
Peter Eastman's avatar
Peter Eastman committed
444
445
            q1[i] = (float) charge;
            q2[i] = 1.0f;
446
        }
Peter Eastman's avatar
Peter Eastman committed
447
        gpuSetLJ14Parameters(gpu, 138.935485f, 1.0f, particle1, particle2, c6, c12, q1, q2);
448
449
450
    }
}

451
void CudaCalcNonbondedForceKernel::executeForces(ContextImpl& context) {
452
453
}

454
double CudaCalcNonbondedForceKernel::executeEnergy(ContextImpl& context) {
455
    return 0.0;
456
457
}

458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
CudaCalcCustomNonbondedForceKernel::~CudaCalcCustomNonbondedForceKernel() {
}

void CudaCalcCustomNonbondedForceKernel::initialize(const System& system, const CustomNonbondedForce& force) {
    data.hasCustomNonbonded = true;
    numParticles = force.getNumParticles();
    _gpuContext* gpu = data.gpu;

    // Identify which exceptions are actual interactions.

    vector<pair<int, int> > exclusions;
    vector<int> exceptions;
    {
        vector<double> parameters;
        for (int i = 0; i < force.getNumExceptions(); i++) {
            int particle1, particle2;
            force.getExceptionParameters(i, particle1, particle2, parameters);
            exclusions.push_back(pair<int, int>(particle1, particle2));
            if (parameters.size() > 0)
                exceptions.push_back(i);
        }
    }

    // Initialize nonbonded interactions.

    vector<int> particle(numParticles);
    vector<vector<double> > parameters(numParticles);
    vector<vector<int> > exclusionList(numParticles);
    for (int i = 0; i < numParticles; i++) {
        force.getParticleParameters(i, parameters[i]);
        particle[i] = i;
        exclusionList[i].push_back(i);
    }
    for (int i = 0; i < (int)exclusions.size(); i++) {
        exclusionList[exclusions[i].first].push_back(exclusions[i].second);
        exclusionList[exclusions[i].second].push_back(exclusions[i].first);
    }
495
496
497
    Vec3 boxVectors[3];
    system.getPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
    gpuSetPeriodicBoxSize(gpu, (float)boxVectors[0][0], (float)boxVectors[1][1], (float)boxVectors[2][2]);
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
    CudaNonbondedMethod method = NO_CUTOFF;
    if (force.getNonbondedMethod() != CustomNonbondedForce::NoCutoff)
        method = CUTOFF;
    if (force.getNonbondedMethod() == CustomNonbondedForce::CutoffPeriodic) {
        method = PERIODIC;
    }
    data.customNonbondedMethod = method;

    // Initialize exceptions.

    int numExceptions = exceptions.size();
    vector<int> exceptionParticle1(numExceptions);
    vector<int> exceptionParticle2(numExceptions);
    vector<vector<double> > exceptionParams(numExceptions);
    for (int i = 0; i < numExceptions; i++)
        force.getExceptionParameters(exceptions[i], exceptionParticle1[i], exceptionParticle2[i], exceptionParams[i]);

515
516
517
518
519
520
521
522
523
524
525
    // Record the tabulated functions.

    for (int i = 0; i < force.getNumFunctions(); i++) {
        string name;
        vector<double> values;
        double min, max;
        bool interpolating;
        force.getFunctionParameters(i, name, values, min, max, interpolating);
        gpuSetTabulatedFunction(gpu, i, name, values, min, max, interpolating);
    }

526
527
528
529
530
531
532
533
    // Record information for the expressions.

    vector<string> paramNames;
    vector<string> combiningRules;
    for (int i = 0; i < force.getNumParameters(); i++) {
        paramNames.push_back(force.getParameterName(i));
        combiningRules.push_back(force.getParameterCombiningRule(i));
    }
534
535
536
537
    globalParamNames.resize(force.getNumGlobalParameters());
    globalParamValues.resize(force.getNumGlobalParameters());
    for (int i = 0; i < force.getNumGlobalParameters(); i++) {
        globalParamNames[i] = force.getGlobalParameterName(i);
538
        globalParamValues[i] = (float) force.getGlobalParameterDefaultValue(i);
539
    }
540
    gpuSetCustomNonbondedParameters(gpu, parameters, exclusionList, exceptionParticle1, exceptionParticle2, exceptionParams, method,
541
542
            (float)force.getCutoffDistance(), force.getEnergyFunction(), combiningRules, paramNames, globalParamNames);
    if (globalParamValues.size() > 0)
543
        SetCustomNonbondedGlobalParams(&globalParamValues[0]);
544
545
546
}

void CudaCalcCustomNonbondedForceKernel::executeForces(ContextImpl& context) {
547
    updateGlobalParams(context);
548
549
550
}

double CudaCalcCustomNonbondedForceKernel::executeEnergy(ContextImpl& context) {
551
    updateGlobalParams(context);
552
553
554
    return 0.0;
}

555
556
void CudaCalcCustomNonbondedForceKernel::updateGlobalParams(ContextImpl& context) {
    bool changed = false;
557
    for (int i = 0; i < (int) globalParamNames.size(); i++) {
558
559
560
561
562
563
        float value = (float) context.getParameter(globalParamNames[i]);
        if (value != globalParamValues[i])
            changed = true;
        globalParamValues[i] = value;
    }
    if (changed)
564
        SetCustomNonbondedGlobalParams(&globalParamValues[0]);
565
566
}

567
CudaCalcGBSAOBCForceKernel::~CudaCalcGBSAOBCForceKernel() {
568
569
}

570
void CudaCalcGBSAOBCForceKernel::initialize(const System& system, const GBSAOBCForce& force) {
571

Peter Eastman's avatar
Peter Eastman committed
572
    int numParticles = system.getNumParticles();
573
    _gpuContext* gpu = data.gpu;
Peter Eastman's avatar
Peter Eastman committed
574
575
    vector<float> radius(numParticles);
    vector<float> scale(numParticles);
576
    vector<float> charge(numParticles);
Peter Eastman's avatar
Peter Eastman committed
577
    for (int i = 0; i < numParticles; i++) {
578
579
        double particleCharge, particleRadius, scalingFactor;
        force.getParticleParameters(i, particleCharge, particleRadius, scalingFactor);
Peter Eastman's avatar
Peter Eastman committed
580
        radius[i] = (float) particleRadius;
581
        scale[i] = (float) scalingFactor;
582
        charge[i] = (float) particleCharge;
583
    }
584
    gpuSetObcParameters(gpu, (float) force.getSoluteDielectric(), (float) force.getSolventDielectric(), radius, scale, charge);
585
586
}

587
void CudaCalcGBSAOBCForceKernel::executeForces(ContextImpl& context) {
588
589
}

Mark Friedrichs's avatar
Mark Friedrichs committed
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
CudaCalcGBVIForceKernel::~CudaCalcGBVIForceKernel() {
}

void CudaCalcGBVIForceKernel::initialize(const System& system, const GBVIForce& force, const std::vector<double> & inputScaledRadii) {

    int numParticles = system.getNumParticles();
    _gpuContext* gpu = data.gpu;

    vector<int> particle(numParticles);
    vector<float> radius(numParticles);
    vector<float> scaledRadii(numParticles);
    vector<float> gammas(numParticles);

    for (int i = 0; i < numParticles; i++) {
        double charge, particleRadius, gamma, bornRadiusScaleFactor;
        force.getParticleParameters(i, charge, particleRadius, gamma );
        particle[i]                  = i;
        radius[i]                    = (float) particleRadius;
        gammas[i]                    = (float) gamma;
        scaledRadii[i]               = (float) inputScaledRadii[i];
    }
    gpuSetGBVIParameters(gpu, (float) force.getSoluteDielectric(), (float) force.getSolventDielectric(), particle,
                         radius, gammas, scaledRadii );
}

void CudaCalcGBVIForceKernel::executeForces(ContextImpl& context) {
}

double CudaCalcGBVIForceKernel::executeEnergy(ContextImpl& context) {
    return 0.0;
}

622
static void initializeIntegration(const System& system, CudaPlatform::PlatformData& data, const Integrator& integrator) {
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639

    // Initialize any terms that haven't already been handled by a Force.

    _gpuContext* gpu = data.gpu;
    if (!data.hasBonds)
        gpuSetBondParameters(gpu, vector<int>(), vector<int>(), vector<float>(), vector<float>());
    if (!data.hasAngles)
        gpuSetBondAngleParameters(gpu, vector<int>(), vector<int>(), vector<int>(), vector<float>(), vector<float>());
    if (!data.hasPeriodicTorsions)
        gpuSetDihedralParameters(gpu, vector<int>(), vector<int>(), vector<int>(), vector<int>(), vector<float>(), vector<float>(), vector<int>());
    if (!data.hasRB)
        gpuSetRbDihedralParameters(gpu, vector<int>(), vector<int>(), vector<int>(), vector<int>(), vector<float>(), vector<float>(),
                vector<float>(), vector<float>(), vector<float>(), vector<float>());
    if (!data.hasNonbonded) {
        gpuSetCoulombParameters(gpu, 138.935485f, vector<int>(), vector<float>(), vector<float>(), vector<float>(), vector<char>(), vector<vector<int> >(), NO_CUTOFF);
        gpuSetLJ14Parameters(gpu, 138.935485f, 1.0f, vector<int>(), vector<int>(), vector<float>(), vector<float>(), vector<float>(), vector<float>());
    }
640
641
642
    
    // Set masses.
    
Peter Eastman's avatar
Peter Eastman committed
643
644
645
646
    int numParticles = system.getNumParticles();
    vector<float> mass(numParticles);
    for (int i = 0; i < numParticles; i++)
        mass[i] = (float) system.getParticleMass(i);
647
648
649
650
    gpuSetMass(gpu, mass);
    
    // Set constraints.
    
651
    int numConstraints = system.getNumConstraints();
Peter Eastman's avatar
Peter Eastman committed
652
653
    vector<int> particle1(numConstraints);
    vector<int> particle2(numConstraints);
654
655
656
657
    vector<float> distance(numConstraints);
    vector<float> invMass1(numConstraints);
    vector<float> invMass2(numConstraints);
    for (int i = 0; i < numConstraints; i++) {
Peter Eastman's avatar
Peter Eastman committed
658
        int particle1Index, particle2Index;
659
        double constraintDistance;
Peter Eastman's avatar
Peter Eastman committed
660
661
662
        system.getConstraintParameters(i, particle1Index, particle2Index, constraintDistance);
        particle1[i] = particle1Index;
        particle2[i] = particle2Index;
663
        distance[i] = (float) constraintDistance;
Peter Eastman's avatar
Peter Eastman committed
664
665
        invMass1[i] = 1.0f/mass[particle1Index];
        invMass2[i] = 1.0f/mass[particle2Index];
666
    }
667
    gpuSetConstraintParameters(gpu, particle1, particle2, distance, invMass1, invMass2, (float)integrator.getConstraintTolerance());
668
669
    
    // Finish initialization.
670

671
672
673
    gpuBuildThreadBlockWorkList(gpu);
    gpuBuildExclusionList(gpu);
    gpuBuildOutputBuffers(gpu);
674
    gpuSetConstants(gpu);
675
676
677
    kClearBornForces(gpu);
    kClearForces(gpu);
    cudaThreadSynchronize();
678
679
}

680
double CudaCalcGBSAOBCForceKernel::executeEnergy(ContextImpl& context) {
681
	return 0.0;
682
683
684
685
686
687
}

CudaIntegrateVerletStepKernel::~CudaIntegrateVerletStepKernel() {
}

void CudaIntegrateVerletStepKernel::initialize(const System& system, const VerletIntegrator& integrator) {
688
    initializeIntegration(system, data, integrator);
689
690
691
    prevStepSize = -1.0;
}

692
void CudaIntegrateVerletStepKernel::execute(ContextImpl& context, const VerletIntegrator& integrator) {
693
694
695
696
697
    _gpuContext* gpu = data.gpu;
    double stepSize = integrator.getStepSize();
    if (stepSize != prevStepSize) {
        // Initialize the GPU parameters.
        
698
        gpuSetVerletIntegrationParameters(gpu, (float) stepSize, 0.0f);
699
700
701
702
703
        gpuSetConstants(gpu);
        prevStepSize = stepSize;
    }
    kVerletUpdatePart1(gpu);
    kApplyFirstShake(gpu);
704
    kApplyFirstSettle(gpu);
705
    kApplyFirstCCMA(gpu);
706
707
    if (data.removeCM)
        if (data.stepCount%data.cmMotionFrequency == 0)
708
709
            gpu->bCalculateCM = true;
    kVerletUpdatePart2(gpu);
710
    data.time += stepSize;
711
    data.stepCount++;
712
713
714
715
716
717
}

CudaIntegrateLangevinStepKernel::~CudaIntegrateLangevinStepKernel() {
}

void CudaIntegrateLangevinStepKernel::initialize(const System& system, const LangevinIntegrator& integrator) {
718
    initializeIntegration(system, data, integrator);
719
720
721
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) integrator.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
722
723
    prevTemp = -1.0;
    prevFriction = -1.0;
724
725
726
    prevStepSize = -1.0;
}

727
void CudaIntegrateLangevinStepKernel::execute(ContextImpl& context, const LangevinIntegrator& integrator) {
728
    _gpuContext* gpu = data.gpu;
729
730
731
    double temperature = integrator.getTemperature();
    double friction = integrator.getFriction();
    double stepSize = integrator.getStepSize();
732
733
    if (temperature != prevTemp || friction != prevFriction || stepSize != prevStepSize) {
        // Initialize the GPU parameters.
734
        
735
        double tau = (friction == 0.0 ? 0.0 : 1.0/friction);
736
        gpuSetLangevinIntegrationParameters(gpu, (float) tau, (float) stepSize, (float) temperature, 0.0f);
737
738
739
740
741
742
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFriction = friction;
        prevStepSize = stepSize;
    }
743
    kLangevinUpdatePart1(gpu);
744
    kApplyFirstShake(gpu);
745
    kApplyFirstSettle(gpu);
746
    kApplyFirstCCMA(gpu);
747
748
    if (data.removeCM)
        if (data.stepCount%data.cmMotionFrequency == 0)
749
            gpu->bCalculateCM = true;
750
    kLangevinUpdatePart2(gpu);
751
    kApplySecondShake(gpu);
752
    kApplySecondSettle(gpu);
753
    kApplySecondCCMA(gpu);
754
    data.time += stepSize;
755
    data.stepCount++;
756
}
757
758
759
760
761

CudaIntegrateBrownianStepKernel::~CudaIntegrateBrownianStepKernel() {
}

void CudaIntegrateBrownianStepKernel::initialize(const System& system, const BrownianIntegrator& integrator) {
762
    initializeIntegration(system, data, integrator);
763
764
765
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) integrator.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
766
767
    prevTemp = -1.0;
    prevFriction = -1.0;
768
769
770
    prevStepSize = -1.0;
}

771
void CudaIntegrateBrownianStepKernel::execute(ContextImpl& context, const BrownianIntegrator& integrator) {
772
773
774
775
776
777
778
779
    _gpuContext* gpu = data.gpu;
    double temperature = integrator.getTemperature();
    double friction = integrator.getFriction();
    double stepSize = integrator.getStepSize();
    if (temperature != prevTemp || friction != prevFriction || stepSize != prevStepSize) {
        // Initialize the GPU parameters.
        
        double tau = (friction == 0.0 ? 0.0 : 1.0/friction);
780
        gpuSetBrownianIntegrationParameters(gpu, (float) tau, (float) stepSize, (float) temperature);
781
782
783
784
785
786
787
788
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFriction = friction;
        prevStepSize = stepSize;
    }
    kBrownianUpdatePart1(gpu);
    kApplyFirstShake(gpu);
789
    kApplyFirstSettle(gpu);
790
    kApplyFirstCCMA(gpu);
791
792
    if (data.removeCM)
        if (data.stepCount%data.cmMotionFrequency == 0)
793
794
            gpu->bCalculateCM = true;
    kBrownianUpdatePart2(gpu);
795
    data.time += stepSize;
796
797
798
799
800
801
802
803
804
805
806
    data.stepCount++;
}

CudaIntegrateVariableVerletStepKernel::~CudaIntegrateVariableVerletStepKernel() {
}

void CudaIntegrateVariableVerletStepKernel::initialize(const System& system, const VariableVerletIntegrator& integrator) {
    initializeIntegration(system, data, integrator);
    prevErrorTol = -1.0;
}

807
void CudaIntegrateVariableVerletStepKernel::execute(ContextImpl& context, const VariableVerletIntegrator& integrator, double maxTime) {
808
809
810
811
812
813
814
815
816
    _gpuContext* gpu = data.gpu;
    double errorTol = integrator.getErrorTolerance();
    if (errorTol != prevErrorTol) {
        // Initialize the GPU parameters.

        gpuSetVerletIntegrationParameters(gpu, 0.0f, (float) errorTol);
        gpuSetConstants(gpu);
        prevErrorTol = errorTol;
    }
817
    float maxStepSize = (float)(maxTime-data.time);
818
    kSelectVerletStepSize(gpu, maxStepSize);
819
820
821
822
823
824
825
826
827
828
    kVerletUpdatePart1(gpu);
    kApplyFirstShake(gpu);
    kApplyFirstSettle(gpu);
    kApplyFirstCCMA(gpu);
    if (data.removeCM)
        if (data.stepCount%data.cmMotionFrequency == 0)
            gpu->bCalculateCM = true;
    kVerletUpdatePart2(gpu);
    gpu->psStepSize->Download();
    data.time += (*gpu->psStepSize)[0].y;
829
830
    if ((*gpu->psStepSize)[0].y == maxStepSize)
        data.time = maxTime; // Avoid round-off error
831
    data.stepCount++;
832
}
833

834
835
836
837
838
839
840
841
CudaIntegrateVariableLangevinStepKernel::~CudaIntegrateVariableLangevinStepKernel() {
}

void CudaIntegrateVariableLangevinStepKernel::initialize(const System& system, const VariableLangevinIntegrator& integrator) {
    initializeIntegration(system, data, integrator);
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) integrator.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
842
843
    prevTemp = -1.0;
    prevFriction = -1.0;
844
845
846
847
848
849
850
851
852
853
854
855
    prevErrorTol = -1.0;
}

void CudaIntegrateVariableLangevinStepKernel::execute(ContextImpl& context, const VariableLangevinIntegrator& integrator, double maxTime) {
    _gpuContext* gpu = data.gpu;
    double temperature = integrator.getTemperature();
    double friction = integrator.getFriction();
    double errorTol = integrator.getErrorTolerance();
    if (temperature != prevTemp || friction != prevFriction || errorTol != prevErrorTol) {
        // Initialize the GPU parameters.

        double tau = (friction == 0.0 ? 0.0 : 1.0/friction);
856
        gpuSetLangevinIntegrationParameters(gpu, (float) tau, 0.0f, (float) temperature, (float) errorTol);
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFriction = friction;
        prevErrorTol = errorTol;
    }
    float maxStepSize = (float)(maxTime-data.time);
    kSelectLangevinStepSize(gpu, maxStepSize);
    kLangevinUpdatePart1(gpu);
    kApplyFirstShake(gpu);
    kApplyFirstSettle(gpu);
    kApplyFirstCCMA(gpu);
    if (data.removeCM)
        if (data.stepCount%data.cmMotionFrequency == 0)
            gpu->bCalculateCM = true;
    kLangevinUpdatePart2(gpu);
    kApplySecondShake(gpu);
    kApplySecondSettle(gpu);
    kApplySecondCCMA(gpu);
    gpu->psStepSize->Download();
    data.time += (*gpu->psStepSize)[0].y;
    if ((*gpu->psStepSize)[0].y == maxStepSize)
        data.time = maxTime; // Avoid round-off error
    data.stepCount++;
}

883
884
885
886
CudaApplyAndersenThermostatKernel::~CudaApplyAndersenThermostatKernel() {
}

void CudaApplyAndersenThermostatKernel::initialize(const System& system, const AndersenThermostat& thermostat) {
887
888
889
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) thermostat.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
890
891
    prevTemp = -1.0;
    prevFrequency = -1.0;
892
893
894
    prevStepSize = -1.0;
}

895
void CudaApplyAndersenThermostatKernel::execute(ContextImpl& context) {
896
    _gpuContext* gpu = data.gpu;
897
898
    double temperature = context.getParameter(AndersenThermostat::Temperature());
    double frequency = context.getParameter(AndersenThermostat::CollisionFrequency());
899
900
901
902
    double stepSize = context.getIntegrator().getStepSize();
    if (temperature != prevTemp || frequency != prevFrequency || stepSize != prevStepSize) {
        // Initialize the GPU parameters.
        
903
        gpuSetAndersenThermostatParameters(gpu, (float) temperature, (float) frequency);
904
905
906
907
908
909
910
911
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFrequency = frequency;
        prevStepSize = stepSize;
    }
    kCalculateAndersenThermostat(gpu);
}
912

913
void CudaCalcKineticEnergyKernel::initialize(const System& system) {
Peter Eastman's avatar
Peter Eastman committed
914
915
    int numParticles = system.getNumParticles();
    masses.resize(numParticles);
916
    for (int i = 0; i < numParticles; ++i)
Peter Eastman's avatar
Peter Eastman committed
917
        masses[i] = system.getParticleMass(i);
918
919
}

920
double CudaCalcKineticEnergyKernel::execute(ContextImpl& context) {
921
922
923
    // We don't currently have a GPU kernel to do this, so we retrieve the velocities and calculate the energy
    // on the CPU.
    
924
925
    _gpuContext* gpu = data.gpu;
    gpu->psVelm4->Download();
926
    double energy = 0.0;
927
928
929
930
    for (int i = 0; i < (int) masses.size(); ++i) {
        float4 v = (*gpu->psVelm4)[i];
        energy += masses[i]*(v.x*v.x+v.y*v.y+v.z*v.z);
    }
931
932
    return 0.5*energy;
}
933

934
void CudaRemoveCMMotionKernel::initialize(const System& system, const CMMotionRemover& force) {
935
    data.removeCM = true;
936
    data.cmMotionFrequency = force.getFrequency();
937
938
}

939
void CudaRemoveCMMotionKernel::execute(ContextImpl& context) {
940
}