CudaKernels.cpp 22.9 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
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
 * Authors: Peter Eastman                                                     *
 * Contributors:                                                              *
 *                                                                            *
 * Permission is hereby granted, free of charge, to any person obtaining a    *
 * copy of this software and associated documentation files (the "Software"), *
 * to deal in the Software without restriction, including without limitation  *
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,   *
 * and/or sell copies of the Software, and to permit persons to whom the      *
 * Software is furnished to do so, subject to the following conditions:       *
 *                                                                            *
 * The above copyright notice and this permission notice shall be included in *
 * all copies or substantial portions of the Software.                        *
 *                                                                            *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,   *
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL    *
 * THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,    *
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR      *
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE  *
 * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
 * -------------------------------------------------------------------------- */

#include "CudaKernels.h"
#include "CudaStreamImpl.h"
34
35
#include "openmm/LangevinIntegrator.h"
#include "openmm/OpenMMContext.h"
36
#include "ReferencePlatform.h"
37
#include "openmm/internal/OpenMMContextImpl.h"
38
#include "kernels/gputypes.h"
39
#include "kernels/cudaKernels.h"
40
41
42
43
44
45
46
#include <cmath>

extern "C" int gpuSetConstants( gpuContext gpu );

using namespace OpenMM;
using namespace std;

47
48
static void calcForces(OpenMMContextImpl& context, CudaPlatform::PlatformData& data) {
    _gpuContext* gpu = data.gpu;
49
    if (data.nonbondedMethod != NO_CUTOFF && data.stepCount%100 == 0)
50
51
52
53
        gpuReorderAtoms(gpu);
    data.stepCount++;
    kClearForces(gpu);
    if (gpu->bIncludeGBSA) {
54
        gpu->bRecalculateBornRadii = true;
55
56
57
58
        kCalculateCDLJObcGbsaForces1(gpu);
        kReduceObcGbsaBornForces(gpu);
        kCalculateObcGbsaForces2(gpu);
    }
59
    else if (data.hasNonbonded) {
60
61
62
        kCalculateCDLJForces(gpu);
    }
    kCalculateLocalForces(gpu);
63
64
65
66
    if (gpu->bIncludeGBSA)
        kReduceBornSumAndForces(gpu);
    else
        kReduceForces(gpu);
67
68
}

69
70
71
static double calcEnergy(OpenMMContextImpl& context, System& system) {
    // We don't currently have GPU kernels to calculate energy, so instead we have the reference
    // platform do it.  This is VERY slow.
72

73
74
75
76
77
78
79
    LangevinIntegrator integrator(0.0, 1.0, 0.0);
    ReferencePlatform platform;
    OpenMMContext refContext(system, integrator, platform);
    const Stream& positions = context.getPositions();
    double* posData = new double[positions.getSize()*3];
    positions.saveToArray(posData);
    vector<Vec3> pos(positions.getSize());
80
    for (int i = 0; i < (int)pos.size(); i++)
81
82
83
84
85
86
        pos[i] = Vec3(posData[3*i], posData[3*i+1], posData[3*i+2]);
    delete[] posData;
    refContext.setPositions(pos);
    return refContext.getState(State::Energy).getPotentialEnergy();
}

87
88
89
void CudaInitializeForcesKernel::initialize(const System& system) {
}

90
void CudaInitializeForcesKernel::execute(OpenMMContextImpl& context) {
91
92
}

93
94
95
96
97
98
99
CudaCalcHarmonicBondForceKernel::~CudaCalcHarmonicBondForceKernel() {
}

void CudaCalcHarmonicBondForceKernel::initialize(const System& system, const HarmonicBondForce& force) {
    if (data.primaryKernel == NULL)
        data.primaryKernel = this;
    data.hasBonds = true;
100
    numBonds = force.getNumBonds();
Peter Eastman's avatar
Peter Eastman committed
101
102
    vector<int> particle1(numBonds);
    vector<int> particle2(numBonds);
103
104
105
106
    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
107
        force.getBondParameters(i, particle1[i], particle2[i], lengthValue, kValue);
108
109
110
        length[i] = (float) lengthValue;
        k[i] = (float) kValue;
    }
Peter Eastman's avatar
Peter Eastman committed
111
    gpuSetBondParameters(data.gpu, particle1, particle2, length, k);
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
}

void CudaCalcHarmonicBondForceKernel::executeForces(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        calcForces(context, data);
}

double CudaCalcHarmonicBondForceKernel::executeEnergy(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        return calcEnergy(context, system);
    return 0.0;
}

CudaCalcHarmonicAngleForceKernel::~CudaCalcHarmonicAngleForceKernel() {
}

void CudaCalcHarmonicAngleForceKernel::initialize(const System& system, const HarmonicAngleForce& force) {
    if (data.primaryKernel == NULL)
        data.primaryKernel = this;
    data.hasAngles = true;
132
    numAngles = force.getNumAngles();
133
    const float RadiansToDegrees = (float) (180.0/3.14159265);
Peter Eastman's avatar
Peter Eastman committed
134
135
136
    vector<int> particle1(numAngles);
    vector<int> particle2(numAngles);
    vector<int> particle3(numAngles);
137
138
139
140
    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
141
        force.getAngleParameters(i, particle1[i], particle2[i], particle3[i], angleValue, kValue);
142
143
        angle[i] = (float) (angleValue*RadiansToDegrees);
        k[i] = (float) kValue;
144
    }
Peter Eastman's avatar
Peter Eastman committed
145
    gpuSetBondAngleParameters(data.gpu, particle1, particle2, particle3, angle, k);
146
}
147

148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
void CudaCalcHarmonicAngleForceKernel::executeForces(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        calcForces(context, data);
}

double CudaCalcHarmonicAngleForceKernel::executeEnergy(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        return calcEnergy(context, system);
    return 0.0;
}

CudaCalcPeriodicTorsionForceKernel::~CudaCalcPeriodicTorsionForceKernel() {
}

void CudaCalcPeriodicTorsionForceKernel::initialize(const System& system, const PeriodicTorsionForce& force) {
    if (data.primaryKernel == NULL)
        data.primaryKernel = this;
    data.hasPeriodicTorsions = true;
    numTorsions = force.getNumTorsions();
167
    const float RadiansToDegrees = (float)(180.0/3.14159265);
Peter Eastman's avatar
Peter Eastman committed
168
169
170
171
    vector<int> particle1(numTorsions);
    vector<int> particle2(numTorsions);
    vector<int> particle3(numTorsions);
    vector<int> particle4(numTorsions);
172
173
174
175
176
    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
177
        force.getTorsionParameters(i, particle1[i], particle2[i], particle3[i], particle4[i], periodicity[i], phaseValue, kValue);
178
179
        k[i] = (float) kValue;
        phase[i] = (float) (phaseValue*RadiansToDegrees);
180
    }
Peter Eastman's avatar
Peter Eastman committed
181
    gpuSetDihedralParameters(data.gpu, particle1, particle2, particle3, particle4, k, phase, periodicity);
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
}

void CudaCalcPeriodicTorsionForceKernel::executeForces(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        calcForces(context, data);
}

double CudaCalcPeriodicTorsionForceKernel::executeEnergy(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        return calcEnergy(context, system);
    return 0.0;
}

CudaCalcRBTorsionForceKernel::~CudaCalcRBTorsionForceKernel() {
}

void CudaCalcRBTorsionForceKernel::initialize(const System& system, const RBTorsionForce& force) {
    if (data.primaryKernel == NULL)
        data.primaryKernel = this;
    data.hasRB = true;
    numTorsions = force.getNumTorsions();
Peter Eastman's avatar
Peter Eastman committed
203
204
205
206
    vector<int> particle1(numTorsions);
    vector<int> particle2(numTorsions);
    vector<int> particle3(numTorsions);
    vector<int> particle4(numTorsions);
207
208
209
210
211
212
213
214
    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
215
        force.getTorsionParameters(i, particle1[i], particle2[i], particle3[i], particle4[i], c[0], c[1], c[2], c[3], c[4], c[5]);
216
217
218
219
220
221
        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];
222
    }
Peter Eastman's avatar
Peter Eastman committed
223
    gpuSetRbDihedralParameters(data.gpu, particle1, particle2, particle3, particle4, c0, c1, c2, c3, c4, c5);
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
}

void CudaCalcRBTorsionForceKernel::executeForces(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        calcForces(context, data);
}

double CudaCalcRBTorsionForceKernel::executeEnergy(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        return calcEnergy(context, system);
    return 0.0;
}

CudaCalcNonbondedForceKernel::~CudaCalcNonbondedForceKernel() {
}

240
void CudaCalcNonbondedForceKernel::initialize(const System& system, const NonbondedForce& force) {
241
242
243
    if (data.primaryKernel == NULL)
        data.primaryKernel = this;
    data.hasNonbonded = true;
Peter Eastman's avatar
Peter Eastman committed
244
    numParticles = force.getNumParticles();
245
    _gpuContext* gpu = data.gpu;
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260

    // Identify which exceptions are 1-4 interactions.

    vector<pair<int, int> > exclusions;
    vector<int> nb14s;
    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)
            nb14s.push_back(i);
    }
    num14 = nb14s.size();

261
262
    // Initialize nonbonded interactions.
    
263
    {
Peter Eastman's avatar
Peter Eastman committed
264
265
266
267
        vector<int> particle(numParticles);
        vector<float> c6(numParticles);
        vector<float> c12(numParticles);
        vector<float> q(numParticles);
268
        vector<char> symbol;
Peter Eastman's avatar
Peter Eastman committed
269
270
        vector<vector<int> > exclusionList(numParticles);
        for (int i = 0; i < numParticles; i++) {
271
            double charge, radius, depth;
Peter Eastman's avatar
Peter Eastman committed
272
273
            force.getParticleParameters(i, charge, radius, depth);
            particle[i] = i;
274
275
276
            q[i] = (float) charge;
            c6[i] = (float) (4*depth*pow(radius, 6.0));
            c12[i] = (float) (4*depth*pow(radius, 12.0));
277
278
            exclusionList[i].push_back(i);
        }
279
        for (int i = 0; i < (int)exclusions.size(); i++) {
280
281
282
            exclusionList[exclusions[i].first].push_back(exclusions[i].second);
            exclusionList[exclusions[i].second].push_back(exclusions[i].first);
        }
283
284
        CudaNonbondedMethod method = NO_CUTOFF;
        if (force.getNonbondedMethod() != NonbondedForce::NoCutoff) {
285
            gpuSetNonbondedCutoff(gpu, (float)force.getCutoffDistance(), 78.3f);
286
287
288
289
290
            method = CUTOFF;
        }
        if (force.getNonbondedMethod() == NonbondedForce::CutoffPeriodic) {
            Vec3 boxVectors[3];
            force.getPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
291
            gpuSetPeriodicBoxSize(gpu, (float)boxVectors[0][0], (float)boxVectors[1][1], (float)boxVectors[2][2]);
292
293
            method = PERIODIC;
        }
294
295
296
297

        if (force.getNonbondedMethod() == NonbondedForce::Ewald) {
            Vec3 boxVectors[3];
            force.getPeriodicBoxVectors(boxVectors[0], boxVectors[1], boxVectors[2]);
298
            gpuSetPeriodicBoxSize(gpu, (float)boxVectors[0][0], (float)boxVectors[1][1], (float)boxVectors[2][2]);
299
300
301
            method = EWALD;

        }
302
303
        data.nonbondedMethod = method;
        gpuSetCoulombParameters(gpu, 138.935485f, particle, c6, c12, q, symbol, exclusionList, method);
304
305
306
307
    }

    // Initialize 1-4 nonbonded interactions.
    
308
    {
Peter Eastman's avatar
Peter Eastman committed
309
310
        vector<int> particle1(num14);
        vector<int> particle2(num14);
311
312
313
314
315
        vector<float> c6(num14);
        vector<float> c12(num14);
        vector<float> q1(num14);
        vector<float> q2(num14);
        for (int i = 0; i < num14; i++) {
316
            double charge, sig, eps;
317
            force.getExceptionParameters(nb14s[i], particle1[i], particle2[i], charge, sig, eps);
318
319
            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
320
321
            q1[i] = (float) charge;
            q2[i] = 1.0f;
322
        }
Peter Eastman's avatar
Peter Eastman committed
323
        gpuSetLJ14Parameters(gpu, 138.935485f, 1.0f, particle1, particle2, c6, c12, q1, q2);
324
325
326
    }
}

327
328
329
void CudaCalcNonbondedForceKernel::executeForces(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        calcForces(context, data);
330
331
}

332
333
334
335
double CudaCalcNonbondedForceKernel::executeEnergy(OpenMMContextImpl& context) {
    if (data.primaryKernel == this)
        return calcEnergy(context, system);
    return 0.0;
336
337
}

338
CudaCalcGBSAOBCForceKernel::~CudaCalcGBSAOBCForceKernel() {
339
340
}

341
void CudaCalcGBSAOBCForceKernel::initialize(const System& system, const GBSAOBCForce& force) {
342

Peter Eastman's avatar
Peter Eastman committed
343
    int numParticles = system.getNumParticles();
344
    _gpuContext* gpu = data.gpu;
Peter Eastman's avatar
Peter Eastman committed
345
346
    vector<float> radius(numParticles);
    vector<float> scale(numParticles);
347
    vector<float> charge(numParticles);
Peter Eastman's avatar
Peter Eastman committed
348
    for (int i = 0; i < numParticles; i++) {
349
350
        double particleCharge, particleRadius, scalingFactor;
        force.getParticleParameters(i, particleCharge, particleRadius, scalingFactor);
Peter Eastman's avatar
Peter Eastman committed
351
        radius[i] = (float) particleRadius;
352
        scale[i] = (float) scalingFactor;
353
        charge[i] = (float) particleCharge;
354
    }
355
    gpuSetObcParameters(gpu, (float) force.getSoluteDielectric(), (float) force.getSolventDielectric(), radius, scale, charge);
356
357
}

358
void CudaCalcGBSAOBCForceKernel::executeForces(OpenMMContextImpl& context) {
359
360
}

361
static void initializeIntegration(const System& system, CudaPlatform::PlatformData& data, const Integrator& integrator) {
362
363
364
    
    // Set masses.
    
365
    _gpuContext* gpu = data.gpu;
Peter Eastman's avatar
Peter Eastman committed
366
367
368
369
    int numParticles = system.getNumParticles();
    vector<float> mass(numParticles);
    for (int i = 0; i < numParticles; i++)
        mass[i] = (float) system.getParticleMass(i);
370
371
372
373
    gpuSetMass(gpu, mass);
    
    // Set constraints.
    
374
    int numConstraints = system.getNumConstraints();
Peter Eastman's avatar
Peter Eastman committed
375
376
    vector<int> particle1(numConstraints);
    vector<int> particle2(numConstraints);
377
378
379
380
    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
381
        int particle1Index, particle2Index;
382
        double constraintDistance;
Peter Eastman's avatar
Peter Eastman committed
383
384
385
        system.getConstraintParameters(i, particle1Index, particle2Index, constraintDistance);
        particle1[i] = particle1Index;
        particle2[i] = particle2Index;
386
        distance[i] = (float) constraintDistance;
Peter Eastman's avatar
Peter Eastman committed
387
388
        invMass1[i] = 1.0f/mass[particle1Index];
        invMass2[i] = 1.0f/mass[particle2Index];
389
    }
390
    gpuSetConstraintParameters(gpu, particle1, particle2, distance, invMass1, invMass2, (float)integrator.getConstraintTolerance(), 4);
391
392
393
394
395
396
397
398
399
400
401
402
403
    
    // Initialize any terms that haven't already been handled by a Force.
    
    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) {
404
        gpuSetCoulombParameters(gpu, 138.935485f, vector<int>(), vector<float>(), vector<float>(), vector<float>(), vector<char>(), vector<vector<int> >(), NO_CUTOFF);
405
406
407
408
        gpuSetLJ14Parameters(gpu, 138.935485f, 1.0f, vector<int>(), vector<int>(), vector<float>(), vector<float>(), vector<float>(), vector<float>());
    }
    
    // Finish initialization.
409

410
411
412
    gpuBuildThreadBlockWorkList(gpu);
    gpuBuildExclusionList(gpu);
    gpuBuildOutputBuffers(gpu);
413
    gpuSetConstants(gpu);
414
415
416
    kClearBornForces(gpu);
    kClearForces(gpu);
    cudaThreadSynchronize();
417
418
}

419
double CudaCalcGBSAOBCForceKernel::executeEnergy(OpenMMContextImpl& context) {
420
	return 0.0;
421
422
423
424
425
426
}

CudaIntegrateVerletStepKernel::~CudaIntegrateVerletStepKernel() {
}

void CudaIntegrateVerletStepKernel::initialize(const System& system, const VerletIntegrator& integrator) {
427
    initializeIntegration(system, data, integrator);
428
429
430
431
432
433
434
435
436
    prevStepSize = -1.0;
}

void CudaIntegrateVerletStepKernel::execute(OpenMMContextImpl& context, const VerletIntegrator& integrator) {
    _gpuContext* gpu = data.gpu;
    double stepSize = integrator.getStepSize();
    if (stepSize != prevStepSize) {
        // Initialize the GPU parameters.
        
437
        gpuSetVerletIntegrationParameters(gpu, (float) stepSize);
438
439
440
441
442
443
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevStepSize = stepSize;
    }
    kVerletUpdatePart1(gpu);
    kApplyFirstShake(gpu);
444
    kApplyFirstSettle(gpu);
445
    kApplyFirstLincs(gpu);
446
    if (data.removeCM) {
447
        int step = (int) (context.getTime()/stepSize);
448
449
450
451
452
453
454
455
456
457
        if (step%data.cmMotionFrequency == 0)
            gpu->bCalculateCM = true;
    }
    kVerletUpdatePart2(gpu);
}

CudaIntegrateLangevinStepKernel::~CudaIntegrateLangevinStepKernel() {
}

void CudaIntegrateLangevinStepKernel::initialize(const System& system, const LangevinIntegrator& integrator) {
458
    initializeIntegration(system, data, integrator);
459
460
461
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) integrator.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
462
463
464
    prevStepSize = -1.0;
}

465
void CudaIntegrateLangevinStepKernel::execute(OpenMMContextImpl& context, const LangevinIntegrator& integrator) {
466
    _gpuContext* gpu = data.gpu;
467
468
469
    double temperature = integrator.getTemperature();
    double friction = integrator.getFriction();
    double stepSize = integrator.getStepSize();
470
471
472
473
    if (temperature != prevTemp || friction != prevFriction || stepSize != prevStepSize) {
        // Initialize the GPU parameters.
        
        double tau = (friction == 0.0 ? 0.0 : 1.0/friction);
474
        gpuSetIntegrationParameters(gpu, (float) tau, (float) stepSize, (float) temperature);
475
476
477
478
479
480
481
482
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFriction = friction;
        prevStepSize = stepSize;
    }
    kUpdatePart1(gpu);
    kApplyFirstShake(gpu);
483
    kApplyFirstSettle(gpu);
484
    kApplyFirstLincs(gpu);
485
    if (data.removeCM) {
486
        int step = (int) (context.getTime()/stepSize);
487
488
489
        if (step%data.cmMotionFrequency == 0)
            gpu->bCalculateCM = true;
    }
490
491
    kUpdatePart2(gpu);
    kApplySecondShake(gpu);
492
    kApplySecondSettle(gpu);
493
    kApplySecondLincs(gpu);
494
}
495
496
497
498
499

CudaIntegrateBrownianStepKernel::~CudaIntegrateBrownianStepKernel() {
}

void CudaIntegrateBrownianStepKernel::initialize(const System& system, const BrownianIntegrator& integrator) {
500
    initializeIntegration(system, data, integrator);
501
502
503
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) integrator.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
504
505
506
507
508
509
510
511
512
513
514
515
    prevStepSize = -1.0;
}

void CudaIntegrateBrownianStepKernel::execute(OpenMMContextImpl& context, const BrownianIntegrator& integrator) {
    _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);
516
        gpuSetBrownianIntegrationParameters(gpu, (float) tau, (float) stepSize, (float) temperature);
517
518
519
520
521
522
523
524
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFriction = friction;
        prevStepSize = stepSize;
    }
    kBrownianUpdatePart1(gpu);
    kApplyFirstShake(gpu);
525
    kApplyFirstSettle(gpu);
526
    kApplyFirstLincs(gpu);
527
    if (data.removeCM) {
528
        int step = (int) (context.getTime()/stepSize);
529
530
531
532
533
        if (step%data.cmMotionFrequency == 0)
            gpu->bCalculateCM = true;
    }
    kBrownianUpdatePart2(gpu);
}
534
535
536
537
538

CudaApplyAndersenThermostatKernel::~CudaApplyAndersenThermostatKernel() {
}

void CudaApplyAndersenThermostatKernel::initialize(const System& system, const AndersenThermostat& thermostat) {
539
540
541
    _gpuContext* gpu = data.gpu;
    gpu->seed = (unsigned long) thermostat.getRandomNumberSeed();
    gpuInitializeRandoms(gpu);
542
543
544
545
546
    prevStepSize = -1.0;
}

void CudaApplyAndersenThermostatKernel::execute(OpenMMContextImpl& context) {
    _gpuContext* gpu = data.gpu;
547
548
    double temperature = context.getParameter(AndersenThermostat::Temperature());
    double frequency = context.getParameter(AndersenThermostat::CollisionFrequency());
549
550
551
552
    double stepSize = context.getIntegrator().getStepSize();
    if (temperature != prevTemp || frequency != prevFrequency || stepSize != prevStepSize) {
        // Initialize the GPU parameters.
        
553
        gpuSetAndersenThermostatParameters(gpu, (float) temperature, (float) (frequency*stepSize));
554
555
556
557
558
559
560
561
        gpuSetConstants(gpu);
        kGenerateRandoms(gpu);
        prevTemp = temperature;
        prevFrequency = frequency;
        prevStepSize = stepSize;
    }
    kCalculateAndersenThermostat(gpu);
}
562

563
void CudaCalcKineticEnergyKernel::initialize(const System& system) {
Peter Eastman's avatar
Peter Eastman committed
564
565
    int numParticles = system.getNumParticles();
    masses.resize(numParticles);
566
    for (int i = 0; i < numParticles; ++i)
Peter Eastman's avatar
Peter Eastman committed
567
        masses[i] = system.getParticleMass(i);
568
569
}

570
double CudaCalcKineticEnergyKernel::execute(OpenMMContextImpl& context) {
571
572
573
    // We don't currently have a GPU kernel to do this, so we retrieve the velocities and calculate the energy
    // on the CPU.
    
574
    const Stream& velocities = context.getVelocities();
575
576
577
578
579
580
581
582
    double* v = new double[velocities.getSize()*3];
    velocities.saveToArray(v);
    double energy = 0.0;
    for (size_t i = 0; i < masses.size(); ++i)
        energy += masses[i]*(v[i*3]*v[i*3]+v[i*3+1]*v[i*3+1]+v[i*3+2]*v[i*3+2]);
    delete v;
    return 0.5*energy;
}
583

584
void CudaRemoveCMMotionKernel::initialize(const System& system, const CMMotionRemover& force) {
585
    data.removeCM = true;
586
    data.cmMotionFrequency = force.getFrequency();
587
588
}

589
void CudaRemoveCMMotionKernel::execute(OpenMMContextImpl& context) {
590
}