Skip to content

GitLab

Unverified Commit c456dd54 authored by Peter Eastman's avatar Peter Eastman Committed by GitHub
Browse files

Use cuCtxPushCurrent() and cuCtxPopCurrent() for selecting CUDA context (#3258) 

* Use cuCtxPushCurrent() and cuCtxPopCurrent() for selecting CUDA context

* Fixed errors in amoeba coda

* Fixed more errors in context selection
parent b2c35a8b
Hide whitespace changes
Inline Side-by-side
Showing with 292 additions and 122 deletions
platforms/common/include/openmm/common/ComputeContext.h
View file @ c456dd54
......@@ -93,9 +93,30 @@ public:
* doing any computation when you do not know what other code has just been executing on
* the thread. Platforms that rely on binding contexts to threads (such as CUDA) need to
* implement this.
*
* @deprecated It is recommended to use pushAsCurrent() and popAsCurrent() instead, or even better to create a ContextSelector.
* This provides better interoperability with other libraries that use CUDA and create
* their own contexts.
*/
virtual void setAsCurrent() {
}
/**
* Set this as the current context for the calling thread, maintaining any previous context
* on a stack. This should be called before doing any computation when you do not know what
* other code has just been executing on the thread. It must be paired with popAsCurrent()
* when you are done to restore the previous context. Alternatively, you can create a
* ContextSelector object to automate this for a block of code.
*
* Platforms that rely on binding contexts to threads (such as CUDA) need to implement this.
*/
virtual void pushAsCurrent() {
}
/**
* Restore a previous context that was replaced by pushAsCurrent(). Platforms that rely on binding
* contexts to threads (such as CUDA) need to implement this.
*/
virtual void popAsCurrent() {
}
/**
* Get the number of contexts being used for the current simulation.
* This is relevant when a simulation is parallelized across multiple devices. In that case,
......
platforms/common/include/openmm/common/ContextSelector.h 0 → 100644
View file @ c456dd54
#ifndef OPENMM_CONTEXTSELECTOR_H_
#define OPENMM_CONTEXTSELECTOR_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) 2021 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 <http://www.gnu.org/licenses/>. *
* -------------------------------------------------------------------------- */
#include "ComputeContext.h"
namespace OpenMM {
/**
* This class provides a safe and easy way to select a ComputeContext as current
* for a block of code. The constructor calls pushAsCurrent() on the context.
* When it goes out of scope, the destructor calls popAsCurrent() on it. Simply
* create a local variable of this class, and the context will be current for
* the remainder of the code block in which it is declared.
*/
class OPENMM_EXPORT_COMMON ContextSelector {
public:
ContextSelector(ComputeContext& context) : context(context) {
context.pushAsCurrent();
}
~ContextSelector() {
context.popAsCurrent();
}
private:
ComputeContext& context;
};
} // namespace OpenMM
#endif /*OPENMM_CONTEXTSELECTOR_H_*/
platforms/common/src/CommonKernels.cpp
View file @ c456dd54
......@@ -25,6 +25,7 @@
* -------------------------------------------------------------------------- */
#include "openmm/common/CommonKernels.h"
#include "openmm/common/ContextSelector.h"
#include "openmm/common/ExpressionUtilities.h"
#include "openmm/Context.h"
#include "openmm/internal/AndersenThermostatImpl.h"
......@@ -110,7 +111,7 @@ void CommonApplyConstraintsKernel::initialize(const System& system) {
}
void CommonApplyConstraintsKernel::apply(ContextImpl& context, double tol) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (!hasInitializedKernel) {
hasInitializedKernel = true;
map<string, string> defines;
......@@ -167,7 +168,7 @@ private:
};
void CommonCalcHarmonicBondForceKernel::initialize(const System& system, const HarmonicBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumBonds()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumBonds()/numContexts;
......@@ -197,7 +198,7 @@ double CommonCalcHarmonicBondForceKernel::execute(ContextImpl& context, bool inc
}
void CommonCalcHarmonicBondForceKernel::copyParametersToContext(ContextImpl& context, const HarmonicBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumBonds()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumBonds()/numContexts;
......@@ -251,13 +252,13 @@ private:
};
CommonCalcCustomBondForceKernel::~CommonCalcCustomBondForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
}
void CommonCalcCustomBondForceKernel::initialize(const System& system, const CustomBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumBonds()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumBonds()/numContexts;
......@@ -332,6 +333,7 @@ void CommonCalcCustomBondForceKernel::initialize(const System& system, const Cus
}
double CommonCalcCustomBondForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -347,7 +349,7 @@ double CommonCalcCustomBondForceKernel::execute(ContextImpl& context, bool inclu
}
void CommonCalcCustomBondForceKernel::copyParametersToContext(ContextImpl& context, const CustomBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumBonds()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumBonds()/numContexts;
......@@ -402,7 +404,7 @@ private:
};
void CommonCalcHarmonicAngleForceKernel::initialize(const System& system, const HarmonicAngleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumAngles()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumAngles()/numContexts;
......@@ -433,7 +435,7 @@ double CommonCalcHarmonicAngleForceKernel::execute(ContextImpl& context, bool in
}
void CommonCalcHarmonicAngleForceKernel::copyParametersToContext(ContextImpl& context, const HarmonicAngleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumAngles()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumAngles()/numContexts;
......@@ -489,13 +491,13 @@ private:
};
CommonCalcCustomAngleForceKernel::~CommonCalcCustomAngleForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
}
void CommonCalcCustomAngleForceKernel::initialize(const System& system, const CustomAngleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumAngles()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumAngles()/numContexts;
......@@ -570,6 +572,7 @@ void CommonCalcCustomAngleForceKernel::initialize(const System& system, const Cu
}
double CommonCalcCustomAngleForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -585,7 +588,7 @@ double CommonCalcCustomAngleForceKernel::execute(ContextImpl& context, bool incl
}
void CommonCalcCustomAngleForceKernel::copyParametersToContext(ContextImpl& context, const CustomAngleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumAngles()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumAngles()/numContexts;
......@@ -641,7 +644,7 @@ private:
};
void CommonCalcPeriodicTorsionForceKernel::initialize(const System& system, const PeriodicTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -672,7 +675,7 @@ double CommonCalcPeriodicTorsionForceKernel::execute(ContextImpl& context, bool
}
void CommonCalcPeriodicTorsionForceKernel::copyParametersToContext(ContextImpl& context, const PeriodicTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -726,7 +729,7 @@ private:
};
void CommonCalcRBTorsionForceKernel::initialize(const System& system, const RBTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -762,7 +765,7 @@ double CommonCalcRBTorsionForceKernel::execute(ContextImpl& context, bool includ
}
void CommonCalcRBTorsionForceKernel::copyParametersToContext(ContextImpl& context, const RBTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -827,7 +830,7 @@ CommonCalcCustomTorsionForceKernel::~CommonCalcCustomTorsionForceKernel() {
}
void CommonCalcCustomTorsionForceKernel::initialize(const System& system, const CustomTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -902,6 +905,7 @@ void CommonCalcCustomTorsionForceKernel::initialize(const System& system, const
}
double CommonCalcCustomTorsionForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -917,7 +921,7 @@ double CommonCalcCustomTorsionForceKernel::execute(ContextImpl& context, bool in
}
void CommonCalcCustomTorsionForceKernel::copyParametersToContext(ContextImpl& context, const CustomTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -975,7 +979,7 @@ private:
};
void CommonCalcCMAPTorsionForceKernel::initialize(const System& system, const CMAPTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsions()/numContexts;
......@@ -1038,6 +1042,7 @@ void CommonCalcCMAPTorsionForceKernel::copyParametersToContext(ContextImpl& cont
// Update the maps.
ContextSelector selector(cc);
vector<mm_float4> coeffVec;
vector<double> energy;
vector<vector<double> > c;
......@@ -1100,13 +1105,13 @@ private:
};
CommonCalcCustomExternalForceKernel::~CommonCalcCustomExternalForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
}
void CommonCalcCustomExternalForceKernel::initialize(const System& system, const CustomExternalForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumParticles()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumParticles()/numContexts;
......@@ -1182,6 +1187,7 @@ void CommonCalcCustomExternalForceKernel::initialize(const System& system, const
}
double CommonCalcCustomExternalForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -1197,7 +1203,7 @@ double CommonCalcCustomExternalForceKernel::execute(ContextImpl& context, bool i
}
void CommonCalcCustomExternalForceKernel::copyParametersToContext(ContextImpl& context, const CustomExternalForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumParticles()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumParticles()/numContexts;
......@@ -1250,13 +1256,13 @@ private:
};
CommonCalcCustomCompoundBondForceKernel::~CommonCalcCustomCompoundBondForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
}
void CommonCalcCustomCompoundBondForceKernel::initialize(const System& system, const CustomCompoundBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumBonds()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumBonds()/numContexts;
......@@ -1367,6 +1373,7 @@ void CommonCalcCustomCompoundBondForceKernel::initialize(const System& system, c
}
double CommonCalcCustomCompoundBondForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -1382,7 +1389,7 @@ double CommonCalcCustomCompoundBondForceKernel::execute(ContextImpl& context, bo
}
void CommonCalcCustomCompoundBondForceKernel::copyParametersToContext(ContextImpl& context, const CustomCompoundBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumBonds()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumBonds()/numContexts;
......@@ -1453,13 +1460,13 @@ private:
};
CommonCalcCustomCentroidBondForceKernel::~CommonCalcCustomCentroidBondForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
}
void CommonCalcCustomCentroidBondForceKernel::initialize(const System& system, const CustomCentroidBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
numBonds = force.getNumBonds();
if (numBonds == 0)
return;
......@@ -1678,6 +1685,7 @@ void CommonCalcCustomCentroidBondForceKernel::initialize(const System& system, c
double CommonCalcCustomCentroidBondForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
if (numBonds == 0)
return 0.0;
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -1701,7 +1709,7 @@ double CommonCalcCustomCentroidBondForceKernel::execute(ContextImpl& context, bo
}
void CommonCalcCustomCentroidBondForceKernel::copyParametersToContext(ContextImpl& context, const CustomCentroidBondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (numBonds != force.getNumBonds())
throw OpenMMException("updateParametersInContext: The number of bonds has changed");
if (numBonds == 0)
......@@ -1814,7 +1822,7 @@ private:
};
CommonCalcCustomNonbondedForceKernel::~CommonCalcCustomNonbondedForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
if (forceCopy != NULL)
......@@ -1822,7 +1830,7 @@ CommonCalcCustomNonbondedForceKernel::~CommonCalcCustomNonbondedForceKernel() {
}
void CommonCalcCustomNonbondedForceKernel::initialize(const System& system, const CustomNonbondedForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int forceIndex;
for (forceIndex = 0; forceIndex < system.getNumForces() && &system.getForce(forceIndex) != &force; ++forceIndex)
;
......@@ -2243,6 +2251,7 @@ double CommonCalcCustomNonbondedForceKernel::execute(ContextImpl& context, bool
// When using a neighbor list, run the whole calculation on a single device.
return 0.0;
}
ContextSelector selector(cc);
bool recomputeLongRangeCorrection = !hasInitializedLongRangeCorrection;
if (globals.isInitialized()) {
bool changed = false;
......@@ -2311,7 +2320,7 @@ double CommonCalcCustomNonbondedForceKernel::execute(ContextImpl& context, bool
}
void CommonCalcCustomNonbondedForceKernel::copyParametersToContext(ContextImpl& context, const CustomNonbondedForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
if (numParticles != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -2359,7 +2368,7 @@ private:
};
void CommonCalcGBSAOBCForceKernel::initialize(const System& system, const GBSAOBCForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (cc.getNumContexts() > 1)
throw OpenMMException("GBSAOBCForce does not support using multiple devices");
int forceIndex;
......@@ -2418,6 +2427,7 @@ void CommonCalcGBSAOBCForceKernel::initialize(const System& system, const GBSAOB
}
double CommonCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
NonbondedUtilities& nb = cc.getNonbondedUtilities();
bool deviceIsCpu = cc.getIsCPU();
if (!hasCreatedKernels) {
......@@ -2534,7 +2544,7 @@ double CommonCalcGBSAOBCForceKernel::execute(ContextImpl& context, bool includeF
void CommonCalcGBSAOBCForceKernel::copyParametersToContext(ContextImpl& context, const GBSAOBCForce& force) {
// Make sure the new parameters are acceptable.
cc.setAsCurrent();
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
if (numParticles != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -2593,7 +2603,7 @@ private:
};
CommonCalcCustomGBForceKernel::~CommonCalcCustomGBForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
if (computedValues != NULL)
......@@ -2607,7 +2617,7 @@ CommonCalcCustomGBForceKernel::~CommonCalcCustomGBForceKernel() {
}
void CommonCalcCustomGBForceKernel::initialize(const System& system, const CustomGBForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (cc.getNumContexts() > 1)
throw OpenMMException("CustomGBForce does not support using multiple devices");
NonbondedUtilities& nb = cc.getNonbondedUtilities();
......@@ -3426,6 +3436,7 @@ void CommonCalcCustomGBForceKernel::initialize(const System& system, const Custo
}
double CommonCalcCustomGBForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
bool deviceIsCpu = cc.getIsCPU();
NonbondedUtilities& nb = cc.getNonbondedUtilities();
int elementSize = (cc.getUseDoublePrecision() ? sizeof(double) : sizeof(float));
......@@ -3634,7 +3645,7 @@ double CommonCalcCustomGBForceKernel::execute(ContextImpl& context, bool include
}
void CommonCalcCustomGBForceKernel::copyParametersToContext(ContextImpl& context, const CustomGBForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
if (numParticles != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -3732,7 +3743,7 @@ private:
};
CommonCalcCustomHbondForceKernel::~CommonCalcCustomHbondForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (donorParams != NULL)
delete donorParams;
if (acceptorParams != NULL)
......@@ -3756,7 +3767,7 @@ static void applyDonorAndAcceptorForces(stringstream& applyToDonor, stringstream
void CommonCalcCustomHbondForceKernel::initialize(const System& system, const CustomHbondForce& force) {
// Record the lists of donors and acceptors, and the parameters for each one.
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumDonors()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumDonors()/numContexts;
......@@ -4084,6 +4095,7 @@ void CommonCalcCustomHbondForceKernel::initialize(const System& system, const Cu
double CommonCalcCustomHbondForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
if (numDonors == 0 || numAcceptors == 0)
return 0.0;
ContextSelector selector(cc);
if (globals.isInitialized()) {
bool changed = false;
for (int i = 0; i < (int) globalParamNames.size(); i++) {
......@@ -4148,7 +4160,7 @@ double CommonCalcCustomHbondForceKernel::execute(ContextImpl& context, bool incl
}
void CommonCalcCustomHbondForceKernel::copyParametersToContext(ContextImpl& context, const CustomHbondForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumDonors()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumDonors()/numContexts;
......@@ -4226,13 +4238,13 @@ private:
};
CommonCalcCustomManyParticleForceKernel::~CommonCalcCustomManyParticleForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (params != NULL)
delete params;
}
void CommonCalcCustomManyParticleForceKernel::initialize(const System& system, const CustomManyParticleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (!cc.getSupports64BitGlobalAtomics())
throw OpenMMException("CustomManyParticleForce requires a device that supports 64 bit atomic operations");
int numParticles = force.getNumParticles();
......@@ -4548,6 +4560,7 @@ void CommonCalcCustomManyParticleForceKernel::initialize(const System& system, c
}
double CommonCalcCustomManyParticleForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (!hasInitializedKernel) {
hasInitializedKernel = true;
......@@ -4674,7 +4687,7 @@ double CommonCalcCustomManyParticleForceKernel::execute(ContextImpl& context, bo
}
void CommonCalcCustomManyParticleForceKernel::copyParametersToContext(ContextImpl& context, const CustomManyParticleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
if (numParticles != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -4766,6 +4779,7 @@ void CommonCalcGayBerneForceKernel::initialize(const System& system, const GayBe
// Initialize interactions.
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
sigParams.initialize<mm_float4>(cc, cc.getPaddedNumAtoms(), "sigParams");
epsParams.initialize<mm_float2>(cc, cc.getPaddedNumAtoms(), "epsParams");
......@@ -4877,6 +4891,7 @@ void CommonCalcGayBerneForceKernel::initialize(const System& system, const GayBe
}
double CommonCalcGayBerneForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (!hasInitializedKernels) {
hasInitializedKernels = true;
sortAtoms();
......@@ -4993,6 +5008,7 @@ void CommonCalcGayBerneForceKernel::copyParametersToContext(ContextImpl& context
// Record the per-particle parameters.
ContextSelector selector(cc);
vector<mm_float4> sigParamsVector(cc.getPaddedNumAtoms(), mm_float4(0, 0, 0, 0));
vector<mm_float2> epsParamsVector(cc.getPaddedNumAtoms(), mm_float2(0, 0));
vector<mm_float4> scaleVector(cc.getPaddedNumAtoms(), mm_float4(0, 0, 0, 0));
......@@ -5111,6 +5127,7 @@ private:
};
void CommonCalcCustomCVForceKernel::initialize(const System& system, const CustomCVForce& force, ContextImpl& innerContext) {
ContextSelector selector(cc);
int numCVs = force.getNumCollectiveVariables();
for (int i = 0; i < force.getNumGlobalParameters(); i++)
globalParameterNames.push_back(force.getGlobalParameterName(i));
......@@ -5203,6 +5220,7 @@ void CommonCalcCustomCVForceKernel::initialize(const System& system, const Custo
}
double CommonCalcCustomCVForceKernel::execute(ContextImpl& context, ContextImpl& innerContext, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
copyState(context, innerContext);
int numCVs = variableNames.size();
int numAtoms = cc.getNumAtoms();
......@@ -5248,6 +5266,7 @@ double CommonCalcCustomCVForceKernel::execute(ContextImpl& context, ContextImpl&
}
void CommonCalcCustomCVForceKernel::copyState(ContextImpl& context, ContextImpl& innerContext) {
ContextSelector selector(cc);
int numAtoms = cc.getNumAtoms();
ComputeContext& cc2 = getInnerComputeContext(innerContext);
if (!hasInitializedListeners) {
......@@ -5295,14 +5314,14 @@ void CommonCalcCustomCVForceKernel::copyParametersToContext(ContextImpl& context
void CommonIntegrateVerletStepKernel::initialize(const System& system, const VerletIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
ComputeProgram program = cc.compileProgram(CommonKernelSources::verlet);
kernel1 = program->createKernel("integrateVerletPart1");
kernel2 = program->createKernel("integrateVerletPart2");
}
void CommonIntegrateVerletStepKernel::execute(ContextImpl& context, const VerletIntegrator& integrator) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int paddedNumAtoms = cc.getPaddedNumAtoms();
......@@ -5360,7 +5379,7 @@ double CommonIntegrateVerletStepKernel::computeKineticEnergy(ContextImpl& contex
void CommonIntegrateLangevinStepKernel::initialize(const System& system, const LangevinIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
cc.getIntegrationUtilities().initRandomNumberGenerator(integrator.getRandomNumberSeed());
ComputeProgram program = cc.compileProgram(CommonKernelSources::langevin);
kernel1 = program->createKernel("integrateLangevinPart1");
......@@ -5370,7 +5389,7 @@ void CommonIntegrateLangevinStepKernel::initialize(const System& system, const L
}
void CommonIntegrateLangevinStepKernel::execute(ContextImpl& context, const LangevinIntegrator& integrator) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int paddedNumAtoms = cc.getPaddedNumAtoms();
......@@ -5447,7 +5466,7 @@ double CommonIntegrateLangevinStepKernel::computeKineticEnergy(ContextImpl& cont
void CommonIntegrateLangevinMiddleStepKernel::initialize(const System& system, const LangevinMiddleIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
cc.getIntegrationUtilities().initRandomNumberGenerator(integrator.getRandomNumberSeed());
ComputeProgram program = cc.compileProgram(CommonKernelSources::langevinMiddle);
kernel1 = program->createKernel("integrateLangevinMiddlePart1");
......@@ -5465,6 +5484,7 @@ void CommonIntegrateLangevinMiddleStepKernel::initialize(const System& system, c
}
void CommonIntegrateLangevinMiddleStepKernel::execute(ContextImpl& context, const LangevinMiddleIntegrator& integrator) {
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int paddedNumAtoms = cc.getPaddedNumAtoms();
......@@ -5540,6 +5560,7 @@ double CommonIntegrateLangevinMiddleStepKernel::computeKineticEnergy(ContextImpl
void CommonIntegrateNoseHooverStepKernel::initialize(const System& system, const NoseHooverIntegrator& integrator) {
cc.initializeContexts();
ContextSelector selector(cc);
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
map<string, string> defines;
defines["BOLTZ"] = cc.doubleToString(BOLTZ);
......@@ -5600,6 +5621,7 @@ void CommonIntegrateNoseHooverStepKernel::initialize(const System& system, const
}
void CommonIntegrateNoseHooverStepKernel::execute(ContextImpl& context, const NoseHooverIntegrator& integrator, bool &forcesAreValid) {
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int paddedNumAtoms = cc.getPaddedNumAtoms();
double dt = integrator.getStepSize();
......@@ -5897,9 +5919,8 @@ std::pair<double, double> CommonIntegrateNoseHooverStepKernel::propagateChain(Co
}
double CommonIntegrateNoseHooverStepKernel::computeHeatBathEnergy(ContextImpl& context, const NoseHooverChain &nhc) {
ContextSelector selector(cc);
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
int chainID = nhc.getChainID();
int chainLength = nhc.getChainLength();
......@@ -5980,9 +6001,8 @@ double CommonIntegrateNoseHooverStepKernel::computeHeatBathEnergy(ContextImpl& c
}
std::pair<double, double> CommonIntegrateNoseHooverStepKernel::computeMaskedKineticEnergy(ContextImpl& context, const NoseHooverChain &nhc, bool downloadValue) {
ContextSelector selector(cc);
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
int chainID = nhc.getChainID();
const auto & nhcAtoms = nhc.getThermostatedAtoms();
const auto & nhcPairs = nhc.getThermostatedPairs();
......@@ -6106,6 +6126,7 @@ void CommonIntegrateNoseHooverStepKernel::scaleVelocities(ContextImpl& context,
}
void CommonIntegrateNoseHooverStepKernel::createCheckpoint(ContextImpl& context, ostream& stream) const {
ContextSelector selector(cc);
int numChains = chainState.size();
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
stream.write((char*) &numChains, sizeof(int));
......@@ -6128,6 +6149,7 @@ void CommonIntegrateNoseHooverStepKernel::createCheckpoint(ContextImpl& context,
}
void CommonIntegrateNoseHooverStepKernel::loadCheckpoint(ContextImpl& context, istream& stream) {
ContextSelector selector(cc);
int numChains;
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
stream.read((char*) &numChains, sizeof(int));
......@@ -6154,6 +6176,7 @@ void CommonIntegrateNoseHooverStepKernel::loadCheckpoint(ContextImpl& context, i
}
void CommonIntegrateNoseHooverStepKernel::getChainStates(ContextImpl& context, vector<vector<double> >& positions, vector<vector<double> >& velocities) const {
ContextSelector selector(cc);
int numChains = chainState.size();
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
positions.clear();
......@@ -6182,6 +6205,7 @@ void CommonIntegrateNoseHooverStepKernel::getChainStates(ContextImpl& context, v
}
void CommonIntegrateNoseHooverStepKernel::setChainStates(ContextImpl& context, const vector<vector<double> >& positions, const vector<vector<double> >& velocities) {
ContextSelector selector(cc);
int numChains = positions.size();
bool useDouble = cc.getUseDoublePrecision() || cc.getUseMixedPrecision();
chainState.clear();
......@@ -6207,7 +6231,7 @@ void CommonIntegrateNoseHooverStepKernel::setChainStates(ContextImpl& context, c
void CommonIntegrateBrownianStepKernel::initialize(const System& system, const BrownianIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
cc.getIntegrationUtilities().initRandomNumberGenerator(integrator.getRandomNumberSeed());
ComputeProgram program = cc.compileProgram(CommonKernelSources::brownian);
kernel1 = program->createKernel("integrateBrownianPart1");
......@@ -6216,7 +6240,7 @@ void CommonIntegrateBrownianStepKernel::initialize(const System& system, const B
}
void CommonIntegrateBrownianStepKernel::execute(ContextImpl& context, const BrownianIntegrator& integrator) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int paddedNumAtoms = cc.getPaddedNumAtoms();
......@@ -6292,7 +6316,7 @@ double CommonIntegrateBrownianStepKernel::computeKineticEnergy(ContextImpl& cont
void CommonIntegrateVariableVerletStepKernel::initialize(const System& system, const VariableVerletIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
ComputeProgram program = cc.compileProgram(CommonKernelSources::verlet);
kernel1 = program->createKernel("integrateVerletPart1");
kernel2 = program->createKernel("integrateVerletPart2");
......@@ -6301,7 +6325,7 @@ void CommonIntegrateVariableVerletStepKernel::initialize(const System& system, c
}
double CommonIntegrateVariableVerletStepKernel::execute(ContextImpl& context, const VariableVerletIntegrator& integrator, double maxTime) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int paddedNumAtoms = cc.getPaddedNumAtoms();
......@@ -6392,7 +6416,7 @@ double CommonIntegrateVariableVerletStepKernel::computeKineticEnergy(ContextImpl
void CommonIntegrateVariableLangevinStepKernel::initialize(const System& system, const VariableLangevinIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
cc.getIntegrationUtilities().initRandomNumberGenerator(integrator.getRandomNumberSeed());
ComputeProgram program = cc.compileProgram(CommonKernelSources::langevin);
kernel1 = program->createKernel("integrateLangevinPart1");
......@@ -6404,7 +6428,7 @@ void CommonIntegrateVariableLangevinStepKernel::initialize(const System& system,
}
double CommonIntegrateVariableLangevinStepKernel::execute(ContextImpl& context, const VariableLangevinIntegrator& integrator, double maxTime) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int paddedNumAtoms = cc.getPaddedNumAtoms();
......@@ -6573,7 +6597,7 @@ private:
void CommonIntegrateCustomStepKernel::initialize(const System& system, const CustomIntegrator& integrator) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
cc.getIntegrationUtilities().initRandomNumberGenerator(integrator.getRandomNumberSeed());
numGlobalVariables = integrator.getNumGlobalVariables();
int elementSize = (cc.getUseDoublePrecision() || cc.getUseMixedPrecision() ? sizeof(double) : sizeof(float));
......@@ -6633,7 +6657,7 @@ string CommonIntegrateCustomStepKernel::createPerDofComputation(const string& va
}
void CommonIntegrateCustomStepKernel::prepareForComputation(ContextImpl& context, CustomIntegrator& integrator, bool& forcesAreValid) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
int numSteps = integrator.getNumComputations();
......@@ -7098,6 +7122,7 @@ void CommonIntegrateCustomStepKernel::findExpressionsForDerivs(const ExpressionT
}
void CommonIntegrateCustomStepKernel::execute(ContextImpl& context, CustomIntegrator& integrator, bool& forcesAreValid) {
ContextSelector selector(cc);
prepareForComputation(context, integrator, forcesAreValid);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
......@@ -7278,6 +7303,7 @@ bool CommonIntegrateCustomStepKernel::evaluateCondition(int step) {
}
double CommonIntegrateCustomStepKernel::computeKineticEnergy(ContextImpl& context, CustomIntegrator& integrator, bool& forcesAreValid) {
ContextSelector selector(cc);
prepareForComputation(context, integrator, forcesAreValid);
cc.clearBuffer(sumBuffer);
kineticEnergyKernel->setArg(8, cc.getIntegrationUtilities().getRandom());
......@@ -7354,6 +7380,7 @@ void CommonIntegrateCustomStepKernel::setGlobalVariables(ContextImpl& context, c
}
void CommonIntegrateCustomStepKernel::getPerDofVariable(ContextImpl& context, int variable, vector<Vec3>& values) const {
ContextSelector selector(cc);
values.resize(perDofValues[variable].getSize());
const vector<int>& order = cc.getAtomIndex();
if (cc.getUseDoublePrecision() || cc.getUseMixedPrecision()) {
......@@ -7397,7 +7424,7 @@ void CommonIntegrateCustomStepKernel::setPerDofVariable(ContextImpl& context, in
}
void CommonRemoveCMMotionKernel::initialize(const System& system, const CMMotionRemover& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
frequency = force.getFrequency();
int numAtoms = cc.getNumAtoms();
cmMomentum.initialize<mm_float4>(cc, cc.getPaddedNumAtoms(), "cmMomentum");
......@@ -7418,7 +7445,7 @@ void CommonRemoveCMMotionKernel::initialize(const System& system, const CMMotion
}
void CommonRemoveCMMotionKernel::execute(ContextImpl& context) {
cc.setAsCurrent();
ContextSelector selector(cc);
kernel1->execute(cc.getNumAtoms(), 64);
kernel2->execute(cc.getNumAtoms(), 64);
}
......@@ -7446,6 +7473,7 @@ private:
void CommonCalcRMSDForceKernel::initialize(const System& system, const RMSDForce& force) {
// Create data structures.
ContextSelector selector(cc);
bool useDouble = cc.getUseDoublePrecision();
int elementSize = (useDouble ? sizeof(double) : sizeof(float));
int numParticles = force.getParticles().size();
......@@ -7513,6 +7541,7 @@ void CommonCalcRMSDForceKernel::recordParameters(const RMSDForce& force) {
}
double CommonCalcRMSDForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (cc.getUseDoublePrecision())
return executeImpl<double>(context);
return executeImpl<float>(context);
......@@ -7591,6 +7620,7 @@ double CommonCalcRMSDForceKernel::executeImpl(ContextImpl& context) {
}
void CommonCalcRMSDForceKernel::copyParametersToContext(ContextImpl& context, const RMSDForce& force) {
ContextSelector selector(cc);
if (referencePos.getSize() != force.getReferencePositions().size())
throw OpenMMException("updateParametersInContext: The number of reference positions has changed");
int numParticles = force.getParticles().size();
......@@ -7607,7 +7637,7 @@ void CommonCalcRMSDForceKernel::copyParametersToContext(ContextImpl& context, co
}
void CommonApplyAndersenThermostatKernel::initialize(const System& system, const AndersenThermostat& thermostat) {
cc.setAsCurrent();
ContextSelector selector(cc);
randomSeed = thermostat.getRandomNumberSeed();
ComputeProgram program = cc.compileProgram(CommonKernelSources::andersenThermostat);
kernel = program->createKernel("applyAndersenThermostat");
......@@ -7634,7 +7664,7 @@ void CommonApplyAndersenThermostatKernel::initialize(const System& system, const
}
void CommonApplyAndersenThermostatKernel::execute(ContextImpl& context) {
cc.setAsCurrent();
ContextSelector selector(cc);
kernel->setArg(1, (float) context.getParameter(AndersenThermostat::CollisionFrequency()));
kernel->setArg(2, (float) (BOLTZ*context.getParameter(AndersenThermostat::Temperature())));
double stepSize = context.getIntegrator().getStepSize();
......@@ -7647,6 +7677,7 @@ void CommonApplyAndersenThermostatKernel::execute(ContextImpl& context) {
}
void CommonApplyMonteCarloBarostatKernel::initialize(const System& system, const Force& thermostat) {
ContextSelector selector(cc);
savedPositions.initialize(cc, cc.getPaddedNumAtoms(), cc.getUseDoublePrecision() ? sizeof(mm_double4) : sizeof(mm_float4), "savedPositions");
savedLongForces.initialize<long long>(cc, cc.getPaddedNumAtoms()*3, "savedLongForces");
try {
......@@ -7661,6 +7692,7 @@ void CommonApplyMonteCarloBarostatKernel::initialize(const System& system, const
}
void CommonApplyMonteCarloBarostatKernel::scaleCoordinates(ContextImpl& context, double scaleX, double scaleY, double scaleZ) {
ContextSelector selector(cc);
if (!hasInitializedKernels) {
hasInitializedKernels = true;
......@@ -7684,10 +7716,6 @@ void CommonApplyMonteCarloBarostatKernel::scaleCoordinates(ContextImpl& context,
// Initialize the kernel arguments.
//KERNEL void scalePositions(float scaleX, float scaleY, float scaleZ, int numMolecules, real4 periodicBoxSize,
// real4 invPeriodicBoxSize, real4 periodicBoxVecX, real4 periodicBoxVecY, real4 periodicBoxVecZ, GLOBAL real4* RESTRICT posq,
// GLOBAL const int* RESTRICT moleculeAtoms, GLOBAL const int* RESTRICT moleculeStartIndex) {
kernel->addArg();
kernel->addArg();
kernel->addArg();
......@@ -7702,9 +7730,6 @@ void CommonApplyMonteCarloBarostatKernel::scaleCoordinates(ContextImpl& context,
cc.getLongForceBuffer().copyTo(savedLongForces);
if (savedFloatForces.isInitialized())
cc.getFloatForceBuffer().copyTo(savedFloatForces);
// int bytesToCopy = cc.getPosq().getSize()*(cc.getUseDoublePrecision() ? sizeof(mm_double4) : sizeof(mm_float4));
// cc.getQueue().enqueueCopyBuffer(cc.getPosq().getDeviceBuffer(), savedPositions.getDeviceBuffer(), 0, 0, bytesToCopy);
// cc.getQueue().enqueueCopyBuffer(cc.getForce().getDeviceBuffer(), savedForces.getDeviceBuffer(), 0, 0, bytesToCopy);
kernel->setArg(0, (float) scaleX);
kernel->setArg(1, (float) scaleY);
kernel->setArg(2, (float) scaleZ);
......@@ -7716,12 +7741,9 @@ void CommonApplyMonteCarloBarostatKernel::scaleCoordinates(ContextImpl& context,
}
void CommonApplyMonteCarloBarostatKernel::restoreCoordinates(ContextImpl& context) {
ContextSelector selector(cc);
savedPositions.copyTo(cc.getPosq());
savedLongForces.copyTo(cc.getLongForceBuffer());
if (savedFloatForces.isInitialized())
savedFloatForces.copyTo(cc.getFloatForceBuffer());
// int bytesToCopy = cc.getPosq().getSize()*(cc.getUseDoublePrecision() ? sizeof(mm_double4) : sizeof(mm_float4));
// cc.getQueue().enqueueCopyBuffer(savedPositions.getDeviceBuffer(), cc.getPosq().getDeviceBuffer(), 0, 0, bytesToCopy);
// cc.getQueue().enqueueCopyBuffer(savedForces.getDeviceBuffer(), cc.getForce().getDeviceBuffer(), 0, 0, bytesToCopy);
}
platforms/common/src/ComputeContext.cpp
View file @ c456dd54
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2019 Stanford University and the Authors. *
* Portions copyright (c) 2019-2021 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -25,6 +25,7 @@
* -------------------------------------------------------------------------- */
#include "openmm/common/ComputeContext.h"
#include "openmm/common/ContextSelector.h"
#include "openmm/System.h"
#include "openmm/VirtualSite.h"
#include "openmm/internal/ContextImpl.h"
......@@ -362,6 +363,7 @@ bool ComputeContext::invalidateMolecules(ComputeForceInfo* force) {
// atoms to their original order, rebuild the list of identical molecules, and sort them
// again.
ContextSelector selector(*this);
vector<mm_int4> newCellOffsets(numAtoms);
if (getUseDoublePrecision()) {
vector<mm_double4> oldPosq(paddedNumAtoms);
......@@ -598,6 +600,7 @@ void ComputeContext::reorderAtomsImpl() {
// Update the arrays.
ContextSelector selector(*this);
for (int i = 0; i < numAtoms; i++) {
atomIndex[i] = originalIndex[i];
posCellOffsets[i] = newCellOffsets[i];
......
platforms/common/src/IntegrationUtilities.cpp
View file @ c456dd54
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009-2020 Stanford University and the Authors. *
* Portions copyright (c) 2009-2021 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -26,6 +26,7 @@
#include "openmm/common/IntegrationUtilities.h"
#include "openmm/common/ComputeContext.h"
#include "openmm/common/ContextSelector.h"
#include "CommonKernelSources.h"
#include "openmm/internal/OSRngSeed.h"
#include "openmm/HarmonicAngleForce.h"
......@@ -736,6 +737,7 @@ void IntegrationUtilities::applyVelocityConstraints(double tol) {
}
void IntegrationUtilities::computeVirtualSites() {
ContextSelector selector(context);
if (numVsites > 0)
vsitePositionKernel->execute(numVsites);
}
......@@ -812,6 +814,7 @@ void IntegrationUtilities::loadCheckpoint(istream& stream) {
}
double IntegrationUtilities::computeKineticEnergy(double timeShift) {
ContextSelector selector(context);
int numParticles = context.getNumAtoms();
if (timeShift != 0) {
// Copy the velocities into the posDelta array while we temporarily modify them.
......
platforms/cuda/include/CudaContext.h
View file @ c456dd54
......@@ -100,6 +100,16 @@ public:
* valid, this returns without doing anything.
*/
void setAsCurrent();
/**
* Push the CUcontext associated with this object to be the current context. If the context is not
* valid, this returns without doing anything.
*/
void pushAsCurrent();
/**
* Pop the CUcontext associated with this object off the stack of contexts. If the context is not
* valid, this returns without doing anything.
*/
void popAsCurrent();
/**
* Get the CUdevice associated with this object.
*/
......
platforms/cuda/src/CudaArray.cpp
View file @ c456dd54
......@@ -26,6 +26,7 @@
#include "CudaArray.h"
#include "CudaContext.h"
#include "openmm/common/ContextSelector.h"
#include <iostream>
#include <sstream>
#include <vector>
......@@ -41,7 +42,7 @@ CudaArray::CudaArray(CudaContext& context, int size, int elementSize, const std:
CudaArray::~CudaArray() {
if (pointer != 0 && ownsMemory && context->getContextIsValid()) {
context->setAsCurrent();
ContextSelector selector(*context);
CUresult result = cuMemFree(pointer);
if (result != CUDA_SUCCESS) {
std::stringstream str;
......@@ -59,6 +60,7 @@ void CudaArray::initialize(ComputeContext& context, int size, int elementSize, c
this->elementSize = elementSize;
this->name = name;
ownsMemory = true;
ContextSelector selector(*this->context);
CUresult result = cuMemAlloc(&pointer, size*elementSize);
if (result != CUDA_SUCCESS) {
std::stringstream str;
......@@ -72,6 +74,7 @@ void CudaArray::resize(int size) {
throw OpenMMException("CudaArray has not been initialized");
if (!ownsMemory)
throw OpenMMException("Cannot resize an array that does not own its storage");
ContextSelector selector(*context);
CUresult result = cuMemFree(pointer);
if (result != CUDA_SUCCESS) {
std::stringstream str;
......
platforms/cuda/src/CudaContext.cpp
View file @ c456dd54
......@@ -38,6 +38,7 @@
#include "CudaNonbondedUtilities.h"
#include "CudaProgram.h"
#include "openmm/common/ComputeArray.h"
#include "openmm/common/ContextSelector.h"
#include "SHA1.h"
#include "openmm/Platform.h"
#include "openmm/System.h"
......@@ -190,6 +191,8 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
if (cuCtxCreate(&context, flags, device) == CUDA_SUCCESS) {
this->deviceIndex = trialDeviceIndex;
CUcontext popped;
cuCtxPopCurrent(&popped);
break;
}
}
......@@ -231,14 +234,16 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
computeCapability = major+0.1*minor;
contextIsValid = true;
ContextSelector selector(*this);
CHECK_RESULT(cuCtxSetCacheConfig(CU_FUNC_CACHE_PREFER_SHARED));
if (contextIndex > 0) {
int canAccess;
cuDeviceCanAccessPeer(&canAccess, getDevice(), platformData.contexts[0]->getDevice());
if (canAccess) {
platformData.contexts[0]->setAsCurrent();
CHECK_RESULT(cuCtxEnablePeerAccess(getContext(), 0));
setAsCurrent();
{
ContextSelector selector2(*platformData.contexts[0]);
CHECK_RESULT(cuCtxEnablePeerAccess(getContext(), 0));
}
CHECK_RESULT(cuCtxEnablePeerAccess(platformData.contexts[0]->getContext(), 0));
}
}
......@@ -397,7 +402,7 @@ CudaContext::CudaContext(const System& system, int deviceIndex, bool useBlocking
}
CudaContext::~CudaContext() {
setAsCurrent();
pushAsCurrent();
for (auto force : forces)
delete force;
for (auto listener : reorderListeners)
......@@ -416,6 +421,7 @@ CudaContext::~CudaContext() {
delete bonded;
if (nonbonded != NULL)
delete nonbonded;
popAsCurrent();
string errorMessage = "Error deleting Context";
if (contextIsValid && !isLinkedContext) {
cuProfilerStop();
......@@ -425,7 +431,7 @@ CudaContext::~CudaContext() {
}
void CudaContext::initialize() {
cuCtxSetCurrent(context);
ContextSelector selector(*this);
string errorMessage = "Error initializing Context";
int numEnergyBuffers = max(numThreadBlocks*ThreadBlockSize, nonbonded->getNumEnergyBuffers());
if (useDoublePrecision) {
......@@ -478,6 +484,17 @@ void CudaContext::setAsCurrent() {
cuCtxSetCurrent(context);
}
void CudaContext::pushAsCurrent() {
if (contextIsValid)
cuCtxPushCurrent(context);
}
void CudaContext::popAsCurrent() {
CUcontext popped;
if (contextIsValid)
cuCtxPopCurrent(&popped);
}
CUmodule CudaContext::createModule(const string source, const char* optimizationFlags) {
return createModule(source, map<string, string>(), optimizationFlags);
}
......
platforms/cuda/src/CudaIntegrationUtilities.cpp
View file @ c456dd54
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2009-2020 Stanford University and the Authors. *
* Portions copyright (c) 2009-2021 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -26,6 +26,7 @@
#include "CudaIntegrationUtilities.h"
#include "CudaContext.h"
#include "openmm/common/ContextSelector.h"
using namespace OpenMM;
using namespace std;
......@@ -46,7 +47,7 @@ CudaIntegrationUtilities::CudaIntegrationUtilities(CudaContext& context, const S
}
CudaIntegrationUtilities::~CudaIntegrationUtilities() {
context.setAsCurrent();
ContextSelector selector(context);
if (ccmaConvergedMemory != NULL) {
cuMemFreeHost(ccmaConvergedMemory);
cuEventDestroy(ccmaEvent);
......@@ -66,6 +67,7 @@ CudaArray& CudaIntegrationUtilities::getStepSize() {
}
void CudaIntegrationUtilities::applyConstraintsImpl(bool constrainVelocities, double tol) {
ContextSelector selector(context);
ComputeKernel settleKernel, shakeKernel, ccmaForceKernel;
if (constrainVelocities) {
settleKernel = settleVelKernel;
......@@ -131,6 +133,7 @@ void CudaIntegrationUtilities::applyConstraintsImpl(bool constrainVelocities, do
}
void CudaIntegrationUtilities::distributeForcesFromVirtualSites() {
ContextSelector selector(context);
if (numVsites > 0) {
vsiteForceKernel->setArg(2, context.getLongForceBuffer());
vsiteForceKernel->execute(numVsites);
......
platforms/cuda/src/CudaKernels.cpp
View file @ c456dd54
......@@ -29,6 +29,7 @@
#include "openmm/Context.h"
#include "openmm/internal/ContextImpl.h"
#include "openmm/internal/NonbondedForceImpl.h"
#include "openmm/common/ContextSelector.h"
#include "CommonKernelSources.h"
#include "CudaBondedUtilities.h"
#include "CudaExpressionUtilities.h"
......@@ -58,7 +59,7 @@ void CudaCalcForcesAndEnergyKernel::initialize(const System& system) {
void CudaCalcForcesAndEnergyKernel::beginComputation(ContextImpl& context, bool includeForces, bool includeEnergy, int groups) {
cu.setForcesValid(true);
cu.setAsCurrent();
ContextSelector selector(cu);
cu.clearAutoclearBuffers();
for (auto computation : cu.getPreComputations())
computation->computeForceAndEnergy(includeForces, includeEnergy, groups);
......@@ -71,7 +72,7 @@ void CudaCalcForcesAndEnergyKernel::beginComputation(ContextImpl& context, bool
}
double CudaCalcForcesAndEnergyKernel::finishComputation(ContextImpl& context, bool includeForces, bool includeEnergy, int groups, bool& valid) {
cu.setAsCurrent();
ContextSelector selector(cu);
cu.getBondedUtilities().computeInteractions(groups);
cu.getNonbondedUtilities().computeInteractions(groups, includeForces, includeEnergy);
double sum = 0.0;
......@@ -109,7 +110,7 @@ void CudaUpdateStateDataKernel::setStepCount(const ContextImpl& context, long lo
}
void CudaUpdateStateDataKernel::getPositions(ContextImpl& context, vector<Vec3>& positions) {
cu.setAsCurrent();
ContextSelector selector(cu);
int numParticles = context.getSystem().getNumParticles();
positions.resize(numParticles);
vector<float4> posCorrection;
......@@ -170,7 +171,7 @@ void CudaUpdateStateDataKernel::getPositions(ContextImpl& context, vector<Vec3>&
}
void CudaUpdateStateDataKernel::setPositions(ContextImpl& context, const vector<Vec3>& positions) {
cu.setAsCurrent();
ContextSelector selector(cu);
const vector<int>& order = cu.getAtomIndex();
int numParticles = context.getSystem().getNumParticles();
if (cu.getUseDoublePrecision()) {
......@@ -221,7 +222,7 @@ void CudaUpdateStateDataKernel::setPositions(ContextImpl& context, const vector<
}
void CudaUpdateStateDataKernel::getVelocities(ContextImpl& context, vector<Vec3>& velocities) {
cu.setAsCurrent();
ContextSelector selector(cu);
const vector<int>& order = cu.getAtomIndex();
int numParticles = context.getSystem().getNumParticles();
velocities.resize(numParticles);
......@@ -246,7 +247,7 @@ void CudaUpdateStateDataKernel::getVelocities(ContextImpl& context, vector<Vec3>
}
void CudaUpdateStateDataKernel::setVelocities(ContextImpl& context, const vector<Vec3>& velocities) {
cu.setAsCurrent();
ContextSelector selector(cu);
const vector<int>& order = cu.getAtomIndex();
int numParticles = context.getSystem().getNumParticles();
if (cu.getUseDoublePrecision() || cu.getUseMixedPrecision()) {
......@@ -280,7 +281,7 @@ void CudaUpdateStateDataKernel::setVelocities(ContextImpl& context, const vector
}
void CudaUpdateStateDataKernel::getForces(ContextImpl& context, vector<Vec3>& forces) {
cu.setAsCurrent();
ContextSelector selector(cu);
long long* force = (long long*) cu.getPinnedBuffer();
cu.getForce().download(force);
const vector<int>& order = cu.getAtomIndex();
......@@ -293,6 +294,7 @@ void CudaUpdateStateDataKernel::getForces(ContextImpl& context, vector<Vec3>& fo
}
void CudaUpdateStateDataKernel::getEnergyParameterDerivatives(ContextImpl& context, map<string, double>& derivs) {
ContextSelector selector(cu);
const vector<string>& paramDerivNames = cu.getEnergyParamDerivNames();
int numDerivs = paramDerivNames.size();
if (numDerivs == 0)
......@@ -346,7 +348,7 @@ void CudaUpdateStateDataKernel::setPeriodicBoxVectors(ContextImpl& context, cons
}
void CudaUpdateStateDataKernel::createCheckpoint(ContextImpl& context, ostream& stream) {
cu.setAsCurrent();
ContextSelector selector(cu);
int version = 3;
stream.write((char*) &version, sizeof(int));
int precision = (cu.getUseDoublePrecision() ? 2 : cu.getUseMixedPrecision() ? 1 : 0);
......@@ -376,7 +378,7 @@ void CudaUpdateStateDataKernel::createCheckpoint(ContextImpl& context, ostream&
}
void CudaUpdateStateDataKernel::loadCheckpoint(ContextImpl& context, istream& stream) {
cu.setAsCurrent();
ContextSelector selector(cu);
int version;
stream.read((char*) &version, sizeof(int));
if (version != 3)
......@@ -458,7 +460,7 @@ public:
forceTemp.initialize<float4>(cu, cu.getNumAtoms(), "PmeForce");
}
float* getPosq() {
cu.setAsCurrent();
ContextSelector selector(cu);
cu.getPosq().download(posq);
return (float*) &posq[0];
}
......@@ -542,7 +544,7 @@ private:
};
CudaCalcNonbondedForceKernel::~CudaCalcNonbondedForceKernel() {
cu.setAsCurrent();
ContextSelector selector(cu);
if (sort != NULL)
delete sort;
if (fft != NULL)
......@@ -569,7 +571,7 @@ CudaCalcNonbondedForceKernel::~CudaCalcNonbondedForceKernel() {
}
void CudaCalcNonbondedForceKernel::initialize(const System& system, const NonbondedForce& force) {
cu.setAsCurrent();
ContextSelector selector(cu);
int forceIndex;
for (forceIndex = 0; forceIndex < system.getNumForces() && &system.getForce(forceIndex) != &force; ++forceIndex)
;
......@@ -1129,6 +1131,7 @@ void CudaCalcNonbondedForceKernel::initialize(const System& system, const Nonbon
double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy, bool includeDirect, bool includeReciprocal) {
// Update particle and exception parameters.
ContextSelector selector(cu);
bool paramChanged = false;
for (int i = 0; i < paramNames.size(); i++) {
double value = context.getParameter(paramNames[i]);
......@@ -1364,7 +1367,7 @@ double CudaCalcNonbondedForceKernel::execute(ContextImpl& context, bool includeF
void CudaCalcNonbondedForceKernel::copyParametersToContext(ContextImpl& context, const NonbondedForce& force) {
// Make sure the new parameters are acceptable.
cu.setAsCurrent();
ContextSelector selector(cu);
if (force.getNumParticles() != cu.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
if (!hasCoulomb || !hasLJ) {
......
platforms/cuda/src/CudaParallelKernels.cpp
View file @ c456dd54
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2011-2019 Stanford University and the Authors. *
* Portions copyright (c) 2011-2021 Stanford University and the Authors. *
* Authors: Peter Eastman *
* Contributors: *
* *
......@@ -26,6 +26,7 @@
#include "CudaParallelKernels.h"
#include "CudaKernelSources.h"
#include "openmm/common/ContextSelector.h"
using namespace OpenMM;
using namespace std;
......@@ -69,7 +70,7 @@ public:
void execute() {
// Copy coordinates over to this device and execute the kernel.
cu.setAsCurrent();
ContextSelector selector(cu);
if (cu.getContextIndex() > 0) {
cuStreamWaitEvent(cu.getCurrentStream(), event, 0);
if (!cu.getPlatformData().peerAccessSupported)
......@@ -100,6 +101,7 @@ public:
void execute() {
// Execute the kernel, then download forces.
ContextSelector selector(cu);
energy += kernel.finishComputation(context, includeForce, includeEnergy, groups, valid);
if (cu.getComputeForceCount() < 200) {
// Record timing information for load balancing. Since this takes time, only do it at the start of the simulation.
......@@ -148,7 +150,7 @@ CudaParallelCalcForcesAndEnergyKernel::CudaParallelCalcForcesAndEnergyKernel(str
}
CudaParallelCalcForcesAndEnergyKernel::~CudaParallelCalcForcesAndEnergyKernel() {
data.contexts[0]->setAsCurrent();
ContextSelector selector(*data.contexts[0]);
if (pinnedPositionBuffer != NULL)
cuMemFreeHost(pinnedPositionBuffer);
if (pinnedForceBuffer != NULL)
......@@ -161,7 +163,7 @@ CudaParallelCalcForcesAndEnergyKernel::~CudaParallelCalcForcesAndEnergyKernel()
void CudaParallelCalcForcesAndEnergyKernel::initialize(const System& system) {
CudaContext& cu = *data.contexts[0];
cu.setAsCurrent();
ContextSelector selector(cu);
CUmodule module = cu.createModule(CudaKernelSources::parallel);
sumKernel = cu.getKernel(module, "sumForces");
int numContexts = data.contexts.size();
......@@ -176,7 +178,7 @@ void CudaParallelCalcForcesAndEnergyKernel::initialize(const System& system) {
void CudaParallelCalcForcesAndEnergyKernel::beginComputation(ContextImpl& context, bool includeForce, bool includeEnergy, int groups) {
CudaContext& cu = *data.contexts[0];
cu.setAsCurrent();
ContextSelector selector(cu);
if (!contextForces.isInitialized()) {
contextForces.initialize<long long>(cu, 3*(data.contexts.size()-1)*cu.getPaddedNumAtoms(), "contextForces");
CHECK_RESULT(cuMemHostAlloc((void**) &pinnedForceBuffer, 3*(data.contexts.size()-1)*cu.getPaddedNumAtoms()*sizeof(long long), CU_MEMHOSTALLOC_PORTABLE), "Error allocating pinned memory");
......@@ -219,6 +221,7 @@ double CudaParallelCalcForcesAndEnergyKernel::finishComputation(ContextImpl& con
// Sum the forces from all devices.
CudaContext& cu = *data.contexts[0];
ContextSelector selector(cu);
if (!cu.getPlatformData().peerAccessSupported)
contextForces.upload(pinnedForceBuffer, false);
int bufferSize = 3*cu.getPaddedNumAtoms();
......
platforms/cuda/tests/TestCudaFFT3D.cpp
View file @ c456dd54
......@@ -59,6 +59,7 @@ void testTransform(bool realToComplex, int xsize, int ysize, int zsize) {
platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", true, 1, NULL);
CudaContext& context = *platformData.contexts[0];
context.initialize();
context.setAsCurrent();
OpenMM_SFMT::SFMT sfmt;
init_gen_rand(0, sfmt);
vector<Real2> original(xsize*ysize*zsize);
......
platforms/cuda/tests/TestCudaNonbondedForce.cpp
View file @ c456dd54
......@@ -172,7 +172,7 @@ bool canRunHugeTest() {
// Only run the huge test if the device has at least 4 GB of memory.
return (memory >= 4*(1<<30));
return (memory >= 4L*(1<<30));
}
void runPlatformTests() {
......
platforms/cuda/tests/TestCudaRandom.cpp
View file @ c456dd54
......@@ -59,6 +59,7 @@ void testGaussian() {
platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", true, 1, NULL);
CudaContext& context = *platformData.contexts[0];
context.initialize();
context.setAsCurrent();
context.getIntegrationUtilities().initRandomNumberGenerator(0);
CudaArray& random = context.getIntegrationUtilities().getRandom();
context.getIntegrationUtilities().prepareRandomNumbers(random.getSize());
......
platforms/cuda/tests/TestCudaSort.cpp
View file @ c456dd54
......@@ -69,6 +69,7 @@ void verifySorting(vector<float> array) {
platform.getPropertyDefaultValue(CudaPlatform::CudaHostCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaDisablePmeStream()), "false", true, 1, NULL);
CudaContext& context = *platformData.contexts[0];
context.initialize();
context.setAsCurrent();
CudaArray data(context, array.size(), 4, "sortData");
data.upload(array);
CudaSort sort(context, new SortTrait(), array.size());
......
plugins/amoeba/platforms/common/src/AmoebaCommonKernels.cpp
View file @ c456dd54
......@@ -29,6 +29,7 @@
#endif
#include "AmoebaCommonKernels.h"
#include "CommonAmoebaKernelSources.h"
#include "openmm/common/ContextSelector.h"
#include "openmm/internal/ContextImpl.h"
#include "openmm/internal/AmoebaGeneralizedKirkwoodForceImpl.h"
#include "openmm/internal/AmoebaMultipoleForceImpl.h"
......@@ -122,7 +123,7 @@ CommonCalcAmoebaTorsionTorsionForceKernel::CommonCalcAmoebaTorsionTorsionForceKe
}
void CommonCalcAmoebaTorsionTorsionForceKernel::initialize(const System& system, const AmoebaTorsionTorsionForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int numContexts = cc.getNumContexts();
int startIndex = cc.getContextIndex()*force.getNumTorsionTorsions()/numContexts;
int endIndex = (cc.getContextIndex()+1)*force.getNumTorsionTorsions()/numContexts;
......@@ -230,11 +231,10 @@ CommonCalcAmoebaMultipoleForceKernel::CommonCalcAmoebaMultipoleForceKernel(const
}
CommonCalcAmoebaMultipoleForceKernel::~CommonCalcAmoebaMultipoleForceKernel() {
cc.setAsCurrent();
}
void CommonCalcAmoebaMultipoleForceKernel::initialize(const System& system, const AmoebaMultipoleForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (!cc.getSupports64BitGlobalAtomics())
throw OpenMMException("AmoebaMultipoleForce requires a device that supports 64 bit atomic operations");
......@@ -1045,6 +1045,7 @@ void CommonCalcAmoebaMultipoleForceKernel::initializeScaleFactors() {
}
double CommonCalcAmoebaMultipoleForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (!hasInitializedScaleFactors) {
initializeScaleFactors();
for (auto impl : context.getForceImpls()) {
......@@ -1412,6 +1413,7 @@ void CommonCalcAmoebaMultipoleForceKernel::ensureMultipolesValid(ContextImpl& co
}
void CommonCalcAmoebaMultipoleForceKernel::getLabFramePermanentDipoles(ContextImpl& context, vector<Vec3>& dipoles) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
int numParticles = cc.getNumAtoms();
dipoles.resize(numParticles);
......@@ -1432,6 +1434,7 @@ void CommonCalcAmoebaMultipoleForceKernel::getLabFramePermanentDipoles(ContextIm
void CommonCalcAmoebaMultipoleForceKernel::getInducedDipoles(ContextImpl& context, vector<Vec3>& dipoles) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
int numParticles = cc.getNumAtoms();
dipoles.resize(numParticles);
......@@ -1452,6 +1455,7 @@ void CommonCalcAmoebaMultipoleForceKernel::getInducedDipoles(ContextImpl& contex
void CommonCalcAmoebaMultipoleForceKernel::getTotalDipoles(ContextImpl& context, vector<Vec3>& dipoles) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
int numParticles = cc.getNumAtoms();
dipoles.resize(numParticles);
......@@ -1493,6 +1497,7 @@ void CommonCalcAmoebaMultipoleForceKernel::getTotalDipoles(ContextImpl& context,
}
void CommonCalcAmoebaMultipoleForceKernel::getElectrostaticPotential(ContextImpl& context, const vector<Vec3>& inputGrid, vector<double>& outputElectrostaticPotential) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
int numPoints = inputGrid.size();
int elementSize = (cc.getUseDoublePrecision() ? sizeof(double) : sizeof(float));
......@@ -1647,6 +1652,7 @@ void CommonCalcAmoebaMultipoleForceKernel::computeSystemMultipoleMoments(Context
void CommonCalcAmoebaMultipoleForceKernel::getSystemMultipoleMoments(ContextImpl& context, vector<double>& outputMultipoleMoments) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
if (cc.getUseDoublePrecision())
computeSystemMultipoleMoments<double, mm_double4, mm_double4>(context, outputMultipoleMoments);
......@@ -1659,7 +1665,7 @@ void CommonCalcAmoebaMultipoleForceKernel::getSystemMultipoleMoments(ContextImpl
void CommonCalcAmoebaMultipoleForceKernel::copyParametersToContext(ContextImpl& context, const AmoebaMultipoleForce& force) {
// Make sure the new parameters are acceptable.
cc.setAsCurrent();
ContextSelector selector(cc);
if (force.getNumMultipoles() != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of multipoles has changed");
......@@ -1749,7 +1755,7 @@ CommonCalcAmoebaGeneralizedKirkwoodForceKernel::CommonCalcAmoebaGeneralizedKirkw
}
void CommonCalcAmoebaGeneralizedKirkwoodForceKernel::initialize(const System& system, const AmoebaGeneralizedKirkwoodForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (cc.getNumContexts() > 1)
throw OpenMMException("AmoebaGeneralizedKirkwoodForce does not support using multiple devices");
const AmoebaMultipoleForce* multipoles = NULL;
......@@ -1976,7 +1982,7 @@ void CommonCalcAmoebaGeneralizedKirkwoodForceKernel::finishComputation() {
void CommonCalcAmoebaGeneralizedKirkwoodForceKernel::copyParametersToContext(ContextImpl& context, const AmoebaGeneralizedKirkwoodForce& force) {
// Make sure the new parameters are acceptable.
cc.setAsCurrent();
ContextSelector selector(cc);
if (force.getNumParticles() != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -2017,13 +2023,13 @@ CommonCalcAmoebaVdwForceKernel::CommonCalcAmoebaVdwForceKernel(const std::string
}
CommonCalcAmoebaVdwForceKernel::~CommonCalcAmoebaVdwForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (nonbonded != NULL)
delete nonbonded;
}
void CommonCalcAmoebaVdwForceKernel::initialize(const System& system, const AmoebaVdwForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
int paddedNumAtoms = cc.getPaddedNumAtoms();
bondReductionAtoms.initialize<int>(cc, paddedNumAtoms, "bondReductionAtoms");
bondReductionFactors.initialize<float>(cc, paddedNumAtoms, "bondReductionFactors");
......@@ -2131,6 +2137,7 @@ void CommonCalcAmoebaVdwForceKernel::initialize(const System& system, const Amoe
}
double CommonCalcAmoebaVdwForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
if (!hasInitializedNonbonded) {
hasInitializedNonbonded = true;
nonbonded->initialize(system);
......@@ -2160,7 +2167,7 @@ double CommonCalcAmoebaVdwForceKernel::execute(ContextImpl& context, bool includ
void CommonCalcAmoebaVdwForceKernel::copyParametersToContext(ContextImpl& context, const AmoebaVdwForce& force) {
// Make sure the new parameters are acceptable.
cc.setAsCurrent();
ContextSelector selector(cc);
if (force.getNumParticles() != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -2229,6 +2236,7 @@ void CommonCalcAmoebaWcaDispersionForceKernel::initialize(const System& system,
// Record parameters.
ContextSelector selector(cc);
vector<mm_float2> radiusEpsilonVec(paddedNumAtoms, mm_float2(0, 0));
for (int i = 0; i < numParticles; i++) {
double radius, epsilon;
......@@ -2272,6 +2280,7 @@ void CommonCalcAmoebaWcaDispersionForceKernel::initialize(const System& system,
}
double CommonCalcAmoebaWcaDispersionForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
NonbondedUtilities& nb = cc.getNonbondedUtilities();
int startTileIndex = nb.getStartTileIndex();
int numTileIndices = nb.getNumTiles();
......@@ -2285,7 +2294,7 @@ double CommonCalcAmoebaWcaDispersionForceKernel::execute(ContextImpl& context, b
void CommonCalcAmoebaWcaDispersionForceKernel::copyParametersToContext(ContextImpl& context, const AmoebaWcaDispersionForce& force) {
// Make sure the new parameters are acceptable.
cc.setAsCurrent();
ContextSelector selector(cc);
if (force.getNumParticles() != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......@@ -2374,7 +2383,7 @@ CommonCalcHippoNonbondedForceKernel::CommonCalcHippoNonbondedForceKernel(const s
}
void CommonCalcHippoNonbondedForceKernel::initialize(const System& system, const HippoNonbondedForce& force) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (!cc.getSupports64BitGlobalAtomics())
throw OpenMMException("HippoNonbondedForce requires a device that supports 64 bit atomic operations");
extrapolationCoefficients = force.getExtrapolationCoefficients();
......@@ -3170,6 +3179,7 @@ void CommonCalcHippoNonbondedForceKernel::createFieldKernel(const string& intera
}
double CommonCalcHippoNonbondedForceKernel::execute(ContextImpl& context, bool includeForces, bool includeEnergy) {
ContextSelector selector(cc);
NonbondedUtilities& nb = cc.getNonbondedUtilities();
if (!hasInitializedKernels) {
hasInitializedKernels = true;
......@@ -3385,6 +3395,7 @@ void CommonCalcHippoNonbondedForceKernel::addTorquesToForces() {
}
void CommonCalcHippoNonbondedForceKernel::getInducedDipoles(ContextImpl& context, vector<Vec3>& dipoles) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
int numParticles = cc.getNumAtoms();
dipoles.resize(numParticles);
......@@ -3432,6 +3443,7 @@ void CommonCalcHippoNonbondedForceKernel::ensureMultipolesValid(ContextImpl& con
}
void CommonCalcHippoNonbondedForceKernel::getLabFramePermanentDipoles(ContextImpl& context, vector<Vec3>& dipoles) {
ContextSelector selector(cc);
ensureMultipolesValid(context);
int numParticles = cc.getNumAtoms();
dipoles.resize(numParticles);
......@@ -3453,7 +3465,7 @@ void CommonCalcHippoNonbondedForceKernel::getLabFramePermanentDipoles(ContextImp
void CommonCalcHippoNonbondedForceKernel::copyParametersToContext(ContextImpl& context, const HippoNonbondedForce& force) {
// Make sure the new parameters are acceptable.
cc.setAsCurrent();
ContextSelector selector(cc);
if (force.getNumParticles() != cc.getNumAtoms())
throw OpenMMException("updateParametersInContext: The number of particles has changed");
......
plugins/amoeba/platforms/cuda/src/AmoebaCudaKernels.cpp
View file @ c456dd54
......@@ -6,7 +6,7 @@
* Biological Structures at Stanford, funded under the NIH Roadmap for *
* Medical Research, grant U54 GM072970. See https://simtk.org. *
* *
* Portions copyright (c) 2008-2020 Stanford University and the Authors. *
* Portions copyright (c) 2008-2021 Stanford University and the Authors. *
* Authors: Peter Eastman, Mark Friedrichs *
* Contributors: *
* *
......@@ -29,6 +29,7 @@
#endif
#include "AmoebaCudaKernels.h"
#include "CudaAmoebaKernelSources.h"
#include "openmm/common/ContextSelector.h"
#include "openmm/internal/ContextImpl.h"
#include "openmm/internal/AmoebaGeneralizedKirkwoodForceImpl.h"
#include "openmm/internal/AmoebaMultipoleForceImpl.h"
......@@ -83,7 +84,7 @@ static void setPeriodicBoxArgs(ComputeContext& cc, ComputeKernel kernel, int ind
* -------------------------------------------------------------------------- */
CudaCalcAmoebaMultipoleForceKernel::~CudaCalcAmoebaMultipoleForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (hasInitializedFFT)
cufftDestroy(fft);
}
......@@ -91,6 +92,7 @@ CudaCalcAmoebaMultipoleForceKernel::~CudaCalcAmoebaMultipoleForceKernel() {
void CudaCalcAmoebaMultipoleForceKernel::initialize(const System& system, const AmoebaMultipoleForce& force) {
CommonCalcAmoebaMultipoleForceKernel::initialize(system, force);
if (usePME) {
ContextSelector selector(cc);
cufftResult result = cufftPlan3d(&fft, gridSizeX, gridSizeY, gridSizeZ, cc.getUseDoublePrecision() ? CUFFT_Z2Z : CUFFT_C2C);
if (result != CUFFT_SUCCESS)
throw OpenMMException("Error initializing FFT: "+cc.intToString(result));
......@@ -120,7 +122,7 @@ void CudaCalcAmoebaMultipoleForceKernel::computeFFT(bool forward) {
* -------------------------------------------------------------------------- */
CudaCalcHippoNonbondedForceKernel::~CudaCalcHippoNonbondedForceKernel() {
cc.setAsCurrent();
ContextSelector selector(cc);
if (sort != NULL)
delete sort;
if (hasInitializedFFT) {
......@@ -134,6 +136,7 @@ CudaCalcHippoNonbondedForceKernel::~CudaCalcHippoNonbondedForceKernel() {
void CudaCalcHippoNonbondedForceKernel::initialize(const System& system, const HippoNonbondedForce& force) {
CommonCalcHippoNonbondedForceKernel::initialize(system, force);
if (usePME) {
ContextSelector selector(cc);
CudaContext& cu = dynamic_cast<CudaContext&>(cc);
sort = new CudaSort(cu, new SortTrait(), cc.getNumAtoms());
cufftResult result = cufftPlan3d(&fftForward, gridSizeX, gridSizeY, gridSizeZ, cc.getUseDoublePrecision() ? CUFFT_D2Z : CUFFT_R2C);
......
plugins/drude/platforms/common/src/CommonDrudeKernels.cpp
View file @ c456dd54
......@@ -34,6 +34,7 @@
#include "openmm/internal/ContextImpl.h"
#include "openmm/common/BondedUtilities.h"
#include "openmm/common/ComputeForceInfo.h"
#include "openmm/common/ContextSelector.h"
#include "openmm/common/IntegrationUtilities.h"
#include "CommonKernelSources.h"
#include "SimTKOpenMMRealType.h"
......@@ -101,9 +102,9 @@ private:
};
void CommonCalcDrudeForceKernel::initialize(const System& system, const DrudeForce& force) {
cc.setAsCurrent();
if (cc.getContextIndex() != 0)
return; // This is run entirely on one device
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
if (numParticles > 0) {
// Create the harmonic interaction .
......@@ -173,6 +174,7 @@ void CommonCalcDrudeForceKernel::copyParametersToContext(ContextImpl& context, c
// Set the particle parameters.
ContextSelector selector(cc);
int numParticles = force.getNumParticles();
if (numParticles > 0) {
if (!particleParams.isInitialized() || numParticles != particleParams.getSize())
......@@ -222,6 +224,7 @@ void CommonCalcDrudeForceKernel::copyParametersToContext(ContextImpl& context, c
void CommonIntegrateDrudeLangevinStepKernel::initialize(const System& system, const DrudeLangevinIntegrator& integrator, const DrudeForce& force) {
cc.initializeContexts();
ContextSelector selector(cc);
cc.getIntegrationUtilities().initRandomNumberGenerator((unsigned int) integrator.getRandomNumberSeed());
// Identify particle pairs and ordinary particles.
......@@ -263,7 +266,7 @@ void CommonIntegrateDrudeLangevinStepKernel::initialize(const System& system, co
}
void CommonIntegrateDrudeLangevinStepKernel::execute(ContextImpl& context, const DrudeLangevinIntegrator& integrator) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
if (!hasInitializedKernels) {
......@@ -378,7 +381,7 @@ CommonIntegrateDrudeSCFStepKernel::~CommonIntegrateDrudeSCFStepKernel() {
void CommonIntegrateDrudeSCFStepKernel::initialize(const System& system, const DrudeSCFIntegrator& integrator, const DrudeForce& force) {
cc.initializeContexts();
cc.setAsCurrent();
ContextSelector selector(cc);
// Identify Drude particles.
......@@ -406,7 +409,7 @@ void CommonIntegrateDrudeSCFStepKernel::initialize(const System& system, const D
}
void CommonIntegrateDrudeSCFStepKernel::execute(ContextImpl& context, const DrudeSCFIntegrator& integrator) {
cc.setAsCurrent();
ContextSelector selector(cc);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
int numAtoms = cc.getNumAtoms();
double dt = integrator.getStepSize();
......
plugins/rpmd/platforms/common/src/CommonRpmdKernels.cpp
View file @ c456dd54
......@@ -32,6 +32,7 @@
#include "CommonRpmdKernels.h"
#include "CommonRpmdKernelSources.h"
#include "openmm/internal/ContextImpl.h"
#include "openmm/common/ContextSelector.h"
#include "openmm/common/IntegrationUtilities.h"
#include "openmm/common/ExpressionUtilities.h"
#include "openmm/common/NonbondedUtilities.h"
......@@ -63,6 +64,7 @@ static int findFFTDimension(int minimum) {
void CommonIntegrateRPMDStepKernel::initialize(const System& system, const RPMDIntegrator& integrator) {
cc.initializeContexts();
ContextSelector selector(cc);
numCopies = integrator.getNumCopies();
numParticles = system.getNumParticles();
workgroupSize = numCopies;
......@@ -213,7 +215,7 @@ void CommonIntegrateRPMDStepKernel::initializeKernels(ContextImpl& context) {
}
void CommonIntegrateRPMDStepKernel::execute(ContextImpl& context, const RPMDIntegrator& integrator, bool forcesAreValid) {
cc.setAsCurrent();
ContextSelector selector(cc);
if (!hasInitializedKernels)
initializeKernels(context);
IntegrationUtilities& integration = cc.getIntegrationUtilities();
......@@ -364,6 +366,7 @@ void CommonIntegrateRPMDStepKernel::setPositions(int copy, const vector<Vec3>& p
// Record the positions.
ContextSelector selector(cc);
if (cc.getUseDoublePrecision()) {
vector<mm_double4> posq(cc.getPaddedNumAtoms());
cc.getPosq().download(posq);
......@@ -393,6 +396,7 @@ void CommonIntegrateRPMDStepKernel::setVelocities(int copy, const vector<Vec3>&
throw OpenMMException("RPMDIntegrator: Cannot set velocities before the integrator is added to a Context");
if (vel.size() != numParticles)
throw OpenMMException("RPMDIntegrator: wrong number of values passed to setVelocities()");
ContextSelector selector(cc);
if (cc.getUseDoublePrecision() || cc.getUseMixedPrecision()) {
vector<mm_double4> velm(cc.getPaddedNumAtoms());
cc.getVelm().download(velm);
......@@ -410,6 +414,7 @@ void CommonIntegrateRPMDStepKernel::setVelocities(int copy, const vector<Vec3>&
}
void CommonIntegrateRPMDStepKernel::copyToContext(int copy, ContextImpl& context) {
ContextSelector selector(cc);
if (!hasInitializedKernels)
initializeKernels(context);
copyToContextKernel->setArg(2, positions);
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment