Merged parallel code (#4649)

* Unified lots of parallel computation code between platforms

* Unified test code between platforms

* Eliminated duplicated timing code

platforms/cuda/tests/TestCudaPeriodicTorsionForce.cpp

@@ -32,49 +32,6 @@
 #include "CudaTests.h"
 #include "TestPeriodicTorsionForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    PeriodicTorsionForce* force = new PeriodicTorsionForce();
-    for (int i = 3; i < numParticles; i++)
-        force->addTorsion(i-3, i-2, i-1, i, 2, 1.1, i);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, i%3);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, CudaPlatform::CudaDeviceIndex());
-    map<string, string> props;
-    props[CudaPlatform::CudaDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-
-    // Try updating some parameters and see if they still match.
-
-    for (int i = 95; i < 102; i++) {
-        int p1, p2, p3, p4, periodicity;
-        double phase, k;
-        force->getTorsionParameters(i, p1, p2, p3, p4, periodicity, phase, k);
-        force->setTorsionParameters(i, p1, p2, p3, p4, periodicity, phase+0.1, 2*k);
-    }
-    force->updateParametersInContext(context1);
-    force->updateParametersInContext(context2);
-    State state3 = context1.getState(State::Energy);
-    State state4 = context2.getState(State::Energy);
-    ASSERT_EQUAL_TOL(state3.getPotentialEnergy(), state4.getPotentialEnergy(), 1e-5);
-    ASSERT(fabs(state1.getPotentialEnergy()-state3.getPotentialEnergy()) > 0.1);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/cuda/tests/TestCudaRBTorsionForce.cpp

@@ -32,34 +32,6 @@
 #include "CudaTests.h"
 #include "TestRBTorsionForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    RBTorsionForce* force = new RBTorsionForce();
-    for (int i = 3; i < numParticles; i++)
-        force->addTorsion(i-3, i-2, i-1, i, 2, 0.1*i, 0.5*i, i, 1, 1);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, i%3);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, CudaPlatform::CudaDeviceIndex());
-    map<string, string> props;
-    props[CudaPlatform::CudaDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/include/HipContext.h

@@ -157,6 +157,11 @@ public:
      * one ComputeContext is created for each device.
      */
     std::vector<ComputeContext*> getAllContexts();
+    /**
+     * Get a workspace used for accumulating energy when a simulation is parallelized across
+     * multiple devices.
+     */
+    double& getEnergyWorkspace();
     /**
      * Get the stream currently being used for execution.
      */

platforms/hip/src/HipContext.cpp

@@ -672,6 +672,10 @@ vector<ComputeContext*> HipContext::getAllContexts() {
     return result;
 }
 
+double& HipContext::getEnergyWorkspace() {
+    return platformData.contextEnergy[contextIndex];
+}
+
 hipStream_t HipContext::getCurrentStream() {
     return currentStream;
 }

platforms/hip/src/HipKernelFactory.cpp

@@ -30,6 +30,7 @@
 #include "HipParallelKernels.h"
 #include "HipPlatform.h"
 #include "openmm/common/CommonKernels.h"
+#include "openmm/common/CommonParallelKernels.h"
 #include "openmm/internal/ContextImpl.h"
 #include "openmm/OpenMMException.h"
 
@@ -37,39 +38,39 @@ using namespace OpenMM;
 
 KernelImpl* HipKernelFactory::createKernelImpl(std::string name, const Platform& platform, ContextImpl& context) const {
     HipPlatform::PlatformData& data = *static_cast<HipPlatform::PlatformData*>(context.getPlatformData());
+    HipContext& cu = *data.contexts[0];
     if (data.contexts.size() > 1) {
         // We are running in parallel on multiple devices, so we may want to create a parallel kernel.
 
         if (name == CalcForcesAndEnergyKernel::Name())
             return new HipParallelCalcForcesAndEnergyKernel(name, platform, data);
         if (name == CalcHarmonicBondForceKernel::Name())
-            return new HipParallelCalcHarmonicBondForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcHarmonicBondForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCustomBondForceKernel::Name())
-            return new HipParallelCalcCustomBondForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomBondForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcHarmonicAngleForceKernel::Name())
-            return new HipParallelCalcHarmonicAngleForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcHarmonicAngleForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCustomAngleForceKernel::Name())
-            return new HipParallelCalcCustomAngleForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomAngleForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcPeriodicTorsionForceKernel::Name())
-            return new HipParallelCalcPeriodicTorsionForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcPeriodicTorsionForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcRBTorsionForceKernel::Name())
-            return new HipParallelCalcRBTorsionForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcRBTorsionForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCMAPTorsionForceKernel::Name())
-            return new HipParallelCalcCMAPTorsionForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCMAPTorsionForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCustomTorsionForceKernel::Name())
-            return new HipParallelCalcCustomTorsionForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomTorsionForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcNonbondedForceKernel::Name())
             return new HipParallelCalcNonbondedForceKernel(name, platform, data, context.getSystem());
         if (name == CalcCustomNonbondedForceKernel::Name())
-            return new HipParallelCalcCustomNonbondedForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomNonbondedForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCustomExternalForceKernel::Name())
-            return new HipParallelCalcCustomExternalForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomExternalForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCustomHbondForceKernel::Name())
-            return new HipParallelCalcCustomHbondForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomHbondForceKernel(name, platform, cu, context.getSystem());
         if (name == CalcCustomCompoundBondForceKernel::Name())
-            return new HipParallelCalcCustomCompoundBondForceKernel(name, platform, data, context.getSystem());
+            return new CommonParallelCalcCustomCompoundBondForceKernel(name, platform, cu, context.getSystem());
     }
-    HipContext& cu = *data.contexts[0];
     if (name == CalcForcesAndEnergyKernel::Name())
         return new HipCalcForcesAndEnergyKernel(name, platform, cu);
     if (name == UpdateStateDataKernel::Name())

platforms/hip/tests/TestHipCustomAngleForce.cpp

@@ -33,35 +33,6 @@
 #include "HipTests.h"
 #include "TestCustomAngleForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    CustomAngleForce* force = new CustomAngleForce("(theta-1.1)^2");
-    vector<double> params;
-    for (int i = 2; i < numParticles; i++)
-        force->addAngle(i-2, i-1, i, params);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, 0);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipCustomBondForce.cpp

@@ -33,35 +33,6 @@
 #include "HipTests.h"
 #include "TestCustomBondForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    CustomBondForce* force = new CustomBondForce(("(r-1.1)^2"));
-    vector<double> params;
-    for (int i = 1; i < numParticles; i++)
-        force->addBond(i-1, i, params);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, 0, 0);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipCustomCompoundBondForce.cpp

@@ -33,39 +33,6 @@
 #include "HipTests.h"
 #include "TestCustomCompoundBondForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    CustomCompoundBondForce* force = new CustomCompoundBondForce(2, ("(distance(p1,p2)-1.1)^2"));
-    vector<int> particles(2);
-    vector<double> params;
-    for (int i = 1; i < numParticles; i++) {
-        particles[0] = i-1;
-        particles[1] = i;
-        force->addBond(particles, params);
-    }
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, 0, 0);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipCustomExternalForce.cpp

@@ -34,37 +34,6 @@
 #include "TestCustomExternalForce.h"
 #include "sfmt/SFMT.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    CustomExternalForce* force = new CustomExternalForce("x^2+y^2+z^2");
-    vector<double> params;
-    for (int i = 0; i < numParticles; i++)
-        force->addParticle(i, params);
-    system.addForce(force);
-    OpenMM_SFMT::SFMT sfmt;
-    init_gen_rand(0, sfmt);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(5*genrand_real2(sfmt), 5*genrand_real2(sfmt), 5*genrand_real2(sfmt));
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipCustomNonbondedForce.cpp

@@ -33,43 +33,6 @@
 #include "HipTests.h"
 #include "TestCustomNonbondedForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    CustomNonbondedForce* force = new CustomNonbondedForce("4*eps*((sigma/r)^12-(sigma/r)^6); sigma=0.5; eps=1");
-    vector<double> params;
-    for (int i = 0; i < numParticles; i++)
-        force->addParticle(params);
-    system.addForce(force);
-    OpenMM_SFMT::SFMT sfmt;
-    init_gen_rand(0, sfmt);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(5*genrand_real2(sfmt), 5*genrand_real2(sfmt), 5*genrand_real2(sfmt));
-    for (int i = 0; i < numParticles; ++i)
-        for (int j = 0; j < i; ++j) {
-            Vec3 delta = positions[i]-positions[j];
-            if (delta.dot(delta) < 0.1)
-                force->addExclusion(i, j);
-        }
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipCustomTorsionForce.cpp

@@ -33,35 +33,6 @@
 #include "HipTests.h"
 #include "TestCustomTorsionForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    CustomTorsionForce* force = new CustomTorsionForce("sin(theta-1.1)");
-    vector<double> params;
-    for (int i = 3; i < numParticles; i++)
-        force->addTorsion(i-3, i-2, i-1, i, params);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, i%3);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipHarmonicAngleForce.cpp

@@ -33,35 +33,6 @@
 
 #include "HipTests.h"
 #include "TestHarmonicAngleForce.h"
-#include <map>
-
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    HarmonicAngleForce* force = new HarmonicAngleForce();
-    for (int i = 2; i < numParticles; i++)
-        force->addAngle(i-2, i-1, i, 1.1, i);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, 0);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
 
 void runPlatformTests() {
     testParallelComputation();

platforms/hip/tests/TestHipHarmonicBondForce.cpp

@@ -32,35 +32,6 @@
 
 #include "HipTests.h"
 #include "TestHarmonicBondForce.h"
-#include <map>
-
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    HarmonicBondForce* force = new HarmonicBondForce();
-    for (int i = 1; i < numParticles; i++)
-        force->addBond(i-1, i, 1.1, i);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, 0, 0);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
 
 void runPlatformTests() {
     testParallelComputation();

platforms/hip/tests/TestHipNonbondedForce.cpp

@@ -34,98 +34,6 @@
 #include "TestNonbondedForce.h"
 #include <hip/hip_runtime.h>
 
-void testParallelComputation(NonbondedForce::NonbondedMethod method) {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    NonbondedForce* force = new NonbondedForce();
-    for (int i = 0; i < numParticles; i++)
-        force->addParticle(i%2-0.5, 0.5, 1.0);
-    force->setNonbondedMethod(method);
-    system.addForce(force);
-    system.setDefaultPeriodicBoxVectors(Vec3(5,0,0), Vec3(0,5,0), Vec3(0,0,5));
-    OpenMM_SFMT::SFMT sfmt;
-    init_gen_rand(0, sfmt);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(5*genrand_real2(sfmt), 5*genrand_real2(sfmt), 5*genrand_real2(sfmt));
-    force->addGlobalParameter("scale", 0.5);
-    for (int i = 0; i < numParticles; ++i)
-        for (int j = 0; j < i; ++j) {
-            Vec3 delta = positions[i]-positions[j];
-            if (delta.dot(delta) < 0.1) {
-                force->addException(i, j, 0, 1, 0);
-            }
-            else if (delta.dot(delta) < 0.2) {
-                int index = force->addException(i, j, 0.5, 1, 1.0);
-                force->addExceptionParameterOffset("scale", index, 0.5, 0.4, 0.3);
-            }
-        }
-    
-    // Create two contexts, one with a single device and one with two devices.
-
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-
-    // See if they agree.
-
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-
-    // Modify some particle parameters and see if they still agree.
-
-    for (int i = 0; i < numParticles; i += 5) {
-        double charge, sigma, epsilon;
-        force->getParticleParameters(i, charge, sigma, epsilon);
-        force->setParticleParameters(i, 0.9*charge, sigma, epsilon);
-    }
-    force->updateParametersInContext(context1);
-    force->updateParametersInContext(context2);
-    state1 = context1.getState(State::Forces | State::Energy);
-    state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
-void testReordering() {
-    // Check that reordering of atoms doesn't alter their positions.
-
-    const int numParticles = 200;
-    System system;
-    system.setDefaultPeriodicBoxVectors(Vec3(6, 0, 0), Vec3(2.1, 6, 0), Vec3(-1.5, -0.5, 6));
-    NonbondedForce *nonbonded = new NonbondedForce();
-    nonbonded->setNonbondedMethod(NonbondedForce::PME);
-    system.addForce(nonbonded);
-    vector<Vec3> positions;
-    OpenMM_SFMT::SFMT sfmt;
-    init_gen_rand(0, sfmt);
-    for (int i = 0; i < numParticles; i++) {
-        system.addParticle(1.0);
-        nonbonded->addParticle(0.0, 0.0, 0.0);
-        positions.push_back(Vec3(genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5, genrand_real2(sfmt)-0.5)*20);
-    }
-    VerletIntegrator integrator(0.001);
-    Context context(system, integrator, platform);
-    context.setPositions(positions);
-    integrator.step(1);
-    State state = context.getState(State::Positions | State::Velocities);
-    for (int i = 0; i < numParticles; i++) {
-        ASSERT_EQUAL_VEC(positions[i], state.getPositions()[i], 1e-6);
-    }
-}
-
 void testDeterministicForces() {
     // Check that the HipDeterministicForces property works correctly.
 

platforms/hip/tests/TestHipPeriodicTorsionForce.cpp

@@ -33,34 +33,6 @@
 #include "HipTests.h"
 #include "TestPeriodicTorsionForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    PeriodicTorsionForce* force = new PeriodicTorsionForce();
-    for (int i = 3; i < numParticles; i++)
-        force->addTorsion(i-3, i-2, i-1, i, 2, 1.1, i);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, i%3);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/hip/tests/TestHipRBTorsionForce.cpp

@@ -33,34 +33,6 @@
 #include "HipTests.h"
 #include "TestRBTorsionForce.h"
 
-void testParallelComputation() {
-    System system;
-    const int numParticles = 200;
-    for (int i = 0; i < numParticles; i++)
-        system.addParticle(1.0);
-    RBTorsionForce* force = new RBTorsionForce();
-    for (int i = 3; i < numParticles; i++)
-        force->addTorsion(i-3, i-2, i-1, i, 2, 0.1*i, 0.5*i, i, 1, 1);
-    system.addForce(force);
-    vector<Vec3> positions(numParticles);
-    for (int i = 0; i < numParticles; i++)
-        positions[i] = Vec3(i, i%2, i%3);
-    VerletIntegrator integrator1(0.01);
-    Context context1(system, integrator1, platform);
-    context1.setPositions(positions);
-    State state1 = context1.getState(State::Forces | State::Energy);
-    VerletIntegrator integrator2(0.01);
-    string deviceIndex = platform.getPropertyValue(context1, HipPlatform::HipDeviceIndex());
-    map<string, string> props;
-    props[HipPlatform::HipDeviceIndex()] = deviceIndex+","+deviceIndex;
-    Context context2(system, integrator2, platform, props);
-    context2.setPositions(positions);
-    State state2 = context2.getState(State::Forces | State::Energy);
-    ASSERT_EQUAL_TOL(state1.getPotentialEnergy(), state2.getPotentialEnergy(), 1e-5);
-    for (int i = 0; i < numParticles; i++)
-        ASSERT_EQUAL_VEC(state1.getForces()[i], state2.getForces()[i], 1e-5);
-}
-
 void runPlatformTests() {
     testParallelComputation();
 }

platforms/opencl/include/OpenCLContext.h

@@ -197,6 +197,11 @@ public:
      * one ComputeContext is created for each device.
      */
     std::vector<ComputeContext*> getAllContexts();
+    /**
+     * Get a workspace used for accumulating energy when a simulation is parallelized across
+     * multiple devices.
+     */
+    double& getEnergyWorkspace();
     /**
      * Get the cl::CommandQueue currently being used for execution.
      */