Merge branch 'upstream' into fork

platforms/cuda/tests/TestCudaMultipleForces.cpp

@@ -44,7 +44,7 @@
 #include "openmm/RBTorsionForce.h"
 #include "openmm/System.h"
 #include "openmm/VerletIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include "sfmt/SFMT.h"
 #include <iostream>
 #include <vector>

platforms/cuda/tests/TestCudaNonbondedForce.cpp

@@ -45,7 +45,7 @@
 #include "openmm/internal/ContextImpl.h"
 #include "CudaArray.h"
 #include "CudaNonbondedUtilities.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include "sfmt/SFMT.h"
 #include <iostream>
 #include <vector>

platforms/cuda/tests/TestCudaPeriodicTorsionForce.cpp

@@ -39,7 +39,7 @@
 #include "openmm/PeriodicTorsionForce.h"
 #include "openmm/System.h"
 #include "openmm/VerletIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include <iostream>
 #include <vector>
 

platforms/cuda/tests/TestCudaRBTorsionForce.cpp

@@ -39,7 +39,7 @@
 #include "openmm/RBTorsionForce.h"
 #include "openmm/System.h"
 #include "openmm/VerletIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include <iostream>
 #include <vector>
 

platforms/cuda/tests/TestCudaRandom.cpp

@@ -34,14 +34,14 @@
  */
 
 #include "openmm/internal/AssertionUtilities.h"
-#include "../src/CudaArray.h"
-#include "../src/CudaContext.h"
-#include "../src/CudaIntegrationUtilities.h"
+#include "CudaArray.h"
+#include "CudaContext.h"
+#include "CudaIntegrationUtilities.h"
 #include "openmm/System.h"
 #include "openmm/Context.h"
 #include "CudaPlatform.h"
 #include "openmm/VerletIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include <iostream>
 
 using namespace OpenMM;
@@ -54,7 +54,7 @@ void testGaussian() {
     System system;
     for (int i = 0; i < numAtoms; i++)
         system.addParticle(1.0);
-    CudaPlatform::PlatformData platformData(system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"),
+    CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()));
     CudaContext& context = *platformData.contexts[0];
     context.initialize();

platforms/cuda/tests/TestCudaSort.cpp

@@ -34,9 +34,9 @@
  */
 
 #include "openmm/internal/AssertionUtilities.h"
-#include "../src/CudaArray.h"
-#include "../src/CudaContext.h"
-#include "../src/CudaSort.h"
+#include "CudaArray.h"
+#include "CudaContext.h"
+#include "CudaSort.h"
 #include "sfmt/SFMT.h"
 #include "openmm/System.h"
 #include <iostream>
@@ -64,7 +64,7 @@ void verifySorting(vector<float> array) {
 
     System system;
     system.addParticle(0.0);
-    CudaPlatform::PlatformData platformData(system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"),
+    CudaPlatform::PlatformData platformData(NULL, system, "", "true", platform.getPropertyDefaultValue("CudaPrecision"), "false",
             platform.getPropertyDefaultValue(CudaPlatform::CudaCompiler()), platform.getPropertyDefaultValue(CudaPlatform::CudaTempDirectory()));
     CudaContext& context = *platformData.contexts[0];
     context.initialize();

platforms/cuda/tests/TestCudaVariableLangevinIntegrator.cpp

@@ -40,7 +40,7 @@
 #include "openmm/NonbondedForce.h"
 #include "openmm/System.h"
 #include "openmm/VariableLangevinIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include "sfmt/SFMT.h"
 #include <iostream>
 #include <vector>

platforms/cuda/tests/TestCudaVariableVerletIntegrator.cpp

@@ -40,7 +40,7 @@
 #include "openmm/NonbondedForce.h"
 #include "openmm/System.h"
 #include "openmm/VariableVerletIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include "sfmt/SFMT.h"
 #include <iostream>
 #include <vector>

platforms/cuda/tests/TestCudaVerletIntegrator.cpp

@@ -40,7 +40,7 @@
 #include "openmm/NonbondedForce.h"
 #include "openmm/System.h"
 #include "openmm/VerletIntegrator.h"
-#include "../src/SimTKUtilities/SimTKOpenMMRealType.h"
+#include "SimTKOpenMMRealType.h"
 #include "sfmt/SFMT.h"
 #include <iostream>
 #include <vector>

platforms/opencl/CMakeLists.txt

@@ -102,4 +102,9 @@ SET(CL_KERNELS_H ${CMAKE_CURRENT_BINARY_DIR}/src/${CL_SOURCE_CLASS}.h)
 SET(SOURCE_FILES ${SOURCE_FILES} ${CL_KERNELS_CPP} ${CL_KERNELS_H})
 INCLUDE_DIRECTORIES(BEFORE ${CMAKE_CURRENT_BINARY_DIR}/src)
 
+# Install headers
+
+FILE(GLOB CORE_HEADERS include/*.h)
+INSTALL_FILES(/include/openmm/opencl FILES ${CORE_HEADERS})
+
 SUBDIRS (sharedTarget)

platforms/opencl/src/OpenCLContext.h → platforms/opencl/include/OpenCLContext.h

@@ -158,6 +158,8 @@ public:
     class WorkTask;
     class WorkThread;
     class ReorderListener;
+    class ForcePreComputation;
+    class ForcePostComputation;
     static const int ThreadBlockSize;
     static const int TileSize;
     OpenCLContext(const System& system, int platformIndex, int deviceIndex, const std::string& precision, OpenCLPlatform::PlatformData& platformData);
@@ -554,6 +556,28 @@ public:
     std::vector<ReorderListener*>& getReorderListeners() {
         return reorderListeners;
     }
+    /**
+     * Add a pre-computation that should be called at the very start of force and energy evaluations.
+     * The OpenCLContext assumes ownership of the object, and deletes it when the context itself is deleted.
+     */
+    void addPreComputation(ForcePreComputation* computation);
+    /**
+     * Get the list of ForcePreComputations.
+     */
+    std::vector<ForcePreComputation*>& getPreComputations() {
+        return preComputations;
+    }
+    /**
+     * Add a post-computation that should be called at the very end of force and energy evaluations.
+     * The OpenCLContext assumes ownership of the object, and deletes it when the context itself is deleted.
+     */
+    void addPostComputation(ForcePostComputation* computation);
+    /**
+     * Get the list of ForcePostComputations.
+     */
+    std::vector<ForcePostComputation*>& getPostComputations() {
+        return postComputations;
+    }
     /**
      * Mark that the current molecule definitions (and hence the atom order) may be invalid.
      * This should be called whenever force field parameters change.  It will cause the definitions
@@ -625,6 +649,8 @@ private:
     std::vector<cl::Memory*> autoclearBuffers;
     std::vector<int> autoclearBufferSizes;
     std::vector<ReorderListener*> reorderListeners;
+    std::vector<ForcePreComputation*> preComputations;
+    std::vector<ForcePostComputation*> postComputations;
     OpenCLIntegrationUtilities* integration;
     OpenCLExpressionUtilities* expression;
     OpenCLBondedUtilities* bonded;
@@ -686,7 +712,7 @@ private:
 
 /**
  * This abstract class defines a function to be executed whenever atoms get reordered.
- * Objects that need to know when reordering happens should create a reorderListener
+ * Objects that need to know when reordering happens should create a ReorderListener
  * and register it by calling addReorderListener().
  */
 class OpenCLContext::ReorderListener {
@@ -696,6 +722,39 @@ public:
     }
 };
 
+/**
+ * This abstract class defines a function to be executed at the very beginning of force and
+ * energy evaluation, before any other calculation has been done.  It is useful for operations
+ * that need to be performed at a nonstandard point in the process.  After creating a
+ * ForcePreComputation, register it by calling addForcePreComputation().
+ */
+class OpenCLContext::ForcePreComputation {
+public:
+    /**
+     * @param includeForce  true if forces should be computed
+     * @param includeEnergy true if potential energy should be computed
+     * @param groups        a set of bit flags for which force groups to include
+     */
+    virtual void computeForceAndEnergy(bool includeForces, bool includeEnergy, int groups) = 0;
+};
+
+/**
+ * This abstract class defines a function to be executed at the very end of force and
+ * energy evaluation, after all other calculations have been done.  It is useful for operations
+ * that need to be performed at a nonstandard point in the process.  After creating a
+ * ForcePostComputation, register it by calling addForcePostComputation().
+ */
+class OpenCLContext::ForcePostComputation {
+public:
+    /**
+     * @param includeForce  true if forces should be computed
+     * @param includeEnergy true if potential energy should be computed
+     * @param groups        a set of bit flags for which force groups to include
+     * @return an optional contribution to add to the potential energy.
+     */
+    virtual double computeForceAndEnergy(bool includeForces, bool includeEnergy, int groups) = 0;
+};
+
 } // namespace OpenMM
 
 #endif /*OPENMM_OPENCLCONTEXT_H_*/

platforms/opencl/src/OpenCLKernels.h → platforms/opencl/include/OpenCLKernels.h

@@ -557,7 +557,7 @@ public:
     OpenCLCalcNonbondedForceKernel(std::string name, const Platform& platform, OpenCLContext& cl, const System& system) : CalcNonbondedForceKernel(name, platform),
             hasInitializedKernel(false), cl(cl), sigmaEpsilon(NULL), exceptionParams(NULL), cosSinSums(NULL), pmeGrid(NULL),
             pmeGrid2(NULL), pmeBsplineModuliX(NULL), pmeBsplineModuliY(NULL), pmeBsplineModuliZ(NULL), pmeBsplineTheta(NULL),
-            pmeAtomRange(NULL), pmeAtomGridIndex(NULL), sort(NULL), fft(NULL) {
+            pmeAtomRange(NULL), pmeAtomGridIndex(NULL), sort(NULL), fft(NULL), pmeio(NULL) {
     }
     ~OpenCLCalcNonbondedForceKernel();
     /**
@@ -596,6 +596,9 @@ private:
         const char* getMaxValue() const {return "(int2) (INT_MAX, INT_MAX)";}
         const char* getSortKey() const {return "value.y";}
     };
+    class PmeIO;
+    class PmePreComputation;
+    class PmePostComputation;
     OpenCLContext& cl;
     bool hasInitializedKernel;
     OpenCLArray* sigmaEpsilon;
@@ -611,6 +614,8 @@ private:
     OpenCLArray* pmeAtomGridIndex;
     OpenCLSort* sort;
     OpenCLFFT3D* fft;
+    Kernel cpuPme;
+    PmeIO* pmeio;
     cl::Kernel ewaldSumsKernel;
     cl::Kernel ewaldForcesKernel;
     cl::Kernel pmeGridIndexKernel;