Merge branch 'master' of github.com:SimTk/openmm

CMakeLists.txt

@@ -111,6 +111,11 @@ IF (APPLE AND NOT CMAKE_OSX_DEPLOYMENT_TARGET)
     SET (CMAKE_OSX_DEPLOYMENT_TARGET "10.5" CACHE STRING "The minimum version of OS X to support" FORCE)
 ENDIF (APPLE AND NOT CMAKE_OSX_DEPLOYMENT_TARGET)
 
+# Improve the linking behavior of Mac libraries
+IF (APPLE)
+    SET (CMAKE_INSTALL_NAME_DIR "@rpath")
+ENDIF (APPLE)
+
 IF(UNIX AND NOT CMAKE_BUILD_TYPE)
     SET(CMAKE_BUILD_TYPE Release CACHE STRING "Debug or Release build" FORCE)
 ENDIF (UNIX AND NOT CMAKE_BUILD_TYPE)
@@ -247,6 +252,7 @@ FOREACH(subdir ${OPENMM_SOURCE_SUBDIRS})
     ## OpenMM was previously installed there.
     INCLUDE_DIRECTORIES(BEFORE ${CMAKE_CURRENT_SOURCE_DIR}/${subdir}/include)
 ENDFOREACH(subdir)
+SET_SOURCE_FILES_PROPERTIES(${CMAKE_SOURCE_DIR}/libraries/sfmt/src/SFMT.cpp PROPERTIES COMPILE_FLAGS "-msse2 -DHAVE_SSE2=1")
 
 # If API wrappers are being generated, and add them to the build.
 FIND_PROGRAM(GCCXML_PATH gccxml PATH
@@ -369,6 +375,13 @@ IF(OPENMM_BUILD_OPENCL_LIB)
     ADD_SUBDIRECTORY(platforms/opencl)
 ENDIF(OPENMM_BUILD_OPENCL_LIB)
 
+# Optimized CPU platform
+
+SET(OPENMM_BUILD_CPU_LIB ON CACHE BOOL "Build optimized CPU platform")
+IF(OPENMM_BUILD_CPU_LIB)
+   ADD_SUBDIRECTORY(platforms/cpu)
+ENDIF(OPENMM_BUILD_CPU_LIB)
+
 # Amoeba plugin
 
 SET(OPENMM_BUILD_AMOEBA_PLUGIN ON CACHE BOOL "Build Amoeba plugin")

libraries/lepton/include/Lepton.h

@@ -32,6 +32,7 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
 
+#include "lepton/CompiledExpression.h"
 #include "lepton/CustomFunction.h"
 #include "lepton/ExpressionProgram.h"
 #include "lepton/ExpressionTreeNode.h"

libraries/lepton/include/lepton/CompiledExpression.h

+#ifndef LEPTON_COMPILED_EXPRESSION_H_
+#define LEPTON_COMPILED_EXPRESSION_H_
+
+/* -------------------------------------------------------------------------- *
+ *                                   Lepton                                   *
+ * -------------------------------------------------------------------------- *
+ * This is part of the Lepton expression parser 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) 2013 Stanford University and the Authors.           *
+ * Authors: Peter Eastman                                                     *
+ * Contributors:                                                              *
+ *                                                                            *
+ * Permission is hereby granted, free of charge, to any person obtaining a    *
+ * copy of this software and associated documentation files (the "Software"), *
+ * to deal in the Software without restriction, including without limitation  *
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,   *
+ * and/or sell copies of the Software, and to permit persons to whom the      *
+ * Software is furnished to do so, subject to the following conditions:       *
+ *                                                                            *
+ * The above copyright notice and this permission notice shall be included in *
+ * all copies or substantial portions of the Software.                        *
+ *                                                                            *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,   *
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL    *
+ * THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,    *
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR      *
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE  *
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
+ * -------------------------------------------------------------------------- */
+
+#include "ExpressionTreeNode.h"
+#include "windowsIncludes.h"
+#include <map>
+#include <set>
+#include <string>
+#include <vector>
+
+namespace Lepton {
+
+class Operation;
+class ParsedExpression;
+
+/**
+ * A CompiledExpression is a highly optimized representation of an expression for cases when you want to evaluate
+ * it many times as quickly as possible.  You should treat it as an opaque object; none of the internal representation
+ * is visible.
+ * 
+ * A CompiledExpression is created by calling createCompiledExpression() on a ParsedExpression.
+ * 
+ * WARNING: CompiledExpression is NOT thread safe.  You should never access a CompiledExpression from two threads at
+ * the same time.
+ */
+
+class LEPTON_EXPORT CompiledExpression {
+public:
+    CompiledExpression();
+    CompiledExpression(const CompiledExpression& expression);
+    ~CompiledExpression();
+    CompiledExpression& operator=(const CompiledExpression& expression);
+    /**
+     * Get the names of all variables used by this expression.
+     */
+    const std::set<std::string>& getVariables() const;
+    /**
+     * Get a reference to the memory location where the value of a particular variable is stored.  This can be used
+     * to set the value of the variable before calling evaluate().
+     */
+    double& getVariableReference(const std::string& name);
+    /**
+     * Evaluate the expression.  The values of all variables should have been set before calling this.
+     */
+    double evaluate() const;
+private:
+    friend class ParsedExpression;
+    CompiledExpression(const ParsedExpression& expression);
+    void compileExpression(const ExpressionTreeNode& node, std::vector<std::pair<ExpressionTreeNode, int> >& temps);
+    int findTempIndex(const ExpressionTreeNode& node, std::vector<std::pair<ExpressionTreeNode, int> >& temps);
+    std::vector<std::vector<int> > arguments;
+    std::vector<int> target;
+    std::vector<Operation*> operation;
+    std::map<std::string, int> variableIndices;
+    std::set<std::string> variableNames;
+    mutable std::vector<double> workspace;
+    mutable std::vector<double> argValues;
+    std::map<std::string, double> dummyVariables;
+};
+
+} // namespace Lepton
+
+#endif /*LEPTON_COMPILED_EXPRESSION_H_*/

libraries/lepton/include/lepton/ExpressionProgram.h

@@ -48,9 +48,7 @@ class ParsedExpression;
  * evaluated and the result is pushed back onto the stack.  At the end, the stack contains a single value,
  * which is the value of the expression.
  *
- * An ExpressionProgram is created by calling createProgram() on a ParsedExpression.  It can generally be evaluated
- * more quickly than the ParsedExpression itself, so when you need to evaluate an expression many times, it is
- * most efficient to create an ExpressionProgram from it.
+ * An ExpressionProgram is created by calling createProgram() on a ParsedExpression.
  */
 
 class LEPTON_EXPORT ExpressionProgram {

libraries/lepton/include/lepton/Operation.h

@@ -9,7 +9,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2009 Stanford University and the Authors.           *
+ * Portions copyright (c) 2009-2013 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -965,6 +965,8 @@ private:
 class LEPTON_EXPORT Operation::PowerConstant : public Operation {
 public:
     PowerConstant(double value) : value(value) {
+        intValue = (int) value;
+        isIntPower = (intValue == value);
     }
     std::string getName() const {
         std::stringstream name;
@@ -981,6 +983,25 @@ public:
         return new PowerConstant(value);
     }
     double evaluate(double* args, const std::map<std::string, double>& variables) const {
+        if (isIntPower) {
+            // Integer powers can be computed much more quickly by repeated multiplication.
+            
+            int exponent = intValue;
+            double base = args[0];
+            if (exponent < 0) {
+                exponent = -exponent;
+                base = 1.0/base;
+            }
+            double result = 1.0;
+            while (exponent != 0) {
+                if ((exponent&1) == 1)
+                    result *= base;
+                base *= base;
+                exponent = exponent>>1;
+           }
+           return result;
+        }
+        else
             return std::pow(args[0], value);
     }
     ExpressionTreeNode differentiate(const std::vector<ExpressionTreeNode>& children, const std::vector<ExpressionTreeNode>& childDerivs, const std::string& variable) const;
@@ -996,6 +1017,8 @@ public:
     }
 private:
     double value;
+    int intValue;
+    bool isIntPower;
 };
 
 class LEPTON_EXPORT Operation::Min : public Operation {

libraries/lepton/include/lepton/ParsedExpression.h

@@ -9,7 +9,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2009 Stanford University and the Authors.           *
+ * Portions copyright (c) 2009=2013 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -39,6 +39,7 @@
 
 namespace Lepton {
 
+class CompiledExpression;
 class ExpressionProgram;
 
 /**
@@ -97,6 +98,10 @@ public:
      * Create an ExpressionProgram that represents the same calculation as this expression.
      */
     ExpressionProgram createProgram() const;
+    /**
+     * Create a CompiledExpression that represents the same calculation as this expression.
+     */
+    CompiledExpression createCompiledExpression() const;
     /**
      * Create a new ParsedExpression which is identical to this one, except that the names of some
      * variables have been changed.

libraries/lepton/src/CompiledExpression.cpp

+/* -------------------------------------------------------------------------- *
+ *                                   Lepton                                   *
+ * -------------------------------------------------------------------------- *
+ * This is part of the Lepton expression parser 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) 2013 Stanford University and the Authors.           *
+ * Authors: Peter Eastman                                                     *
+ * Contributors:                                                              *
+ *                                                                            *
+ * Permission is hereby granted, free of charge, to any person obtaining a    *
+ * copy of this software and associated documentation files (the "Software"), *
+ * to deal in the Software without restriction, including without limitation  *
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,   *
+ * and/or sell copies of the Software, and to permit persons to whom the      *
+ * Software is furnished to do so, subject to the following conditions:       *
+ *                                                                            *
+ * The above copyright notice and this permission notice shall be included in *
+ * all copies or substantial portions of the Software.                        *
+ *                                                                            *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,   *
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL    *
+ * THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,    *
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR      *
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE  *
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
+ * -------------------------------------------------------------------------- */
+
+#include "lepton/CompiledExpression.h"
+#include "lepton/Operation.h"
+#include "lepton/ParsedExpression.h"
+#include <utility>
+
+using namespace Lepton;
+using namespace std;
+
+CompiledExpression::CompiledExpression() {
+}
+
+CompiledExpression::CompiledExpression(const ParsedExpression& expression) {
+    ParsedExpression expr = expression.optimize(); // Just in case it wasn't already optimized.
+    vector<pair<ExpressionTreeNode, int> > temps;
+    compileExpression(expr.getRootNode(), temps);
+}
+
+CompiledExpression::~CompiledExpression() {
+    for (int i = 0; i < (int) operation.size(); i++)
+        if (operation[i] != NULL)
+            delete operation[i];
+}
+
+CompiledExpression::CompiledExpression(const CompiledExpression& expression) {
+    *this = expression;
+}
+
+CompiledExpression& CompiledExpression::operator=(const CompiledExpression& expression) {
+    arguments = expression.arguments;
+    target = expression.target;
+    variableIndices = expression.variableIndices;
+    variableNames = expression.variableNames;
+    workspace.resize(expression.workspace.size());
+    argValues.resize(expression.argValues.size());
+    operation.resize(expression.operation.size());
+    for (int i = 0; i < (int) operation.size(); i++)
+        operation[i] = expression.operation[i]->clone();
+    return *this;
+}
+
+void CompiledExpression::compileExpression(const ExpressionTreeNode& node, vector<pair<ExpressionTreeNode, int> >& temps) {
+    if (findTempIndex(node, temps) != -1)
+        return; // We have already processed a node identical to this one.
+    
+    // Process the child nodes.
+    
+    vector<int> args;
+    for (int i = 0; i < node.getChildren().size(); i++) {
+        compileExpression(node.getChildren()[i], temps);
+        args.push_back(findTempIndex(node.getChildren()[i], temps));
+    }
+    
+    // Process this node.
+    
+    if (node.getOperation().getId() == Operation::VARIABLE) {
+        variableIndices[node.getOperation().getName()] = workspace.size();
+        variableNames.insert(node.getOperation().getName());
+    }
+    else {
+        int stepIndex = arguments.size();
+        arguments.push_back(vector<int>());
+        target.push_back(workspace.size());
+        operation.push_back(node.getOperation().clone());
+        if (args.size() == 0)
+            arguments[stepIndex].push_back(0); // The value won't actually be used.  We just need something there.
+        else {
+            // If the arguments are sequential, we can just pass a pointer to the first one.
+            
+            bool sequential = true;
+            for (int i = 1; i < args.size(); i++)
+                if (args[i] != args[i-1]+1)
+                    sequential = false;
+            if (sequential)
+                arguments[stepIndex].push_back(args[0]);
+            else {
+                arguments[stepIndex] = args;
+                if (args.size() > argValues.size())
+                    argValues.resize(args.size(), 0.0);
+            }
+        }
+    }
+    temps.push_back(make_pair(node, workspace.size()));
+    workspace.push_back(0.0);
+}
+
+int CompiledExpression::findTempIndex(const ExpressionTreeNode& node, vector<pair<ExpressionTreeNode, int> >& temps) {
+    for (int i = 0; i < (int) temps.size(); i++)
+        if (temps[i].first == node)
+            return i;
+    return -1;
+}
+
+const set<string>& CompiledExpression::getVariables() const {
+    return variableNames;
+}
+
+double& CompiledExpression::getVariableReference(const string& name) {
+    map<string, int>::iterator index = variableIndices.find(name);
+    if (index == variableIndices.end())
+        throw Exception("getVariableReference: Unknown variable '"+name+"'");
+    return workspace[index->second];
+}
+
+double CompiledExpression::evaluate() const {
+    // Loop over the operations and evaluate each one.
+    
+    for (int step = 0; step < operation.size(); step++) {
+        const vector<int>& args = arguments[step];
+        if (args.size() == 1)
+            workspace[target[step]] = operation[step]->evaluate(&workspace[args[0]], dummyVariables);
+        else {
+            for (int i = 0; i < args.size(); i++)
+                argValues[i] = workspace[args[i]];
+            workspace[target[step]] = operation[step]->evaluate(&argValues[0], dummyVariables);
+        }
+    }
+    return workspace[workspace.size()-1];
+}

libraries/lepton/src/ExpressionProgram.cpp

@@ -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 Stanford University and the Authors.           *
+ * Portions copyright (c) 2009-2013 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -93,14 +93,13 @@ double ExpressionProgram::evaluate() const {
 }
 
 double ExpressionProgram::evaluate(const std::map<std::string, double>& variables) const {
-    vector<double> args(max(maxArgs, 1));
-    vector<double> stack(stackSize);
-    int stackPointer = 0;
+    vector<double> stack(stackSize+1);
+    int stackPointer = stackSize;
     for (int i = 0; i < (int) operations.size(); i++) {
         int numArgs = operations[i]->getNumArguments();
-        for (int j = 0; j < numArgs; j++)
-            args[j] = stack[--stackPointer];
-        stack[stackPointer++] = operations[i]->evaluate(&args[0], variables);
+        double result = operations[i]->evaluate(&stack[stackPointer], variables);
+        stackPointer += numArgs-1;
+        stack[stackPointer] = result;
     }
-    return stack[0];
+    return stack[stackSize-1];
 }

libraries/lepton/src/ParsedExpression.cpp

@@ -30,6 +30,7 @@
  * -------------------------------------------------------------------------- */
 
 #include "lepton/ParsedExpression.h"
+#include "lepton/CompiledExpression.h"
 #include "lepton/ExpressionProgram.h"
 #include "lepton/Operation.h"
 #include <limits>
@@ -294,6 +295,10 @@ ExpressionProgram ParsedExpression::createProgram() const {
     return ExpressionProgram(*this);
 }
 
+CompiledExpression ParsedExpression::createCompiledExpression() const {
+    return CompiledExpression(*this);
+}
+
 ParsedExpression ParsedExpression::renameVariables(const map<string, string>& replacements) const {
     return ParsedExpression(renameNodeVariables(getRootNode(), replacements));
 }

libraries/sfmt/include/sfmt/SFMT-sse2.h

@@ -49,22 +49,23 @@ PRE_ALWAYS static __m128i mm_recursion(__m128i *a, __m128i *b,
  * This function fills the internal state array with pseudorandom
  * integers.
  */
-inline static void gen_rand_all(void) {
+inline static void gen_rand_all(SFMT& sfmt) {
     int i;
     __m128i r, r1, r2, mask;
     mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
 
-    r1 = _mm_load_si128(&sfmt[N - 2].si);
-    r2 = _mm_load_si128(&sfmt[N - 1].si);
+    SFMTData& data = *sfmt.data;
+    r1 = _mm_load_si128(&data.sfmt[N - 2].si);
+    r2 = _mm_load_si128(&data.sfmt[N - 1].si);
     for (i = 0; i < N - POS1; i++) {
-	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
-	_mm_store_si128(&sfmt[i].si, r);
+	r = mm_recursion(&data.sfmt[i].si, &data.sfmt[i + POS1].si, r1, r2, mask);
+	_mm_store_si128(&data.sfmt[i].si, r);
 	r1 = r2;
 	r2 = r;
     }
     for (; i < N; i++) {
-	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1 - N].si, r1, r2, mask);
-	_mm_store_si128(&sfmt[i].si, r);
+	r = mm_recursion(&data.sfmt[i].si, &data.sfmt[i + POS1 - N].si, r1, r2, mask);
+	_mm_store_si128(&data.sfmt[i].si, r);
 	r1 = r2;
 	r2 = r;
     }
@@ -77,21 +78,22 @@ inline static void gen_rand_all(void) {
  * @param array an 128-bit array to be filled by pseudorandom numbers.  
  * @param size number of 128-bit pesudorandom numbers to be generated.
  */
-inline static void gen_rand_array(w128_t *array, int size) {
+inline static void gen_rand_array(w128_t *array, int size, SFMT& sfmt) {
     int i, j;
     __m128i r, r1, r2, mask;
     mask = _mm_set_epi32(MSK4, MSK3, MSK2, MSK1);
 
-    r1 = _mm_load_si128(&sfmt[N - 2].si);
-    r2 = _mm_load_si128(&sfmt[N - 1].si);
+    SFMTData& data = *sfmt.data;
+    r1 = _mm_load_si128(&data.sfmt[N - 2].si);
+    r2 = _mm_load_si128(&data.sfmt[N - 1].si);
     for (i = 0; i < N - POS1; i++) {
-	r = mm_recursion(&sfmt[i].si, &sfmt[i + POS1].si, r1, r2, mask);
+	r = mm_recursion(&data.sfmt[i].si, &data.sfmt[i + POS1].si, r1, r2, mask);
 	_mm_store_si128(&array[i].si, r);
 	r1 = r2;
 	r2 = r;
     }
     for (; i < N; i++) {
-	r = mm_recursion(&sfmt[i].si, &array[i + POS1 - N].si, r1, r2, mask);
+	r = mm_recursion(&data.sfmt[i].si, &array[i + POS1 - N].si, r1, r2, mask);
 	_mm_store_si128(&array[i].si, r);
 	r1 = r2;
 	r2 = r;
@@ -106,13 +108,13 @@ inline static void gen_rand_array(w128_t *array, int size) {
     }
     for (j = 0; j < 2 * N - size; j++) {
 	r = _mm_load_si128(&array[j + size - N].si);
-	_mm_store_si128(&sfmt[j].si, r);
+	_mm_store_si128(&data.sfmt[j].si, r);
     }
     for (; i < size; i++) {
 	r = mm_recursion(&array[i - N].si, &array[i + POS1 - N].si, r1, r2,
 			 mask);
 	_mm_store_si128(&array[i].si, r);
-	_mm_store_si128(&sfmt[j++].si, r);
+	_mm_store_si128(&data.sfmt[j++].si, r);
 	r1 = r2;
 	r2 = r;
     }

libraries/sfmt/src/SFMT.cpp

@@ -144,9 +144,9 @@ inline static void swap(w128_t *array, int size);
 #endif
 
 #if defined(HAVE_ALTIVEC)
-  #include "SFMT-alti.h"
+  #include "sfmt/SFMT-alti.h"
 #elif defined(HAVE_SSE2)
-  #include "SFMT-sse2.h"
+  #include "sfmt/SFMT-sse2.h"
 #endif
 
 /**

openmmapi/include/openmm/internal/CustomNonbondedForceImpl.h

@@ -9,7 +9,7 @@
  * Biological Structures at Stanford, funded under the NIH Roadmap for        *
  * Medical Research, grant U54 GM072970. See https://simtk.org.               *
  *                                                                            *
- * Portions copyright (c) 2008-2012 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2013 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -35,7 +35,7 @@
 #include "ForceImpl.h"
 #include "openmm/CustomNonbondedForce.h"
 #include "openmm/Kernel.h"
-#include "lepton/ExpressionProgram.h"
+#include "lepton/CompiledExpression.h"
 #include <utility>
 #include <map>
 #include <string>
@@ -68,7 +68,7 @@ public:
     static double calcLongRangeCorrection(const CustomNonbondedForce& force, const Context& context);
 private:
     class TabulatedFunction;
-    static double integrateInteraction(const Lepton::ExpressionProgram& expression, const std::vector<double>& params1, const std::vector<double>& params2,
+    static double integrateInteraction(Lepton::CompiledExpression& expression, const std::vector<double>& params1, const std::vector<double>& params2,
             const CustomNonbondedForce& force, const Context& context);
     const CustomNonbondedForce& owner;
     Kernel kernel;

openmmapi/include/openmm/internal/hardware.h

+#ifndef OPENMM_HARDWARE_H_
+#define OPENMM_HARDWARE_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) 2013 Stanford University and the Authors.           *
+ * Authors: Peter Eastman                                                     *
+ * Contributors:                                                              *
+ *                                                                            *
+ * Permission is hereby granted, free of charge, to any person obtaining a    *
+ * copy of this software and associated documentation files (the "Software"), *
+ * to deal in the Software without restriction, including without limitation  *
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,   *
+ * and/or sell copies of the Software, and to permit persons to whom the      *
+ * Software is furnished to do so, subject to the following conditions:       *
+ *                                                                            *
+ * The above copyright notice and this permission notice shall be included in *
+ * all copies or substantial portions of the Software.                        *
+ *                                                                            *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,   *
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL    *
+ * THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,    *
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR      *
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE  *
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
+ * -------------------------------------------------------------------------- */
+
+/**
+ * This file defines a collection of functions for querying the specific hardware being used.
+ */
+
+/**
+ * Get the number of CPU cores available.
+ */
+#ifdef __APPLE__
+   #include <sys/sysctl.h>
+   #include <dlfcn.h>
+#else
+   #ifdef WIN32
+      #include <windows.h>
+   #else
+      #include <dlfcn.h>
+      #include <unistd.h>
+   #endif
+#endif
+
+static int getNumProcessors() {
+#ifdef __APPLE__
+    int ncpu;
+    size_t len = 4;
+    if (sysctlbyname("hw.logicalcpu", &ncpu, &len, NULL, 0) == 0)
+       return ncpu;
+    else
+       return 1;
+#else
+#ifdef WIN32
+    SYSTEM_INFO siSysInfo;
+    int ncpu;
+    GetSystemInfo(&siSysInfo);
+    ncpu = siSysInfo.dwNumberOfProcessors;
+    if (ncpu < 1)
+        ncpu = 1;
+    return ncpu;
+#else
+    long nProcessorsOnline = sysconf(_SC_NPROCESSORS_ONLN);
+    if (nProcessorsOnline == -1)
+        return 1;
+    else
+        return (int) nProcessorsOnline;
+#endif
+#endif
+}
+
+/**
+ * Get a description of the CPU's capabilities.
+ */
+#ifdef _WIN32
+#define cpuid __cpuid
+#else
+static void cpuid(int cpuInfo[4], int infoType){
+#ifdef __LP64__
+    __asm__ __volatile__ (
+        "cpuid":
+        "=a" (cpuInfo[0]),
+        "=b" (cpuInfo[1]),
+        "=c" (cpuInfo[2]),
+        "=d" (cpuInfo[3]) :
+        "a" (infoType)
+    );
+#else
+    __asm__ __volatile__ (
+        "pushl %%ebx\n"
+        "cpuid\n"
+        "movl %%ebx, %1\n"
+        "popl %%ebx\n" :
+        "=a" (cpuInfo[0]),
+        "=r" (cpuInfo[1]),
+        "=c" (cpuInfo[2]),
+        "=d" (cpuInfo[3]) :
+        "a" (infoType)
+    );
+#endif
+}
+#endif
+
+#endif // OPENMM_HARDWARE_H_

openmmapi/include/openmm/internal/vectorize.h

+#ifndef OPENMM_VECTORIZE_H_
+#define OPENMM_VECTORIZE_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) 2013 Stanford University and the Authors.           *
+ * Authors: Peter Eastman                                                     *
+ * Contributors:                                                              *
+ *                                                                            *
+ * Permission is hereby granted, free of charge, to any person obtaining a    *
+ * copy of this software and associated documentation files (the "Software"), *
+ * to deal in the Software without restriction, including without limitation  *
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,   *
+ * and/or sell copies of the Software, and to permit persons to whom the      *
+ * Software is furnished to do so, subject to the following conditions:       *
+ *                                                                            *
+ * The above copyright notice and this permission notice shall be included in *
+ * all copies or substantial portions of the Software.                        *
+ *                                                                            *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR *
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,   *
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL    *
+ * THE AUTHORS, CONTRIBUTORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,    *
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR      *
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE  *
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
+ * -------------------------------------------------------------------------- */
+
+#include <smmintrin.h>
+
+
+// This file defines classes and functions to simplify vectorizing code with SSE.
+
+class ivec4;
+
+/**
+ * A four element vector of floats.
+ */
+class fvec4 {
+public:
+    __m128 val;
+    
+    fvec4() {}
+    fvec4(float v) : val(_mm_set1_ps(v)) {}
+    fvec4(float v1, float v2, float v3, float v4) : val(_mm_set_ps(v4, v3, v2, v1)) {}
+    fvec4(__m128 v) : val(v) {}
+    fvec4(const float* v) : val(_mm_loadu_ps(v)) {}
+    operator __m128() const {
+        return val;
+    }
+    float operator[](int i) const {
+        int resultBits = _mm_extract_ps(val, i);
+        return *((float*) &resultBits);
+    }
+    void store(float* v) const {
+        _mm_storeu_ps(v, val);
+    }
+    fvec4 operator+(fvec4 other) const {
+        return _mm_add_ps(val, other);
+    }
+    fvec4 operator-(fvec4 other) const {
+        return _mm_sub_ps(val, other);
+    }
+    fvec4 operator*(fvec4 other) const {
+        return _mm_mul_ps(val, other);
+    }
+    fvec4 operator/(fvec4 other) const {
+        return _mm_div_ps(val, other);
+    }
+    void operator+=(fvec4 other) {
+        val = _mm_add_ps(val, other);
+    }
+    void operator-=(fvec4 other) {
+        val = _mm_sub_ps(val, other);
+    }
+    void operator*=(fvec4 other) {
+        val = _mm_mul_ps(val, other);
+    }
+    void operator/=(fvec4 other) {
+        val = _mm_div_ps(val, other);
+    }
+    fvec4 operator-() const {
+        return _mm_sub_ps(_mm_set1_ps(0.0f), val);
+    }
+    fvec4 operator&(fvec4 other) const {
+        return _mm_and_ps(val, other);
+    }
+    fvec4 operator==(fvec4 other) const {
+        return _mm_cmpeq_ps(val, other);
+    }
+    fvec4 operator!=(fvec4 other) const {
+        return _mm_cmpneq_ps(val, other);
+    }
+    fvec4 operator>(fvec4 other) const {
+        return _mm_cmpgt_ps(val, other);
+    }
+    fvec4 operator<(fvec4 other) const {
+        return _mm_cmplt_ps(val, other);
+    }
+    fvec4 operator>=(fvec4 other) const {
+        return _mm_cmpge_ps(val, other);
+    }
+    fvec4 operator<=(fvec4 other) const {
+        return _mm_cmple_ps(val, other);
+    }
+    operator ivec4() const;
+};
+
+/**
+ * A four element vector of ints.
+ */
+class ivec4 {
+public:
+    __m128i val;
+    
+    ivec4() {}
+    ivec4(int v) : val(_mm_set1_epi32(v)) {}
+    ivec4(int v1, int v2, int v3, int v4) : val(_mm_set_epi32(v4, v3, v2, v1)) {}
+    ivec4(__m128i v) : val(v) {}
+    ivec4(const int* v) : val(_mm_loadu_si128((const __m128i*) v)) {}
+    operator __m128i() const {
+        return val;
+    }
+    int operator[](int i) const {
+        return _mm_extract_epi32(val, i);
+    }
+    void store(int* v) const {
+        _mm_storeu_si128((__m128i*) v, val);
+    }
+    ivec4 operator+(ivec4 other) const {
+        return _mm_add_epi32(val, other);
+    }
+    ivec4 operator-(ivec4 other) const {
+        return _mm_sub_epi32(val, other);
+    }
+    ivec4 operator*(ivec4 other) const {
+        return _mm_mul_epi32(val, other);
+    }
+    void operator+=(ivec4 other) {
+        val = _mm_add_epi32(val, other);
+    }
+    void operator-=(ivec4 other) {
+        val = _mm_sub_epi32(val, other);
+    }
+    void operator*=(ivec4 other) {
+        val = _mm_mul_epi32(val, other);
+    }
+    ivec4 operator-() const {
+        return _mm_sub_epi32(_mm_set1_epi32(0), val);
+    }
+    ivec4 operator&(ivec4 other) const {
+        return _mm_and_si128(val, other);
+    }
+    ivec4 operator==(ivec4 other) const {
+        return _mm_cmpeq_epi32(val, other);
+    }
+    operator fvec4() const;
+};
+
+// Conversion operators.
+
+inline fvec4::operator ivec4() const {
+    return _mm_cvttps_epi32(val);
+}
+
+inline ivec4::operator fvec4() const {
+    return _mm_cvtepi32_ps(val);
+}
+
+// Functions that operate on fvec4s.
+
+static inline fvec4 floor(fvec4 v) {
+    return fvec4(_mm_floor_ps(v.val));
+}
+
+static inline fvec4 ceil(fvec4 v) {
+    return fvec4(_mm_ceil_ps(v.val));
+}
+
+static inline fvec4 round(fvec4 v) {
+    return fvec4(_mm_round_ps(v.val, _MM_FROUND_TO_NEAREST_INT));
+}
+
+static inline fvec4 min(fvec4 v1, fvec4 v2) {
+    return fvec4(_mm_min_ps(v1.val, v2.val));
+}
+
+static inline fvec4 max(fvec4 v1, fvec4 v2) {
+    return fvec4(_mm_max_ps(v1.val, v2.val));
+}
+
+static inline fvec4 abs(fvec4 v) {
+    static const __m128 mask = _mm_castsi128_ps(_mm_set1_epi32(0x7FFFFFFF));
+    return fvec4(_mm_and_ps(v.val, mask));
+}
+
+static inline fvec4 sqrt(fvec4 v) {
+    return fvec4(_mm_sqrt_ps(v.val));
+}
+
+static inline float dot3(fvec4 v1, fvec4 v2) {
+    return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0x71));
+}
+
+static inline float dot4(fvec4 v1, fvec4 v2) {
+    return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0xF1));
+}
+
+// Functions that operate on ivec4s.
+
+static inline ivec4 min(ivec4 v1, ivec4 v2) {
+    return ivec4(_mm_min_epi32(v1.val, v2.val));
+}
+
+static inline ivec4 max(ivec4 v1, ivec4 v2) {
+    return ivec4(_mm_max_epi32(v1.val, v2.val));
+}
+
+static inline ivec4 abs(ivec4 v) {
+    return ivec4(_mm_abs_epi32(v.val));
+}
+
+// Mathematical operators involving a scalar and a vector.
+
+static inline fvec4 operator+(float v1, fvec4 v2) {
+    return fvec4(v1)+v2;
+}
+
+static inline fvec4 operator-(float v1, fvec4 v2) {
+    return fvec4(v1)-v2;
+}
+
+static inline fvec4 operator*(float v1, fvec4 v2) {
+    return fvec4(v1)*v2;
+}
+
+static inline fvec4 operator/(float v1, fvec4 v2) {
+    return fvec4(v1)/v2;
+}
+
+#endif /*OPENMM_VECTORIZE_H_*/
+

openmmapi/src/ContextImpl.cpp

@@ -69,7 +69,9 @@ ContextImpl::ContextImpl(Context& owner, const System& system, Integrator& integ
         int particle1, particle2;
         double distance;
         system.getConstraintParameters(i, particle1, particle2, distance);
-        if (system.getParticleMass(particle1) == 0.0 || system.getParticleMass(particle2) == 0.0)
+        double mass1 = system.getParticleMass(particle1);
+        double mass2 = system.getParticleMass(particle2);
+        if ((mass1 == 0.0 && mass2 != 0.0) || (mass2 == 0.0 && mass1 != 0.0))
             throw OpenMMException("A constraint cannot involve a massless particle");
     }
     

openmmapi/src/CustomNonbondedForceImpl.cpp

@@ -182,7 +182,7 @@ double CustomNonbondedForceImpl::calcLongRangeCorrection(const CustomNonbondedFo
         force.getFunctionParameters(i, name, values, min, max);
         functions[name] = new TabulatedFunction(min, max, values);
     }
-    Lepton::ExpressionProgram expression = Lepton::Parser::parse(force.getEnergyFunction(), functions).optimize().createProgram();
+    Lepton::CompiledExpression expression = Lepton::Parser::parse(force.getEnergyFunction(), functions).createCompiledExpression();
     
     // Identify all particle classes (defined by parameters), and record the class of each particle.
     
@@ -251,25 +251,35 @@ double CustomNonbondedForceImpl::calcLongRangeCorrection(const CustomNonbondedFo
     return 2*M_PI*numParticles*numParticles*sum;
 }
 
-double CustomNonbondedForceImpl::integrateInteraction(const Lepton::ExpressionProgram& expression, const vector<double>& params1, const vector<double>& params2,
+double CustomNonbondedForceImpl::integrateInteraction(Lepton::CompiledExpression& expression, const vector<double>& params1, const vector<double>& params2,
         const CustomNonbondedForce& force, const Context& context) {
-    map<std::string, double> variables;
+    const set<string>& variables = expression.getVariables();
     for (int i = 0; i < force.getNumPerParticleParameters(); i++) {
         stringstream name1, name2;
         name1 << force.getPerParticleParameterName(i) << 1;
         name2 << force.getPerParticleParameterName(i) << 2;
-        variables[name1.str()] = params1[i];
-        variables[name2.str()] = params2[i];
+        if (variables.find(name1.str()) != variables.end())
+            expression.getVariableReference(name1.str()) = params1[i];
+        if (variables.find(name2.str()) != variables.end())
+            expression.getVariableReference(name2.str()) = params2[i];
     }
     for (int i = 0; i < force.getNumGlobalParameters(); i++) {
         const string& name = force.getGlobalParameterName(i);
-        variables[name] = context.getParameter(name);
+        if (variables.find(name) != variables.end())
+            expression.getVariableReference(name) = context.getParameter(name);
     }
     
     // To integrate from r_cutoff to infinity, make the change of variables x=r_cutoff/r and integrate from 0 to 1.
     // This introduces another r^2 into the integral, which along with the r^2 in the formula for the correction
     // means we multiply the function by r^4.  Use the midpoint method.
 
+    double* rPointer;
+    try {
+        rPointer = &expression.getVariableReference("r");
+    }
+    catch (exception& ex) {
+        throw OpenMMException("CustomNonbondedForce: Cannot use long range correction with a force that does not depend on r.");
+    }
     double cutoff = force.getCutoffDistance();
     double sum = 0;
     int numPoints = 1;
@@ -281,9 +291,9 @@ double CustomNonbondedForceImpl::integrateInteraction(const Lepton::ExpressionPr
                 continue;
             double x = (i+0.5)/numPoints;
             double r = cutoff/x;
-            variables["r"] = r;
+            *rPointer = r;
             double r2 = r*r;
-            newSum += expression.evaluate(variables)*r2*r2;
+            newSum += expression.evaluate()*r2*r2;
         }
         sum = newSum/numPoints + oldSum/3;
         if (iteration > 2 && (fabs((sum-oldSum)/sum) < 1e-5 || sum == 0))
@@ -309,8 +319,8 @@ double CustomNonbondedForceImpl::integrateInteraction(const Lepton::ExpressionPr
                 double x = (i+0.5)/numPoints;
                 double r = rswitch+x*(cutoff-rswitch);
                 double switchValue = x*x*x*(10+x*(-15+x*6));
-                variables["r"] = r;
-                newSum += switchValue*expression.evaluate(variables)*r*r;
+                *rPointer = r;
+                newSum += switchValue*expression.evaluate()*r*r;
             }
             sum2 = newSum/numPoints + oldSum/3;
             if (iteration > 2 && (fabs((sum2-oldSum)/sum2) < 1e-5 || sum2 == 0))