Merge remote-tracking branch 'origin/master' into charmm

.travis.yml

-language: cpp
-compiler:
-  - clang
+language: c
 
-before_install:
-  - sudo apt-get update -qq
-  - sudo apt-get install -qq libpcre3 libpcre3-dev gromacs
-  - sudo apt-get install -qq swig doxygen llvm-3.3
-  - sudo apt-get install -qq python-numpy python-scipy python-nose
-  - export ASAN_SYMBOLIZER_PATH=/usr/bin/llvm-symbolizer-3.3
+install:
+  - source tools/ci/install.sh
+  - export PYTHONUNBUFFERED=true
 
 script:
-  - cmake -DCMAKE_INSTALL_PREFIX=$HOME/OpenMM .
-  - make
-  - make install
-  - sudo make PythonInstall
-  - # run all of the tests
-  - ctest -V
-  - # get a list of all of the failed tests into this stupid ctest format
-  - python -c 'fn = "Testing/Temporary/LastTestsFailed.log"; import os; os.path.exists(fn) or exit(0); l = [line.split(":")[0] for line in open(fn)]; triplets = zip(l, l, [","]*len(l)); print "".join(",".join(t) for t in triplets)' > FailedTests.log
-  - # rerun all of the failed tests
-  - if [ -s FailedTests.log ]; then ctest -V -I FailedTests.log; fi;
-  - # run the python tests too
-  - cd python/tests
-  - nosetests -vv
+  - export CC="clang++"
+  - source deactivate
+  - conda install --yes conda-build
+  - # Build the conda package, testing build before packaging.
+  - conda build tools/conda-recipe
+  - # Install the conda package locally.
+  - source activate $python
+  - conda install $HOME/miniconda/conda-bld/linux-64/openmm-dev-*
+  - conda list -e
+  - # Run the Python tests.
+  - pushd .
+  - cd wrappers/python/tests
+  - nosetests -vv --processes=-1 --process-timeout=200
+  - popd
+
+env:
+  global:
+    # encrypted BINSTAR_TOKEN for push of dev package to binstar
+    - secure: Qz3pEYXXFnNQ/WK+15ad4cdbLJvzgCIZRwKD9fLiS3CDO2ldAQWxzaz8RQOwqbFtZUWu7lQpr+GukNJz5p0w18QEto+BxLYG9aW5mjoc+F2vCjyWFjkwnJ/Z/3uBKTcr5x9Y7HKaPGivaJ4BNACifjt7cCpeVJzV6u2+bBgSoHc=
+
+  matrix:
+    - python=2.7  CONDA_PY=27
+    #- python=3.3  CONDA_PY=33
+
+after_success:
+  - echo "after_success"
+  - source tools/ci/after_success.sh
+
+

CMakeLists.txt

@@ -61,6 +61,11 @@ ELSE(WIN32)
     ENDIF(NOT OPENMM_INSTALL_PREFIX)
 ENDIF(WIN32)
 
+# Include CPU-Features for Android
+IF (ANDROID)
+    INCLUDE_DIRECTORIES(${ANDROID_NDK}/sources/cpufeatures)
+ENDIF (ANDROID)
+
 # It seems that on linux and mac, everything is trying to be installed in /usr/local/openmm
 # But if every install target is prefixed with /openmm/, on Windows the install files
 # end up in C:/Program Files/OpenMM/openmm/ which is ugly.
@@ -87,9 +92,11 @@ IF(WIN32)
     SET(PTHREADS_LIB pthreadVC2)
     SET(PTHREADS_LIB_STATIC pthreadVC2_static_mt)
 ELSE(WIN32)
-    SET(PTHREADS_LIB pthread)
-    # in linux, even in static builds we link against the dynamic object (since its tied to libc versions)
-    SET(PTHREADS_LIB_STATIC pthread)
+    IF (NOT ANDROID)
+        SET(PTHREADS_LIB pthread)
+        # in linux, even in static builds we link against the dynamic object (since its tied to libc versions)
+        SET(PTHREADS_LIB_STATIC pthread)
+    ENDIF (NOT ANDROID)
 ENDIF(WIN32)
 
 # The build system will set ARCH64 for 64 bit builds, which require
@@ -121,11 +128,11 @@ IF (APPLE)
     SET (CMAKE_INSTALL_NAME_DIR "@rpath")
     SET(EXTRA_COMPILE_FLAGS "-msse2 -stdlib=libc++")
 ELSE (APPLE)
-    IF (MSVC)
+    IF (MSVC OR ANDROID)
         SET(EXTRA_COMPILE_FLAGS)
-    ELSE (MSVC)
+    ELSE (MSVC OR ANDROID)
         SET(EXTRA_COMPILE_FLAGS "-msse2")
-    ENDIF (MSVC)
+    ENDIF (MSVC OR ANDROID)
 ENDIF (APPLE)
 
 IF(UNIX AND NOT CMAKE_BUILD_TYPE)
@@ -137,8 +144,13 @@ IF (NOT CMAKE_CXX_FLAGS_DEBUG)
 ENDIF (NOT CMAKE_CXX_FLAGS_DEBUG)
 
 IF (NOT CMAKE_CXX_FLAGS_RELEASE)
-    SET(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG" CACHE STRING 
-          "To use when CMAKE_BUILD_TYPE=Release" FORCE)
+    IF (ANDROID)
+        SET(CMAKE_CXX_FLAGS_RELEASE "-mfloat-abi=softfp -march=armv7-a -mfpu=neon -funsafe-math-optimizations -O3 -DNDEBUG" CACHE STRING
+        "To use when CMAKE_BUILD_TYPE=Release" FORCE)
+    ELSE (ANDROID)
+        SET(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG" CACHE STRING 
+        "To use when CMAKE_BUILD_TYPE=Release" FORCE)
+    ENDIF (ANDROID)
 ENDIF (NOT CMAKE_CXX_FLAGS_RELEASE)
 
 
@@ -252,7 +264,11 @@ 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(${CMAKE_SOURCE_DIR}/libraries/sfmt/src/SFMT.cpp PROPERTIES COMPILE_FLAGS "-DHAVE_SSE2=1")
+IF (ANDROID)
+    SET_SOURCE_FILES_PROPERTIES(${CMAKE_SOURCE_DIR}/libraries/sfmt/src/SFMT.cpp PROPERTIES COMPILE_FLAGS "-UHAVE_SSE2")
+ELSE (ANDROID)
+    SET_SOURCE_FILES_PROPERTIES(${CMAKE_SOURCE_DIR}/libraries/sfmt/src/SFMT.cpp PROPERTIES COMPILE_FLAGS "-DHAVE_SSE2=1")
+ENDIF(ANDROID)
 
 # If API wrappers are being generated, and add them to the build.
 SET(OPENMM_BUILD_C_AND_FORTRAN_WRAPPERS ON CACHE BOOL "Build wrappers for C and Fortran")
@@ -287,13 +303,17 @@ ENDIF(OPENMM_BUILD_C_AND_FORTRAN_WRAPPERS)
 # On Linux need to link to libdl
 FIND_LIBRARY(DL_LIBRARY dl)
 IF(DL_LIBRARY)
-  TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${DL_LIBRARY} ${PTHREADS_LIB})
-  IF(OPENMM_BUILD_STATIC_LIB)
-    TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${DL_LIBRARY} ${PTHREADS_LIB})
-  ENDIF(OPENMM_BUILD_STATIC_LIB)
-  MARK_AS_ADVANCED(DL_LIBRARY)
+    TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${DL_LIBRARY} ${PTHREADS_LIB})
+    IF(OPENMM_BUILD_STATIC_LIB)
+        TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${DL_LIBRARY} ${PTHREADS_LIB})
+    ENDIF(OPENMM_BUILD_STATIC_LIB)
+    MARK_AS_ADVANCED(DL_LIBRARY)
 ELSE(DL_LIBRARY)
-  TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${PTHREADS_LIB})
+    IF (ANDROID)
+        TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${PTHREADS_LIB} cpufeatures)
+    ELSE (ANDROID)
+        TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${PTHREADS_LIB})
+    ENDIF (ANDROID)
 ENDIF(DL_LIBRARY)
 
 ADD_SUBDIRECTORY(platforms/reference/tests)

docs/usersguide/application.rst

@@ -2187,6 +2187,49 @@ second atom has class OS and the third has class P:
 
     <Proper class1="" class2="OS" class3="P" class4="" per="3" phase="0.0" k="0.66944"/>
 
+<CustomNonbondedForce>
+===============
+
+To add a CustomNonbondedForce to the System, include a tag that looks like this:
+
+.. code-block:: xml
+
+    <CustomNonbondedForce energy="scale*epsilon1*epsilon2*((sigma1+sigma2)/r)^12" bondCutoff="3">
+     <GlobalParameter name="scale" defaultValue="1"/>
+     <PerParticleParameter name="sigma"/>
+     <PerParticleParameter name="epsilon"/>
+     <Atom type="0" sigma="0.3249" epsilon="0.7112"/>
+     <Atom type="1" sigma="0.1069" epsilon="0.0656"/>
+     <Atom type="2" sigma="0.3399" epsilon="0.4577"/>
+     ...
+    </CustomNonbondedForce>
+
+The energy expression for the CustomNonbondedForce is specified by the
+:code:`energy` attribute.  This is a mathematical expression that gives the
+energy of each pairwise interaction as a function of the distance *r*\ .  It
+also may depend on an arbitrary list of global or per-particle parameters.  Use
+a :code:`<GlobalParameter>` tag to define a global parameter, and a
+:code:`<PerParticleParameter>` tag to define a per-particle parameter.
+
+Exclusions are created automatically based on the :code:`bondCutoff` attribute.
+After setting the nonbonded parameters for all atoms, the force field calls
+:code:`createExclusionsFromBonds()` on the CustomNonbondedForce, passing in this
+value as its argument.  To avoid creating exclusions, set :code:`bondCutoff` to 0.
+
+Each :code:`<Atom>` tag specifies the parameters for one atom type
+(specified with the :code:`type` attribute) or atom class (specified with
+the :code:`class` attribute).  It is fine to mix these two methods, having
+some tags specify a type and others specify a class.  However you do it, you
+must make sure that a unique set of parameters is defined for every atom type.
+The remaining attributes are the values to use for the per-atom parameters. All
+per-atom parameters must be specified for every :code:`<Atom>` tag, and the
+attribute name must match the name of the parameter.  For instance, if there is
+a per-atom parameter with the name “radius”, then every :code:`<Atom>` tag
+must include an attribute called :code:`radius`\ .
+
+CustomNonbondedForce also allows you to define tabulated functions.  See section
+:ref:`tabulated-functions` for details.
+
 <CustomGBForce>
 ===============
 
@@ -2244,30 +2287,8 @@ attribute name must match the name of the parameter.  For instance, if there is
 a per-atom parameter with the name “radius”, then every :code:`<Atom>` tag
 must include an attribute called :code:`radius`\ .
 
-CustomGBForce also allows you to define tabulated functions.  To define a
-function, include a :code:`<Function>` tag inside the
-:code:`<CustomGBForce>` tag:
-
-.. code-block:: xml
-
-    <Function name="myfn" min="-5" max="5">
-    0.983674857694 -0.980096396266 -0.975743130031 -0.970451936613 -0.964027580076
-    -0.956237458128 -0.946806012846 -0.935409070603 -0.921668554406 -0.905148253645
-    -0.885351648202 -0.861723159313 -0.833654607012 -0.800499021761 -0.761594155956
-    -0.716297870199 -0.664036770268 -0.604367777117 -0.537049566998 -0.46211715726
-    -0.379948962255 -0.291312612452 -0.197375320225 -0.099667994625 0.0
-    0.099667994625 0.197375320225 0.291312612452 0.379948962255 0.46211715726
-    0.537049566998 0.604367777117 0.664036770268 0.716297870199 0.761594155956
-    0.800499021761 0.833654607012 0.861723159313 0.885351648202 0.905148253645
-    0.921668554406 0.935409070603 0.946806012846 0.956237458128 0.964027580076
-    0.970451936613 0.975743130031 0.980096396266 0.983674857694 0.986614298151
-    0.989027402201
-    </Function>
-
-The tag’s attributes define the name of the function and the range of values for
-which it is defined.  The tabulated values are listed inside the body of the
-tag, with successive values separated by white space.  Again, see the API
-documentation for more details.
+CustomGBForce also allows you to define tabulated functions.  See section
+:ref:`tabulated-functions` for details.
 
 Writing Custom Expressions
 ==========================
@@ -2304,6 +2325,55 @@ is exactly equivalent to
 The definition of an intermediate value may itself involve other intermediate
 values.  All uses of a value must appear *before* that value’s definition.
 
+.. _tabulated-functions:
+
+TabulatedFunctions
+==================
+
+Some forces, such as CustomNonbondedForce and CustomGBForce, allow you to define
+tabulated functions.  To define a function, include a :code:`<Function>` tag inside the
+:code:`<CustomNonbondedForce>` or :code:`<CustomGBForce>` tag:
+
+.. code-block:: xml
+
+    <Function name="myfn" type="Continuous1D" min="-5" max="5">
+    0.983674857694 -0.980096396266 -0.975743130031 -0.970451936613 -0.964027580076
+    -0.956237458128 -0.946806012846 -0.935409070603 -0.921668554406 -0.905148253645
+    -0.885351648202 -0.861723159313 -0.833654607012 -0.800499021761 -0.761594155956
+    -0.716297870199 -0.664036770268 -0.604367777117 -0.537049566998 -0.46211715726
+    -0.379948962255 -0.291312612452 -0.197375320225 -0.099667994625 0.0
+    0.099667994625 0.197375320225 0.291312612452 0.379948962255 0.46211715726
+    0.537049566998 0.604367777117 0.664036770268 0.716297870199 0.761594155956
+    0.800499021761 0.833654607012 0.861723159313 0.885351648202 0.905148253645
+    0.921668554406 0.935409070603 0.946806012846 0.956237458128 0.964027580076
+    0.970451936613 0.975743130031 0.980096396266 0.983674857694 0.986614298151
+    0.989027402201
+    </Function>
+
+The tag’s attributes define the name of the function, the type of function, and
+the range of values for which it is defined.  The required set of attributed
+depends on the function type:
+
+.. tabularcolumns:: |l|L|
+
+============  =======================================================
+Type          Required Attributes
+============  =======================================================
+Continuous1D  min, max
+Continuous2D  xmin, ymin, xmax, ymax, xsize, ysize
+Continuous3D  xmin, ymin, zmin, xmax, ymax, zmax, xsize, ysize, zsize
+Discrete1D
+Discrete2D    xsize, ysize
+Discrete3D    xsize, ysize, zsize
+============  =======================================================
+
+
+The "min" and "max" attributes define the range of the independent variables for
+a continuous function.  The "size" attributes define the size of the table along
+each axis.  The tabulated values are listed inside the body of the tag, with
+successive values separated by white space.  See the API documentation for more
+details.
+
 
 Using Multiple Files
 ********************

libraries/lepton/src/CompiledExpression.cpp

@@ -84,13 +84,13 @@ void CompiledExpression::compileExpression(const ExpressionTreeNode& node, vecto
     // Process this node.
     
     if (node.getOperation().getId() == Operation::VARIABLE) {
-        variableIndices[node.getOperation().getName()] = workspace.size();
+        variableIndices[node.getOperation().getName()] = (int) workspace.size();
         variableNames.insert(node.getOperation().getName());
     }
     else {
-        int stepIndex = arguments.size();
+        int stepIndex = (int) arguments.size();
         arguments.push_back(vector<int>());
-        target.push_back(workspace.size());
+        target.push_back((int) 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.

libraries/lepton/src/ExpressionProgram.cpp

@@ -71,13 +71,13 @@ ExpressionProgram& ExpressionProgram::operator=(const ExpressionProgram& program
 }
 
 void ExpressionProgram::buildProgram(const ExpressionTreeNode& node) {
-    for (int i = node.getChildren().size()-1; i >= 0; i--)
+    for (int i = (int) node.getChildren().size()-1; i >= 0; i--)
         buildProgram(node.getChildren()[i]);
     operations.push_back(node.getOperation().clone());
 }
 
 int ExpressionProgram::getNumOperations() const {
-    return operations.size();
+    return (int) operations.size();
 }
 
 const Operation& ExpressionProgram::getOperation(int index) const {

libraries/lepton/src/ParsedExpression.cpp

@@ -60,7 +60,7 @@ double ParsedExpression::evaluate(const map<string, double>& variables) const {
 }
 
 double ParsedExpression::evaluate(const ExpressionTreeNode& node, const map<string, double>& variables) {
-    int numArgs = node.getChildren().size();
+    int numArgs = (int) node.getChildren().size();
     vector<double> args(max(numArgs, 1));
     for (int i = 0; i < numArgs; i++)
         args[i] = evaluate(node.getChildren()[i], variables);
@@ -269,7 +269,15 @@ ExpressionTreeNode ParsedExpression::substituteSimplerExpression(const Expressio
                 return ExpressionTreeNode(new Operation::Constant(dynamic_cast<const Operation::MultiplyConstant*>(&node.getOperation())->getValue()*getConstantValue(children[0])));
             if (children[0].getOperation().getId() == Operation::NEGATE) // Combine a multiply and a negate into a single multiply
                 return ExpressionTreeNode(new Operation::MultiplyConstant(-dynamic_cast<const Operation::MultiplyConstant*>(&node.getOperation())->getValue()), children[0].getChildren()[0]);
+            break;
         }
+        default:
+        {
+            // If operation ID is not one of the above,
+            // we don't substitute a simpler expression.
+            break;
+        }
+
     }
     return ExpressionTreeNode(node.getOperation().clone(), children);
 }

libraries/lepton/src/Parser.cpp

@@ -70,7 +70,7 @@ string Parser::trim(const string& expression) {
     int start, end;
     for (start = 0; start < (int) expression.size() && isspace(expression[start]); start++)
         ;
-    for (end = expression.size()-1; end > start && isspace(expression[end]); end--)
+    for (end = (int) expression.size()-1; end > start && isspace(expression[end]); end--)
         ;
     if (start == end && isspace(expression[end]))
         return "";
@@ -140,7 +140,7 @@ vector<ParseToken> Parser::tokenize(const string& expression) {
         ParseToken token = getNextToken(expression, pos);
         if (token.getType() != ParseToken::Whitespace)
             tokens.push_back(token);
-        pos += token.getText().size();
+        pos += (int) token.getText().size();
     }
     return tokens;
 }
@@ -257,7 +257,7 @@ ExpressionTreeNode Parser::parsePrecedence(const vector<ParseToken>& tokens, int
 
     while (pos < (int) tokens.size() && tokens[pos].getType() == ParseToken::Operator) {
         token = tokens[pos];
-        int opIndex = Operators.find(token.getText());
+        int opIndex = (int) Operators.find(token.getText());
         int opPrecedence = Precedence[opIndex];
         if (opPrecedence < precedence)
             return result;

openmmapi/include/openmm/CustomNonbondedForce.h

@@ -343,6 +343,8 @@ public:
     void setParticleParameters(int index, const std::vector<double>& parameters);
     /**
      * Add a particle pair to the list of interactions that should be excluded.
+     * 
+     * In many cases, you can use createExclusionsFromBonds() rather than adding each exclusion explicitly.
      *
      * @param particle1  the index of the first particle in the pair
      * @param particle2  the index of the second particle in the pair
@@ -365,6 +367,15 @@ public:
      * @param particle2  the index of the second particle in the pair
      */
     void setExclusionParticles(int index, int particle1, int particle2);
+    /**
+     * Identify exclusions based on the molecular topology.  Particles which are separated by up to a specified number of
+     * bonds are added as exclusions.
+     *
+     * @param bonds       the set of bonds based on which to construct exclusions.  Each element specifies the indices of
+     *                    two particles that are bonded to each other.
+     * @param bondCutoff  pairs of particles that are separated by this many bonds or fewer are added to the list of exclusions
+     */
+    void createExclusionsFromBonds(const std::vector<std::pair<int, int> >& bonds, int bondCutoff);
     /**
      * Add a tabulated function that may appear in the energy expression.
      *

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) 2008-2013 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2014 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -67,7 +67,6 @@ public:
      */
     static double calcLongRangeCorrection(const CustomNonbondedForce& force, const Context& context);
 private:
-    class TabulatedFunction;
     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;

openmmapi/include/openmm/internal/hardware.h

@@ -47,8 +47,12 @@
       #define NOMINMAX
       #include <windows.h>
    #else
-      #include <dlfcn.h>
-      #include <unistd.h>
+      #ifdef __ANDROID__
+        #include <cpu-features.h>
+      #else
+        #include <dlfcn.h>
+        #include <unistd.h>
+      #endif
    #endif
 #endif
 
@@ -70,11 +74,15 @@ static int getNumProcessors() {
         ncpu = 1;
     return ncpu;
 #else
-    long nProcessorsOnline = sysconf(_SC_NPROCESSORS_ONLN);
-    if (nProcessorsOnline == -1)
-        return 1;
-    else
-        return (int) nProcessorsOnline;
+    #ifdef __ANDROID__
+        return android_getCpuCount();
+    #else
+      long nProcessorsOnline = sysconf(_SC_NPROCESSORS_ONLN);
+      if (nProcessorsOnline == -1)
+          return 1;
+      else
+          return (int) nProcessorsOnline;
+    #endif
 #endif
 #endif
 }
@@ -85,30 +93,32 @@ static int getNumProcessors() {
 #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
-}
+#ifndef __ANDROID__
+    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
 
 #endif // OPENMM_HARDWARE_H_

openmmapi/include/openmm/internal/vectorize.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) 2013 Stanford University and the Authors.           *
+ * Portions copyright (c) 2014 Stanford University and the Authors.           *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -31,256 +31,11 @@
  * 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 {
-        float result[4];
-        store(result);
-        return result[i];
-    }
-    void store(float* v) const {
-        _mm_storeu_ps(v, val);
-    }
-    fvec4 operator+(const fvec4& other) const {
-        return _mm_add_ps(val, other);
-    }
-    fvec4 operator-(const fvec4& other) const {
-        return _mm_sub_ps(val, other);
-    }
-    fvec4 operator*(const fvec4& other) const {
-        return _mm_mul_ps(val, other);
-    }
-    fvec4 operator/(const fvec4& other) const {
-        return _mm_div_ps(val, other);
-    }
-    void operator+=(const fvec4& other) {
-        val = _mm_add_ps(val, other);
-    }
-    void operator-=(const fvec4& other) {
-        val = _mm_sub_ps(val, other);
-    }
-    void operator*=(const fvec4& other) {
-        val = _mm_mul_ps(val, other);
-    }
-    void operator/=(const fvec4& other) {
-        val = _mm_div_ps(val, other);
-    }
-    fvec4 operator-() const {
-        return _mm_sub_ps(_mm_set1_ps(0.0f), val);
-    }
-    fvec4 operator&(const fvec4& other) const {
-        return _mm_and_ps(val, other);
-    }
-    fvec4 operator|(const fvec4& other) const {
-        return _mm_or_ps(val, other);
-    }
-    fvec4 operator==(const fvec4& other) const {
-        return _mm_cmpeq_ps(val, other);
-    }
-    fvec4 operator!=(const fvec4& other) const {
-        return _mm_cmpneq_ps(val, other);
-    }
-    fvec4 operator>(const fvec4& other) const {
-        return _mm_cmpgt_ps(val, other);
-    }
-    fvec4 operator<(const fvec4& other) const {
-        return _mm_cmplt_ps(val, other);
-    }
-    fvec4 operator>=(const fvec4& other) const {
-        return _mm_cmpge_ps(val, other);
-    }
-    fvec4 operator<=(const 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 {
-        int result[4];
-        store(result);
-        return result[i];
-    }
-    void store(int* v) const {
-        _mm_storeu_si128((__m128i*) v, val);
-    }
-    ivec4 operator+(const ivec4& other) const {
-        return _mm_add_epi32(val, other);
-    }
-    ivec4 operator-(const ivec4& other) const {
-        return _mm_sub_epi32(val, other);
-    }
-    ivec4 operator*(const ivec4& other) const {
-        return _mm_mul_epi32(val, other);
-    }
-    void operator+=(const ivec4& other) {
-        val = _mm_add_epi32(val, other);
-    }
-    void operator-=(const ivec4& other) {
-        val = _mm_sub_epi32(val, other);
-    }
-    void operator*=(const ivec4& other) {
-        val = _mm_mul_epi32(val, other);
-    }
-    ivec4 operator-() const {
-        return _mm_sub_epi32(_mm_set1_epi32(0), val);
-    }
-    ivec4 operator&(const ivec4& other) const {
-        return _mm_and_si128(val, other);
-    }
-    ivec4 operator|(const ivec4& other) const {
-        return _mm_or_si128(val, other);
-    }
-    ivec4 operator==(const ivec4& other) const {
-        return _mm_cmpeq_epi32(val, other);
-    }
-    ivec4 operator!=(const ivec4& other) const {
-        return _mm_xor_si128(*this==other, _mm_set1_epi32(0xFFFFFFFF));
-    }
-    ivec4 operator>(const ivec4& other) const {
-        return _mm_cmpgt_epi32(val, other);
-    }
-    ivec4 operator<(const ivec4& other) const {
-        return _mm_cmplt_epi32(val, other);
-    }
-    ivec4 operator>=(const ivec4& other) const {
-        return _mm_xor_si128(_mm_cmplt_epi32(val, other), _mm_set1_epi32(0xFFFFFFFF));
-    }
-    ivec4 operator<=(const ivec4& other) const {
-        return _mm_xor_si128(_mm_cmpgt_epi32(val, other), _mm_set1_epi32(0xFFFFFFFF));
-    }
-    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(const fvec4& v) {
-    return fvec4(_mm_floor_ps(v.val));
-}
-
-static inline fvec4 ceil(const fvec4& v) {
-    return fvec4(_mm_ceil_ps(v.val));
-}
-
-static inline fvec4 round(const fvec4& v) {
-    return fvec4(_mm_round_ps(v.val, _MM_FROUND_TO_NEAREST_INT));
-}
-
-static inline fvec4 min(const fvec4& v1, const fvec4& v2) {
-    return fvec4(_mm_min_ps(v1.val, v2.val));
-}
-
-static inline fvec4 max(const fvec4& v1, const fvec4& v2) {
-    return fvec4(_mm_max_ps(v1.val, v2.val));
-}
-
-static inline fvec4 abs(const 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(const fvec4& v) {
-    return fvec4(_mm_sqrt_ps(v.val));
-}
-
-static inline float dot3(const fvec4& v1, const fvec4& v2) {
-    return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0x71));
-}
-
-static inline float dot4(const fvec4& v1, const fvec4& v2) {
-    return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0xF1));
-}
-
-static inline void transpose(fvec4& v1, fvec4& v2, fvec4& v3, fvec4& v4) {
-    _MM_TRANSPOSE4_PS(v1, v2, v3, v4);
-}
-
-// Functions that operate on ivec4s.
-
-static inline ivec4 min(const ivec4& v1, const ivec4& v2) {
-    return ivec4(_mm_min_epi32(v1.val, v2.val));
-}
-
-static inline ivec4 max(const ivec4& v1, const ivec4& v2) {
-    return ivec4(_mm_max_epi32(v1.val, v2.val));
-}
-
-static inline ivec4 abs(const ivec4& v) {
-    return ivec4(_mm_abs_epi32(v.val));
-}
-
-static inline bool any(const ivec4& v) {
-    return !_mm_test_all_zeros(v, _mm_set1_epi32(0xFFFFFFFF));
-}
-
-// Mathematical operators involving a scalar and a vector.
-
-static inline fvec4 operator+(float v1, const fvec4& v2) {
-    return fvec4(v1)+v2;
-}
-
-static inline fvec4 operator-(float v1, const fvec4& v2) {
-    return fvec4(v1)-v2;
-}
-
-static inline fvec4 operator*(float v1, const fvec4& v2) {
-    return fvec4(v1)*v2;
-}
-
-static inline fvec4 operator/(float v1, const fvec4& v2) {
-    return fvec4(v1)/v2;
-}
-
-// Operations for blending fvec4s based on an ivec4.
-
-static inline fvec4 blend(const fvec4& v1, const fvec4& v2, const ivec4& mask) {
-    return fvec4(_mm_blendv_ps(v1.val, v2.val, _mm_castsi128_ps(mask.val)));
-}
+        
+#if defined(__ANDROID__)
+    #include "vectorize_neon.h"
+#else
+    #include "vectorize_sse.h"
+#endif
 
 #endif /*OPENMM_VECTORIZE_H_*/
-

openmmapi/include/openmm/internal/vectorize_neon.h

+#ifndef OPENMM_VECTORIZE_NEON_H_
+#define OPENMM_VECTORIZE_NEON_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) 2013-2014 Stanford University and the Authors.      *
+ * Authors: Mateus Lima, 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 <cpu-features.h>
+#include <arm_neon.h>
+#include <cmath>
+
+typedef int int32_t;
+
+// This file defines classes and functions to simplify vectorizing code with NEON.
+
+class ivec4;
+
+/**
+ * A four element vector of floats.
+ */
+class fvec4 {
+public:
+    float32x4_t val;
+
+    fvec4() {}
+    fvec4(float v) : val(vdupq_n_f32(v)) {}
+    fvec4(float v1, float v2, float v3, float v4) {
+        float v[] = {v1, v2, v3, v4};
+        val = vld1q_f32(v);
+    }
+    fvec4(float32x4_t v) : val(v) {}
+    fvec4(const float* v) : val(vld1q_f32(v)) {}
+    operator float32x4_t() const {
+        return val;
+    }
+    float operator[](int i) const {
+        switch (i) {
+            case 0:
+                return vgetq_lane_f32(val, 0);
+            case 1:
+                return vgetq_lane_f32(val, 1);
+            case 2:
+                return vgetq_lane_f32(val, 2);
+            case 3:
+                return vgetq_lane_f32(val, 3);
+        }
+        return 0.0f;
+    }
+    void store(float* v) const {
+        vst1q_f32(v, val);
+    }
+    fvec4 operator+(const fvec4& other) const {
+        return vaddq_f32(val, other);
+    }
+    fvec4 operator-(const fvec4& other) const {
+        return vsubq_f32(val, other);
+    }
+    fvec4 operator*(const fvec4& other) const {
+        return vmulq_f32(val, other);
+    }
+    fvec4 operator/(const fvec4& other) const {
+        // NEON does not have a divide float-point operator, so we get the reciprocal and multiply.
+
+        float32x4_t reciprocal = vrecpeq_f32(other);
+        reciprocal = vmulq_f32(vrecpsq_f32(other, reciprocal), reciprocal);
+        reciprocal = vmulq_f32(vrecpsq_f32(other, reciprocal), reciprocal);
+        fvec4 result = vmulq_f32(val,reciprocal);
+        return result;
+    }
+    void operator+=(const fvec4& other) {
+        val = vaddq_f32(val, other);
+    }
+    void operator-=(const fvec4& other) {
+        val = vsubq_f32(val, other);
+    }
+    void operator*=(const fvec4& other) {
+        val = vmulq_f32(val, other);
+    }
+    void operator/=(const fvec4& other) {
+        val = *this/other;
+    }
+    fvec4 operator-() const {
+        return vnegq_f32(val);
+    }
+    fvec4 operator&(const fvec4& other) const {
+        return vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(val), vreinterpretq_u32_f32(other)));
+    }
+    fvec4 operator|(const fvec4& other) const {
+        return vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(val), vreinterpretq_u32_f32(other)));
+    }
+    fvec4 operator==(const fvec4& other) const {
+        return vcvtq_f32_s32(vreinterpretq_s32_u32(vceqq_f32(val, other)));
+    }
+    fvec4 operator!=(const fvec4& other) const {
+        return vcvtq_f32_s32(vreinterpretq_s32_u32(vmvnq_u32(vceqq_f32(val, other)))); // not(equals(val, other))
+    }
+    fvec4 operator>(const fvec4& other) const {
+        return vcvtq_f32_s32(vreinterpretq_s32_u32(vcgtq_f32(val, other)));
+    }
+    fvec4 operator<(const fvec4& other) const {
+        return vcvtq_f32_s32(vreinterpretq_s32_u32(vcltq_f32(val, other)));
+    }
+    fvec4 operator>=(const fvec4& other) const {
+        return vcvtq_f32_s32(vreinterpretq_s32_u32(vcgeq_f32(val, other)));
+    }
+    fvec4 operator<=(const fvec4& other) const {
+        return vcvtq_f32_s32(vreinterpretq_s32_u32(vcleq_f32(val, other)));
+    }
+    operator ivec4() const;
+};
+
+/**
+ * A four element vector of ints.
+ */
+class ivec4 {
+public:
+    
+    int32x4_t val;
+
+    ivec4() {}
+    ivec4(int v) : val(vdupq_n_s32(v)) {}
+    ivec4(int v1, int v2, int v3, int v4) {
+        int v[] = {v1, v2, v3, v4};
+        val = vld1q_s32(v);
+    }
+    ivec4(int32x4_t v) : val(v) {}
+    ivec4(const int* v) : val(vld1q_s32(v)) {}
+    operator int32x4_t() const {
+        return val;
+    }
+    int operator[](int i) const {
+        switch (i) {
+            case 0:
+                return vgetq_lane_s32(val, 0);
+            case 1:
+                return vgetq_lane_s32(val, 1);
+            case 2:
+                return vgetq_lane_s32(val, 2);
+            case 3:
+                return vgetq_lane_s32(val, 3);
+        }
+        return 0;
+    }
+    void store(int* v) const {
+        vst1q_s32(v, val);
+    }
+    ivec4 operator+(const ivec4& other) const {
+        return vaddq_s32(val, other);
+    }
+    ivec4 operator-(const ivec4& other) const {
+        return vsubq_s32(val, other);
+    }
+    ivec4 operator*(const ivec4& other) const {
+        return vmulq_s32(val, other);
+    }
+    void operator+=(const ivec4& other) {
+        val = vaddq_s32(val, other);
+    }
+    void operator-=(const ivec4& other) {
+        val = vsubq_s32(val, other);
+    }
+    void operator*=(const ivec4& other) {
+        val = vmulq_s32(val, other);
+    }
+    ivec4 operator-() const {
+        return vnegq_s32(val);
+    }
+    ivec4 operator&(const ivec4& other) const {
+        return vandq_s32(val, other);
+    }
+    ivec4 operator|(const ivec4& other) const {
+        return vorrq_s32(val, other);
+    }
+    ivec4 operator==(const ivec4& other) const {
+        return vreinterpretq_s32_u32(vceqq_s32(val, other));
+    }
+    ivec4 operator!=(const ivec4& other) const {
+        return vreinterpretq_s32_u32(vmvnq_u32(vceqq_s32(val, other))); // not(equal(val, other))
+    }
+    ivec4 operator>(const ivec4& other) const {
+        return vreinterpretq_s32_u32(vcgtq_s32(val, other));
+    }
+    ivec4 operator<(const ivec4& other) const {
+        return vreinterpretq_s32_u32(vcltq_s32(val, other));
+    }
+    ivec4 operator>=(const ivec4& other) const {
+        return vreinterpretq_s32_u32(vcgeq_s32(val, other));
+    }
+    ivec4 operator<=(const ivec4& other) const {
+        return vreinterpretq_s32_u32(vcleq_s32(val, other));
+    }
+    operator fvec4() const;
+};
+
+// Conversion operators.
+
+inline fvec4::operator ivec4() const {
+    return ivec4(vcvtq_s32_f32(val));
+}
+
+inline ivec4::operator fvec4() const {
+    return fvec4(vcvtq_f32_s32(val));
+}
+
+// Functions that operate on fvec4s.
+
+static inline fvec4 min(const fvec4& v1, const fvec4& v2) {
+    return vminq_f32(v1, v2);
+}
+
+static inline fvec4 max(const fvec4& v1, const fvec4& v2) {
+    return vmaxq_f32(v1, v2);
+}
+
+static inline fvec4 abs(const fvec4& v) {
+    return vabsq_f32(v);
+}
+
+static inline fvec4 sqrt(const fvec4& v) {
+    float32x4_t recipSqrt = vrsqrteq_f32(v);
+    recipSqrt = vmulq_f32(recipSqrt, vrsqrtsq_f32(vmulq_f32(recipSqrt, v), recipSqrt));
+    recipSqrt = vmulq_f32(recipSqrt, vrsqrtsq_f32(vmulq_f32(recipSqrt, v), recipSqrt));
+    return vmulq_f32(v, recipSqrt);
+}
+
+static inline float dot3(const fvec4& v1, const fvec4& v2) {
+    fvec4 result = v1*v2;
+    return vgetq_lane_f32(result, 0) + vgetq_lane_f32(result, 1) + vgetq_lane_f32(result, 2);
+}
+
+static inline float dot4(const fvec4& v1, const fvec4& v2) {
+    fvec4 result = v1*v2;
+    return vgetq_lane_f32(result, 0) + vgetq_lane_f32(result, 1) + vgetq_lane_f32(result, 2) + vgetq_lane_f32(result,3);
+}
+
+static inline void transpose(fvec4& v1, fvec4& v2, fvec4& v3, fvec4& v4) {
+    float32x4x2_t t1 = vuzpq_f32(v1, v3);
+    float32x4x2_t t2 = vuzpq_f32(v2, v4);
+    float32x4x2_t t3 = vtrnq_f32(t1.val[0], t2.val[0]);
+    float32x4x2_t t4 = vtrnq_f32(t1.val[1], t2.val[1]);
+    v1 = t3.val[0];
+    v2 = t4.val[0];
+    v3 = t3.val[1];
+    v4 = t4.val[1];
+}
+
+// Functions that operate on ivec4s.
+
+static inline ivec4 min(const ivec4& v1, const ivec4& v2) {
+    return vminq_s32(v1, v2);
+}
+
+static inline ivec4 max(const ivec4& v1, const ivec4& v2) {
+    return vmaxq_s32(v1, v2);
+}
+
+static inline ivec4 abs(const ivec4& v) {
+    return vabdq_s32(v, ivec4(0));
+}
+
+static inline bool any(const ivec4& v) {
+    return (vgetq_lane_s32(v, 0) != 0 || vgetq_lane_s32(v, 1) != 0 || vgetq_lane_s32(v, 2) != 0 || vgetq_lane_s32(v, 3) != 0);
+}
+
+// Mathematical operators involving a scalar and a vector.
+
+static inline fvec4 operator+(float v1, const fvec4& v2) {
+    return fvec4(v1)+v2;
+}
+
+static inline fvec4 operator-(float v1, const fvec4& v2) {
+    return fvec4(v1)-v2;
+}
+
+static inline fvec4 operator*(float v1, const fvec4& v2) {
+    return fvec4(v1)*v2;
+}
+
+static inline fvec4 operator/(float v1, const fvec4& v2) {
+    return fvec4(v1)/v2;
+}
+
+// Operations for blending fvec4s based on an ivec4.
+
+static inline fvec4 blend(const fvec4& v1, const fvec4& v2, const ivec4& mask) {
+    return vbslq_f32(vreinterpretq_u32_s32(mask), v2, v1);
+}
+
+// These are at the end since they involve other functions defined above.
+
+static inline fvec4 round(const fvec4& v) {
+    fvec4 shift(0x1.0p23f);
+    fvec4 absResult = (abs(v)+shift)-shift;
+    return blend(v, absResult, ivec4(0x7FFFFFFF));
+}
+
+static inline fvec4 floor(const fvec4& v) {
+    fvec4 rounded = round(v);
+    return rounded + blend(0.0f, -1.0f, rounded>v);
+}
+
+static inline fvec4 ceil(const fvec4& v) {
+    fvec4 rounded = round(v);
+    return rounded + blend(0.0f, 1.0f, rounded<v);
+}
+
+#endif /*OPENMM_VECTORIZE_NEON_H_*/

openmmapi/include/openmm/internal/vectorize_sse.h

+#ifndef OPENMM_VECTORIZE_SSE_H_
+#define OPENMM_VECTORIZE_SSE_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 {
+        float result[4];
+        store(result);
+        return result[i];
+    }
+    void store(float* v) const {
+        _mm_storeu_ps(v, val);
+    }
+    fvec4 operator+(const fvec4& other) const {
+        return _mm_add_ps(val, other);
+    }
+    fvec4 operator-(const fvec4& other) const {
+        return _mm_sub_ps(val, other);
+    }
+    fvec4 operator*(const fvec4& other) const {
+        return _mm_mul_ps(val, other);
+    }
+    fvec4 operator/(const fvec4& other) const {
+        return _mm_div_ps(val, other);
+    }
+    void operator+=(const fvec4& other) {
+        val = _mm_add_ps(val, other);
+    }
+    void operator-=(const fvec4& other) {
+        val = _mm_sub_ps(val, other);
+    }
+    void operator*=(const fvec4& other) {
+        val = _mm_mul_ps(val, other);
+    }
+    void operator/=(const fvec4& other) {
+        val = _mm_div_ps(val, other);
+    }
+    fvec4 operator-() const {
+        return _mm_sub_ps(_mm_set1_ps(0.0f), val);
+    }
+    fvec4 operator&(const fvec4& other) const {
+        return _mm_and_ps(val, other);
+    }
+    fvec4 operator|(const fvec4& other) const {
+        return _mm_or_ps(val, other);
+    }
+    fvec4 operator==(const fvec4& other) const {
+        return _mm_cmpeq_ps(val, other);
+    }
+    fvec4 operator!=(const fvec4& other) const {
+        return _mm_cmpneq_ps(val, other);
+    }
+    fvec4 operator>(const fvec4& other) const {
+        return _mm_cmpgt_ps(val, other);
+    }
+    fvec4 operator<(const fvec4& other) const {
+        return _mm_cmplt_ps(val, other);
+    }
+    fvec4 operator>=(const fvec4& other) const {
+        return _mm_cmpge_ps(val, other);
+    }
+    fvec4 operator<=(const 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 {
+        int result[4];
+        store(result);
+        return result[i];
+    }
+    void store(int* v) const {
+        _mm_storeu_si128((__m128i*) v, val);
+    }
+    ivec4 operator+(const ivec4& other) const {
+        return _mm_add_epi32(val, other);
+    }
+    ivec4 operator-(const ivec4& other) const {
+        return _mm_sub_epi32(val, other);
+    }
+    ivec4 operator*(const ivec4& other) const {
+        return _mm_mul_epi32(val, other);
+    }
+    void operator+=(const ivec4& other) {
+        val = _mm_add_epi32(val, other);
+    }
+    void operator-=(const ivec4& other) {
+        val = _mm_sub_epi32(val, other);
+    }
+    void operator*=(const ivec4& other) {
+        val = _mm_mul_epi32(val, other);
+    }
+    ivec4 operator-() const {
+        return _mm_sub_epi32(_mm_set1_epi32(0), val);
+    }
+    ivec4 operator&(const ivec4& other) const {
+        return _mm_and_si128(val, other);
+    }
+    ivec4 operator|(const ivec4& other) const {
+        return _mm_or_si128(val, other);
+    }
+    ivec4 operator==(const ivec4& other) const {
+        return _mm_cmpeq_epi32(val, other);
+    }
+    ivec4 operator!=(const ivec4& other) const {
+        return _mm_xor_si128(*this==other, _mm_set1_epi32(0xFFFFFFFF));
+    }
+    ivec4 operator>(const ivec4& other) const {
+        return _mm_cmpgt_epi32(val, other);
+    }
+    ivec4 operator<(const ivec4& other) const {
+        return _mm_cmplt_epi32(val, other);
+    }
+    ivec4 operator>=(const ivec4& other) const {
+        return _mm_xor_si128(_mm_cmplt_epi32(val, other), _mm_set1_epi32(0xFFFFFFFF));
+    }
+    ivec4 operator<=(const ivec4& other) const {
+        return _mm_xor_si128(_mm_cmpgt_epi32(val, other), _mm_set1_epi32(0xFFFFFFFF));
+    }
+    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(const fvec4& v) {
+    return fvec4(_mm_floor_ps(v.val));
+}
+
+static inline fvec4 ceil(const fvec4& v) {
+    return fvec4(_mm_ceil_ps(v.val));
+}
+
+static inline fvec4 round(const fvec4& v) {
+    return fvec4(_mm_round_ps(v.val, _MM_FROUND_TO_NEAREST_INT));
+}
+
+static inline fvec4 min(const fvec4& v1, const fvec4& v2) {
+    return fvec4(_mm_min_ps(v1.val, v2.val));
+}
+
+static inline fvec4 max(const fvec4& v1, const fvec4& v2) {
+    return fvec4(_mm_max_ps(v1.val, v2.val));
+}
+
+static inline fvec4 abs(const 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(const fvec4& v) {
+    return fvec4(_mm_sqrt_ps(v.val));
+}
+
+static inline float dot3(const fvec4& v1, const fvec4& v2) {
+    return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0x71));
+}
+
+static inline float dot4(const fvec4& v1, const fvec4& v2) {
+    return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0xF1));
+}
+
+static inline void transpose(fvec4& v1, fvec4& v2, fvec4& v3, fvec4& v4) {
+    _MM_TRANSPOSE4_PS(v1, v2, v3, v4);
+}
+
+// Functions that operate on ivec4s.
+
+static inline ivec4 min(const ivec4& v1, const ivec4& v2) {
+    return ivec4(_mm_min_epi32(v1.val, v2.val));
+}
+
+static inline ivec4 max(const ivec4& v1, const ivec4& v2) {
+    return ivec4(_mm_max_epi32(v1.val, v2.val));
+}
+
+static inline ivec4 abs(const ivec4& v) {
+    return ivec4(_mm_abs_epi32(v.val));
+}
+
+static inline bool any(const ivec4& v) {
+    return !_mm_test_all_zeros(v, _mm_set1_epi32(0xFFFFFFFF));
+}
+
+// Mathematical operators involving a scalar and a vector.
+
+static inline fvec4 operator+(float v1, const fvec4& v2) {
+    return fvec4(v1)+v2;
+}
+
+static inline fvec4 operator-(float v1, const fvec4& v2) {
+    return fvec4(v1)-v2;
+}
+
+static inline fvec4 operator*(float v1, const fvec4& v2) {
+    return fvec4(v1)*v2;
+}
+
+static inline fvec4 operator/(float v1, const fvec4& v2) {
+    return fvec4(v1)/v2;
+}
+
+// Operations for blending fvec4s based on an ivec4.
+
+static inline fvec4 blend(const fvec4& v1, const fvec4& v2, const ivec4& mask) {
+    return fvec4(_mm_blendv_ps(v1.val, v2.val, _mm_castsi128_ps(mask.val)));
+}
+
+#endif /*OPENMM_VECTORIZE_SSE_H_*/
+

openmmapi/src/CustomNonbondedForce.cpp

@@ -174,6 +174,33 @@ void CustomNonbondedForce::setExclusionParticles(int index, int particle1, int p
     exclusions[index].particle1 = particle1;
     exclusions[index].particle2 = particle2;
 }
+
+void CustomNonbondedForce::createExclusionsFromBonds(const vector<pair<int, int> >& bonds, int bondCutoff) {
+    if (bondCutoff < 1)
+        return;
+    vector<set<int> > exclusions(particles.size());
+    vector<set<int> > bonded12(exclusions.size());
+    for (int i = 0; i < (int) bonds.size(); ++i) {
+        int p1 = bonds[i].first;
+        int p2 = bonds[i].second;
+        exclusions[p1].insert(p2);
+        exclusions[p2].insert(p1);
+        bonded12[p1].insert(p2);
+        bonded12[p2].insert(p1);
+    }
+    for (int level = 0; level < bondCutoff-1; level++) {
+        vector<set<int> > currentExclusions = exclusions;
+        for (int i = 0; i < (int) particles.size(); i++) {
+            for (set<int>::const_iterator iter = currentExclusions[i].begin(); iter != currentExclusions[i].end(); ++iter)
+                exclusions[*iter].insert(bonded12[i].begin(), bonded12[i].end());
+        }
+    }
+    for (int i = 0; i < (int) exclusions.size(); ++i)
+        for (set<int>::const_iterator iter = exclusions[i].begin(); iter != exclusions[i].end(); ++iter)
+            if (*iter < i)
+                addExclusion(*iter, i);
+}
+
 int CustomNonbondedForce::addTabulatedFunction(const std::string& name, TabulatedFunction* function) {
     functions.push_back(FunctionInfo(name, function));
     return functions.size()-1;

openmmapi/src/CustomNonbondedForceImpl.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) 2008-2013 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2014 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -37,7 +37,7 @@
 #include "openmm/internal/CustomNonbondedForceImpl.h"
 #include "openmm/internal/SplineFitter.h"
 #include "openmm/kernels.h"
-#include "lepton/CustomFunction.h"
+#include "ReferenceTabulatedFunction.h"
 #include "lepton/ParsedExpression.h"
 #include "lepton/Parser.h"
 #include <cmath>
@@ -137,38 +137,6 @@ void CustomNonbondedForceImpl::updateParametersInContext(ContextImpl& context) {
     kernel.getAs<CalcCustomNonbondedForceKernel>().copyParametersToContext(context, owner);
 }
 
-class CustomNonbondedForceImpl::TabulatedFunction : public Lepton::CustomFunction {
-public:
-    TabulatedFunction(double min, double max, const vector<double>& values) :
-            min(min), max(max), values(values) {
-        int numValues = values.size();
-        x.resize(numValues);
-        for (int i = 0; i < numValues; i++)
-            x[i] = min+i*(max-min)/(numValues-1);
-        SplineFitter::createNaturalSpline(x, values, derivs);
-    }
-    int getNumArguments() const {
-        return 1;
-    }
-    double evaluate(const double* arguments) const {
-        double t = arguments[0];
-        if (t < min || t > max)
-            return 0.0;
-        return SplineFitter::evaluateSpline(x, values, derivs, t);
-    }
-    double evaluateDerivative(const double* arguments, const int* derivOrder) const {
-        double t = arguments[0];
-        if (t < min || t > max)
-            return 0.0;
-        return SplineFitter::evaluateSplineDerivative(x, values, derivs, t);
-    }
-    CustomFunction* clone() const {
-        return new TabulatedFunction(min, max, values);
-    }
-    double min, max;
-    vector<double> x, values, derivs;
-};
-
 double CustomNonbondedForceImpl::calcLongRangeCorrection(const CustomNonbondedForce& force, const Context& context) {
     if (force.getNonbondedMethod() == CustomNonbondedForce::NoCutoff || force.getNonbondedMethod() == CustomNonbondedForce::CutoffNonPeriodic)
         return 0.0;
@@ -176,13 +144,8 @@ double CustomNonbondedForceImpl::calcLongRangeCorrection(const CustomNonbondedFo
     // Parse the energy expression.
     
     map<string, Lepton::CustomFunction*> functions;
-    for (int i = 0; i < force.getNumFunctions(); i++) {
-        string name;
-        vector<double> values;
-        double min, max;
-        force.getFunctionParameters(i, name, values, min, max);
-        functions[name] = new TabulatedFunction(min, max, values);
-    }
+    for (int i = 0; i < force.getNumFunctions(); i++)
+        functions[force.getTabulatedFunctionName(i)] = createReferenceTabulatedFunction(force.getTabulatedFunction(i));
     Lepton::CompiledExpression expression = Lepton::Parser::parse(force.getEnergyFunction(), functions).createCompiledExpression();
     
     // Identify all particle classes (defined by parameters), and record the class of each particle.

openmmapi/src/OSRngSeed.cpp

@@ -29,7 +29,6 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
 
-#include <stdexcept>
 #if defined(_WIN32) || defined(__CYGWIN__)
 #include <windows.h>
 static HCRYPTPROV hCryptProv = 0;
@@ -38,28 +37,31 @@ static HCRYPTPROV hCryptProv = 0;
 #include <fcntl.h>
 #include <unistd.h>
 #endif
+#include "openmm/OpenMMException.h"
 #include "openmm/internal/OSRngSeed.h"
 
+using OpenMM::OpenMMException;
+
 int osrngseed(void) {
     int value;
 #if defined(_WIN32) || defined(__CYGWIN__)
     if (!::CryptAcquireContextW(&hCryptProv, 0, 0, PROV_RSA_FULL, CRYPT_VERIFYCONTEXT | CRYPT_SILENT)) {
-        throw std::runtime_error("Failed to initialize Windows random API (CryptoGen)");
+        throw OpenMMException("Failed to initialize Windows random API (CryptoGen)");
     }
     if (!CryptGenRandom(hCryptProv, sizeof(int), (BYTE*) &value)) {
         ::CryptReleaseContext(hCryptProv, 0);
-        throw std::runtime_error("Failed to get random numbers");
+        throw OpenMMException("Failed to get random numbers");
     }
     if (!::CryptReleaseContext(hCryptProv, 0)) {
-        throw std::runtime_error("Failed to release Windows random API context");
+        throw OpenMMException("Failed to release Windows random API context");
     }
 #else
     int m_fd = open("/dev/urandom", O_RDONLY);
     if (m_fd == -1) {
-        throw std::runtime_error("Failed to open /dev/urandom");
+        throw OpenMMException("Failed to open /dev/urandom");
     }
     if (read(m_fd, &value, sizeof(int)) != sizeof(int)) {
-        throw std::runtime_error("Failed to read bytes from /dev/urandom");
+        throw OpenMMException("Failed to read bytes from /dev/urandom");
     }
     close(m_fd);
 #endif

platforms/cpu/sharedTarget/CMakeLists.txt

 FOREACH(file ${SOURCE_FILES})
     IF (file MATCHES ".*Vec8.*")
-		IF (MSVC)
+        IF (MSVC)
             SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} /arch:AVX /D__AVX__")
-		ELSE (MSVC)
-            SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -msse4.1 -mavx")
-		ENDIF (MSVC)
+        ELSE (MSVC)
+            IF (NOT ANDROID)
+                SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -msse4.1 -mavx")
+            ENDIF (NOT ANDROID)
+        ENDIF (MSVC)
     ELSE (file MATCHES ".*Vec8.*")
-		IF (NOT MSVC)
-            SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -msse4.1")
-		ENDIF (NOT MSVC)
+        IF (NOT MSVC)
+            IF (NOT ANDROID)
+                SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -msse4.1")
+            ENDIF (NOT ANDROID)
+        ENDIF (NOT MSVC)
     ENDIF (file MATCHES ".*Vec8.*")
 ENDFOREACH(file)
 ADD_LIBRARY(${SHARED_TARGET} SHARED ${SOURCE_FILES} ${SOURCE_INCLUDE_FILES} ${API_ABS_INCLUDE_FILES})

platforms/cpu/src/CpuNeighborList.cpp

@@ -37,7 +37,6 @@
 #include <set>
 #include <map>
 #include <cmath>
-#include <smmintrin.h>
 
 using namespace std;
 

platforms/cpu/src/CpuNonbondedForceVec4.cpp

@@ -103,7 +103,7 @@ void CpuNonbondedForceVec4::calculateBlockIxn(int blockIndex, float* forces, dou
             dEdR = epsSig6*(12.0f*sig6 - 6.0f);
             energy = epsSig6*(sig6-1.0f);
             if (useSwitch) {
-                fvec4 t = (r>switchingDistance) & ((r-switchingDistance)*invSwitchingInterval);
+                fvec4 t = blend(0.0f, (r-switchingDistance)*invSwitchingInterval, r>switchingDistance);
                 fvec4 switchValue = 1+t*t*t*(-10.0f+t*(15.0f-t*6.0f));
                 fvec4 switchDeriv = t*t*(-30.0f+t*(60.0f-t*30.0f))*invSwitchingInterval;
                 dEdR = switchValue*dEdR - energy*switchDeriv*r;
@@ -214,7 +214,7 @@ void CpuNonbondedForceVec4::calculateBlockEwaldIxn(int blockIndex, float* forces
             dEdR = epsSig6*(12.0f*sig6 - 6.0f);
             energy = epsSig6*(sig6-1.0f);
             if (useSwitch) {
-                fvec4 t = (r>switchingDistance) & ((r-switchingDistance)*invSwitchingInterval);
+                fvec4 t = blend(0.0f, (r-switchingDistance)*invSwitchingInterval, r>switchingDistance);
                 fvec4 switchValue = 1+t*t*t*(-10.0f+t*(15.0f-t*6.0f));
                 fvec4 switchDeriv = t*t*(-30.0f+t*(60.0f-t*30.0f))*invSwitchingInterval;
                 dEdR = switchValue*dEdR - energy*switchDeriv*r;

platforms/cpu/src/CpuPlatform.cpp

@@ -36,6 +36,7 @@
 #include "ReferenceConstraints.h"
 #include "openmm/internal/hardware.h"
 #include <sstream>
+#include <stdlib.h>
 
 using namespace OpenMM;
 using namespace std;
@@ -93,15 +94,20 @@ bool CpuPlatform::supportsDoublePrecision() const {
 }
 
 bool CpuPlatform::isProcessorSupported() {
-    // Make sure the CPU supports SSE 4.1.
-    
-    int cpuInfo[4];
-    cpuid(cpuInfo, 0);
-    if (cpuInfo[0] >= 1) {
-        cpuid(cpuInfo, 1);
-        return ((cpuInfo[2] & ((int) 1 << 19)) != 0);
-    }
-    return false;
+    // Make sure the CPU supports SSE 4.1 or NEON.
+        
+    #ifdef __ANDROID__
+        uint64_t features = android_getCpuFeatures();
+        return (features & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
+    #else
+        int cpuInfo[4];
+        cpuid(cpuInfo, 0);
+        if (cpuInfo[0] >= 1) {
+            cpuid(cpuInfo, 1);
+            return ((cpuInfo[2] & ((int) 1 << 19)) != 0);
+        }
+        return false;
+    #endif
 }
 
 void CpuPlatform::contextCreated(ContextImpl& context, const map<string, string>& properties) const {