CPU platform works on PNaCl

openmmapi/include/openmm/internal/vectorize_pnacl.h

@@ -33,6 +33,7 @@
  * -------------------------------------------------------------------------- */
 
 #include <cmath>
+#include <cstdlib>
 
 // This file defines classes and functions to simplify vectorizing code with portable SIMD vectors.
 
@@ -276,7 +277,7 @@ static inline ivec4 max(const ivec4& v1, const ivec4& v2) {
 }
 
 static inline ivec4 abs(const ivec4& v) {
-    return ivec4(std::abs(v[0]), std::abs(v[1]), std::abs(v[2]), std::abs(v[3]));
+    return ivec4(abs(v[0]), abs(v[1]), abs(v[2]), abs(v[3]));
 }
 
 static inline bool any(const ivec4& v) {

platforms/cpu/CMakeLists.txt

@@ -76,7 +76,9 @@ INCLUDE_DIRECTORIES(BEFORE ${CMAKE_CURRENT_SOURCE_DIR}/src)
 FILE(GLOB CORE_HEADERS include/*.h)
 INSTALL_FILES(/include/openmm/cpu FILES ${CORE_HEADERS})
 
-SUBDIRS (sharedTarget)
+IF(OPENMM_BUILD_SHARED_LIB)
+    SUBDIRS (sharedTarget)
+ENDIF(OPENMM_BUILD_SHARED_LIB)
 IF(OPENMM_BUILD_STATIC_LIB)
     SUBDIRS (staticTarget)
 ENDIF(OPENMM_BUILD_STATIC_LIB)

platforms/cpu/include/CpuGBSAOBCForce.h

@@ -98,9 +98,9 @@ private:
     float periodicBoxSize[3];
     float cutoffDistance, soluteDielectric, solventDielectric;
     std::vector<std::pair<float, float> > particleParams;        
-    std::vector<float> bornRadii;
-    std::vector<std::vector<float> > threadBornForces;
-    std::vector<float> obcChain;
+    AlignedArray<float> bornRadii;
+    std::vector<AlignedArray<float> > threadBornForces;
+    AlignedArray<float> obcChain;
     std::vector<double> threadEnergy;
     std::vector<float> logTable;
     float logDX, logDXInv;

platforms/cpu/src/CpuGBSAOBCForce.cpp

@@ -213,7 +213,7 @@ void CpuGBSAOBCForce::threadComputeForce(ThreadPool& threads, int threadIndex) {
     const float probeRadius = 0.14f;
     const float surfaceAreaFactor = 28.3919551;
     double energy = 0.0;
-    vector<float>& bornForces = threadBornForces[threadIndex];
+    AlignedArray<float>& bornForces = threadBornForces[threadIndex];
     for (int i = 0; i < numParticles; i++)
         bornForces[i] = 0.0f;
     while (true) {
@@ -410,10 +410,18 @@ fvec4 CpuGBSAOBCForce::fastLog(const fvec4& x) {
         return fvec4(logf(x[0]), logf(x[1]), logf(x[2]), logf(x[3]));
     fvec4 coeff2 = x1-index;
     fvec4 coeff1 = 1.0f-coeff2;
+#ifdef __PNACL__
+    // PNaCl crashes on unaligned loads.
+    fvec4 t1(logTable[index[0]], logTable[index[0]+1], logTable[index[0]+2], logTable[index[0]+3]);
+    fvec4 t2(logTable[index[1]], logTable[index[1]+1], logTable[index[1]+2], logTable[index[1]+3]);
+    fvec4 t3(logTable[index[2]], logTable[index[2]+1], logTable[index[2]+2], logTable[index[2]+3]);
+    fvec4 t4(logTable[index[3]], logTable[index[3]+1], logTable[index[3]+2], logTable[index[3]+3]);
+#else
     fvec4 t1(&logTable[index[0]]);
     fvec4 t2(&logTable[index[1]]);
     fvec4 t3(&logTable[index[2]]);
     fvec4 t4(&logTable[index[3]]);
+#endif
     transpose(t1, t2, t3, t4);
     return coeff1*t1 + coeff2*t2;
 }

platforms/cpu/src/CpuNonbondedForceVec4.cpp

@@ -292,10 +292,18 @@ fvec4 CpuNonbondedForceVec4::ewaldScaleFunction(const fvec4& x) {
     ivec4 index = min(floor(x1), NUM_TABLE_POINTS);
     fvec4 coeff2 = x1-index;
     fvec4 coeff1 = 1.0f-coeff2;
+#ifdef __PNACL__
+    // PNaCl crashes on unaligned loads.
+    fvec4 t1(ewaldScaleTable[index[0]], ewaldScaleTable[index[0]+1], ewaldScaleTable[index[0]+2], ewaldScaleTable[index[0]+3]);
+    fvec4 t2(ewaldScaleTable[index[1]], ewaldScaleTable[index[1]+1], ewaldScaleTable[index[1]+2], ewaldScaleTable[index[1]+3]);
+    fvec4 t3(ewaldScaleTable[index[2]], ewaldScaleTable[index[2]+1], ewaldScaleTable[index[2]+2], ewaldScaleTable[index[2]+3]);
+    fvec4 t4(ewaldScaleTable[index[3]], ewaldScaleTable[index[3]+1], ewaldScaleTable[index[3]+2], ewaldScaleTable[index[3]+3]);
+#else
     fvec4 t1(&ewaldScaleTable[index[0]]);
     fvec4 t2(&ewaldScaleTable[index[1]]);
     fvec4 t3(&ewaldScaleTable[index[2]]);
     fvec4 t4(&ewaldScaleTable[index[3]]);
+#endif
     transpose(t1, t2, t3, t4);
     return coeff1*t1 + coeff2*t2;
 }

platforms/cpu/src/CpuRandom.cpp

@@ -33,7 +33,7 @@ CpuRandom::CpuRandom() : hasInitialized(false) {
 }
 
 CpuRandom::~CpuRandom() {
-    for (int i = 0; i < threadRandom.size(); i++)
+    for (int i = 0; i < (int) threadRandom.size(); i++)
         delete threadRandom[i];
 }
 

platforms/cpu/staticTarget/CMakeLists.txt

@@ -2,18 +2,20 @@ FOREACH(file ${SOURCE_FILES})
     IF (file MATCHES ".*Vec8.*")
 		IF (MSVC)
             SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} /arch:AVX /D__AVX__")
+        ELSEIF (PNACL)
+            SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS}")
 		ELSE (MSVC)
             SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -msse4.1 -mavx")
 		ENDIF (MSVC)
     ELSE (file MATCHES ".*Vec8.*")
-		IF (NOT MSVC)
+		IF (NOT (MSVC OR ANDROID OR PNACL))
             SET_SOURCE_FILES_PROPERTIES(${file} PROPERTIES COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -msse4.1")
-		ENDIF (NOT MSVC)
+		ENDIF (NOT (MSVC OR ANDROID OR PNACL))
     ENDIF (file MATCHES ".*Vec8.*")
 ENDFOREACH(file)
 ADD_LIBRARY(${STATIC_TARGET} STATIC ${SOURCE_FILES} ${SOURCE_INCLUDE_FILES} ${API_ABS_INCLUDE_FILES})
 
-TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${OPENMM_LIBRARY_NAME} ${PTHREADS_LIB_STATIC})
+TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${OPENMM_LIBRARY_NAME}_static ${PTHREADS_LIB_STATIC})
 #-DPTW32_STATIC_LIB only works for the windows pthreads.
 SET_TARGET_PROPERTIES(${STATIC_TARGET} PROPERTIES LINK_FLAGS "${EXTRA_COMPILE_FLAGS}" COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS} -DOPENMM_CPU_BUILDING_STATIC_LIBRARY -DPTW32_STATIC_LIB")
 

platforms/cpu/tests/CMakeLists.txt

@@ -17,9 +17,12 @@ FILE(GLOB TEST_PROGS "*Test*.cpp")
 FOREACH(TEST_PROG ${TEST_PROGS})
     GET_FILENAME_COMPONENT(TEST_ROOT ${TEST_PROG} NAME_WE)
 
-    # Link with shared library
     ADD_EXECUTABLE(${TEST_ROOT} ${TEST_PROG})
+    IF (OPENMM_BUILD_SHARED_LIB)
         TARGET_LINK_LIBRARIES(${TEST_ROOT} ${SHARED_TARGET})
+    ELSE (OPENMM_BUILD_SHARED_LIB)
+        TARGET_LINK_LIBRARIES(${TEST_ROOT} ${STATIC_TARGET})
+    ENDIF (OPENMM_BUILD_SHARED_LIB)
     SET_TARGET_PROPERTIES(${TEST_ROOT} PROPERTIES LINK_FLAGS "${EXTRA_COMPILE_FLAGS}" COMPILE_FLAGS "${EXTRA_COMPILE_FLAGS}")
     ADD_TEST(${TEST_ROOT} ${EXECUTABLE_OUTPUT_PATH}/${TEST_ROOT} single)