Merge pull request #277 from peastman/winvec

Fixed lots of compilation errors on Windows

CMakeLists.txt

@@ -299,6 +299,8 @@ IF(DL_LIBRARY)
   IF(OPENMM_BUILD_STATIC_LIB)
     TARGET_LINK_LIBRARIES(${STATIC_TARGET} ${DL_LIBRARY} ${PTHREADS_LIB})
   ENDIF(OPENMM_BUILD_STATIC_LIB)
+ELSE(DL_LIBRARY)
+  TARGET_LINK_LIBRARIES(${SHARED_TARGET} ${PTHREADS_LIB})
 ENDIF(DL_LIBRARY)
 IF(WIN32)
     MARK_AS_ADVANCED(DL_LIBRARY)

openmmapi/include/openmm/internal/ThreadPool.h

@@ -32,6 +32,7 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
 
+#define NOMINMAX
 #include "windowsExport.h"
 #include <pthread.h>
 #include <vector>

openmmapi/include/openmm/internal/hardware.h

@@ -44,6 +44,7 @@
    #include <dlfcn.h>
 #else
    #ifdef WIN32
+      #define NOMINMAX
       #include <windows.h>
    #else
       #include <dlfcn.h>

openmmapi/include/openmm/internal/vectorize.h

@@ -62,55 +62,55 @@ public:
     void store(float* v) const {
         _mm_storeu_ps(v, val);
     }
-    fvec4 operator+(fvec4 other) const {
+    fvec4 operator+(const fvec4& other) const {
         return _mm_add_ps(val, other);
     }
-    fvec4 operator-(fvec4 other) const {
+    fvec4 operator-(const fvec4& other) const {
         return _mm_sub_ps(val, other);
     }
-    fvec4 operator*(fvec4 other) const {
+    fvec4 operator*(const fvec4& other) const {
         return _mm_mul_ps(val, other);
     }
-    fvec4 operator/(fvec4 other) const {
+    fvec4 operator/(const fvec4& other) const {
         return _mm_div_ps(val, other);
     }
-    void operator+=(fvec4 other) {
+    void operator+=(const fvec4& other) {
         val = _mm_add_ps(val, other);
     }
-    void operator-=(fvec4 other) {
+    void operator-=(const fvec4& other) {
         val = _mm_sub_ps(val, other);
     }
-    void operator*=(fvec4 other) {
+    void operator*=(const fvec4& other) {
         val = _mm_mul_ps(val, other);
     }
-    void operator/=(fvec4 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&(fvec4 other) const {
+    fvec4 operator&(const fvec4& other) const {
         return _mm_and_ps(val, other);
     }
-    fvec4 operator|(fvec4 other) const {
+    fvec4 operator|(const fvec4& other) const {
         return _mm_or_ps(val, other);
     }
-    fvec4 operator==(fvec4 other) const {
+    fvec4 operator==(const fvec4& other) const {
         return _mm_cmpeq_ps(val, other);
     }
-    fvec4 operator!=(fvec4 other) const {
+    fvec4 operator!=(const fvec4& other) const {
         return _mm_cmpneq_ps(val, other);
     }
-    fvec4 operator>(fvec4 other) const {
+    fvec4 operator>(const fvec4& other) const {
         return _mm_cmpgt_ps(val, other);
     }
-    fvec4 operator<(fvec4 other) const {
+    fvec4 operator<(const fvec4& other) const {
         return _mm_cmplt_ps(val, other);
     }
-    fvec4 operator>=(fvec4 other) const {
+    fvec4 operator>=(const fvec4& other) const {
         return _mm_cmpge_ps(val, other);
     }
-    fvec4 operator<=(fvec4 other) const {
+    fvec4 operator<=(const fvec4& other) const {
         return _mm_cmple_ps(val, other);
     }
     operator ivec4() const;
@@ -139,49 +139,49 @@ public:
     void store(int* v) const {
         _mm_storeu_si128((__m128i*) v, val);
     }
-    ivec4 operator+(ivec4 other) const {
+    ivec4 operator+(const ivec4& other) const {
         return _mm_add_epi32(val, other);
     }
-    ivec4 operator-(ivec4 other) const {
+    ivec4 operator-(const ivec4& other) const {
         return _mm_sub_epi32(val, other);
     }
-    ivec4 operator*(ivec4 other) const {
+    ivec4 operator*(const ivec4& other) const {
         return _mm_mul_epi32(val, other);
     }
-    void operator+=(ivec4 other) {
+    void operator+=(const ivec4& other) {
         val = _mm_add_epi32(val, other);
     }
-    void operator-=(ivec4 other) {
+    void operator-=(const ivec4& other) {
         val = _mm_sub_epi32(val, other);
     }
-    void operator*=(ivec4 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&(ivec4 other) const {
+    ivec4 operator&(const ivec4& other) const {
         return _mm_and_si128(val, other);
     }
-    ivec4 operator|(ivec4 other) const {
+    ivec4 operator|(const ivec4& other) const {
         return _mm_or_si128(val, other);
     }
-    ivec4 operator==(ivec4 other) const {
+    ivec4 operator==(const ivec4& other) const {
         return _mm_cmpeq_epi32(val, other);
     }
-    ivec4 operator!=(ivec4 other) const {
+    ivec4 operator!=(const ivec4& other) const {
         return _mm_xor_si128(*this==other, _mm_set1_epi32(0xFFFFFFFF));
     }
-    ivec4 operator>(ivec4 other) const {
+    ivec4 operator>(const ivec4& other) const {
         return _mm_cmpgt_epi32(val, other);
     }
-    ivec4 operator<(ivec4 other) const {
+    ivec4 operator<(const ivec4& other) const {
         return _mm_cmplt_epi32(val, other);
     }
-    ivec4 operator>=(ivec4 other) const {
+    ivec4 operator>=(const ivec4& other) const {
         return _mm_xor_si128(_mm_cmplt_epi32(val, other), _mm_set1_epi32(0xFFFFFFFF));
     }
-    ivec4 operator<=(ivec4 other) const {
+    ivec4 operator<=(const ivec4& other) const {
         return _mm_xor_si128(_mm_cmpgt_epi32(val, other), _mm_set1_epi32(0xFFFFFFFF));
     }
     operator fvec4() const;
@@ -199,40 +199,40 @@ inline ivec4::operator fvec4() const {
 
 // Functions that operate on fvec4s.
 
-static inline fvec4 floor(fvec4 v) {
+static inline fvec4 floor(const fvec4& v) {
     return fvec4(_mm_floor_ps(v.val));
 }
 
-static inline fvec4 ceil(fvec4 v) {
+static inline fvec4 ceil(const fvec4& v) {
     return fvec4(_mm_ceil_ps(v.val));
 }
 
-static inline fvec4 round(fvec4 v) {
+static inline fvec4 round(const fvec4& v) {
     return fvec4(_mm_round_ps(v.val, _MM_FROUND_TO_NEAREST_INT));
 }
 
-static inline fvec4 min(fvec4 v1, fvec4 v2) {
+static inline fvec4 min(const fvec4& v1, const fvec4& v2) {
     return fvec4(_mm_min_ps(v1.val, v2.val));
 }
 
-static inline fvec4 max(fvec4 v1, fvec4 v2) {
+static inline fvec4 max(const fvec4& v1, const fvec4& v2) {
     return fvec4(_mm_max_ps(v1.val, v2.val));
 }
 
-static inline fvec4 abs(fvec4 v) {
+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(fvec4 v) {
+static inline fvec4 sqrt(const fvec4& v) {
     return fvec4(_mm_sqrt_ps(v.val));
 }
 
-static inline float dot3(fvec4 v1, fvec4 v2) {
+static inline float dot3(const fvec4& v1, const fvec4& v2) {
     return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0x71));
 }
 
-static inline float dot4(fvec4 v1, fvec4 v2) {
+static inline float dot4(const fvec4& v1, const fvec4& v2) {
     return _mm_cvtss_f32(_mm_dp_ps(v1, v2, 0xF1));
 }
 
@@ -242,43 +242,43 @@ static inline void transpose(fvec4& v1, fvec4& v2, fvec4& v3, fvec4& v4) {
 
 // Functions that operate on ivec4s.
 
-static inline ivec4 min(ivec4 v1, ivec4 v2) {
+static inline ivec4 min(const ivec4& v1, const ivec4& v2) {
     return ivec4(_mm_min_epi32(v1.val, v2.val));
 }
 
-static inline ivec4 max(ivec4 v1, ivec4 v2) {
+static inline ivec4 max(const ivec4& v1, const ivec4& v2) {
     return ivec4(_mm_max_epi32(v1.val, v2.val));
 }
 
-static inline ivec4 abs(ivec4 v) {
+static inline ivec4 abs(const ivec4& v) {
     return ivec4(_mm_abs_epi32(v.val));
 }
 
-static inline bool any(ivec4 v) {
+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, fvec4 v2) {
+static inline fvec4 operator+(float v1, const fvec4& v2) {
     return fvec4(v1)+v2;
 }
 
-static inline fvec4 operator-(float v1, fvec4 v2) {
+static inline fvec4 operator-(float v1, const fvec4& v2) {
     return fvec4(v1)-v2;
 }
 
-static inline fvec4 operator*(float v1, fvec4 v2) {
+static inline fvec4 operator*(float v1, const fvec4& v2) {
     return fvec4(v1)*v2;
 }
 
-static inline fvec4 operator/(float v1, fvec4 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(fvec4 v1, fvec4 v2, ivec4 mask) {
+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)));
 }
 

platforms/cpu/include/CpuGBSAOBCForce.h

@@ -123,7 +123,7 @@ private:
     /**
      * Evaluate log(x) using a lookup table for speed.
      */
-    fvec4 fastLog(fvec4 x);
+    fvec4 fastLog(const fvec4& x);
 };
 
 } // namespace OpenMM

platforms/cpu/include/CpuNonbondedForceVec4.h

@@ -74,12 +74,12 @@ protected:
       /**
        * Compute a fast approximation to erfc(x).
        */
-      static fvec4 erfcApprox(fvec4 x);
+      static fvec4 erfcApprox(const fvec4& x);
       
       /**
        * Evaluate the scale factor used with Ewald and PME: erfc(alpha*r) + 2*alpha*r*exp(-alpha*alpha*r*r)/sqrt(PI)
        */
-      fvec4 ewaldScaleFunction(fvec4 x);
+      fvec4 ewaldScaleFunction(const fvec4& x);
 };
 
 } // namespace OpenMM

platforms/cpu/include/CpuNonbondedForceVec8.h

@@ -25,9 +25,10 @@
 #ifndef OPENMM_CPU_NONBONDED_FORCE_VEC8_H__
 #define OPENMM_CPU_NONBONDED_FORCE_VEC8_H__
 
+#include "CpuNonbondedForce.h"
+
 #ifdef __AVX__
 
-#include "CpuNonbondedForce.h"
 #include "openmm/internal/vectorize8.h"
 
 // ---------------------------------------------------------------------------------------
@@ -72,12 +73,12 @@ protected:
       /**
        * Compute a fast approximation to erfc(x).
        */
-      static fvec8 erfcApprox(fvec8 x);
+      static fvec8 erfcApprox(const fvec8& x);
       
       /**
        * Evaluate the scale factor used with Ewald and PME: erfc(alpha*r) + 2*alpha*r*exp(-alpha*alpha*r*r)/sqrt(PI)
        */
-      fvec8 ewaldScaleFunction(fvec8 x);
+      fvec8 ewaldScaleFunction(const fvec8& x);
 };
 
 } // namespace OpenMM

platforms/cpu/include/CpuRandom.h

@@ -33,6 +33,7 @@
  * -------------------------------------------------------------------------- */
 
 #include "sfmt/SFMT.h"
+#include "windowsExportCpu.h"
 #include <vector>
 
 namespace OpenMM {
@@ -40,7 +41,7 @@ namespace OpenMM {
 /**
  * This class provides a multithreaded random number generator.
  */
-class OPENMM_EXPORT CpuRandom {
+class OPENMM_EXPORT_CPU CpuRandom {
 public:
     CpuRandom();
     ~CpuRandom();

platforms/cpu/include/CpuSETTLE.h

@@ -33,6 +33,7 @@
  * -------------------------------------------------------------------------- */
 
 #include "ReferenceSETTLEAlgorithm.h"
+#include "windowsExportCpu.h"
 #include "openmm/System.h"
 #include "openmm/internal/ThreadPool.h"
 #include <vector>
@@ -42,7 +43,7 @@ namespace OpenMM {
 /**
  * This class uses multiple ReferenceSETTLEAlgorithm objects to execute the algorithm in parallel.
  */
-class OPENMM_EXPORT CpuSETTLE : public ReferenceConstraintAlgorithm {
+class OPENMM_EXPORT_CPU CpuSETTLE : public ReferenceConstraintAlgorithm {
 public:
     class ApplyToPositionsTask;
     class ApplyToVelocitiesTask;

platforms/cpu/src/CpuGBSAOBCForce.cpp

@@ -26,7 +26,9 @@
 #include "SimTKOpenMMRealType.h"
 #include "openmm/internal/vectorize.h"
 #include "gmx_atomic.h"
+#include <algorithm>
 #include <cmath>
+#include <cstdlib>
 
 using namespace std;
 using namespace OpenMM;
@@ -391,7 +393,7 @@ void CpuGBSAOBCForce::getDeltaR(const fvec4& posI, const fvec4& x, const fvec4&
     r2 = dx*dx + dy*dy + dz*dz;
 }
 
-fvec4 CpuGBSAOBCForce::fastLog(fvec4 x) {
+fvec4 CpuGBSAOBCForce::fastLog(const fvec4& x) {
     // Evaluate log(x) using a lookup table for speed.
 
     if (any((x < TABLE_MIN) | (x >= TABLE_MAX)))

platforms/cpu/src/CpuNeighborList.cpp

@@ -154,7 +154,7 @@ public:
         return VoxelIndex(x, y);
     }
 
-    void getNeighbors(vector<int>& neighbors, int blockIndex, fvec4 blockCenter, fvec4 blockWidth, const vector<int>& sortedAtoms, vector<char>& exclusions, float maxDistance, const vector<int>& blockAtoms, const float* atomLocations, const vector<VoxelIndex>& atomVoxelIndex) const {
+    void getNeighbors(vector<int>& neighbors, int blockIndex, const fvec4& blockCenter, const fvec4& blockWidth, const vector<int>& sortedAtoms, vector<char>& exclusions, float maxDistance, const vector<int>& blockAtoms, const float* atomLocations, const vector<VoxelIndex>& atomVoxelIndex) const {
         neighbors.resize(0);
         exclusions.resize(0);
         fvec4 boxSize(periodicBoxSize[0], periodicBoxSize[1], periodicBoxSize[2], 0);

platforms/cpu/src/CpuNonbondedForce.cpp

@@ -30,6 +30,7 @@
 #include "ReferenceForce.h"
 #include "ReferencePME.h"
 #include "gmx_atomic.h"
+#include <algorithm>
 
 // In case we're using some primitive version of Visual Studio this will
 // make sure that erf() and erfc() are defined.
@@ -182,7 +183,7 @@ void CpuNonbondedForce::calculateReciprocalIxn(int numberOfAtoms, float* posq, c
 
     static const float epsilon     =  1.0;
 
-    int kmax                            = (ewald ? std::max(numRx, std::max(numRy,numRz)) : 0);
+    int kmax                       = (ewald ? max(numRx, max(numRy,numRz)) : 0);
     float factorEwald              = -1 / (4*alphaEwald*alphaEwald);
     float TWO_PI                   = 2.0 * PI_M;
     float recipCoeff               = (float)(ONE_4PI_EPS0*4*PI_M/(periodicBoxSize[0] * periodicBoxSize[1] * periodicBoxSize[2]) /epsilon);

platforms/cpu/src/CpuNonbondedForceVec4.cpp

@@ -273,7 +273,7 @@ void CpuNonbondedForceVec4::getDeltaR(const float* posI, const fvec4& x, const f
     r2 = dx*dx + dy*dy + dz*dz;
 }
 
-fvec4 CpuNonbondedForceVec4::erfcApprox(fvec4 x) {
+fvec4 CpuNonbondedForceVec4::erfcApprox(const fvec4& x) {
     // This approximation for erfc is from Abramowitz and Stegun (1964) p. 299.  They cite the following as
     // the original source: C. Hastings, Jr., Approximations for Digital Computers (1955).  It has a maximum
     // error of 3e-7.
@@ -285,7 +285,7 @@ fvec4 CpuNonbondedForceVec4::erfcApprox(fvec4 x) {
     return 1.0f/(t*t);
 }
 
-fvec4 CpuNonbondedForceVec4::ewaldScaleFunction(fvec4 x) {
+fvec4 CpuNonbondedForceVec4::ewaldScaleFunction(const fvec4& x) {
     // Compute the tabulated Ewald scale factor: erfc(alpha*r) + 2*alpha*r*exp(-alpha*alpha*r*r)/sqrt(PI)
 
     fvec4 x1 = x*ewaldDXInv;

platforms/cpu/src/CpuNonbondedForceVec8.cpp

@@ -25,6 +25,7 @@
 #include "SimTKOpenMMCommon.h"
 #include "SimTKOpenMMUtilities.h"
 #include "CpuNonbondedForceVec8.h"
+#include "openmm/OpenMMException.h"
 #include "openmm/internal/hardware.h"
 
 using namespace std;
@@ -296,7 +297,7 @@ void CpuNonbondedForceVec8::getDeltaR(const float* posI, const fvec8& x, const f
     r2 = dx*dx + dy*dy + dz*dz;
 }
 
-fvec8 CpuNonbondedForceVec8::erfcApprox(fvec8 x) {
+fvec8 CpuNonbondedForceVec8::erfcApprox(const fvec8& x) {
     // This approximation for erfc is from Abramowitz and Stegun (1964) p. 299.  They cite the following as
     // the original source: C. Hastings, Jr., Approximations for Digital Computers (1955).  It has a maximum
     // error of 3e-7.
@@ -308,7 +309,7 @@ fvec8 CpuNonbondedForceVec8::erfcApprox(fvec8 x) {
     return 1.0f/(t*t);
 }
 
-fvec8 CpuNonbondedForceVec8::ewaldScaleFunction(fvec8 x) {
+fvec8 CpuNonbondedForceVec8::ewaldScaleFunction(const fvec8& x) {
     // Compute the tabulated Ewald scale factor: erfc(alpha*r) + 2*alpha*r*exp(-alpha*alpha*r*r)/sqrt(PI)
 
     fvec8 x1 = x*ewaldDXInv;

platforms/reference/include/ReferenceKernelFactory.h

@@ -40,7 +40,7 @@ namespace OpenMM {
  * This KernelFactory creates all kernels for ReferencePlatform.
  */
 
-class ReferenceKernelFactory : public KernelFactory {
+class OPENMM_EXPORT ReferenceKernelFactory : public KernelFactory {
 public:
     KernelImpl* createKernelImpl(std::string name, const Platform& platform, ContextImpl& context) const;
 };

platforms/reference/include/ReferenceLJCoulomb14.h

@@ -26,10 +26,11 @@
 #define __ReferenceLJCoulomb14_H__
 
 #include "ReferenceBondIxn.h"
+#include "openmm/internal/windowsExport.h"
 
 // ---------------------------------------------------------------------------------------
 
-class ReferenceLJCoulomb14 : public ReferenceBondIxn {
+class OPENMM_EXPORT ReferenceLJCoulomb14 : public ReferenceBondIxn {
 
    public:
 

platforms/reference/include/ReferencePME.h

@@ -30,6 +30,7 @@
  */
 
 #include "SimTKOpenMMCommon.h"
+#include "openmm/internal/windowsExport.h"
 #include <vector>
 
 typedef RealOpenMM rvec[3];
@@ -52,7 +53,7 @@ pme_t;
  * pme_order   Interpolation order, almost always 4
  * epsilon_r   Dielectric coefficient, typically 1.0.
  */
-int 
+int OPENMM_EXPORT
 pme_init(pme_t *       ppme,
          RealOpenMM    ewaldcoeff,
          int           natoms,
@@ -73,7 +74,7 @@ pme_init(pme_t *       ppme,
  * energy      Total energy (will be written in units of kJ/mol)
  * pme_virial  Long-range part of the virial, output.
  */
-int
+int OPENMM_EXPORT
 pme_exec(pme_t       pme,
          const std::vector<OpenMM::RealVec>& atomCoordinates,
          std::vector<OpenMM::RealVec>& forces,
@@ -85,5 +86,5 @@ pme_exec(pme_t       pme,
 
 
 /* Release all memory in pme structure */
-int
+int OPENMM_EXPORT
 pme_destroy(pme_t    pme);

platforms/reference/include/ReferenceStochasticDynamics.h

@@ -26,10 +26,11 @@
 #define __ReferenceStochasticDynamics_H__
 
 #include "ReferenceDynamics.h"
+#include "openmm/internal/windowsExport.h"
 
 // ---------------------------------------------------------------------------------------
 
-class ReferenceStochasticDynamics : public ReferenceDynamics {
+class OPENMM_EXPORT ReferenceStochasticDynamics : public ReferenceDynamics {
 
    protected:
 

plugins/amoeba/serialization/include/openmm/serialization/AmoebaAngleForceProxy.h

@@ -32,7 +32,7 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
 
-#include "openmm/serialization/internal/windowsExportAmoebaSerialization.h"
+#include "openmm/internal/windowsExportAmoeba.h"
 #include "openmm/serialization/SerializationProxy.h"
 
 namespace OpenMM {
@@ -41,7 +41,7 @@ namespace OpenMM {
  * This is a proxy for serializing AmoebaAngleForce objects.
  */
 
-class OPENMM_EXPORT_AMOEBA_SERIALIZATION AmoebaAngleForceProxy : public SerializationProxy {
+class OPENMM_EXPORT_AMOEBA AmoebaAngleForceProxy : public SerializationProxy {
 public:
     AmoebaAngleForceProxy();
     void serialize(const void* object, SerializationNode& node) const;

plugins/amoeba/serialization/include/openmm/serialization/AmoebaBondForceProxy.h

@@ -32,7 +32,7 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.                                     *
  * -------------------------------------------------------------------------- */
 
-#include "openmm/serialization/internal/windowsExportAmoebaSerialization.h"
+#include "openmm/internal/windowsExportAmoeba.h"
 #include "openmm/serialization/SerializationProxy.h"
 
 namespace OpenMM {
@@ -41,7 +41,7 @@ namespace OpenMM {
  * This is a proxy for serializing AmoebaBondForce objects.
  */
 
-class OPENMM_EXPORT_AMOEBA_SERIALIZATION AmoebaBondForceProxy : public SerializationProxy {
+class OPENMM_EXPORT_AMOEBA AmoebaBondForceProxy : public SerializationProxy {
 public:
     AmoebaBondForceProxy();
     void serialize(const void* object, SerializationNode& node) const;