Bug fixes to PNaCl vectors

openmmapi/include/openmm/internal/vectorize_pnacl.h

@@ -44,8 +44,8 @@ static bool isVec4Supported() {
     return true;
 }
 
-typedef float __m128 __attribute__((vector_size(16)));
-typedef int __m128i __attribute__((vector_size(16)));
+typedef float __m128 __attribute__((vector_size(16), aligned(4)));
+typedef int __m128i __attribute__((vector_size(16), aligned(4)));
 
 class ivec4;
 
@@ -280,7 +280,7 @@ static inline ivec4 abs(const ivec4& v) {
     return ivec4(abs(v[0]), abs(v[1]), abs(v[2]), abs(v[3]));
 }
 
-static inline bool any(const ivec4& v) {
+static inline bool any(const __m128i& v) {
     return (v[0] || v[1] || v[2] || v[3]);
 }
 
@@ -304,7 +304,7 @@ static inline fvec4 operator/(float v1, const fvec4& v2) {
 
 // Operations for blending fvec4s based on an ivec4.
 
-static inline fvec4 blend(const fvec4& v1, const fvec4& v2, const ivec4& mask) {
+static inline fvec4 blend(const fvec4& v1, const fvec4& v2, const __m128i& mask) {
     return fvec4(mask[0] ? v2[0] : v1[0], mask[1] ? v2[1] : v1[1], mask[2] ? v2[2] : v1[2], mask[3] ? v2[3] : v1[3]);
 }
 

platforms/cpu/src/CpuGBSAOBCForce.cpp

@@ -410,18 +410,10 @@ 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

@@ -292,18 +292,10 @@ 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;
 }