Merge branch 'master' into hardwall

CMakeLists.txt

@@ -82,7 +82,7 @@ ENDIF(${CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT})
 
 # The source is organized into subdirectories, but we handle them all from
 # this CMakeLists file rather than letting CMake visit them as SUBDIRS.
-SET(OPENMM_SOURCE_SUBDIRS . openmmapi olla libraries/jama libraries/quern libraries/lepton libraries/sfmt libraries/lbfgs libraries/hilbert libraries/csha1 platforms/reference serialization libraries/validate libraries/irrxml)
+SET(OPENMM_SOURCE_SUBDIRS . openmmapi olla libraries/jama libraries/quern libraries/lepton libraries/sfmt libraries/lbfgs libraries/hilbert libraries/csha1 platforms/reference serialization libraries/validate libraries/irrxml libraries/vecmath)
 IF(WIN32)
     SET(OPENMM_SOURCE_SUBDIRS ${OPENMM_SOURCE_SUBDIRS} libraries/pthreads)
 ELSE(WIN32)
@@ -159,8 +159,8 @@ ENDIF (NOT CMAKE_CXX_FLAGS_RELEASE)
 # and make it available to the code so it can be built into the binaries.
 
 SET(OPENMM_LIBRARY_NAME OpenMM)
-SET(OPENMM_MAJOR_VERSION 6)
-SET(OPENMM_MINOR_VERSION 3)
+SET(OPENMM_MAJOR_VERSION 7)
+SET(OPENMM_MINOR_VERSION 0)
 SET(OPENMM_BUILD_VERSION 0)
 
 SET(OPENMM_COPYRIGHT_YEARS "2008-2015")
@@ -340,6 +340,7 @@ ELSE(DL_LIBRARY)
 ENDIF(DL_LIBRARY)
 
 IF(BUILD_TESTING)
+    INCLUDE_DIRECTORIES(${CMAKE_SOURCE_DIR}/tests)
     ADD_SUBDIRECTORY(platforms/reference/tests)
 ENDIF(BUILD_TESTING)
 

devtools/forcefield-scripts/processTinkerForceField.py

@@ -1017,7 +1017,6 @@ if( isAmoeba ):
     torsionTorsionUnit      = 1.0
     outputString            = """ <AmoebaTorsionTorsionForce >"""
     tinkerXmlFile.write( "%s\n" % (outputString ) )
-    conversion              = 41.84/radian
     torsionTorsions         = forces['tortors']
     for (index, torsionTorsion) in enumerate(torsionTorsions):
        torInfo       = torsionTorsion[0]

libraries/sfmt/src/SFMT.cpp

@@ -124,11 +124,13 @@ public:
 };
 
 void SFMT::createCheckpoint(std::ostream& stream) {
+    stream.write((char*) &data->baseData, sizeof(data->baseData));
     stream.write((char*) &data->sfmt, sizeof(data->sfmt));
     stream.write((char*) &data->idx, sizeof(data->idx));
 }
 
 void SFMT::loadCheckpoint(std::istream& stream) {
+    stream.read((char*) &data->baseData, sizeof(data->baseData));
     stream.read((char*) &data->sfmt, sizeof(data->sfmt));
     stream.read((char*) &data->idx, sizeof(data->idx));
 }

libraries/vecmath/include/neon_mathfun.h

+/* NEON implementation of sin, cos, exp and log
+
+   Inspired by Intel Approximate Math library, and based on the
+   corresponding algorithms of the cephes math library
+*/
+
+/* Copyright (C) 2011  Julien Pommier
+
+  This software is provided 'as-is', without any express or implied
+  warranty.  In no event will the authors be held liable for any damages
+  arising from the use of this software.
+
+  Permission is granted to anyone to use this software for any purpose,
+  including commercial applications, and to alter it and redistribute it
+  freely, subject to the following restrictions:
+
+  1. The origin of this software must not be misrepresented; you must not
+     claim that you wrote the original software. If you use this software
+     in a product, an acknowledgment in the product documentation would be
+     appreciated but is not required.
+  2. Altered source versions must be plainly marked as such, and must not be
+     misrepresented as being the original software.
+  3. This notice may not be removed or altered from any source distribution.
+
+  (this is the zlib license)
+*/
+
+#include <arm_neon.h>
+#include "openmm/internal/windowsExport.h"
+
+typedef float32x4_t v4sf;  // vector of 4 float
+typedef uint32x4_t v4su;  // vector of 4 uint32
+typedef int32x4_t v4si;  // vector of 4 uint32
+
+#define c_inv_mant_mask ~0x7f800000u
+#define c_cephes_SQRTHF 0.707106781186547524
+#define c_cephes_log_p0 7.0376836292E-2
+#define c_cephes_log_p1 - 1.1514610310E-1
+#define c_cephes_log_p2 1.1676998740E-1
+#define c_cephes_log_p3 - 1.2420140846E-1
+#define c_cephes_log_p4 + 1.4249322787E-1
+#define c_cephes_log_p5 - 1.6668057665E-1
+#define c_cephes_log_p6 + 2.0000714765E-1
+#define c_cephes_log_p7 - 2.4999993993E-1
+#define c_cephes_log_p8 + 3.3333331174E-1
+#define c_cephes_log_q1 -2.12194440e-4
+#define c_cephes_log_q2 0.693359375
+
+/* natural logarithm computed for 4 simultaneous float 
+   return NaN for x <= 0
+*/
+OPENMM_EXPORT v4sf log_ps(v4sf x) {
+  v4sf one = vdupq_n_f32(1);
+
+  x = vmaxq_f32(x, vdupq_n_f32(0)); /* force flush to zero on denormal values */
+  v4su invalid_mask = vcleq_f32(x, vdupq_n_f32(0));
+
+  v4si ux = vreinterpretq_s32_f32(x);
+  
+  v4si emm0 = vshrq_n_s32(ux, 23);
+
+  /* keep only the fractional part */
+  ux = vandq_s32(ux, vdupq_n_s32(c_inv_mant_mask));
+  ux = vorrq_s32(ux, vreinterpretq_s32_f32(vdupq_n_f32(0.5f)));
+  x = vreinterpretq_f32_s32(ux);
+
+  emm0 = vsubq_s32(emm0, vdupq_n_s32(0x7f));
+  v4sf e = vcvtq_f32_s32(emm0);
+
+  e = vaddq_f32(e, one);
+
+  /* part2: 
+     if( x < SQRTHF ) {
+       e -= 1;
+       x = x + x - 1.0;
+     } else { x = x - 1.0; }
+  */
+  v4su mask = vcltq_f32(x, vdupq_n_f32(c_cephes_SQRTHF));
+  v4sf tmp = vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(x), mask));
+  x = vsubq_f32(x, one);
+  e = vsubq_f32(e, vreinterpretq_f32_u32(vandq_u32(vreinterpretq_u32_f32(one), mask)));
+  x = vaddq_f32(x, tmp);
+
+  v4sf z = vmulq_f32(x,x);
+
+  v4sf y = vdupq_n_f32(c_cephes_log_p0);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p1));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p2));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p3));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p4));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p5));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p6));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p7));
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, vdupq_n_f32(c_cephes_log_p8));
+  y = vmulq_f32(y, x);
+
+  y = vmulq_f32(y, z);
+  
+
+  tmp = vmulq_f32(e, vdupq_n_f32(c_cephes_log_q1));
+  y = vaddq_f32(y, tmp);
+
+
+  tmp = vmulq_f32(z, vdupq_n_f32(0.5f));
+  y = vsubq_f32(y, tmp);
+
+  tmp = vmulq_f32(e, vdupq_n_f32(c_cephes_log_q2));
+  x = vaddq_f32(x, y);
+  x = vaddq_f32(x, tmp);
+  x = vreinterpretq_f32_u32(vorrq_u32(vreinterpretq_u32_f32(x), invalid_mask)); // negative arg will be NAN
+  return x;
+}
+
+#define c_exp_hi 88.3762626647949f
+#define c_exp_lo -88.3762626647949f
+
+#define c_cephes_LOG2EF 1.44269504088896341
+#define c_cephes_exp_C1 0.693359375
+#define c_cephes_exp_C2 -2.12194440e-4
+
+#define c_cephes_exp_p0 1.9875691500E-4
+#define c_cephes_exp_p1 1.3981999507E-3
+#define c_cephes_exp_p2 8.3334519073E-3
+#define c_cephes_exp_p3 4.1665795894E-2
+#define c_cephes_exp_p4 1.6666665459E-1
+#define c_cephes_exp_p5 5.0000001201E-1
+
+/* exp() computed for 4 float at once */
+OPENMM_EXPORT v4sf exp_ps(v4sf x) {
+  v4sf tmp, fx;
+
+  v4sf one = vdupq_n_f32(1);
+  x = vminq_f32(x, vdupq_n_f32(c_exp_hi));
+  x = vmaxq_f32(x, vdupq_n_f32(c_exp_lo));
+
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = vmlaq_f32(vdupq_n_f32(0.5f), x, vdupq_n_f32(c_cephes_LOG2EF));
+
+  /* perform a floorf */
+  tmp = vcvtq_f32_s32(vcvtq_s32_f32(fx));
+
+  /* if greater, substract 1 */
+  v4su mask = vcgtq_f32(tmp, fx);    
+  mask = vandq_u32(mask, vreinterpretq_u32_f32(one));
+
+
+  fx = vsubq_f32(tmp, vreinterpretq_f32_u32(mask));
+
+  tmp = vmulq_f32(fx, vdupq_n_f32(c_cephes_exp_C1));
+  v4sf z = vmulq_f32(fx, vdupq_n_f32(c_cephes_exp_C2));
+  x = vsubq_f32(x, tmp);
+  x = vsubq_f32(x, z);
+
+  static const float cephes_exp_p[6] = { c_cephes_exp_p0, c_cephes_exp_p1, c_cephes_exp_p2, c_cephes_exp_p3, c_cephes_exp_p4, c_cephes_exp_p5 };
+  v4sf y = vld1q_dup_f32(cephes_exp_p+0);
+  v4sf c1 = vld1q_dup_f32(cephes_exp_p+1); 
+  v4sf c2 = vld1q_dup_f32(cephes_exp_p+2); 
+  v4sf c3 = vld1q_dup_f32(cephes_exp_p+3); 
+  v4sf c4 = vld1q_dup_f32(cephes_exp_p+4); 
+  v4sf c5 = vld1q_dup_f32(cephes_exp_p+5);
+
+  y = vmulq_f32(y, x);
+  z = vmulq_f32(x,x);
+  y = vaddq_f32(y, c1);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, c2);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, c3);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, c4);
+  y = vmulq_f32(y, x);
+  y = vaddq_f32(y, c5);
+  
+  y = vmulq_f32(y, z);
+  y = vaddq_f32(y, x);
+  y = vaddq_f32(y, one);
+
+  /* build 2^n */
+  int32x4_t mm;
+  mm = vcvtq_s32_f32(fx);
+  mm = vaddq_s32(mm, vdupq_n_s32(0x7f));
+  mm = vshlq_n_s32(mm, 23);
+  v4sf pow2n = vreinterpretq_f32_s32(mm);
+
+  y = vmulq_f32(y, pow2n);
+  return y;
+}
+
+#define c_minus_cephes_DP1 -0.78515625
+#define c_minus_cephes_DP2 -2.4187564849853515625e-4
+#define c_minus_cephes_DP3 -3.77489497744594108e-8
+#define c_sincof_p0 -1.9515295891E-4
+#define c_sincof_p1  8.3321608736E-3
+#define c_sincof_p2 -1.6666654611E-1
+#define c_coscof_p0  2.443315711809948E-005
+#define c_coscof_p1 -1.388731625493765E-003
+#define c_coscof_p2  4.166664568298827E-002
+#define c_cephes_FOPI 1.27323954473516 // 4 / M_PI
+
+/* evaluation of 4 sines & cosines at once.
+
+   The code is the exact rewriting of the cephes sinf function.
+   Precision is excellent as long as x < 8192 (I did not bother to
+   take into account the special handling they have for greater values
+   -- it does not return garbage for arguments over 8192, though, but
+   the extra precision is missing).
+
+   Note that it is such that sinf((float)M_PI) = 8.74e-8, which is the
+   surprising but correct result.
+
+   Note also that when you compute sin(x), cos(x) is available at
+   almost no extra price so both sin_ps and cos_ps make use of
+   sincos_ps..
+  */
+OPENMM_EXPORT void sincos_ps(v4sf x, v4sf *ysin, v4sf *ycos) { // any x
+  v4sf xmm1, xmm2, xmm3, y;
+
+  v4su emm2;
+  
+  v4su sign_mask_sin, sign_mask_cos;
+  sign_mask_sin = vcltq_f32(x, vdupq_n_f32(0));
+  x = vabsq_f32(x);
+
+  /* scale by 4/Pi */
+  y = vmulq_f32(x, vdupq_n_f32(c_cephes_FOPI));
+
+  /* store the integer part of y in mm0 */
+  emm2 = vcvtq_u32_f32(y);
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  emm2 = vaddq_u32(emm2, vdupq_n_u32(1));
+  emm2 = vandq_u32(emm2, vdupq_n_u32(~1));
+  y = vcvtq_f32_u32(emm2);
+
+  /* get the polynom selection mask 
+     there is one polynom for 0 <= x <= Pi/4
+     and another one for Pi/4<x<=Pi/2
+
+     Both branches will be computed.
+  */
+  v4su poly_mask = vtstq_u32(emm2, vdupq_n_u32(2));
+  
+  /* The magic pass: "Extended precision modular arithmetic" 
+     x = ((x - y * DP1) - y * DP2) - y * DP3; */
+  xmm1 = vmulq_n_f32(y, c_minus_cephes_DP1);
+  xmm2 = vmulq_n_f32(y, c_minus_cephes_DP2);
+  xmm3 = vmulq_n_f32(y, c_minus_cephes_DP3);
+  x = vaddq_f32(x, xmm1);
+  x = vaddq_f32(x, xmm2);
+  x = vaddq_f32(x, xmm3);
+
+  sign_mask_sin = veorq_u32(sign_mask_sin, vtstq_u32(emm2, vdupq_n_u32(4)));
+  sign_mask_cos = vtstq_u32(vsubq_u32(emm2, vdupq_n_u32(2)), vdupq_n_u32(4));
+
+  /* Evaluate the first polynom  (0 <= x <= Pi/4) in y1, 
+     and the second polynom      (Pi/4 <= x <= 0) in y2 */
+  v4sf z = vmulq_f32(x,x);
+  v4sf y1, y2;
+
+  y1 = vmulq_n_f32(z, c_coscof_p0);
+  y2 = vmulq_n_f32(z, c_sincof_p0);
+  y1 = vaddq_f32(y1, vdupq_n_f32(c_coscof_p1));
+  y2 = vaddq_f32(y2, vdupq_n_f32(c_sincof_p1));
+  y1 = vmulq_f32(y1, z);
+  y2 = vmulq_f32(y2, z);
+  y1 = vaddq_f32(y1, vdupq_n_f32(c_coscof_p2));
+  y2 = vaddq_f32(y2, vdupq_n_f32(c_sincof_p2));
+  y1 = vmulq_f32(y1, z);
+  y2 = vmulq_f32(y2, z);
+  y1 = vmulq_f32(y1, z);
+  y2 = vmulq_f32(y2, x);
+  y1 = vsubq_f32(y1, vmulq_f32(z, vdupq_n_f32(0.5f)));
+  y2 = vaddq_f32(y2, x);
+  y1 = vaddq_f32(y1, vdupq_n_f32(1));
+
+  /* select the correct result from the two polynoms */  
+  v4sf ys = vbslq_f32(poly_mask, y1, y2);
+  v4sf yc = vbslq_f32(poly_mask, y2, y1);
+  *ysin = vbslq_f32(sign_mask_sin, vnegq_f32(ys), ys);
+  *ycos = vbslq_f32(sign_mask_cos, yc, vnegq_f32(yc));
+}
+
+OPENMM_EXPORT v4sf sin_ps(v4sf x) {
+  v4sf ysin, ycos; 
+  sincos_ps(x, &ysin, &ycos); 
+  return ysin;
+}
+
+OPENMM_EXPORT v4sf cos_ps(v4sf x) {
+  v4sf ysin, ycos; 
+  sincos_ps(x, &ysin, &ycos); 
+  return ycos;
+}
+
+

libraries/vecmath/include/sse_mathfun.h

+/* SIMD (SSE1+MMX or SSE2) implementation of sin, cos, exp and log
+
+   Inspired by Intel Approximate Math library, and based on the
+   corresponding algorithms of the cephes math library
+
+   The default is to use the SSE1 version. If you define USE_SSE2 the
+   the SSE2 intrinsics will be used in place of the MMX intrinsics. Do
+   not expect any significant performance improvement with SSE2.
+*/
+
+/* Copyright (C) 2007  Julien Pommier
+
+  This software is provided 'as-is', without any express or implied
+  warranty.  In no event will the authors be held liable for any damages
+  arising from the use of this software.
+
+  Permission is granted to anyone to use this software for any purpose,
+  including commercial applications, and to alter it and redistribute it
+  freely, subject to the following restrictions:
+
+  1. The origin of this software must not be misrepresented; you must not
+     claim that you wrote the original software. If you use this software
+     in a product, an acknowledgment in the product documentation would be
+     appreciated but is not required.
+  2. Altered source versions must be plainly marked as such, and must not be
+     misrepresented as being the original software.
+  3. This notice may not be removed or altered from any source distribution.
+
+  (this is the zlib license)
+*/
+
+#include <xmmintrin.h>
+#include "openmm/internal/windowsExport.h"
+
+/* yes I know, the top of this file is quite ugly */
+
+#ifdef _MSC_VER /* visual c++ */
+# define ALIGN16_BEG __declspec(align(16))
+# define ALIGN16_END 
+#else /* gcc or icc */
+# define ALIGN16_BEG
+# define ALIGN16_END __attribute__((aligned(16)))
+#endif
+
+/* __m128 is ugly to write */
+typedef __m128 v4sf;  // vector of 4 float (sse1)
+
+#ifdef USE_SSE2
+# include <emmintrin.h>
+typedef __m128i v4si; // vector of 4 int (sse2)
+#else
+typedef __m64 v2si;   // vector of 2 int (mmx)
+#endif
+
+/* declare some SSE constants -- why can't I figure a better way to do that? */
+#define _PS_CONST(Name, Val)                                            \
+  static const ALIGN16_BEG float _ps_##Name[4] ALIGN16_END = { Val, Val, Val, Val }
+#define _PI32_CONST(Name, Val)                                            \
+  static const ALIGN16_BEG int _pi32_##Name[4] ALIGN16_END = { Val, Val, Val, Val }
+#define _PS_CONST_TYPE(Name, Type, Val)                                 \
+  static const ALIGN16_BEG Type _ps_##Name[4] ALIGN16_END = { Val, Val, Val, Val }
+
+_PS_CONST(1  , 1.0f);
+_PS_CONST(0p5, 0.5f);
+/* the smallest non denormalized float number */
+_PS_CONST_TYPE(min_norm_pos, int, 0x00800000);
+_PS_CONST_TYPE(mant_mask, int, 0x7f800000);
+_PS_CONST_TYPE(inv_mant_mask, int, ~0x7f800000);
+
+_PS_CONST_TYPE(sign_mask, int, (int)0x80000000);
+_PS_CONST_TYPE(inv_sign_mask, int, ~0x80000000);
+
+_PI32_CONST(1, 1);
+_PI32_CONST(inv1, ~1);
+_PI32_CONST(2, 2);
+_PI32_CONST(4, 4);
+_PI32_CONST(0x7f, 0x7f);
+
+_PS_CONST(cephes_SQRTHF, 0.707106781186547524);
+_PS_CONST(cephes_log_p0, 7.0376836292E-2);
+_PS_CONST(cephes_log_p1, - 1.1514610310E-1);
+_PS_CONST(cephes_log_p2, 1.1676998740E-1);
+_PS_CONST(cephes_log_p3, - 1.2420140846E-1);
+_PS_CONST(cephes_log_p4, + 1.4249322787E-1);
+_PS_CONST(cephes_log_p5, - 1.6668057665E-1);
+_PS_CONST(cephes_log_p6, + 2.0000714765E-1);
+_PS_CONST(cephes_log_p7, - 2.4999993993E-1);
+_PS_CONST(cephes_log_p8, + 3.3333331174E-1);
+_PS_CONST(cephes_log_q1, -2.12194440e-4);
+_PS_CONST(cephes_log_q2, 0.693359375);
+
+#ifndef USE_SSE2
+typedef union xmm_mm_union {
+  __m128 xmm;
+  __m64 mm[2];
+} xmm_mm_union;
+
+#define COPY_XMM_TO_MM(xmm_, mm0_, mm1_) {          \
+    xmm_mm_union u; u.xmm = xmm_;                   \
+    mm0_ = u.mm[0];                                 \
+    mm1_ = u.mm[1];                                 \
+}
+
+#define COPY_MM_TO_XMM(mm0_, mm1_, xmm_) {                         \
+    xmm_mm_union u; u.mm[0]=mm0_; u.mm[1]=mm1_; xmm_ = u.xmm;      \
+  }
+
+#endif // USE_SSE2
+
+/* natural logarithm computed for 4 simultaneous float 
+   return NaN for x <= 0
+*/
+OPENMM_EXPORT v4sf log_ps(v4sf x) {
+#ifdef USE_SSE2
+  v4si emm0;
+#else
+  v2si mm0, mm1;
+#endif
+  v4sf one = *(v4sf*)_ps_1;
+
+  v4sf invalid_mask = _mm_cmple_ps(x, _mm_setzero_ps());
+
+  x = _mm_max_ps(x, *(v4sf*)_ps_min_norm_pos);  /* cut off denormalized stuff */
+
+#ifndef USE_SSE2
+  /* part 1: x = frexpf(x, &e); */
+  COPY_XMM_TO_MM(x, mm0, mm1);
+  mm0 = _mm_srli_pi32(mm0, 23);
+  mm1 = _mm_srli_pi32(mm1, 23);
+#else
+  emm0 = _mm_srli_epi32(_mm_castps_si128(x), 23);
+#endif
+  /* keep only the fractional part */
+  x = _mm_and_ps(x, *(v4sf*)_ps_inv_mant_mask);
+  x = _mm_or_ps(x, *(v4sf*)_ps_0p5);
+
+#ifndef USE_SSE2
+  /* now e=mm0:mm1 contain the really base-2 exponent */
+  mm0 = _mm_sub_pi32(mm0, *(v2si*)_pi32_0x7f);
+  mm1 = _mm_sub_pi32(mm1, *(v2si*)_pi32_0x7f);
+  v4sf e = _mm_cvtpi32x2_ps(mm0, mm1);
+  _mm_empty(); /* bye bye mmx */
+#else
+  emm0 = _mm_sub_epi32(emm0, *(v4si*)_pi32_0x7f);
+  v4sf e = _mm_cvtepi32_ps(emm0);
+#endif
+
+  e = _mm_add_ps(e, one);
+
+  /* part2: 
+     if( x < SQRTHF ) {
+       e -= 1;
+       x = x + x - 1.0;
+     } else { x = x - 1.0; }
+  */
+  v4sf mask = _mm_cmplt_ps(x, *(v4sf*)_ps_cephes_SQRTHF);
+  v4sf tmp = _mm_and_ps(x, mask);
+  x = _mm_sub_ps(x, one);
+  e = _mm_sub_ps(e, _mm_and_ps(one, mask));
+  x = _mm_add_ps(x, tmp);
+
+
+  v4sf z = _mm_mul_ps(x,x);
+
+  v4sf y = *(v4sf*)_ps_cephes_log_p0;
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p1);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p2);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p3);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p4);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p5);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p6);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p7);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_log_p8);
+  y = _mm_mul_ps(y, x);
+
+  y = _mm_mul_ps(y, z);
+  
+
+  tmp = _mm_mul_ps(e, *(v4sf*)_ps_cephes_log_q1);
+  y = _mm_add_ps(y, tmp);
+
+
+  tmp = _mm_mul_ps(z, *(v4sf*)_ps_0p5);
+  y = _mm_sub_ps(y, tmp);
+
+  tmp = _mm_mul_ps(e, *(v4sf*)_ps_cephes_log_q2);
+  x = _mm_add_ps(x, y);
+  x = _mm_add_ps(x, tmp);
+  x = _mm_or_ps(x, invalid_mask); // negative arg will be NAN
+  return x;
+}
+
+_PS_CONST(exp_hi,	88.3762626647949f);
+_PS_CONST(exp_lo,	-88.3762626647949f);
+
+_PS_CONST(cephes_LOG2EF, 1.44269504088896341);
+_PS_CONST(cephes_exp_C1, 0.693359375);
+_PS_CONST(cephes_exp_C2, -2.12194440e-4);
+
+_PS_CONST(cephes_exp_p0, 1.9875691500E-4);
+_PS_CONST(cephes_exp_p1, 1.3981999507E-3);
+_PS_CONST(cephes_exp_p2, 8.3334519073E-3);
+_PS_CONST(cephes_exp_p3, 4.1665795894E-2);
+_PS_CONST(cephes_exp_p4, 1.6666665459E-1);
+_PS_CONST(cephes_exp_p5, 5.0000001201E-1);
+
+OPENMM_EXPORT v4sf exp_ps(v4sf x) {
+  v4sf tmp = _mm_setzero_ps(), fx;
+#ifdef USE_SSE2
+  v4si emm0;
+#else
+  v2si mm0, mm1;
+#endif
+  v4sf one = *(v4sf*)_ps_1;
+
+  x = _mm_min_ps(x, *(v4sf*)_ps_exp_hi);
+  x = _mm_max_ps(x, *(v4sf*)_ps_exp_lo);
+
+  /* express exp(x) as exp(g + n*log(2)) */
+  fx = _mm_mul_ps(x, *(v4sf*)_ps_cephes_LOG2EF);
+  fx = _mm_add_ps(fx, *(v4sf*)_ps_0p5);
+
+  /* how to perform a floorf with SSE: just below */
+#ifndef USE_SSE2
+  /* step 1 : cast to int */
+  tmp = _mm_movehl_ps(tmp, fx);
+  mm0 = _mm_cvttps_pi32(fx);
+  mm1 = _mm_cvttps_pi32(tmp);
+  /* step 2 : cast back to float */
+  tmp = _mm_cvtpi32x2_ps(mm0, mm1);
+#else
+  emm0 = _mm_cvttps_epi32(fx);
+  tmp  = _mm_cvtepi32_ps(emm0);
+#endif
+  /* if greater, substract 1 */
+  v4sf mask = _mm_cmpgt_ps(tmp, fx);    
+  mask = _mm_and_ps(mask, one);
+  fx = _mm_sub_ps(tmp, mask);
+
+  tmp = _mm_mul_ps(fx, *(v4sf*)_ps_cephes_exp_C1);
+  v4sf z = _mm_mul_ps(fx, *(v4sf*)_ps_cephes_exp_C2);
+  x = _mm_sub_ps(x, tmp);
+  x = _mm_sub_ps(x, z);
+
+  z = _mm_mul_ps(x,x);
+  
+  v4sf y = *(v4sf*)_ps_cephes_exp_p0;
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_exp_p1);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_exp_p2);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_exp_p3);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_exp_p4);
+  y = _mm_mul_ps(y, x);
+  y = _mm_add_ps(y, *(v4sf*)_ps_cephes_exp_p5);
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, x);
+  y = _mm_add_ps(y, one);
+
+  /* build 2^n */
+#ifndef USE_SSE2
+  z = _mm_movehl_ps(z, fx);
+  mm0 = _mm_cvttps_pi32(fx);
+  mm1 = _mm_cvttps_pi32(z);
+  mm0 = _mm_add_pi32(mm0, *(v2si*)_pi32_0x7f);
+  mm1 = _mm_add_pi32(mm1, *(v2si*)_pi32_0x7f);
+  mm0 = _mm_slli_pi32(mm0, 23); 
+  mm1 = _mm_slli_pi32(mm1, 23);
+  
+  v4sf pow2n; 
+  COPY_MM_TO_XMM(mm0, mm1, pow2n);
+  _mm_empty();
+#else
+  emm0 = _mm_cvttps_epi32(fx);
+  emm0 = _mm_add_epi32(emm0, *(v4si*)_pi32_0x7f);
+  emm0 = _mm_slli_epi32(emm0, 23);
+  v4sf pow2n = _mm_castsi128_ps(emm0);
+#endif
+  y = _mm_mul_ps(y, pow2n);
+  return y;
+}
+
+_PS_CONST(minus_cephes_DP1, -0.78515625);
+_PS_CONST(minus_cephes_DP2, -2.4187564849853515625e-4);
+_PS_CONST(minus_cephes_DP3, -3.77489497744594108e-8);
+_PS_CONST(sincof_p0, -1.9515295891E-4);
+_PS_CONST(sincof_p1,  8.3321608736E-3);
+_PS_CONST(sincof_p2, -1.6666654611E-1);
+_PS_CONST(coscof_p0,  2.443315711809948E-005);
+_PS_CONST(coscof_p1, -1.388731625493765E-003);
+_PS_CONST(coscof_p2,  4.166664568298827E-002);
+_PS_CONST(cephes_FOPI, 1.27323954473516); // 4 / M_PI
+
+
+/* evaluation of 4 sines at onces, using only SSE1+MMX intrinsics so
+   it runs also on old athlons XPs and the pentium III of your grand
+   mother.
+
+   The code is the exact rewriting of the cephes sinf function.
+   Precision is excellent as long as x < 8192 (I did not bother to
+   take into account the special handling they have for greater values
+   -- it does not return garbage for arguments over 8192, though, but
+   the extra precision is missing).
+
+   Note that it is such that sinf((float)M_PI) = 8.74e-8, which is the
+   surprising but correct result.
+
+   Performance is also surprisingly good, 1.33 times faster than the
+   macos vsinf SSE2 function, and 1.5 times faster than the
+   __vrs4_sinf of amd's ACML (which is only available in 64 bits). Not
+   too bad for an SSE1 function (with no special tuning) !
+   However the latter libraries probably have a much better handling of NaN,
+   Inf, denormalized and other special arguments..
+
+   On my core 1 duo, the execution of this function takes approximately 95 cycles.
+
+   From what I have observed on the experiments with Intel AMath lib, switching to an
+   SSE2 version would improve the perf by only 10%.
+
+   Since it is based on SSE intrinsics, it has to be compiled at -O2 to
+   deliver full speed.
+*/
+OPENMM_EXPORT v4sf sin_ps(v4sf x) { // any x
+  v4sf xmm1, xmm2 = _mm_setzero_ps(), xmm3, sign_bit, y;
+
+#ifdef USE_SSE2
+  v4si emm0, emm2;
+#else
+  v2si mm0, mm1, mm2, mm3;
+#endif
+  sign_bit = x;
+  /* take the absolute value */
+  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
+  /* extract the sign bit (upper one) */
+  sign_bit = _mm_and_ps(sign_bit, *(v4sf*)_ps_sign_mask);
+  
+  /* scale by 4/Pi */
+  y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
+
+#ifdef USE_SSE2
+  /* store the integer part of y in mm0 */
+  emm2 = _mm_cvttps_epi32(y);
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
+  emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
+  y = _mm_cvtepi32_ps(emm2);
+
+  /* get the swap sign flag */
+  emm0 = _mm_and_si128(emm2, *(v4si*)_pi32_4);
+  emm0 = _mm_slli_epi32(emm0, 29);
+  /* get the polynom selection mask 
+     there is one polynom for 0 <= x <= Pi/4
+     and another one for Pi/4<x<=Pi/2
+
+     Both branches will be computed.
+  */
+  emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_2);
+  emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
+  
+  v4sf swap_sign_bit = _mm_castsi128_ps(emm0);
+  v4sf poly_mask = _mm_castsi128_ps(emm2);
+  sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit);
+  
+#else
+  /* store the integer part of y in mm0:mm1 */
+  xmm2 = _mm_movehl_ps(xmm2, y);
+  mm2 = _mm_cvttps_pi32(y);
+  mm3 = _mm_cvttps_pi32(xmm2);
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  mm2 = _mm_add_pi32(mm2, *(v2si*)_pi32_1);
+  mm3 = _mm_add_pi32(mm3, *(v2si*)_pi32_1);
+  mm2 = _mm_and_si64(mm2, *(v2si*)_pi32_inv1);
+  mm3 = _mm_and_si64(mm3, *(v2si*)_pi32_inv1);
+  y = _mm_cvtpi32x2_ps(mm2, mm3);
+  /* get the swap sign flag */
+  mm0 = _mm_and_si64(mm2, *(v2si*)_pi32_4);
+  mm1 = _mm_and_si64(mm3, *(v2si*)_pi32_4);
+  mm0 = _mm_slli_pi32(mm0, 29);
+  mm1 = _mm_slli_pi32(mm1, 29);
+  /* get the polynom selection mask */
+  mm2 = _mm_and_si64(mm2, *(v2si*)_pi32_2);
+  mm3 = _mm_and_si64(mm3, *(v2si*)_pi32_2);
+  mm2 = _mm_cmpeq_pi32(mm2, _mm_setzero_si64());
+  mm3 = _mm_cmpeq_pi32(mm3, _mm_setzero_si64());
+  v4sf swap_sign_bit, poly_mask;
+  COPY_MM_TO_XMM(mm0, mm1, swap_sign_bit);
+  COPY_MM_TO_XMM(mm2, mm3, poly_mask);
+  sign_bit = _mm_xor_ps(sign_bit, swap_sign_bit);
+  _mm_empty(); /* good-bye mmx */
+#endif
+  
+  /* The magic pass: "Extended precision modular arithmetic" 
+     x = ((x - y * DP1) - y * DP2) - y * DP3; */
+  xmm1 = *(v4sf*)_ps_minus_cephes_DP1;
+  xmm2 = *(v4sf*)_ps_minus_cephes_DP2;
+  xmm3 = *(v4sf*)_ps_minus_cephes_DP3;
+  xmm1 = _mm_mul_ps(y, xmm1);
+  xmm2 = _mm_mul_ps(y, xmm2);
+  xmm3 = _mm_mul_ps(y, xmm3);
+  x = _mm_add_ps(x, xmm1);
+  x = _mm_add_ps(x, xmm2);
+  x = _mm_add_ps(x, xmm3);
+
+  /* Evaluate the first polynom  (0 <= x <= Pi/4) */
+  y = *(v4sf*)_ps_coscof_p0;
+  v4sf z = _mm_mul_ps(x,x);
+
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, *(v4sf*)_ps_coscof_p1);
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, *(v4sf*)_ps_coscof_p2);
+  y = _mm_mul_ps(y, z);
+  y = _mm_mul_ps(y, z);
+  v4sf tmp = _mm_mul_ps(z, *(v4sf*)_ps_0p5);
+  y = _mm_sub_ps(y, tmp);
+  y = _mm_add_ps(y, *(v4sf*)_ps_1);
+  
+  /* Evaluate the second polynom  (Pi/4 <= x <= 0) */
+
+  v4sf y2 = *(v4sf*)_ps_sincof_p0;
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p1);
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p2);
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_mul_ps(y2, x);
+  y2 = _mm_add_ps(y2, x);
+
+  /* select the correct result from the two polynoms */  
+  xmm3 = poly_mask;
+  y2 = _mm_and_ps(xmm3, y2); //, xmm3);
+  y = _mm_andnot_ps(xmm3, y);
+  y = _mm_add_ps(y,y2);
+  /* update the sign */
+  y = _mm_xor_ps(y, sign_bit);
+  return y;
+}
+
+/* almost the same as sin_ps */
+OPENMM_EXPORT v4sf cos_ps(v4sf x) { // any x
+  v4sf xmm1, xmm2 = _mm_setzero_ps(), xmm3, y;
+#ifdef USE_SSE2
+  v4si emm0, emm2;
+#else
+  v2si mm0, mm1, mm2, mm3;
+#endif
+  /* take the absolute value */
+  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
+  
+  /* scale by 4/Pi */
+  y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
+  
+#ifdef USE_SSE2
+  /* store the integer part of y in mm0 */
+  emm2 = _mm_cvttps_epi32(y);
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
+  emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
+  y = _mm_cvtepi32_ps(emm2);
+
+  emm2 = _mm_sub_epi32(emm2, *(v4si*)_pi32_2);
+  
+  /* get the swap sign flag */
+  emm0 = _mm_andnot_si128(emm2, *(v4si*)_pi32_4);
+  emm0 = _mm_slli_epi32(emm0, 29);
+  /* get the polynom selection mask */
+  emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_2);
+  emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
+  
+  v4sf sign_bit = _mm_castsi128_ps(emm0);
+  v4sf poly_mask = _mm_castsi128_ps(emm2);
+#else
+  /* store the integer part of y in mm0:mm1 */
+  xmm2 = _mm_movehl_ps(xmm2, y);
+  mm2 = _mm_cvttps_pi32(y);
+  mm3 = _mm_cvttps_pi32(xmm2);
+
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  mm2 = _mm_add_pi32(mm2, *(v2si*)_pi32_1);
+  mm3 = _mm_add_pi32(mm3, *(v2si*)_pi32_1);
+  mm2 = _mm_and_si64(mm2, *(v2si*)_pi32_inv1);
+  mm3 = _mm_and_si64(mm3, *(v2si*)_pi32_inv1);
+
+  y = _mm_cvtpi32x2_ps(mm2, mm3);
+
+
+  mm2 = _mm_sub_pi32(mm2, *(v2si*)_pi32_2);
+  mm3 = _mm_sub_pi32(mm3, *(v2si*)_pi32_2);
+
+  /* get the swap sign flag in mm0:mm1 and the 
+     polynom selection mask in mm2:mm3 */
+
+  mm0 = _mm_andnot_si64(mm2, *(v2si*)_pi32_4);
+  mm1 = _mm_andnot_si64(mm3, *(v2si*)_pi32_4);
+  mm0 = _mm_slli_pi32(mm0, 29);
+  mm1 = _mm_slli_pi32(mm1, 29);
+
+  mm2 = _mm_and_si64(mm2, *(v2si*)_pi32_2);
+  mm3 = _mm_and_si64(mm3, *(v2si*)_pi32_2);
+
+  mm2 = _mm_cmpeq_pi32(mm2, _mm_setzero_si64());
+  mm3 = _mm_cmpeq_pi32(mm3, _mm_setzero_si64());
+
+  v4sf sign_bit, poly_mask;
+  COPY_MM_TO_XMM(mm0, mm1, sign_bit);
+  COPY_MM_TO_XMM(mm2, mm3, poly_mask);
+  _mm_empty(); /* good-bye mmx */
+#endif
+  /* The magic pass: "Extended precision modular arithmetic" 
+     x = ((x - y * DP1) - y * DP2) - y * DP3; */
+  xmm1 = *(v4sf*)_ps_minus_cephes_DP1;
+  xmm2 = *(v4sf*)_ps_minus_cephes_DP2;
+  xmm3 = *(v4sf*)_ps_minus_cephes_DP3;
+  xmm1 = _mm_mul_ps(y, xmm1);
+  xmm2 = _mm_mul_ps(y, xmm2);
+  xmm3 = _mm_mul_ps(y, xmm3);
+  x = _mm_add_ps(x, xmm1);
+  x = _mm_add_ps(x, xmm2);
+  x = _mm_add_ps(x, xmm3);
+  
+  /* Evaluate the first polynom  (0 <= x <= Pi/4) */
+  y = *(v4sf*)_ps_coscof_p0;
+  v4sf z = _mm_mul_ps(x,x);
+
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, *(v4sf*)_ps_coscof_p1);
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, *(v4sf*)_ps_coscof_p2);
+  y = _mm_mul_ps(y, z);
+  y = _mm_mul_ps(y, z);
+  v4sf tmp = _mm_mul_ps(z, *(v4sf*)_ps_0p5);
+  y = _mm_sub_ps(y, tmp);
+  y = _mm_add_ps(y, *(v4sf*)_ps_1);
+  
+  /* Evaluate the second polynom  (Pi/4 <= x <= 0) */
+
+  v4sf y2 = *(v4sf*)_ps_sincof_p0;
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p1);
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p2);
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_mul_ps(y2, x);
+  y2 = _mm_add_ps(y2, x);
+
+  /* select the correct result from the two polynoms */  
+  xmm3 = poly_mask;
+  y2 = _mm_and_ps(xmm3, y2); //, xmm3);
+  y = _mm_andnot_ps(xmm3, y);
+  y = _mm_add_ps(y,y2);
+  /* update the sign */
+  y = _mm_xor_ps(y, sign_bit);
+
+  return y;
+}
+
+/* since sin_ps and cos_ps are almost identical, sincos_ps could replace both of them..
+   it is almost as fast, and gives you a free cosine with your sine */
+OPENMM_EXPORT void sincos_ps(v4sf x, v4sf *s, v4sf *c) {
+  v4sf xmm1, xmm2, xmm3 = _mm_setzero_ps(), sign_bit_sin, y;
+#ifdef USE_SSE2
+  v4si emm0, emm2, emm4;
+#else
+  v2si mm0, mm1, mm2, mm3, mm4, mm5;
+#endif
+  sign_bit_sin = x;
+  /* take the absolute value */
+  x = _mm_and_ps(x, *(v4sf*)_ps_inv_sign_mask);
+  /* extract the sign bit (upper one) */
+  sign_bit_sin = _mm_and_ps(sign_bit_sin, *(v4sf*)_ps_sign_mask);
+  
+  /* scale by 4/Pi */
+  y = _mm_mul_ps(x, *(v4sf*)_ps_cephes_FOPI);
+    
+#ifdef USE_SSE2
+  /* store the integer part of y in emm2 */
+  emm2 = _mm_cvttps_epi32(y);
+
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  emm2 = _mm_add_epi32(emm2, *(v4si*)_pi32_1);
+  emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_inv1);
+  y = _mm_cvtepi32_ps(emm2);
+
+  emm4 = emm2;
+
+  /* get the swap sign flag for the sine */
+  emm0 = _mm_and_si128(emm2, *(v4si*)_pi32_4);
+  emm0 = _mm_slli_epi32(emm0, 29);
+  v4sf swap_sign_bit_sin = _mm_castsi128_ps(emm0);
+
+  /* get the polynom selection mask for the sine*/
+  emm2 = _mm_and_si128(emm2, *(v4si*)_pi32_2);
+  emm2 = _mm_cmpeq_epi32(emm2, _mm_setzero_si128());
+  v4sf poly_mask = _mm_castsi128_ps(emm2);
+#else
+  /* store the integer part of y in mm2:mm3 */
+  xmm3 = _mm_movehl_ps(xmm3, y);
+  mm2 = _mm_cvttps_pi32(y);
+  mm3 = _mm_cvttps_pi32(xmm3);
+
+  /* j=(j+1) & (~1) (see the cephes sources) */
+  mm2 = _mm_add_pi32(mm2, *(v2si*)_pi32_1);
+  mm3 = _mm_add_pi32(mm3, *(v2si*)_pi32_1);
+  mm2 = _mm_and_si64(mm2, *(v2si*)_pi32_inv1);
+  mm3 = _mm_and_si64(mm3, *(v2si*)_pi32_inv1);
+
+  y = _mm_cvtpi32x2_ps(mm2, mm3);
+
+  mm4 = mm2;
+  mm5 = mm3;
+
+  /* get the swap sign flag for the sine */
+  mm0 = _mm_and_si64(mm2, *(v2si*)_pi32_4);
+  mm1 = _mm_and_si64(mm3, *(v2si*)_pi32_4);
+  mm0 = _mm_slli_pi32(mm0, 29);
+  mm1 = _mm_slli_pi32(mm1, 29);
+  v4sf swap_sign_bit_sin;
+  COPY_MM_TO_XMM(mm0, mm1, swap_sign_bit_sin);
+
+  /* get the polynom selection mask for the sine */
+
+  mm2 = _mm_and_si64(mm2, *(v2si*)_pi32_2);
+  mm3 = _mm_and_si64(mm3, *(v2si*)_pi32_2);
+  mm2 = _mm_cmpeq_pi32(mm2, _mm_setzero_si64());
+  mm3 = _mm_cmpeq_pi32(mm3, _mm_setzero_si64());
+  v4sf poly_mask;
+  COPY_MM_TO_XMM(mm2, mm3, poly_mask);
+#endif
+
+  /* The magic pass: "Extended precision modular arithmetic" 
+     x = ((x - y * DP1) - y * DP2) - y * DP3; */
+  xmm1 = *(v4sf*)_ps_minus_cephes_DP1;
+  xmm2 = *(v4sf*)_ps_minus_cephes_DP2;
+  xmm3 = *(v4sf*)_ps_minus_cephes_DP3;
+  xmm1 = _mm_mul_ps(y, xmm1);
+  xmm2 = _mm_mul_ps(y, xmm2);
+  xmm3 = _mm_mul_ps(y, xmm3);
+  x = _mm_add_ps(x, xmm1);
+  x = _mm_add_ps(x, xmm2);
+  x = _mm_add_ps(x, xmm3);
+
+#ifdef USE_SSE2
+  emm4 = _mm_sub_epi32(emm4, *(v4si*)_pi32_2);
+  emm4 = _mm_andnot_si128(emm4, *(v4si*)_pi32_4);
+  emm4 = _mm_slli_epi32(emm4, 29);
+  v4sf sign_bit_cos = _mm_castsi128_ps(emm4);
+#else
+  /* get the sign flag for the cosine */
+  mm4 = _mm_sub_pi32(mm4, *(v2si*)_pi32_2);
+  mm5 = _mm_sub_pi32(mm5, *(v2si*)_pi32_2);
+  mm4 = _mm_andnot_si64(mm4, *(v2si*)_pi32_4);
+  mm5 = _mm_andnot_si64(mm5, *(v2si*)_pi32_4);
+  mm4 = _mm_slli_pi32(mm4, 29);
+  mm5 = _mm_slli_pi32(mm5, 29);
+  v4sf sign_bit_cos;
+  COPY_MM_TO_XMM(mm4, mm5, sign_bit_cos);
+  _mm_empty(); /* good-bye mmx */
+#endif
+
+  sign_bit_sin = _mm_xor_ps(sign_bit_sin, swap_sign_bit_sin);
+
+  
+  /* Evaluate the first polynom  (0 <= x <= Pi/4) */
+  v4sf z = _mm_mul_ps(x,x);
+  y = *(v4sf*)_ps_coscof_p0;
+
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, *(v4sf*)_ps_coscof_p1);
+  y = _mm_mul_ps(y, z);
+  y = _mm_add_ps(y, *(v4sf*)_ps_coscof_p2);
+  y = _mm_mul_ps(y, z);
+  y = _mm_mul_ps(y, z);
+  v4sf tmp = _mm_mul_ps(z, *(v4sf*)_ps_0p5);
+  y = _mm_sub_ps(y, tmp);
+  y = _mm_add_ps(y, *(v4sf*)_ps_1);
+  
+  /* Evaluate the second polynom  (Pi/4 <= x <= 0) */
+
+  v4sf y2 = *(v4sf*)_ps_sincof_p0;
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p1);
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_add_ps(y2, *(v4sf*)_ps_sincof_p2);
+  y2 = _mm_mul_ps(y2, z);
+  y2 = _mm_mul_ps(y2, x);
+  y2 = _mm_add_ps(y2, x);
+
+  /* select the correct result from the two polynoms */  
+  xmm3 = poly_mask;
+  v4sf ysin2 = _mm_and_ps(xmm3, y2);
+  v4sf ysin1 = _mm_andnot_ps(xmm3, y);
+  y2 = _mm_sub_ps(y2,ysin2);
+  y = _mm_sub_ps(y, ysin1);
+
+  xmm1 = _mm_add_ps(ysin1,ysin2);
+  xmm2 = _mm_add_ps(y,y2);
+ 
+  /* update the sign */
+  *s = _mm_xor_ps(xmm1, sign_bit_sin);
+  *c = _mm_xor_ps(xmm2, sign_bit_cos);
+}
+

libraries/vecmath/src/vecmath.cpp

+#if defined(__ANDROID__)
+    #include "neon_mathfun.h"
+#else
+    #if !defined(__PNACL__)
+        #define USE_SSE2
+        #include "sse_mathfun.h"
+    #endif
+#endif

openmmapi/include/openmm/CustomAngleForce.h

@@ -171,7 +171,7 @@ public:
      * @param parameters    the list of parameters for the new angle
      * @return the index of the angle that was added
      */
-    int addAngle(int particle1, int particle2, int particle3, const std::vector<double>& parameters);
+    int addAngle(int particle1, int particle2, int particle3, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the force field parameters for an angle term.
      *
@@ -191,7 +191,7 @@ public:
      * @param particle3     the index of the third particle connected by the angle
      * @param parameters    the list of parameters for the angle
      */
-    void setAngleParameters(int index, int particle1, int particle2, int particle3, const std::vector<double>& parameters);
+    void setAngleParameters(int index, int particle1, int particle2, int particle3, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Update the per-angle parameters in a Context to match those stored in this Force object.  This method provides
      * an efficient method to update certain parameters in an existing Context without needing to reinitialize it.

openmmapi/include/openmm/CustomBondForce.h

@@ -170,7 +170,7 @@ public:
      * @param parameters    the list of parameters for the new bond
      * @return the index of the bond that was added
      */
-    int addBond(int particle1, int particle2, const std::vector<double>& parameters);
+    int addBond(int particle1, int particle2, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the force field parameters for a bond term.
      *
@@ -188,7 +188,7 @@ public:
      * @param particle2     the index of the second particle connected by the bond
      * @param parameters    the list of parameters for the bond
      */
-    void setBondParameters(int index, int particle1, int particle2, const std::vector<double>& parameters);
+    void setBondParameters(int index, int particle1, int particle2, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Update the per-bond parameters in a Context to match those stored in this Force object.  This method provides
      * an efficient method to update certain parameters in an existing Context without needing to reinitialize it.

openmmapi/include/openmm/CustomCentroidBondForce.h

@@ -237,7 +237,7 @@ public:
      *                    If this is omitted, then particle masses will be used as weights.
      * @return the index of the group that was added
      */
-    int addGroup(const std::vector<int>& particles, const std::vector<double>& weights = std::vector<double>());
+    int addGroup(const std::vector<int>& particles, const std::vector<double>& weights=std::vector<double>());
     /**
      * Get the properties of a group.
      *
@@ -256,7 +256,7 @@ public:
      * @param weights     the weight to use for each particle when computing the center position.
      *                    If this is omitted, then particle masses will be used as weights.
      */
-    void setGroupParameters(int index, const std::vector<int>& particles, const std::vector<double>& weights = std::vector<double>());
+    void setGroupParameters(int index, const std::vector<int>& particles, const std::vector<double>& weights=std::vector<double>());
     /**
      * Add a bond to the force
      *
@@ -264,7 +264,7 @@ public:
      * @param parameters  the list of per-bond parameter values for the new bond
      * @return the index of the bond that was added
      */
-    int addBond(const std::vector<int>& groups, const std::vector<double>& parameters);
+    int addBond(const std::vector<int>& groups, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the properties of a bond.
      *
@@ -280,7 +280,7 @@ public:
      * @param groups      the indices of the groups in the bond
      * @param parameters  the list of per-bond parameter values for the bond
      */
-    void setBondParameters(int index, const std::vector<int>& groups, const std::vector<double>& parameters);
+    void setBondParameters(int index, const std::vector<int>& groups, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Add a tabulated function that may appear in the energy expression.
      *

openmmapi/include/openmm/CustomCompoundBondForce.h

@@ -219,7 +219,7 @@ public:
      * @param parameters  the list of per-bond parameter values for the new bond
      * @return the index of the bond that was added
      */
-    int addBond(const std::vector<int>& particles, const std::vector<double>& parameters);
+    int addBond(const std::vector<int>& particles, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the properties of a bond.
      *
@@ -235,7 +235,7 @@ public:
      * @param particles   the indices of the particles in the bond
      * @param parameters  the list of per-bond parameter values for the bond
      */
-    void setBondParameters(int index, const std::vector<int>& particles, const std::vector<double>& parameters);
+    void setBondParameters(int index, const std::vector<int>& particles, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Add a tabulated function that may appear in the energy expression.
      *

openmmapi/include/openmm/CustomExternalForce.h

@@ -67,6 +67,14 @@ namespace OpenMM {
  * force->addPerParticleParameter("z0");
  * </pre></tt>
  *
+ * Special care is needed in systems that use periodic boundary conditions.  In that case, each particle really represents
+ * an infinite set of particles repeating through space.  The variables x, y, and z contain the coordinates of one of those
+ * periodic copies, but there is no guarantee about which.  It might even change from one time step to the next.  You can handle
+ * this situation by using the function periodicdistance(x1, y1, z1, x2, y2, z2), which returns the minimum distance between
+ * periodic copies of the points (x1, y1, z1) and (x2, y2, z2).  For example, the force given above would be rewritten as
+ *
+ * <tt>CustomExternalForce* force = new CustomExternalForce("k*periodicdistance(x, y, z, x0, y0, z0)^2");</tt>
+ *
  * Expressions may involve the operators + (add), - (subtract), * (multiply), / (divide), and ^ (power), and the following
  * functions: sqrt, exp, log, sin, cos, sec, csc, tan, cot, asin, acos, atan, sinh, cosh, tanh, erf, erfc, min, max, abs, floor, ceil, step, delta, select.  All trigonometric functions
  * are defined in radians, and log is the natural logarithm.  step(x) = 0 if x is less than 0, 1 otherwise.  delta(x) = 1 if x is 0, 0 otherwise.
@@ -172,7 +180,7 @@ public:
      * @param parameters   the list of parameters for the new force term
      * @return the index of the particle term that was added
      */
-    int addParticle(int particle, const std::vector<double>& parameters);
+    int addParticle(int particle, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the force field parameters for a force field term.
      *
@@ -188,7 +196,7 @@ public:
      * @param particle      the index of the particle this term is applied to
      * @param parameters    the list of parameters for the force field term
      */
-    void setParticleParameters(int index, int particle, const std::vector<double>& parameters);
+    void setParticleParameters(int index, int particle, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Update the per-particle parameters in a Context to match those stored in this Force object.  This method provides
      * an efficient method to update certain parameters in an existing Context without needing to reinitialize it.
@@ -206,9 +214,7 @@ public:
      *
      * @returns false
      */
-    bool usesPeriodicBoundaryConditions() const {
-        return false;
-    }
+    bool usesPeriodicBoundaryConditions() const;
 protected:
     ForceImpl* createImpl() const;
 private:

openmmapi/include/openmm/CustomGBForce.h

@@ -319,7 +319,7 @@ public:
      * @param parameters    the list of parameters for the new particle
      * @return the index of the particle that was added
      */
-    int addParticle(const std::vector<double>& parameters);
+    int addParticle(const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the nonbonded force parameters for a particle.
      *

openmmapi/include/openmm/CustomHbondForce.h

@@ -296,7 +296,7 @@ public:
      * @param parameters  the list of per-donor parameter values for the new donor
      * @return the index of the donor that was added
      */
-    int addDonor(int d1, int d2, int d3, const std::vector<double>& parameters);
+    int addDonor(int d1, int d2, int d3, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the properties of a donor group.
      *
@@ -320,7 +320,7 @@ public:
      *                    less than three particles, this must be -1.
      * @param parameters  the list of per-donor parameter values for the donor
      */
-    void setDonorParameters(int index, int d1, int d2, int d3, const std::vector<double>& parameters);
+    void setDonorParameters(int index, int d1, int d2, int d3, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Add an acceptor group to the force
      *
@@ -332,7 +332,7 @@ public:
      * @param parameters  the list of per-acceptor parameter values for the new acceptor
      * @return the index of the acceptor that was added
      */
-    int addAcceptor(int a1, int a2, int a3, const std::vector<double>& parameters);
+    int addAcceptor(int a1, int a2, int a3, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the properties of an acceptor group.
      *
@@ -356,7 +356,7 @@ public:
      *                    less than three particles, this must be -1.
      * @param parameters  the list of per-acceptor parameter values for the acceptor
      */
-    void setAcceptorParameters(int index, int a1, int a2, int a3, const std::vector<double>& parameters);
+    void setAcceptorParameters(int index, int a1, int a2, int a3, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Add a donor-acceptor pair to the list of interactions that should be excluded.
      *

openmmapi/include/openmm/CustomManyParticleForce.h

@@ -348,7 +348,7 @@ public:
      * @param type          the type of the new particle
      * @return the index of the particle that was added
      */
-    int addParticle(const std::vector<double>& parameters, int type=0);
+    int addParticle(const std::vector<double>& parameters=std::vector<double>(), int type=0);
     /**
      * Get the nonbonded force parameters for a particle.
      *

openmmapi/include/openmm/CustomNonbondedForce.h

@@ -328,7 +328,7 @@ public:
      * @param parameters    the list of parameters for the new particle
      * @return the index of the particle that was added
      */
-    int addParticle(const std::vector<double>& parameters);
+    int addParticle(const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the nonbonded force parameters for a particle.
      *

openmmapi/include/openmm/CustomTorsionForce.h

@@ -172,7 +172,7 @@ public:
      * @param parameters    the list of parameters for the new torsion
      * @return the index of the torsion that was added
      */
-    int addTorsion(int particle1, int particle2, int particle3, int particle4, const std::vector<double>& parameters);
+    int addTorsion(int particle1, int particle2, int particle3, int particle4, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Get the force field parameters for a torsion term.
      *
@@ -194,7 +194,7 @@ public:
      * @param particle4     the index of the fourth particle connected by the torsion
      * @param parameters    the list of parameters for the torsion
      */
-    void setTorsionParameters(int index, int particle1, int particle2, int particle3, int particle4, const std::vector<double>& parameters);
+    void setTorsionParameters(int index, int particle1, int particle2, int particle3, int particle4, const std::vector<double>& parameters=std::vector<double>());
     /**
      * Update the per-torsion parameters in a Context to match those stored in this Force object.  This method provides
      * an efficient method to update certain parameters in an existing Context without needing to reinitialize it.

openmmapi/include/openmm/internal/CustomCentroidBondForceImpl.h

@@ -41,6 +41,7 @@
 #include "lepton/ParsedExpression.h"
 #include <utility>
 #include <map>
+#include <set>
 #include <string>
 
 namespace OpenMM {
@@ -93,7 +94,7 @@ private:
     class FunctionPlaceholder;
     static Lepton::ExpressionTreeNode replaceFunctions(const Lepton::ExpressionTreeNode& node, std::map<std::string, int> atoms,
             std::map<std::string, std::vector<int> >& distances, std::map<std::string, std::vector<int> >& angles,
-            std::map<std::string, std::vector<int> >& dihedrals);
+            std::map<std::string, std::vector<int> >& dihedrals, std::set<std::string>& variables);
     void addBondsBetweenGroups(int group1, int group2, std::vector<std::pair<int, int> >& bonds) const;
     const CustomCentroidBondForce& owner;
     Kernel kernel;

openmmapi/include/openmm/internal/CustomCompoundBondForceImpl.h

@@ -40,6 +40,7 @@
 #include "lepton/ParsedExpression.h"
 #include <utility>
 #include <map>
+#include <set>
 #include <string>
 
 namespace OpenMM {
@@ -83,7 +84,7 @@ private:
     class FunctionPlaceholder;
     static Lepton::ExpressionTreeNode replaceFunctions(const Lepton::ExpressionTreeNode& node, std::map<std::string, int> atoms,
             std::map<std::string, std::vector<int> >& distances, std::map<std::string, std::vector<int> >& angles,
-            std::map<std::string, std::vector<int> >& dihedrals);
+            std::map<std::string, std::vector<int> >& dihedrals, std::set<std::string>& variables);
     const CustomCompoundBondForce& owner;
     Kernel kernel;
 };

openmmapi/include/openmm/internal/CustomHbondForceImpl.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) 2008-2010 Stanford University and the Authors.      *
+ * Portions copyright (c) 2008-2015 Stanford University and the Authors.      *
  * Authors: Peter Eastman                                                     *
  * Contributors:                                                              *
  *                                                                            *
@@ -40,6 +40,7 @@
 #include "lepton/ParsedExpression.h"
 #include <utility>
 #include <map>
+#include <set>
 #include <string>
 
 namespace OpenMM {
@@ -84,7 +85,7 @@ private:
     class FunctionPlaceholder;
     static Lepton::ExpressionTreeNode replaceFunctions(const Lepton::ExpressionTreeNode& node, std::map<std::string, int> atoms,
             std::map<std::string, std::vector<int> >& distances, std::map<std::string, std::vector<int> >& angles,
-            std::map<std::string, std::vector<int> >& dihedrals);
+            std::map<std::string, std::vector<int> >& dihedrals, std::set<std::string>& variables);
     const CustomHbondForce& owner;
     Kernel kernel;
 };

openmmapi/include/openmm/internal/CustomManyParticleForceImpl.h

@@ -40,6 +40,7 @@
 #include "lepton/ParsedExpression.h"
 #include <utility>
 #include <map>
+#include <set>
 #include <string>
 
 namespace OpenMM {
@@ -98,7 +99,7 @@ private:
     class FunctionPlaceholder;
     static Lepton::ExpressionTreeNode replaceFunctions(const Lepton::ExpressionTreeNode& node, std::map<std::string, int> atoms,
             std::map<std::string, std::vector<int> >& distances, std::map<std::string, std::vector<int> >& angles,
-            std::map<std::string, std::vector<int> >& dihedrals);
+            std::map<std::string, std::vector<int> >& dihedrals, std::set<std::string>& variables);
     static void generatePermutations(std::vector<int>& values, int numFixed, std::vector<std::vector<int> >& result);
     const CustomManyParticleForce& owner;
     Kernel kernel;