src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  86     real             rswitch_scalar,d_scalar;
  87     __m128           dummy_mask,cutoff_mask;
  88     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  89     __m128           one     = _mm_set1_ps(1.0);
  90     __m128           two     = _mm_set1_ps(2.0);
  91     x                = xx[0];
  92     f                = ff[0];
  93
  94     nri              = nlist->nri;
  95     iinr             = nlist->iinr;
  96     jindex           = nlist->jindex;
  97     jjnr             = nlist->jjnr;
  98     shiftidx         = nlist->shift;
  99     gid              = nlist->gid;
 100     shiftvec         = fr->shift_vec[0];
 101     fshift           = fr->fshift[0];
 102     facel            = _mm_set1_ps(fr->epsfac);
 103     charge           = mdatoms->chargeA;
 104     krf              = _mm_set1_ps(fr->ic->k_rf);
 105     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 106     crf              = _mm_set1_ps(fr->ic->c_rf);
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     /* Setup water-specific parameters */
 112     inr              = nlist->iinr[0];
 113     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 114     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 115     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 116     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 117
 118     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 119     rcutoff_scalar   = fr->rcoulomb;
 120     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 122
 123     rswitch_scalar   = fr->rvdw_switch;
 124     rswitch          = _mm_set1_ps(rswitch_scalar);
 125     /* Setup switch parameters */
 126     d_scalar         = rcutoff_scalar-rswitch_scalar;
 127     d                = _mm_set1_ps(d_scalar);
 128     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 129     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 130     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 131     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 132     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 133     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150         shX              = shiftvec[i_shift_offset+XX];
 151         shY              = shiftvec[i_shift_offset+YY];
 152         shZ              = shiftvec[i_shift_offset+ZZ];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 164         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 165         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 166         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 167         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 168         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 169         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 170         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 171         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 172
 173         fix0             = _mm_setzero_ps();
 174         fiy0             = _mm_setzero_ps();
 175         fiz0             = _mm_setzero_ps();
 176         fix1             = _mm_setzero_ps();
 177         fiy1             = _mm_setzero_ps();
 178         fiz1             = _mm_setzero_ps();
 179         fix2             = _mm_setzero_ps();
 180         fiy2             = _mm_setzero_ps();
 181         fiz2             = _mm_setzero_ps();
 182
 183         /* Reset potential sums */
 184         velecsum         = _mm_setzero_ps();
 185         vvdwsum          = _mm_setzero_ps();
 186
 187         /* Start inner kernel loop */
 188         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 189         {
 190
 191             /* Get j neighbor index, and coordinate index */
 192             jnrA             = jjnr[jidx];
 193             jnrB             = jjnr[jidx+1];
 194             jnrC             = jjnr[jidx+2];
 195             jnrD             = jjnr[jidx+3];
 196
 197             j_coord_offsetA  = DIM*jnrA;
 198             j_coord_offsetB  = DIM*jnrB;
 199             j_coord_offsetC  = DIM*jnrC;
 200             j_coord_offsetD  = DIM*jnrD;
 201
 202             /* load j atom coordinates */
 203             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 204                                               x+j_coord_offsetC,x+j_coord_offsetD,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_ps(ix0,jx0);
 209             dy00             = _mm_sub_ps(iy0,jy0);
 210             dz00             = _mm_sub_ps(iz0,jz0);
 211             dx10             = _mm_sub_ps(ix1,jx0);
 212             dy10             = _mm_sub_ps(iy1,jy0);
 213             dz10             = _mm_sub_ps(iz1,jz0);
 214             dx20             = _mm_sub_ps(ix2,jx0);
 215             dy20             = _mm_sub_ps(iy2,jy0);
 216             dz20             = _mm_sub_ps(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226
 227             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             if (gmx_mm_any_lt(rsq00,rcutoff2))
 244             {
 245
 246             r00              = _mm_mul_ps(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             qq00             = _mm_mul_ps(iq0,jq0);
 250             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 251                                          vdwparam+vdwioffset0+vdwjidx0B,
 252                                          vdwparam+vdwioffset0+vdwjidx0C,
 253                                          vdwparam+vdwioffset0+vdwjidx0D,
 254                                          &c6_00,&c12_00);
 255
 256             /* REACTION-FIELD ELECTROSTATICS */
 257             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 258             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 259
 260             /* LENNARD-JONES DISPERSION/REPULSION */
 261
 262             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 263             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 264             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 265             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 266             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 267
 268             d                = _mm_sub_ps(r00,rswitch);
 269             d                = _mm_max_ps(d,_mm_setzero_ps());
 270             d2               = _mm_mul_ps(d,d);
 271             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 272
 273             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 274
 275             /* Evaluate switch function */
 276             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 277             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 278             vvdw             = _mm_mul_ps(vvdw,sw);
 279             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velec            = _mm_and_ps(velec,cutoff_mask);
 283             velecsum         = _mm_add_ps(velecsum,velec);
 284             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 285             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = _mm_add_ps(felec,fvdw);
 288
 289             fscal            = _mm_and_ps(fscal,cutoff_mask);
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm_mul_ps(fscal,dx00);
 293             ty               = _mm_mul_ps(fscal,dy00);
 294             tz               = _mm_mul_ps(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm_add_ps(fix0,tx);
 298             fiy0             = _mm_add_ps(fiy0,ty);
 299             fiz0             = _mm_add_ps(fiz0,tz);
 300
 301             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 302                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 303                                                    tx,ty,tz);
 304
 305             }
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             if (gmx_mm_any_lt(rsq10,rcutoff2))
 312             {
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm_mul_ps(iq1,jq0);
 316
 317             /* REACTION-FIELD ELECTROSTATICS */
 318             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 319             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 320
 321             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 322
 323             /* Update potential sum for this i atom from the interaction with this j atom. */
 324             velec            = _mm_and_ps(velec,cutoff_mask);
 325             velecsum         = _mm_add_ps(velecsum,velec);
 326
 327             fscal            = felec;
 328
 329             fscal            = _mm_and_ps(fscal,cutoff_mask);
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_ps(fscal,dx10);
 333             ty               = _mm_mul_ps(fscal,dy10);
 334             tz               = _mm_mul_ps(fscal,dz10);
 335
 336             /* Update vectorial force */
 337             fix1             = _mm_add_ps(fix1,tx);
 338             fiy1             = _mm_add_ps(fiy1,ty);
 339             fiz1             = _mm_add_ps(fiz1,tz);
 340
 341             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 342                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 343                                                    tx,ty,tz);
 344
 345             }
 346
 347             /**************************
 348              * CALCULATE INTERACTIONS *
 349              **************************/
 350
 351             if (gmx_mm_any_lt(rsq20,rcutoff2))
 352             {
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq20             = _mm_mul_ps(iq2,jq0);
 356
 357             /* REACTION-FIELD ELECTROSTATICS */
 358             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 359             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 360
 361             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_and_ps(velec,cutoff_mask);
 365             velecsum         = _mm_add_ps(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm_and_ps(fscal,cutoff_mask);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_ps(fscal,dx20);
 373             ty               = _mm_mul_ps(fscal,dy20);
 374             tz               = _mm_mul_ps(fscal,dz20);
 375
 376             /* Update vectorial force */
 377             fix2             = _mm_add_ps(fix2,tx);
 378             fiy2             = _mm_add_ps(fiy2,ty);
 379             fiz2             = _mm_add_ps(fiz2,tz);
 380
 381             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 382                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 383                                                    tx,ty,tz);
 384
 385             }
 386
 387             /* Inner loop uses 142 flops */
 388         }
 389
 390         if(jidx<j_index_end)
 391         {
 392
 393             /* Get j neighbor index, and coordinate index */
 394             jnrA             = jjnr[jidx];
 395             jnrB             = jjnr[jidx+1];
 396             jnrC             = jjnr[jidx+2];
 397             jnrD             = jjnr[jidx+3];
 398
 399             /* Sign of each element will be negative for non-real atoms.
 400              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 401              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 402              */
 403             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 404             jnrA       = (jnrA>=0) ? jnrA : 0;
 405             jnrB       = (jnrB>=0) ? jnrB : 0;
 406             jnrC       = (jnrC>=0) ? jnrC : 0;
 407             jnrD       = (jnrD>=0) ? jnrD : 0;
 408
 409             j_coord_offsetA  = DIM*jnrA;
 410             j_coord_offsetB  = DIM*jnrB;
 411             j_coord_offsetC  = DIM*jnrC;
 412             j_coord_offsetD  = DIM*jnrD;
 413
 414             /* load j atom coordinates */
 415             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 416                                               x+j_coord_offsetC,x+j_coord_offsetD,
 417                                               &jx0,&jy0,&jz0);
 418
 419             /* Calculate displacement vector */
 420             dx00             = _mm_sub_ps(ix0,jx0);
 421             dy00             = _mm_sub_ps(iy0,jy0);
 422             dz00             = _mm_sub_ps(iz0,jz0);
 423             dx10             = _mm_sub_ps(ix1,jx0);
 424             dy10             = _mm_sub_ps(iy1,jy0);
 425             dz10             = _mm_sub_ps(iz1,jz0);
 426             dx20             = _mm_sub_ps(ix2,jx0);
 427             dy20             = _mm_sub_ps(iy2,jy0);
 428             dz20             = _mm_sub_ps(iz2,jz0);
 429
 430             /* Calculate squared distance and things based on it */
 431             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 432             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 433             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 434
 435             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 436             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 437             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 438
 439             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 440             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 441             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 442
 443             /* Load parameters for j particles */
 444             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 445                                                               charge+jnrC+0,charge+jnrD+0);
 446             vdwjidx0A        = 2*vdwtype[jnrA+0];
 447             vdwjidx0B        = 2*vdwtype[jnrB+0];
 448             vdwjidx0C        = 2*vdwtype[jnrC+0];
 449             vdwjidx0D        = 2*vdwtype[jnrD+0];
 450
 451             /**************************
 452              * CALCULATE INTERACTIONS *
 453              **************************/
 454
 455             if (gmx_mm_any_lt(rsq00,rcutoff2))
 456             {
 457
 458             r00              = _mm_mul_ps(rsq00,rinv00);
 459             r00              = _mm_andnot_ps(dummy_mask,r00);
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             qq00             = _mm_mul_ps(iq0,jq0);
 463             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 464                                          vdwparam+vdwioffset0+vdwjidx0B,
 465                                          vdwparam+vdwioffset0+vdwjidx0C,
 466                                          vdwparam+vdwioffset0+vdwjidx0D,
 467                                          &c6_00,&c12_00);
 468
 469             /* REACTION-FIELD ELECTROSTATICS */
 470             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 471             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 472
 473             /* LENNARD-JONES DISPERSION/REPULSION */
 474
 475             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 476             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 477             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 478             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 479             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 480
 481             d                = _mm_sub_ps(r00,rswitch);
 482             d                = _mm_max_ps(d,_mm_setzero_ps());
 483             d2               = _mm_mul_ps(d,d);
 484             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 485
 486             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 487
 488             /* Evaluate switch function */
 489             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 490             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 491             vvdw             = _mm_mul_ps(vvdw,sw);
 492             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm_and_ps(velec,cutoff_mask);
 496             velec            = _mm_andnot_ps(dummy_mask,velec);
 497             velecsum         = _mm_add_ps(velecsum,velec);
 498             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 499             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 500             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 501
 502             fscal            = _mm_add_ps(felec,fvdw);
 503
 504             fscal            = _mm_and_ps(fscal,cutoff_mask);
 505
 506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm_mul_ps(fscal,dx00);
 510             ty               = _mm_mul_ps(fscal,dy00);
 511             tz               = _mm_mul_ps(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm_add_ps(fix0,tx);
 515             fiy0             = _mm_add_ps(fiy0,ty);
 516             fiz0             = _mm_add_ps(fiz0,tz);
 517
 518             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 519                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 520                                                    tx,ty,tz);
 521
 522             }
 523
 524             /**************************
 525              * CALCULATE INTERACTIONS *
 526              **************************/
 527
 528             if (gmx_mm_any_lt(rsq10,rcutoff2))
 529             {
 530
 531             /* Compute parameters for interactions between i and j atoms */
 532             qq10             = _mm_mul_ps(iq1,jq0);
 533
 534             /* REACTION-FIELD ELECTROSTATICS */
 535             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 536             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 537
 538             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 539
 540             /* Update potential sum for this i atom from the interaction with this j atom. */
 541             velec            = _mm_and_ps(velec,cutoff_mask);
 542             velec            = _mm_andnot_ps(dummy_mask,velec);
 543             velecsum         = _mm_add_ps(velecsum,velec);
 544
 545             fscal            = felec;
 546
 547             fscal            = _mm_and_ps(fscal,cutoff_mask);
 548
 549             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm_mul_ps(fscal,dx10);
 553             ty               = _mm_mul_ps(fscal,dy10);
 554             tz               = _mm_mul_ps(fscal,dz10);
 555
 556             /* Update vectorial force */
 557             fix1             = _mm_add_ps(fix1,tx);
 558             fiy1             = _mm_add_ps(fiy1,ty);
 559             fiz1             = _mm_add_ps(fiz1,tz);
 560
 561             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 562                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 563                                                    tx,ty,tz);
 564
 565             }
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             if (gmx_mm_any_lt(rsq20,rcutoff2))
 572             {
 573
 574             /* Compute parameters for interactions between i and j atoms */
 575             qq20             = _mm_mul_ps(iq2,jq0);
 576
 577             /* REACTION-FIELD ELECTROSTATICS */
 578             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 579             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 580
 581             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_and_ps(velec,cutoff_mask);
 585             velec            = _mm_andnot_ps(dummy_mask,velec);
 586             velecsum         = _mm_add_ps(velecsum,velec);
 587
 588             fscal            = felec;
 589
 590             fscal            = _mm_and_ps(fscal,cutoff_mask);
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx20);
 596             ty               = _mm_mul_ps(fscal,dy20);
 597             tz               = _mm_mul_ps(fscal,dz20);
 598
 599             /* Update vectorial force */
 600             fix2             = _mm_add_ps(fix2,tx);
 601             fiy2             = _mm_add_ps(fiy2,ty);
 602             fiz2             = _mm_add_ps(fiz2,tz);
 603
 604             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 605                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 606                                                    tx,ty,tz);
 607
 608             }
 609
 610             /* Inner loop uses 143 flops */
 611         }
 612
 613         /* End of innermost loop */
 614
 615         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 616                                               f+i_coord_offset,fshift+i_shift_offset);
 617
 618         ggid                        = gid[iidx];
 619         /* Update potential energies */
 620         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 621         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 622
 623         /* Increment number of inner iterations */
 624         inneriter                  += j_index_end - j_index_start;
 625
 626         /* Outer loop uses 29 flops */
 627     }
 628
 629     /* Increment number of outer iterations */
 630     outeriter        += nri;
 631
 632     /* Update outer/inner flops */
 633
 634     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*143);
 635 }
 636 /*
 637  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_single
 638  * Electrostatics interaction: ReactionField
 639  * VdW interaction:            LennardJones
 640  * Geometry:                   Water3-Particle
 641  * Calculate force/pot:        Force
 642  */
 643 void
 644 nb_kernel_ElecRFCut_VdwLJSw_GeomW3P1_F_sse2_single
 645                     (t_nblist * gmx_restrict                nlist,
 646                      rvec * gmx_restrict                    xx,
 647                      rvec * gmx_restrict                    ff,
 648                      t_forcerec * gmx_restrict              fr,
 649                      t_mdatoms * gmx_restrict               mdatoms,
 650                      nb_kernel_data_t * gmx_restrict        kernel_data,
 651                      t_nrnb * gmx_restrict                  nrnb)
 652 {
 653     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 654      * just 0 for non-waters.
 655      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 656      * jnr indices corresponding to data put in the four positions in the SIMD register.
 657      */
 658     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 659     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 660     int              jnrA,jnrB,jnrC,jnrD;
 661     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 662     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 663     real             shX,shY,shZ,rcutoff_scalar;
 664     real             *shiftvec,*fshift,*x,*f;
 665     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 666     int              vdwioffset0;
 667     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 668     int              vdwioffset1;
 669     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 670     int              vdwioffset2;
 671     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 672     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 673     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 674     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 675     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 676     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 677     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 678     real             *charge;
 679     int              nvdwtype;
 680     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 681     int              *vdwtype;
 682     real             *vdwparam;
 683     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 684     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 685     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 686     real             rswitch_scalar,d_scalar;
 687     __m128           dummy_mask,cutoff_mask;
 688     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 689     __m128           one     = _mm_set1_ps(1.0);
 690     __m128           two     = _mm_set1_ps(2.0);
 691     x                = xx[0];
 692     f                = ff[0];
 693
 694     nri              = nlist->nri;
 695     iinr             = nlist->iinr;
 696     jindex           = nlist->jindex;
 697     jjnr             = nlist->jjnr;
 698     shiftidx         = nlist->shift;
 699     gid              = nlist->gid;
 700     shiftvec         = fr->shift_vec[0];
 701     fshift           = fr->fshift[0];
 702     facel            = _mm_set1_ps(fr->epsfac);
 703     charge           = mdatoms->chargeA;
 704     krf              = _mm_set1_ps(fr->ic->k_rf);
 705     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 706     crf              = _mm_set1_ps(fr->ic->c_rf);
 707     nvdwtype         = fr->ntype;
 708     vdwparam         = fr->nbfp;
 709     vdwtype          = mdatoms->typeA;
 710
 711     /* Setup water-specific parameters */
 712     inr              = nlist->iinr[0];
 713     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 714     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 715     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 716     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 717
 718     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 719     rcutoff_scalar   = fr->rcoulomb;
 720     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 721     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 722
 723     rswitch_scalar   = fr->rvdw_switch;
 724     rswitch          = _mm_set1_ps(rswitch_scalar);
 725     /* Setup switch parameters */
 726     d_scalar         = rcutoff_scalar-rswitch_scalar;
 727     d                = _mm_set1_ps(d_scalar);
 728     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 729     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 730     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 731     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 732     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 733     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 734
 735     /* Avoid stupid compiler warnings */
 736     jnrA = jnrB = jnrC = jnrD = 0;
 737     j_coord_offsetA = 0;
 738     j_coord_offsetB = 0;
 739     j_coord_offsetC = 0;
 740     j_coord_offsetD = 0;
 741
 742     outeriter        = 0;
 743     inneriter        = 0;
 744
 745     /* Start outer loop over neighborlists */
 746     for(iidx=0; iidx<nri; iidx++)
 747     {
 748         /* Load shift vector for this list */
 749         i_shift_offset   = DIM*shiftidx[iidx];
 750         shX              = shiftvec[i_shift_offset+XX];
 751         shY              = shiftvec[i_shift_offset+YY];
 752         shZ              = shiftvec[i_shift_offset+ZZ];
 753
 754         /* Load limits for loop over neighbors */
 755         j_index_start    = jindex[iidx];
 756         j_index_end      = jindex[iidx+1];
 757
 758         /* Get outer coordinate index */
 759         inr              = iinr[iidx];
 760         i_coord_offset   = DIM*inr;
 761
 762         /* Load i particle coords and add shift vector */
 763         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 764         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 765         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 766         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 767         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 768         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 769         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 770         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 771         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 772
 773         fix0             = _mm_setzero_ps();
 774         fiy0             = _mm_setzero_ps();
 775         fiz0             = _mm_setzero_ps();
 776         fix1             = _mm_setzero_ps();
 777         fiy1             = _mm_setzero_ps();
 778         fiz1             = _mm_setzero_ps();
 779         fix2             = _mm_setzero_ps();
 780         fiy2             = _mm_setzero_ps();
 781         fiz2             = _mm_setzero_ps();
 782
 783         /* Start inner kernel loop */
 784         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 785         {
 786
 787             /* Get j neighbor index, and coordinate index */
 788             jnrA             = jjnr[jidx];
 789             jnrB             = jjnr[jidx+1];
 790             jnrC             = jjnr[jidx+2];
 791             jnrD             = jjnr[jidx+3];
 792
 793             j_coord_offsetA  = DIM*jnrA;
 794             j_coord_offsetB  = DIM*jnrB;
 795             j_coord_offsetC  = DIM*jnrC;
 796             j_coord_offsetD  = DIM*jnrD;
 797
 798             /* load j atom coordinates */
 799             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 800                                               x+j_coord_offsetC,x+j_coord_offsetD,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_ps(ix0,jx0);
 805             dy00             = _mm_sub_ps(iy0,jy0);
 806             dz00             = _mm_sub_ps(iz0,jz0);
 807             dx10             = _mm_sub_ps(ix1,jx0);
 808             dy10             = _mm_sub_ps(iy1,jy0);
 809             dz10             = _mm_sub_ps(iz1,jz0);
 810             dx20             = _mm_sub_ps(ix2,jx0);
 811             dy20             = _mm_sub_ps(iy2,jy0);
 812             dz20             = _mm_sub_ps(iz2,jz0);
 813
 814             /* Calculate squared distance and things based on it */
 815             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 816             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 817             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 818
 819             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 820             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 821             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 822
 823             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 824             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 825             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 826
 827             /* Load parameters for j particles */
 828             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 829                                                               charge+jnrC+0,charge+jnrD+0);
 830             vdwjidx0A        = 2*vdwtype[jnrA+0];
 831             vdwjidx0B        = 2*vdwtype[jnrB+0];
 832             vdwjidx0C        = 2*vdwtype[jnrC+0];
 833             vdwjidx0D        = 2*vdwtype[jnrD+0];
 834
 835             /**************************
 836              * CALCULATE INTERACTIONS *
 837              **************************/
 838
 839             if (gmx_mm_any_lt(rsq00,rcutoff2))
 840             {
 841
 842             r00              = _mm_mul_ps(rsq00,rinv00);
 843
 844             /* Compute parameters for interactions between i and j atoms */
 845             qq00             = _mm_mul_ps(iq0,jq0);
 846             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 847                                          vdwparam+vdwioffset0+vdwjidx0B,
 848                                          vdwparam+vdwioffset0+vdwjidx0C,
 849                                          vdwparam+vdwioffset0+vdwjidx0D,
 850                                          &c6_00,&c12_00);
 851
 852             /* REACTION-FIELD ELECTROSTATICS */
 853             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 854
 855             /* LENNARD-JONES DISPERSION/REPULSION */
 856
 857             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 858             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 859             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 860             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 861             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 862
 863             d                = _mm_sub_ps(r00,rswitch);
 864             d                = _mm_max_ps(d,_mm_setzero_ps());
 865             d2               = _mm_mul_ps(d,d);
 866             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 867
 868             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 869
 870             /* Evaluate switch function */
 871             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 872             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 873             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 874
 875             fscal            = _mm_add_ps(felec,fvdw);
 876
 877             fscal            = _mm_and_ps(fscal,cutoff_mask);
 878
 879             /* Calculate temporary vectorial force */
 880             tx               = _mm_mul_ps(fscal,dx00);
 881             ty               = _mm_mul_ps(fscal,dy00);
 882             tz               = _mm_mul_ps(fscal,dz00);
 883
 884             /* Update vectorial force */
 885             fix0             = _mm_add_ps(fix0,tx);
 886             fiy0             = _mm_add_ps(fiy0,ty);
 887             fiz0             = _mm_add_ps(fiz0,tz);
 888
 889             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 890                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 891                                                    tx,ty,tz);
 892
 893             }
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             if (gmx_mm_any_lt(rsq10,rcutoff2))
 900             {
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq10             = _mm_mul_ps(iq1,jq0);
 904
 905             /* REACTION-FIELD ELECTROSTATICS */
 906             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 907
 908             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 909
 910             fscal            = felec;
 911
 912             fscal            = _mm_and_ps(fscal,cutoff_mask);
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm_mul_ps(fscal,dx10);
 916             ty               = _mm_mul_ps(fscal,dy10);
 917             tz               = _mm_mul_ps(fscal,dz10);
 918
 919             /* Update vectorial force */
 920             fix1             = _mm_add_ps(fix1,tx);
 921             fiy1             = _mm_add_ps(fiy1,ty);
 922             fiz1             = _mm_add_ps(fiz1,tz);
 923
 924             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 925                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 926                                                    tx,ty,tz);
 927
 928             }
 929
 930             /**************************
 931              * CALCULATE INTERACTIONS *
 932              **************************/
 933
 934             if (gmx_mm_any_lt(rsq20,rcutoff2))
 935             {
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq20             = _mm_mul_ps(iq2,jq0);
 939
 940             /* REACTION-FIELD ELECTROSTATICS */
 941             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 942
 943             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 944
 945             fscal            = felec;
 946
 947             fscal            = _mm_and_ps(fscal,cutoff_mask);
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm_mul_ps(fscal,dx20);
 951             ty               = _mm_mul_ps(fscal,dy20);
 952             tz               = _mm_mul_ps(fscal,dz20);
 953
 954             /* Update vectorial force */
 955             fix2             = _mm_add_ps(fix2,tx);
 956             fiy2             = _mm_add_ps(fiy2,ty);
 957             fiz2             = _mm_add_ps(fiz2,tz);
 958
 959             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 960                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 961                                                    tx,ty,tz);
 962
 963             }
 964
 965             /* Inner loop uses 121 flops */
 966         }
 967
 968         if(jidx<j_index_end)
 969         {
 970
 971             /* Get j neighbor index, and coordinate index */
 972             jnrA             = jjnr[jidx];
 973             jnrB             = jjnr[jidx+1];
 974             jnrC             = jjnr[jidx+2];
 975             jnrD             = jjnr[jidx+3];
 976
 977             /* Sign of each element will be negative for non-real atoms.
 978              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 979              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 980              */
 981             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 982             jnrA       = (jnrA>=0) ? jnrA : 0;
 983             jnrB       = (jnrB>=0) ? jnrB : 0;
 984             jnrC       = (jnrC>=0) ? jnrC : 0;
 985             jnrD       = (jnrD>=0) ? jnrD : 0;
 986
 987             j_coord_offsetA  = DIM*jnrA;
 988             j_coord_offsetB  = DIM*jnrB;
 989             j_coord_offsetC  = DIM*jnrC;
 990             j_coord_offsetD  = DIM*jnrD;
 991
 992             /* load j atom coordinates */
 993             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 994                                               x+j_coord_offsetC,x+j_coord_offsetD,
 995                                               &jx0,&jy0,&jz0);
 996
 997             /* Calculate displacement vector */
 998             dx00             = _mm_sub_ps(ix0,jx0);
 999             dy00             = _mm_sub_ps(iy0,jy0);
1000             dz00             = _mm_sub_ps(iz0,jz0);
1001             dx10             = _mm_sub_ps(ix1,jx0);
1002             dy10             = _mm_sub_ps(iy1,jy0);
1003             dz10             = _mm_sub_ps(iz1,jz0);
1004             dx20             = _mm_sub_ps(ix2,jx0);
1005             dy20             = _mm_sub_ps(iy2,jy0);
1006             dz20             = _mm_sub_ps(iz2,jz0);
1007
1008             /* Calculate squared distance and things based on it */
1009             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1010             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1011             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1012
1013             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1014             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1015             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1016
1017             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1018             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1019             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1020
1021             /* Load parameters for j particles */
1022             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1023                                                               charge+jnrC+0,charge+jnrD+0);
1024             vdwjidx0A        = 2*vdwtype[jnrA+0];
1025             vdwjidx0B        = 2*vdwtype[jnrB+0];
1026             vdwjidx0C        = 2*vdwtype[jnrC+0];
1027             vdwjidx0D        = 2*vdwtype[jnrD+0];
1028
1029             /**************************
1030              * CALCULATE INTERACTIONS *
1031              **************************/
1032
1033             if (gmx_mm_any_lt(rsq00,rcutoff2))
1034             {
1035
1036             r00              = _mm_mul_ps(rsq00,rinv00);
1037             r00              = _mm_andnot_ps(dummy_mask,r00);
1038
1039             /* Compute parameters for interactions between i and j atoms */
1040             qq00             = _mm_mul_ps(iq0,jq0);
1041             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1042                                          vdwparam+vdwioffset0+vdwjidx0B,
1043                                          vdwparam+vdwioffset0+vdwjidx0C,
1044                                          vdwparam+vdwioffset0+vdwjidx0D,
1045                                          &c6_00,&c12_00);
1046
1047             /* REACTION-FIELD ELECTROSTATICS */
1048             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1049
1050             /* LENNARD-JONES DISPERSION/REPULSION */
1051
1052             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1053             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1054             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1055             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1056             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1057
1058             d                = _mm_sub_ps(r00,rswitch);
1059             d                = _mm_max_ps(d,_mm_setzero_ps());
1060             d2               = _mm_mul_ps(d,d);
1061             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1062
1063             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1064
1065             /* Evaluate switch function */
1066             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1067             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1068             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1069
1070             fscal            = _mm_add_ps(felec,fvdw);
1071
1072             fscal            = _mm_and_ps(fscal,cutoff_mask);
1073
1074             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm_mul_ps(fscal,dx00);
1078             ty               = _mm_mul_ps(fscal,dy00);
1079             tz               = _mm_mul_ps(fscal,dz00);
1080
1081             /* Update vectorial force */
1082             fix0             = _mm_add_ps(fix0,tx);
1083             fiy0             = _mm_add_ps(fiy0,ty);
1084             fiz0             = _mm_add_ps(fiz0,tz);
1085
1086             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1087                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1088                                                    tx,ty,tz);
1089
1090             }
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq10,rcutoff2))
1097             {
1098
1099             /* Compute parameters for interactions between i and j atoms */
1100             qq10             = _mm_mul_ps(iq1,jq0);
1101
1102             /* REACTION-FIELD ELECTROSTATICS */
1103             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1104
1105             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1106
1107             fscal            = felec;
1108
1109             fscal            = _mm_and_ps(fscal,cutoff_mask);
1110
1111             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1112
1113             /* Calculate temporary vectorial force */
1114             tx               = _mm_mul_ps(fscal,dx10);
1115             ty               = _mm_mul_ps(fscal,dy10);
1116             tz               = _mm_mul_ps(fscal,dz10);
1117
1118             /* Update vectorial force */
1119             fix1             = _mm_add_ps(fix1,tx);
1120             fiy1             = _mm_add_ps(fiy1,ty);
1121             fiz1             = _mm_add_ps(fiz1,tz);
1122
1123             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1124                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1125                                                    tx,ty,tz);
1126
1127             }
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             if (gmx_mm_any_lt(rsq20,rcutoff2))
1134             {
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq20             = _mm_mul_ps(iq2,jq0);
1138
1139             /* REACTION-FIELD ELECTROSTATICS */
1140             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1141
1142             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1143
1144             fscal            = felec;
1145
1146             fscal            = _mm_and_ps(fscal,cutoff_mask);
1147
1148             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1149
1150             /* Calculate temporary vectorial force */
1151             tx               = _mm_mul_ps(fscal,dx20);
1152             ty               = _mm_mul_ps(fscal,dy20);
1153             tz               = _mm_mul_ps(fscal,dz20);
1154
1155             /* Update vectorial force */
1156             fix2             = _mm_add_ps(fix2,tx);
1157             fiy2             = _mm_add_ps(fiy2,ty);
1158             fiz2             = _mm_add_ps(fiz2,tz);
1159
1160             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1161                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1162                                                    tx,ty,tz);
1163
1164             }
1165
1166             /* Inner loop uses 122 flops */
1167         }
1168
1169         /* End of innermost loop */
1170
1171         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1172                                               f+i_coord_offset,fshift+i_shift_offset);
1173
1174         /* Increment number of inner iterations */
1175         inneriter                  += j_index_end - j_index_start;
1176
1177         /* Outer loop uses 27 flops */
1178     }
1179
1180     /* Increment number of outer iterations */
1181     outeriter        += nri;
1182
1183     /* Update outer/inner flops */
1184
1185     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*122);
1186 }