src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSw_VdwNone_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_VF_sse4_1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     __m128i          ewitab;
  83     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  84     real             *ewtab;
  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
 105     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 106     ewtab            = fr->ic->tabq_coul_FDV0;
 107     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 108     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 109
 110     /* Setup water-specific parameters */
 111     inr              = nlist->iinr[0];
 112     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 113     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 114     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 115
 116     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 117     rcutoff_scalar   = fr->rcoulomb;
 118     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 119     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 120
 121     rswitch_scalar   = fr->rcoulomb_switch;
 122     rswitch          = _mm_set1_ps(rswitch_scalar);
 123     /* Setup switch parameters */
 124     d_scalar         = rcutoff_scalar-rswitch_scalar;
 125     d                = _mm_set1_ps(d_scalar);
 126     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 127     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 128     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 129     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 130     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 131     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 132
 133     /* Avoid stupid compiler warnings */
 134     jnrA = jnrB = jnrC = jnrD = 0;
 135     j_coord_offsetA = 0;
 136     j_coord_offsetB = 0;
 137     j_coord_offsetC = 0;
 138     j_coord_offsetD = 0;
 139
 140     outeriter        = 0;
 141     inneriter        = 0;
 142
 143     for(iidx=0;iidx<4*DIM;iidx++)
 144     {
 145         scratch[iidx] = 0.0;
 146     }
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 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         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169         fix1             = _mm_setzero_ps();
 170         fiy1             = _mm_setzero_ps();
 171         fiz1             = _mm_setzero_ps();
 172         fix2             = _mm_setzero_ps();
 173         fiy2             = _mm_setzero_ps();
 174         fiz2             = _mm_setzero_ps();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213
 214             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 215             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217
 218             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 219             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 220             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 221
 222             /* Load parameters for j particles */
 223             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 224                                                               charge+jnrC+0,charge+jnrD+0);
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             if (gmx_mm_any_lt(rsq00,rcutoff2))
 231             {
 232
 233             r00              = _mm_mul_ps(rsq00,rinv00);
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq00             = _mm_mul_ps(iq0,jq0);
 237
 238             /* EWALD ELECTROSTATICS */
 239
 240             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 241             ewrt             = _mm_mul_ps(r00,ewtabscale);
 242             ewitab           = _mm_cvttps_epi32(ewrt);
 243             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 244             ewitab           = _mm_slli_epi32(ewitab,2);
 245             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 246             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 247             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 248             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 249             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 250             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 251             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 252             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 253             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 254
 255             d                = _mm_sub_ps(r00,rswitch);
 256             d                = _mm_max_ps(d,_mm_setzero_ps());
 257             d2               = _mm_mul_ps(d,d);
 258             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)))))));
 259
 260             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 261
 262             /* Evaluate switch function */
 263             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 264             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 265             velec            = _mm_mul_ps(velec,sw);
 266             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             velec            = _mm_and_ps(velec,cutoff_mask);
 270             velecsum         = _mm_add_ps(velecsum,velec);
 271
 272             fscal            = felec;
 273
 274             fscal            = _mm_and_ps(fscal,cutoff_mask);
 275
 276             /* Calculate temporary vectorial force */
 277             tx               = _mm_mul_ps(fscal,dx00);
 278             ty               = _mm_mul_ps(fscal,dy00);
 279             tz               = _mm_mul_ps(fscal,dz00);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_add_ps(fix0,tx);
 283             fiy0             = _mm_add_ps(fiy0,ty);
 284             fiz0             = _mm_add_ps(fiz0,tz);
 285
 286             fjptrA             = f+j_coord_offsetA;
 287             fjptrB             = f+j_coord_offsetB;
 288             fjptrC             = f+j_coord_offsetC;
 289             fjptrD             = f+j_coord_offsetD;
 290             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 291
 292             }
 293
 294             /**************************
 295              * CALCULATE INTERACTIONS *
 296              **************************/
 297
 298             if (gmx_mm_any_lt(rsq10,rcutoff2))
 299             {
 300
 301             r10              = _mm_mul_ps(rsq10,rinv10);
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* EWALD ELECTROSTATICS */
 307
 308             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 309             ewrt             = _mm_mul_ps(r10,ewtabscale);
 310             ewitab           = _mm_cvttps_epi32(ewrt);
 311             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 312             ewitab           = _mm_slli_epi32(ewitab,2);
 313             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 314             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 315             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 316             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 317             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 318             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 319             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 320             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 321             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 322
 323             d                = _mm_sub_ps(r10,rswitch);
 324             d                = _mm_max_ps(d,_mm_setzero_ps());
 325             d2               = _mm_mul_ps(d,d);
 326             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)))))));
 327
 328             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 329
 330             /* Evaluate switch function */
 331             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 332             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 333             velec            = _mm_mul_ps(velec,sw);
 334             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_and_ps(velec,cutoff_mask);
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             fscal            = _mm_and_ps(fscal,cutoff_mask);
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm_mul_ps(fscal,dx10);
 346             ty               = _mm_mul_ps(fscal,dy10);
 347             tz               = _mm_mul_ps(fscal,dz10);
 348
 349             /* Update vectorial force */
 350             fix1             = _mm_add_ps(fix1,tx);
 351             fiy1             = _mm_add_ps(fiy1,ty);
 352             fiz1             = _mm_add_ps(fiz1,tz);
 353
 354             fjptrA             = f+j_coord_offsetA;
 355             fjptrB             = f+j_coord_offsetB;
 356             fjptrC             = f+j_coord_offsetC;
 357             fjptrD             = f+j_coord_offsetD;
 358             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 359
 360             }
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             if (gmx_mm_any_lt(rsq20,rcutoff2))
 367             {
 368
 369             r20              = _mm_mul_ps(rsq20,rinv20);
 370
 371             /* Compute parameters for interactions between i and j atoms */
 372             qq20             = _mm_mul_ps(iq2,jq0);
 373
 374             /* EWALD ELECTROSTATICS */
 375
 376             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 377             ewrt             = _mm_mul_ps(r20,ewtabscale);
 378             ewitab           = _mm_cvttps_epi32(ewrt);
 379             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 380             ewitab           = _mm_slli_epi32(ewitab,2);
 381             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 382             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 383             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 384             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 385             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 386             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 387             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 388             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 389             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 390
 391             d                = _mm_sub_ps(r20,rswitch);
 392             d                = _mm_max_ps(d,_mm_setzero_ps());
 393             d2               = _mm_mul_ps(d,d);
 394             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)))))));
 395
 396             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 397
 398             /* Evaluate switch function */
 399             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 400             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 401             velec            = _mm_mul_ps(velec,sw);
 402             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 403
 404             /* Update potential sum for this i atom from the interaction with this j atom. */
 405             velec            = _mm_and_ps(velec,cutoff_mask);
 406             velecsum         = _mm_add_ps(velecsum,velec);
 407
 408             fscal            = felec;
 409
 410             fscal            = _mm_and_ps(fscal,cutoff_mask);
 411
 412             /* Calculate temporary vectorial force */
 413             tx               = _mm_mul_ps(fscal,dx20);
 414             ty               = _mm_mul_ps(fscal,dy20);
 415             tz               = _mm_mul_ps(fscal,dz20);
 416
 417             /* Update vectorial force */
 418             fix2             = _mm_add_ps(fix2,tx);
 419             fiy2             = _mm_add_ps(fiy2,ty);
 420             fiz2             = _mm_add_ps(fiz2,tz);
 421
 422             fjptrA             = f+j_coord_offsetA;
 423             fjptrB             = f+j_coord_offsetB;
 424             fjptrC             = f+j_coord_offsetC;
 425             fjptrD             = f+j_coord_offsetD;
 426             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 427
 428             }
 429
 430             /* Inner loop uses 195 flops */
 431         }
 432
 433         if(jidx<j_index_end)
 434         {
 435
 436             /* Get j neighbor index, and coordinate index */
 437             jnrlistA         = jjnr[jidx];
 438             jnrlistB         = jjnr[jidx+1];
 439             jnrlistC         = jjnr[jidx+2];
 440             jnrlistD         = jjnr[jidx+3];
 441             /* Sign of each element will be negative for non-real atoms.
 442              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 443              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 444              */
 445             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 446             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 447             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 448             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 449             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 450             j_coord_offsetA  = DIM*jnrA;
 451             j_coord_offsetB  = DIM*jnrB;
 452             j_coord_offsetC  = DIM*jnrC;
 453             j_coord_offsetD  = DIM*jnrD;
 454
 455             /* load j atom coordinates */
 456             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 457                                               x+j_coord_offsetC,x+j_coord_offsetD,
 458                                               &jx0,&jy0,&jz0);
 459
 460             /* Calculate displacement vector */
 461             dx00             = _mm_sub_ps(ix0,jx0);
 462             dy00             = _mm_sub_ps(iy0,jy0);
 463             dz00             = _mm_sub_ps(iz0,jz0);
 464             dx10             = _mm_sub_ps(ix1,jx0);
 465             dy10             = _mm_sub_ps(iy1,jy0);
 466             dz10             = _mm_sub_ps(iz1,jz0);
 467             dx20             = _mm_sub_ps(ix2,jx0);
 468             dy20             = _mm_sub_ps(iy2,jy0);
 469             dz20             = _mm_sub_ps(iz2,jz0);
 470
 471             /* Calculate squared distance and things based on it */
 472             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 473             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 474             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 475
 476             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 477             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 478             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 479
 480             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 481             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 482             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 483
 484             /* Load parameters for j particles */
 485             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 486                                                               charge+jnrC+0,charge+jnrD+0);
 487
 488             /**************************
 489              * CALCULATE INTERACTIONS *
 490              **************************/
 491
 492             if (gmx_mm_any_lt(rsq00,rcutoff2))
 493             {
 494
 495             r00              = _mm_mul_ps(rsq00,rinv00);
 496             r00              = _mm_andnot_ps(dummy_mask,r00);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq00             = _mm_mul_ps(iq0,jq0);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 504             ewrt             = _mm_mul_ps(r00,ewtabscale);
 505             ewitab           = _mm_cvttps_epi32(ewrt);
 506             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 507             ewitab           = _mm_slli_epi32(ewitab,2);
 508             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 509             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 510             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 511             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 512             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 513             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 514             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 515             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 516             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 517
 518             d                = _mm_sub_ps(r00,rswitch);
 519             d                = _mm_max_ps(d,_mm_setzero_ps());
 520             d2               = _mm_mul_ps(d,d);
 521             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)))))));
 522
 523             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 524
 525             /* Evaluate switch function */
 526             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 527             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 528             velec            = _mm_mul_ps(velec,sw);
 529             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 530
 531             /* Update potential sum for this i atom from the interaction with this j atom. */
 532             velec            = _mm_and_ps(velec,cutoff_mask);
 533             velec            = _mm_andnot_ps(dummy_mask,velec);
 534             velecsum         = _mm_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_and_ps(fscal,cutoff_mask);
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx00);
 544             ty               = _mm_mul_ps(fscal,dy00);
 545             tz               = _mm_mul_ps(fscal,dz00);
 546
 547             /* Update vectorial force */
 548             fix0             = _mm_add_ps(fix0,tx);
 549             fiy0             = _mm_add_ps(fiy0,ty);
 550             fiz0             = _mm_add_ps(fiz0,tz);
 551
 552             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 553             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 554             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 555             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 556             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 557
 558             }
 559
 560             /**************************
 561              * CALCULATE INTERACTIONS *
 562              **************************/
 563
 564             if (gmx_mm_any_lt(rsq10,rcutoff2))
 565             {
 566
 567             r10              = _mm_mul_ps(rsq10,rinv10);
 568             r10              = _mm_andnot_ps(dummy_mask,r10);
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq10             = _mm_mul_ps(iq1,jq0);
 572
 573             /* EWALD ELECTROSTATICS */
 574
 575             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 576             ewrt             = _mm_mul_ps(r10,ewtabscale);
 577             ewitab           = _mm_cvttps_epi32(ewrt);
 578             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 579             ewitab           = _mm_slli_epi32(ewitab,2);
 580             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 581             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 582             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 583             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 584             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 585             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 586             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 587             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 588             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 589
 590             d                = _mm_sub_ps(r10,rswitch);
 591             d                = _mm_max_ps(d,_mm_setzero_ps());
 592             d2               = _mm_mul_ps(d,d);
 593             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)))))));
 594
 595             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 596
 597             /* Evaluate switch function */
 598             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 599             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 600             velec            = _mm_mul_ps(velec,sw);
 601             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 602
 603             /* Update potential sum for this i atom from the interaction with this j atom. */
 604             velec            = _mm_and_ps(velec,cutoff_mask);
 605             velec            = _mm_andnot_ps(dummy_mask,velec);
 606             velecsum         = _mm_add_ps(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm_and_ps(fscal,cutoff_mask);
 611
 612             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 613
 614             /* Calculate temporary vectorial force */
 615             tx               = _mm_mul_ps(fscal,dx10);
 616             ty               = _mm_mul_ps(fscal,dy10);
 617             tz               = _mm_mul_ps(fscal,dz10);
 618
 619             /* Update vectorial force */
 620             fix1             = _mm_add_ps(fix1,tx);
 621             fiy1             = _mm_add_ps(fiy1,ty);
 622             fiz1             = _mm_add_ps(fiz1,tz);
 623
 624             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 625             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 626             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 627             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 628             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 629
 630             }
 631
 632             /**************************
 633              * CALCULATE INTERACTIONS *
 634              **************************/
 635
 636             if (gmx_mm_any_lt(rsq20,rcutoff2))
 637             {
 638
 639             r20              = _mm_mul_ps(rsq20,rinv20);
 640             r20              = _mm_andnot_ps(dummy_mask,r20);
 641
 642             /* Compute parameters for interactions between i and j atoms */
 643             qq20             = _mm_mul_ps(iq2,jq0);
 644
 645             /* EWALD ELECTROSTATICS */
 646
 647             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 648             ewrt             = _mm_mul_ps(r20,ewtabscale);
 649             ewitab           = _mm_cvttps_epi32(ewrt);
 650             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 651             ewitab           = _mm_slli_epi32(ewitab,2);
 652             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 653             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 654             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 655             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 656             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 657             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 658             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 659             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 660             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 661
 662             d                = _mm_sub_ps(r20,rswitch);
 663             d                = _mm_max_ps(d,_mm_setzero_ps());
 664             d2               = _mm_mul_ps(d,d);
 665             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)))))));
 666
 667             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 668
 669             /* Evaluate switch function */
 670             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 671             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 672             velec            = _mm_mul_ps(velec,sw);
 673             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 674
 675             /* Update potential sum for this i atom from the interaction with this j atom. */
 676             velec            = _mm_and_ps(velec,cutoff_mask);
 677             velec            = _mm_andnot_ps(dummy_mask,velec);
 678             velecsum         = _mm_add_ps(velecsum,velec);
 679
 680             fscal            = felec;
 681
 682             fscal            = _mm_and_ps(fscal,cutoff_mask);
 683
 684             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 685
 686             /* Calculate temporary vectorial force */
 687             tx               = _mm_mul_ps(fscal,dx20);
 688             ty               = _mm_mul_ps(fscal,dy20);
 689             tz               = _mm_mul_ps(fscal,dz20);
 690
 691             /* Update vectorial force */
 692             fix2             = _mm_add_ps(fix2,tx);
 693             fiy2             = _mm_add_ps(fiy2,ty);
 694             fiz2             = _mm_add_ps(fiz2,tz);
 695
 696             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 697             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 698             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 699             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 700             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 701
 702             }
 703
 704             /* Inner loop uses 198 flops */
 705         }
 706
 707         /* End of innermost loop */
 708
 709         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 710                                               f+i_coord_offset,fshift+i_shift_offset);
 711
 712         ggid                        = gid[iidx];
 713         /* Update potential energies */
 714         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 715
 716         /* Increment number of inner iterations */
 717         inneriter                  += j_index_end - j_index_start;
 718
 719         /* Outer loop uses 19 flops */
 720     }
 721
 722     /* Increment number of outer iterations */
 723     outeriter        += nri;
 724
 725     /* Update outer/inner flops */
 726
 727     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*198);
 728 }
 729 /*
 730  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_sse4_1_single
 731  * Electrostatics interaction: Ewald
 732  * VdW interaction:            None
 733  * Geometry:                   Water3-Particle
 734  * Calculate force/pot:        Force
 735  */
 736 void
 737 nb_kernel_ElecEwSw_VdwNone_GeomW3P1_F_sse4_1_single
 738                     (t_nblist * gmx_restrict                nlist,
 739                      rvec * gmx_restrict                    xx,
 740                      rvec * gmx_restrict                    ff,
 741                      t_forcerec * gmx_restrict              fr,
 742                      t_mdatoms * gmx_restrict               mdatoms,
 743                      nb_kernel_data_t * gmx_restrict        kernel_data,
 744                      t_nrnb * gmx_restrict                  nrnb)
 745 {
 746     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 747      * just 0 for non-waters.
 748      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 749      * jnr indices corresponding to data put in the four positions in the SIMD register.
 750      */
 751     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 752     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 753     int              jnrA,jnrB,jnrC,jnrD;
 754     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 755     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 756     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 757     real             rcutoff_scalar;
 758     real             *shiftvec,*fshift,*x,*f;
 759     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 760     real             scratch[4*DIM];
 761     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 762     int              vdwioffset0;
 763     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 764     int              vdwioffset1;
 765     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 766     int              vdwioffset2;
 767     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 768     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 769     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 770     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 771     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 772     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 773     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 774     real             *charge;
 775     __m128i          ewitab;
 776     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 777     real             *ewtab;
 778     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 779     real             rswitch_scalar,d_scalar;
 780     __m128           dummy_mask,cutoff_mask;
 781     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 782     __m128           one     = _mm_set1_ps(1.0);
 783     __m128           two     = _mm_set1_ps(2.0);
 784     x                = xx[0];
 785     f                = ff[0];
 786
 787     nri              = nlist->nri;
 788     iinr             = nlist->iinr;
 789     jindex           = nlist->jindex;
 790     jjnr             = nlist->jjnr;
 791     shiftidx         = nlist->shift;
 792     gid              = nlist->gid;
 793     shiftvec         = fr->shift_vec[0];
 794     fshift           = fr->fshift[0];
 795     facel            = _mm_set1_ps(fr->epsfac);
 796     charge           = mdatoms->chargeA;
 797
 798     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 799     ewtab            = fr->ic->tabq_coul_FDV0;
 800     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 801     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 802
 803     /* Setup water-specific parameters */
 804     inr              = nlist->iinr[0];
 805     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 806     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 807     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 808
 809     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 810     rcutoff_scalar   = fr->rcoulomb;
 811     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 812     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 813
 814     rswitch_scalar   = fr->rcoulomb_switch;
 815     rswitch          = _mm_set1_ps(rswitch_scalar);
 816     /* Setup switch parameters */
 817     d_scalar         = rcutoff_scalar-rswitch_scalar;
 818     d                = _mm_set1_ps(d_scalar);
 819     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 820     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 821     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 822     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 823     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 824     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 825
 826     /* Avoid stupid compiler warnings */
 827     jnrA = jnrB = jnrC = jnrD = 0;
 828     j_coord_offsetA = 0;
 829     j_coord_offsetB = 0;
 830     j_coord_offsetC = 0;
 831     j_coord_offsetD = 0;
 832
 833     outeriter        = 0;
 834     inneriter        = 0;
 835
 836     for(iidx=0;iidx<4*DIM;iidx++)
 837     {
 838         scratch[iidx] = 0.0;
 839     }
 840
 841     /* Start outer loop over neighborlists */
 842     for(iidx=0; iidx<nri; iidx++)
 843     {
 844         /* Load shift vector for this list */
 845         i_shift_offset   = DIM*shiftidx[iidx];
 846
 847         /* Load limits for loop over neighbors */
 848         j_index_start    = jindex[iidx];
 849         j_index_end      = jindex[iidx+1];
 850
 851         /* Get outer coordinate index */
 852         inr              = iinr[iidx];
 853         i_coord_offset   = DIM*inr;
 854
 855         /* Load i particle coords and add shift vector */
 856         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 857                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 858
 859         fix0             = _mm_setzero_ps();
 860         fiy0             = _mm_setzero_ps();
 861         fiz0             = _mm_setzero_ps();
 862         fix1             = _mm_setzero_ps();
 863         fiy1             = _mm_setzero_ps();
 864         fiz1             = _mm_setzero_ps();
 865         fix2             = _mm_setzero_ps();
 866         fiy2             = _mm_setzero_ps();
 867         fiz2             = _mm_setzero_ps();
 868
 869         /* Start inner kernel loop */
 870         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 871         {
 872
 873             /* Get j neighbor index, and coordinate index */
 874             jnrA             = jjnr[jidx];
 875             jnrB             = jjnr[jidx+1];
 876             jnrC             = jjnr[jidx+2];
 877             jnrD             = jjnr[jidx+3];
 878             j_coord_offsetA  = DIM*jnrA;
 879             j_coord_offsetB  = DIM*jnrB;
 880             j_coord_offsetC  = DIM*jnrC;
 881             j_coord_offsetD  = DIM*jnrD;
 882
 883             /* load j atom coordinates */
 884             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 885                                               x+j_coord_offsetC,x+j_coord_offsetD,
 886                                               &jx0,&jy0,&jz0);
 887
 888             /* Calculate displacement vector */
 889             dx00             = _mm_sub_ps(ix0,jx0);
 890             dy00             = _mm_sub_ps(iy0,jy0);
 891             dz00             = _mm_sub_ps(iz0,jz0);
 892             dx10             = _mm_sub_ps(ix1,jx0);
 893             dy10             = _mm_sub_ps(iy1,jy0);
 894             dz10             = _mm_sub_ps(iz1,jz0);
 895             dx20             = _mm_sub_ps(ix2,jx0);
 896             dy20             = _mm_sub_ps(iy2,jy0);
 897             dz20             = _mm_sub_ps(iz2,jz0);
 898
 899             /* Calculate squared distance and things based on it */
 900             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 901             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 902             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 903
 904             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 905             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 906             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 907
 908             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 909             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 910             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 911
 912             /* Load parameters for j particles */
 913             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 914                                                               charge+jnrC+0,charge+jnrD+0);
 915
 916             /**************************
 917              * CALCULATE INTERACTIONS *
 918              **************************/
 919
 920             if (gmx_mm_any_lt(rsq00,rcutoff2))
 921             {
 922
 923             r00              = _mm_mul_ps(rsq00,rinv00);
 924
 925             /* Compute parameters for interactions between i and j atoms */
 926             qq00             = _mm_mul_ps(iq0,jq0);
 927
 928             /* EWALD ELECTROSTATICS */
 929
 930             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 931             ewrt             = _mm_mul_ps(r00,ewtabscale);
 932             ewitab           = _mm_cvttps_epi32(ewrt);
 933             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 934             ewitab           = _mm_slli_epi32(ewitab,2);
 935             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 936             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 937             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 938             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 939             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 940             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 941             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 942             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 943             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 944
 945             d                = _mm_sub_ps(r00,rswitch);
 946             d                = _mm_max_ps(d,_mm_setzero_ps());
 947             d2               = _mm_mul_ps(d,d);
 948             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)))))));
 949
 950             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 951
 952             /* Evaluate switch function */
 953             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 954             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
 955             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 956
 957             fscal            = felec;
 958
 959             fscal            = _mm_and_ps(fscal,cutoff_mask);
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm_mul_ps(fscal,dx00);
 963             ty               = _mm_mul_ps(fscal,dy00);
 964             tz               = _mm_mul_ps(fscal,dz00);
 965
 966             /* Update vectorial force */
 967             fix0             = _mm_add_ps(fix0,tx);
 968             fiy0             = _mm_add_ps(fiy0,ty);
 969             fiz0             = _mm_add_ps(fiz0,tz);
 970
 971             fjptrA             = f+j_coord_offsetA;
 972             fjptrB             = f+j_coord_offsetB;
 973             fjptrC             = f+j_coord_offsetC;
 974             fjptrD             = f+j_coord_offsetD;
 975             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 976
 977             }
 978
 979             /**************************
 980              * CALCULATE INTERACTIONS *
 981              **************************/
 982
 983             if (gmx_mm_any_lt(rsq10,rcutoff2))
 984             {
 985
 986             r10              = _mm_mul_ps(rsq10,rinv10);
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq10             = _mm_mul_ps(iq1,jq0);
 990
 991             /* EWALD ELECTROSTATICS */
 992
 993             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 994             ewrt             = _mm_mul_ps(r10,ewtabscale);
 995             ewitab           = _mm_cvttps_epi32(ewrt);
 996             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 997             ewitab           = _mm_slli_epi32(ewitab,2);
 998             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 999             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1000             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1001             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1002             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1003             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1004             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1005             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1006             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1007
1008             d                = _mm_sub_ps(r10,rswitch);
1009             d                = _mm_max_ps(d,_mm_setzero_ps());
1010             d2               = _mm_mul_ps(d,d);
1011             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)))))));
1012
1013             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1014
1015             /* Evaluate switch function */
1016             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1017             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1018             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1019
1020             fscal            = felec;
1021
1022             fscal            = _mm_and_ps(fscal,cutoff_mask);
1023
1024             /* Calculate temporary vectorial force */
1025             tx               = _mm_mul_ps(fscal,dx10);
1026             ty               = _mm_mul_ps(fscal,dy10);
1027             tz               = _mm_mul_ps(fscal,dz10);
1028
1029             /* Update vectorial force */
1030             fix1             = _mm_add_ps(fix1,tx);
1031             fiy1             = _mm_add_ps(fiy1,ty);
1032             fiz1             = _mm_add_ps(fiz1,tz);
1033
1034             fjptrA             = f+j_coord_offsetA;
1035             fjptrB             = f+j_coord_offsetB;
1036             fjptrC             = f+j_coord_offsetC;
1037             fjptrD             = f+j_coord_offsetD;
1038             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1039
1040             }
1041
1042             /**************************
1043              * CALCULATE INTERACTIONS *
1044              **************************/
1045
1046             if (gmx_mm_any_lt(rsq20,rcutoff2))
1047             {
1048
1049             r20              = _mm_mul_ps(rsq20,rinv20);
1050
1051             /* Compute parameters for interactions between i and j atoms */
1052             qq20             = _mm_mul_ps(iq2,jq0);
1053
1054             /* EWALD ELECTROSTATICS */
1055
1056             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1057             ewrt             = _mm_mul_ps(r20,ewtabscale);
1058             ewitab           = _mm_cvttps_epi32(ewrt);
1059             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1060             ewitab           = _mm_slli_epi32(ewitab,2);
1061             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1062             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1063             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1064             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1065             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1066             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1067             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1068             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1069             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1070
1071             d                = _mm_sub_ps(r20,rswitch);
1072             d                = _mm_max_ps(d,_mm_setzero_ps());
1073             d2               = _mm_mul_ps(d,d);
1074             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)))))));
1075
1076             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1077
1078             /* Evaluate switch function */
1079             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1080             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1081             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm_and_ps(fscal,cutoff_mask);
1086
1087             /* Calculate temporary vectorial force */
1088             tx               = _mm_mul_ps(fscal,dx20);
1089             ty               = _mm_mul_ps(fscal,dy20);
1090             tz               = _mm_mul_ps(fscal,dz20);
1091
1092             /* Update vectorial force */
1093             fix2             = _mm_add_ps(fix2,tx);
1094             fiy2             = _mm_add_ps(fiy2,ty);
1095             fiz2             = _mm_add_ps(fiz2,tz);
1096
1097             fjptrA             = f+j_coord_offsetA;
1098             fjptrB             = f+j_coord_offsetB;
1099             fjptrC             = f+j_coord_offsetC;
1100             fjptrD             = f+j_coord_offsetD;
1101             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1102
1103             }
1104
1105             /* Inner loop uses 186 flops */
1106         }
1107
1108         if(jidx<j_index_end)
1109         {
1110
1111             /* Get j neighbor index, and coordinate index */
1112             jnrlistA         = jjnr[jidx];
1113             jnrlistB         = jjnr[jidx+1];
1114             jnrlistC         = jjnr[jidx+2];
1115             jnrlistD         = jjnr[jidx+3];
1116             /* Sign of each element will be negative for non-real atoms.
1117              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1118              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1119              */
1120             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1121             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1122             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1123             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1124             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1125             j_coord_offsetA  = DIM*jnrA;
1126             j_coord_offsetB  = DIM*jnrB;
1127             j_coord_offsetC  = DIM*jnrC;
1128             j_coord_offsetD  = DIM*jnrD;
1129
1130             /* load j atom coordinates */
1131             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1132                                               x+j_coord_offsetC,x+j_coord_offsetD,
1133                                               &jx0,&jy0,&jz0);
1134
1135             /* Calculate displacement vector */
1136             dx00             = _mm_sub_ps(ix0,jx0);
1137             dy00             = _mm_sub_ps(iy0,jy0);
1138             dz00             = _mm_sub_ps(iz0,jz0);
1139             dx10             = _mm_sub_ps(ix1,jx0);
1140             dy10             = _mm_sub_ps(iy1,jy0);
1141             dz10             = _mm_sub_ps(iz1,jz0);
1142             dx20             = _mm_sub_ps(ix2,jx0);
1143             dy20             = _mm_sub_ps(iy2,jy0);
1144             dz20             = _mm_sub_ps(iz2,jz0);
1145
1146             /* Calculate squared distance and things based on it */
1147             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1148             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1149             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1150
1151             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1152             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1153             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1154
1155             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1156             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1157             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1158
1159             /* Load parameters for j particles */
1160             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1161                                                               charge+jnrC+0,charge+jnrD+0);
1162
1163             /**************************
1164              * CALCULATE INTERACTIONS *
1165              **************************/
1166
1167             if (gmx_mm_any_lt(rsq00,rcutoff2))
1168             {
1169
1170             r00              = _mm_mul_ps(rsq00,rinv00);
1171             r00              = _mm_andnot_ps(dummy_mask,r00);
1172
1173             /* Compute parameters for interactions between i and j atoms */
1174             qq00             = _mm_mul_ps(iq0,jq0);
1175
1176             /* EWALD ELECTROSTATICS */
1177
1178             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1179             ewrt             = _mm_mul_ps(r00,ewtabscale);
1180             ewitab           = _mm_cvttps_epi32(ewrt);
1181             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1182             ewitab           = _mm_slli_epi32(ewitab,2);
1183             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1184             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1185             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1186             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1187             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1188             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1189             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1190             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
1191             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1192
1193             d                = _mm_sub_ps(r00,rswitch);
1194             d                = _mm_max_ps(d,_mm_setzero_ps());
1195             d2               = _mm_mul_ps(d,d);
1196             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)))))));
1197
1198             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1199
1200             /* Evaluate switch function */
1201             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1202             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(velec,dsw)) );
1203             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1204
1205             fscal            = felec;
1206
1207             fscal            = _mm_and_ps(fscal,cutoff_mask);
1208
1209             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1210
1211             /* Calculate temporary vectorial force */
1212             tx               = _mm_mul_ps(fscal,dx00);
1213             ty               = _mm_mul_ps(fscal,dy00);
1214             tz               = _mm_mul_ps(fscal,dz00);
1215
1216             /* Update vectorial force */
1217             fix0             = _mm_add_ps(fix0,tx);
1218             fiy0             = _mm_add_ps(fiy0,ty);
1219             fiz0             = _mm_add_ps(fiz0,tz);
1220
1221             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1222             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1223             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1224             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1225             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1226
1227             }
1228
1229             /**************************
1230              * CALCULATE INTERACTIONS *
1231              **************************/
1232
1233             if (gmx_mm_any_lt(rsq10,rcutoff2))
1234             {
1235
1236             r10              = _mm_mul_ps(rsq10,rinv10);
1237             r10              = _mm_andnot_ps(dummy_mask,r10);
1238
1239             /* Compute parameters for interactions between i and j atoms */
1240             qq10             = _mm_mul_ps(iq1,jq0);
1241
1242             /* EWALD ELECTROSTATICS */
1243
1244             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1245             ewrt             = _mm_mul_ps(r10,ewtabscale);
1246             ewitab           = _mm_cvttps_epi32(ewrt);
1247             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1248             ewitab           = _mm_slli_epi32(ewitab,2);
1249             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1250             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1251             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1252             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1253             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1254             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1255             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1256             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1257             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1258
1259             d                = _mm_sub_ps(r10,rswitch);
1260             d                = _mm_max_ps(d,_mm_setzero_ps());
1261             d2               = _mm_mul_ps(d,d);
1262             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)))))));
1263
1264             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1265
1266             /* Evaluate switch function */
1267             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1268             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1269             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1270
1271             fscal            = felec;
1272
1273             fscal            = _mm_and_ps(fscal,cutoff_mask);
1274
1275             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1276
1277             /* Calculate temporary vectorial force */
1278             tx               = _mm_mul_ps(fscal,dx10);
1279             ty               = _mm_mul_ps(fscal,dy10);
1280             tz               = _mm_mul_ps(fscal,dz10);
1281
1282             /* Update vectorial force */
1283             fix1             = _mm_add_ps(fix1,tx);
1284             fiy1             = _mm_add_ps(fiy1,ty);
1285             fiz1             = _mm_add_ps(fiz1,tz);
1286
1287             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1288             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1289             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1290             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1291             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1292
1293             }
1294
1295             /**************************
1296              * CALCULATE INTERACTIONS *
1297              **************************/
1298
1299             if (gmx_mm_any_lt(rsq20,rcutoff2))
1300             {
1301
1302             r20              = _mm_mul_ps(rsq20,rinv20);
1303             r20              = _mm_andnot_ps(dummy_mask,r20);
1304
1305             /* Compute parameters for interactions between i and j atoms */
1306             qq20             = _mm_mul_ps(iq2,jq0);
1307
1308             /* EWALD ELECTROSTATICS */
1309
1310             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1311             ewrt             = _mm_mul_ps(r20,ewtabscale);
1312             ewitab           = _mm_cvttps_epi32(ewrt);
1313             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1314             ewitab           = _mm_slli_epi32(ewitab,2);
1315             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1316             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1317             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1318             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1319             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1320             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1321             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1322             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1323             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1324
1325             d                = _mm_sub_ps(r20,rswitch);
1326             d                = _mm_max_ps(d,_mm_setzero_ps());
1327             d2               = _mm_mul_ps(d,d);
1328             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)))))));
1329
1330             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1331
1332             /* Evaluate switch function */
1333             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1334             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1335             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1336
1337             fscal            = felec;
1338
1339             fscal            = _mm_and_ps(fscal,cutoff_mask);
1340
1341             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1342
1343             /* Calculate temporary vectorial force */
1344             tx               = _mm_mul_ps(fscal,dx20);
1345             ty               = _mm_mul_ps(fscal,dy20);
1346             tz               = _mm_mul_ps(fscal,dz20);
1347
1348             /* Update vectorial force */
1349             fix2             = _mm_add_ps(fix2,tx);
1350             fiy2             = _mm_add_ps(fiy2,ty);
1351             fiz2             = _mm_add_ps(fiz2,tz);
1352
1353             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1354             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1355             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1356             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1357             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1358
1359             }
1360
1361             /* Inner loop uses 189 flops */
1362         }
1363
1364         /* End of innermost loop */
1365
1366         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1367                                               f+i_coord_offset,fshift+i_shift_offset);
1368
1369         /* Increment number of inner iterations */
1370         inneriter                  += j_index_end - j_index_start;
1371
1372         /* Outer loop uses 18 flops */
1373     }
1374
1375     /* Increment number of outer iterations */
1376     outeriter        += nri;
1377
1378     /* Update outer/inner flops */
1379
1380     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*189);
1381 }