src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_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_ElecEwSh_VdwNone_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_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              vdwioffset1;
  70     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  71     int              vdwioffset2;
  72     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  73     int              vdwioffset3;
  74     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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           dummy_mask,cutoff_mask;
  86     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  87     __m128           one     = _mm_set1_ps(1.0);
  88     __m128           two     = _mm_set1_ps(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_ps(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102
 103     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 104     ewtab            = fr->ic->tabq_coul_FDV0;
 105     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 106     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 107
 108     /* Setup water-specific parameters */
 109     inr              = nlist->iinr[0];
 110     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 111     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 112     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 113
 114     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 115     rcutoff_scalar   = fr->rcoulomb;
 116     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 117     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 118
 119     /* Avoid stupid compiler warnings */
 120     jnrA = jnrB = jnrC = jnrD = 0;
 121     j_coord_offsetA = 0;
 122     j_coord_offsetB = 0;
 123     j_coord_offsetC = 0;
 124     j_coord_offsetD = 0;
 125
 126     outeriter        = 0;
 127     inneriter        = 0;
 128
 129     for(iidx=0;iidx<4*DIM;iidx++)
 130     {
 131         scratch[iidx] = 0.0;
 132     }
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 150                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 151
 152         fix1             = _mm_setzero_ps();
 153         fiy1             = _mm_setzero_ps();
 154         fiz1             = _mm_setzero_ps();
 155         fix2             = _mm_setzero_ps();
 156         fiy2             = _mm_setzero_ps();
 157         fiz2             = _mm_setzero_ps();
 158         fix3             = _mm_setzero_ps();
 159         fiy3             = _mm_setzero_ps();
 160         fiz3             = _mm_setzero_ps();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_ps();
 164
 165         /* Start inner kernel loop */
 166         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 167         {
 168
 169             /* Get j neighbor index, and coordinate index */
 170             jnrA             = jjnr[jidx];
 171             jnrB             = jjnr[jidx+1];
 172             jnrC             = jjnr[jidx+2];
 173             jnrD             = jjnr[jidx+3];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176             j_coord_offsetC  = DIM*jnrC;
 177             j_coord_offsetD  = DIM*jnrD;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               x+j_coord_offsetC,x+j_coord_offsetD,
 182                                               &jx0,&jy0,&jz0);
 183
 184             /* Calculate displacement vector */
 185             dx10             = _mm_sub_ps(ix1,jx0);
 186             dy10             = _mm_sub_ps(iy1,jy0);
 187             dz10             = _mm_sub_ps(iz1,jz0);
 188             dx20             = _mm_sub_ps(ix2,jx0);
 189             dy20             = _mm_sub_ps(iy2,jy0);
 190             dz20             = _mm_sub_ps(iz2,jz0);
 191             dx30             = _mm_sub_ps(ix3,jx0);
 192             dy30             = _mm_sub_ps(iy3,jy0);
 193             dz30             = _mm_sub_ps(iz3,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 197             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 198             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 199
 200             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 201             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 202             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 203
 204             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 205             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 206             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 207
 208             /* Load parameters for j particles */
 209             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 210                                                               charge+jnrC+0,charge+jnrD+0);
 211
 212             fjx0             = _mm_setzero_ps();
 213             fjy0             = _mm_setzero_ps();
 214             fjz0             = _mm_setzero_ps();
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             if (gmx_mm_any_lt(rsq10,rcutoff2))
 221             {
 222
 223             r10              = _mm_mul_ps(rsq10,rinv10);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq10             = _mm_mul_ps(iq1,jq0);
 227
 228             /* EWALD ELECTROSTATICS */
 229
 230             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 231             ewrt             = _mm_mul_ps(r10,ewtabscale);
 232             ewitab           = _mm_cvttps_epi32(ewrt);
 233             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 234             ewitab           = _mm_slli_epi32(ewitab,2);
 235             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 236             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 237             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 238             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 239             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 240             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 241             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 242             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 243             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 244
 245             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velec            = _mm_and_ps(velec,cutoff_mask);
 249             velecsum         = _mm_add_ps(velecsum,velec);
 250
 251             fscal            = felec;
 252
 253             fscal            = _mm_and_ps(fscal,cutoff_mask);
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_ps(fscal,dx10);
 257             ty               = _mm_mul_ps(fscal,dy10);
 258             tz               = _mm_mul_ps(fscal,dz10);
 259
 260             /* Update vectorial force */
 261             fix1             = _mm_add_ps(fix1,tx);
 262             fiy1             = _mm_add_ps(fiy1,ty);
 263             fiz1             = _mm_add_ps(fiz1,tz);
 264
 265             fjx0             = _mm_add_ps(fjx0,tx);
 266             fjy0             = _mm_add_ps(fjy0,ty);
 267             fjz0             = _mm_add_ps(fjz0,tz);
 268
 269             }
 270
 271             /**************************
 272              * CALCULATE INTERACTIONS *
 273              **************************/
 274
 275             if (gmx_mm_any_lt(rsq20,rcutoff2))
 276             {
 277
 278             r20              = _mm_mul_ps(rsq20,rinv20);
 279
 280             /* Compute parameters for interactions between i and j atoms */
 281             qq20             = _mm_mul_ps(iq2,jq0);
 282
 283             /* EWALD ELECTROSTATICS */
 284
 285             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 286             ewrt             = _mm_mul_ps(r20,ewtabscale);
 287             ewitab           = _mm_cvttps_epi32(ewrt);
 288             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 289             ewitab           = _mm_slli_epi32(ewitab,2);
 290             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 291             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 292             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 293             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 294             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 295             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 296             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 297             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 298             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 299
 300             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 301
 302             /* Update potential sum for this i atom from the interaction with this j atom. */
 303             velec            = _mm_and_ps(velec,cutoff_mask);
 304             velecsum         = _mm_add_ps(velecsum,velec);
 305
 306             fscal            = felec;
 307
 308             fscal            = _mm_and_ps(fscal,cutoff_mask);
 309
 310             /* Calculate temporary vectorial force */
 311             tx               = _mm_mul_ps(fscal,dx20);
 312             ty               = _mm_mul_ps(fscal,dy20);
 313             tz               = _mm_mul_ps(fscal,dz20);
 314
 315             /* Update vectorial force */
 316             fix2             = _mm_add_ps(fix2,tx);
 317             fiy2             = _mm_add_ps(fiy2,ty);
 318             fiz2             = _mm_add_ps(fiz2,tz);
 319
 320             fjx0             = _mm_add_ps(fjx0,tx);
 321             fjy0             = _mm_add_ps(fjy0,ty);
 322             fjz0             = _mm_add_ps(fjz0,tz);
 323
 324             }
 325
 326             /**************************
 327              * CALCULATE INTERACTIONS *
 328              **************************/
 329
 330             if (gmx_mm_any_lt(rsq30,rcutoff2))
 331             {
 332
 333             r30              = _mm_mul_ps(rsq30,rinv30);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq30             = _mm_mul_ps(iq3,jq0);
 337
 338             /* EWALD ELECTROSTATICS */
 339
 340             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 341             ewrt             = _mm_mul_ps(r30,ewtabscale);
 342             ewitab           = _mm_cvttps_epi32(ewrt);
 343             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 344             ewitab           = _mm_slli_epi32(ewitab,2);
 345             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 346             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 347             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 348             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 349             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 350             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 351             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 352             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 353             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 354
 355             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm_and_ps(velec,cutoff_mask);
 359             velecsum         = _mm_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             fscal            = _mm_and_ps(fscal,cutoff_mask);
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_ps(fscal,dx30);
 367             ty               = _mm_mul_ps(fscal,dy30);
 368             tz               = _mm_mul_ps(fscal,dz30);
 369
 370             /* Update vectorial force */
 371             fix3             = _mm_add_ps(fix3,tx);
 372             fiy3             = _mm_add_ps(fiy3,ty);
 373             fiz3             = _mm_add_ps(fiz3,tz);
 374
 375             fjx0             = _mm_add_ps(fjx0,tx);
 376             fjy0             = _mm_add_ps(fjy0,ty);
 377             fjz0             = _mm_add_ps(fjz0,tz);
 378
 379             }
 380
 381             fjptrA             = f+j_coord_offsetA;
 382             fjptrB             = f+j_coord_offsetB;
 383             fjptrC             = f+j_coord_offsetC;
 384             fjptrD             = f+j_coord_offsetD;
 385
 386             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 387
 388             /* Inner loop uses 138 flops */
 389         }
 390
 391         if(jidx<j_index_end)
 392         {
 393
 394             /* Get j neighbor index, and coordinate index */
 395             jnrlistA         = jjnr[jidx];
 396             jnrlistB         = jjnr[jidx+1];
 397             jnrlistC         = jjnr[jidx+2];
 398             jnrlistD         = jjnr[jidx+3];
 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       = (jnrlistA>=0) ? jnrlistA : 0;
 405             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 406             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 407             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 408             j_coord_offsetA  = DIM*jnrA;
 409             j_coord_offsetB  = DIM*jnrB;
 410             j_coord_offsetC  = DIM*jnrC;
 411             j_coord_offsetD  = DIM*jnrD;
 412
 413             /* load j atom coordinates */
 414             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 415                                               x+j_coord_offsetC,x+j_coord_offsetD,
 416                                               &jx0,&jy0,&jz0);
 417
 418             /* Calculate displacement vector */
 419             dx10             = _mm_sub_ps(ix1,jx0);
 420             dy10             = _mm_sub_ps(iy1,jy0);
 421             dz10             = _mm_sub_ps(iz1,jz0);
 422             dx20             = _mm_sub_ps(ix2,jx0);
 423             dy20             = _mm_sub_ps(iy2,jy0);
 424             dz20             = _mm_sub_ps(iz2,jz0);
 425             dx30             = _mm_sub_ps(ix3,jx0);
 426             dy30             = _mm_sub_ps(iy3,jy0);
 427             dz30             = _mm_sub_ps(iz3,jz0);
 428
 429             /* Calculate squared distance and things based on it */
 430             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 431             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 432             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 433
 434             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 435             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 436             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 437
 438             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 439             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 440             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 441
 442             /* Load parameters for j particles */
 443             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 444                                                               charge+jnrC+0,charge+jnrD+0);
 445
 446             fjx0             = _mm_setzero_ps();
 447             fjy0             = _mm_setzero_ps();
 448             fjz0             = _mm_setzero_ps();
 449
 450             /**************************
 451              * CALCULATE INTERACTIONS *
 452              **************************/
 453
 454             if (gmx_mm_any_lt(rsq10,rcutoff2))
 455             {
 456
 457             r10              = _mm_mul_ps(rsq10,rinv10);
 458             r10              = _mm_andnot_ps(dummy_mask,r10);
 459
 460             /* Compute parameters for interactions between i and j atoms */
 461             qq10             = _mm_mul_ps(iq1,jq0);
 462
 463             /* EWALD ELECTROSTATICS */
 464
 465             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 466             ewrt             = _mm_mul_ps(r10,ewtabscale);
 467             ewitab           = _mm_cvttps_epi32(ewrt);
 468             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 469             ewitab           = _mm_slli_epi32(ewitab,2);
 470             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 471             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 472             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 473             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 474             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 475             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 476             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 477             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 478             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 479
 480             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm_and_ps(velec,cutoff_mask);
 484             velec            = _mm_andnot_ps(dummy_mask,velec);
 485             velecsum         = _mm_add_ps(velecsum,velec);
 486
 487             fscal            = felec;
 488
 489             fscal            = _mm_and_ps(fscal,cutoff_mask);
 490
 491             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 492
 493             /* Calculate temporary vectorial force */
 494             tx               = _mm_mul_ps(fscal,dx10);
 495             ty               = _mm_mul_ps(fscal,dy10);
 496             tz               = _mm_mul_ps(fscal,dz10);
 497
 498             /* Update vectorial force */
 499             fix1             = _mm_add_ps(fix1,tx);
 500             fiy1             = _mm_add_ps(fiy1,ty);
 501             fiz1             = _mm_add_ps(fiz1,tz);
 502
 503             fjx0             = _mm_add_ps(fjx0,tx);
 504             fjy0             = _mm_add_ps(fjy0,ty);
 505             fjz0             = _mm_add_ps(fjz0,tz);
 506
 507             }
 508
 509             /**************************
 510              * CALCULATE INTERACTIONS *
 511              **************************/
 512
 513             if (gmx_mm_any_lt(rsq20,rcutoff2))
 514             {
 515
 516             r20              = _mm_mul_ps(rsq20,rinv20);
 517             r20              = _mm_andnot_ps(dummy_mask,r20);
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             qq20             = _mm_mul_ps(iq2,jq0);
 521
 522             /* EWALD ELECTROSTATICS */
 523
 524             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 525             ewrt             = _mm_mul_ps(r20,ewtabscale);
 526             ewitab           = _mm_cvttps_epi32(ewrt);
 527             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 531             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 532             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 533             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 534             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 535             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 536             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 537             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 538
 539             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 540
 541             /* Update potential sum for this i atom from the interaction with this j atom. */
 542             velec            = _mm_and_ps(velec,cutoff_mask);
 543             velec            = _mm_andnot_ps(dummy_mask,velec);
 544             velecsum         = _mm_add_ps(velecsum,velec);
 545
 546             fscal            = felec;
 547
 548             fscal            = _mm_and_ps(fscal,cutoff_mask);
 549
 550             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 551
 552             /* Calculate temporary vectorial force */
 553             tx               = _mm_mul_ps(fscal,dx20);
 554             ty               = _mm_mul_ps(fscal,dy20);
 555             tz               = _mm_mul_ps(fscal,dz20);
 556
 557             /* Update vectorial force */
 558             fix2             = _mm_add_ps(fix2,tx);
 559             fiy2             = _mm_add_ps(fiy2,ty);
 560             fiz2             = _mm_add_ps(fiz2,tz);
 561
 562             fjx0             = _mm_add_ps(fjx0,tx);
 563             fjy0             = _mm_add_ps(fjy0,ty);
 564             fjz0             = _mm_add_ps(fjz0,tz);
 565
 566             }
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             if (gmx_mm_any_lt(rsq30,rcutoff2))
 573             {
 574
 575             r30              = _mm_mul_ps(rsq30,rinv30);
 576             r30              = _mm_andnot_ps(dummy_mask,r30);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq30             = _mm_mul_ps(iq3,jq0);
 580
 581             /* EWALD ELECTROSTATICS */
 582
 583             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 584             ewrt             = _mm_mul_ps(r30,ewtabscale);
 585             ewitab           = _mm_cvttps_epi32(ewrt);
 586             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 587             ewitab           = _mm_slli_epi32(ewitab,2);
 588             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 589             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 590             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 591             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 592             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 593             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 594             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 595             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 596             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 597
 598             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm_and_ps(velec,cutoff_mask);
 602             velec            = _mm_andnot_ps(dummy_mask,velec);
 603             velecsum         = _mm_add_ps(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm_and_ps(fscal,cutoff_mask);
 608
 609             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm_mul_ps(fscal,dx30);
 613             ty               = _mm_mul_ps(fscal,dy30);
 614             tz               = _mm_mul_ps(fscal,dz30);
 615
 616             /* Update vectorial force */
 617             fix3             = _mm_add_ps(fix3,tx);
 618             fiy3             = _mm_add_ps(fiy3,ty);
 619             fiz3             = _mm_add_ps(fiz3,tz);
 620
 621             fjx0             = _mm_add_ps(fjx0,tx);
 622             fjy0             = _mm_add_ps(fjy0,ty);
 623             fjz0             = _mm_add_ps(fjz0,tz);
 624
 625             }
 626
 627             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 628             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 629             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 630             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 631
 632             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 633
 634             /* Inner loop uses 141 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 640                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 641
 642         ggid                        = gid[iidx];
 643         /* Update potential energies */
 644         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 645
 646         /* Increment number of inner iterations */
 647         inneriter                  += j_index_end - j_index_start;
 648
 649         /* Outer loop uses 19 flops */
 650     }
 651
 652     /* Increment number of outer iterations */
 653     outeriter        += nri;
 654
 655     /* Update outer/inner flops */
 656
 657     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
 658 }
 659 /*
 660  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse4_1_single
 661  * Electrostatics interaction: Ewald
 662  * VdW interaction:            None
 663  * Geometry:                   Water4-Particle
 664  * Calculate force/pot:        Force
 665  */
 666 void
 667 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse4_1_single
 668                     (t_nblist * gmx_restrict                nlist,
 669                      rvec * gmx_restrict                    xx,
 670                      rvec * gmx_restrict                    ff,
 671                      t_forcerec * gmx_restrict              fr,
 672                      t_mdatoms * gmx_restrict               mdatoms,
 673                      nb_kernel_data_t * gmx_restrict        kernel_data,
 674                      t_nrnb * gmx_restrict                  nrnb)
 675 {
 676     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 677      * just 0 for non-waters.
 678      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 679      * jnr indices corresponding to data put in the four positions in the SIMD register.
 680      */
 681     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 682     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 683     int              jnrA,jnrB,jnrC,jnrD;
 684     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 685     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 686     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 687     real             rcutoff_scalar;
 688     real             *shiftvec,*fshift,*x,*f;
 689     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 690     real             scratch[4*DIM];
 691     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 692     int              vdwioffset1;
 693     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 694     int              vdwioffset2;
 695     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 696     int              vdwioffset3;
 697     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 698     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 699     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 703     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 704     real             *charge;
 705     __m128i          ewitab;
 706     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 707     real             *ewtab;
 708     __m128           dummy_mask,cutoff_mask;
 709     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 710     __m128           one     = _mm_set1_ps(1.0);
 711     __m128           two     = _mm_set1_ps(2.0);
 712     x                = xx[0];
 713     f                = ff[0];
 714
 715     nri              = nlist->nri;
 716     iinr             = nlist->iinr;
 717     jindex           = nlist->jindex;
 718     jjnr             = nlist->jjnr;
 719     shiftidx         = nlist->shift;
 720     gid              = nlist->gid;
 721     shiftvec         = fr->shift_vec[0];
 722     fshift           = fr->fshift[0];
 723     facel            = _mm_set1_ps(fr->epsfac);
 724     charge           = mdatoms->chargeA;
 725
 726     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 727     ewtab            = fr->ic->tabq_coul_F;
 728     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 729     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 730
 731     /* Setup water-specific parameters */
 732     inr              = nlist->iinr[0];
 733     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 734     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 735     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 736
 737     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 738     rcutoff_scalar   = fr->rcoulomb;
 739     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 740     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 741
 742     /* Avoid stupid compiler warnings */
 743     jnrA = jnrB = jnrC = jnrD = 0;
 744     j_coord_offsetA = 0;
 745     j_coord_offsetB = 0;
 746     j_coord_offsetC = 0;
 747     j_coord_offsetD = 0;
 748
 749     outeriter        = 0;
 750     inneriter        = 0;
 751
 752     for(iidx=0;iidx<4*DIM;iidx++)
 753     {
 754         scratch[iidx] = 0.0;
 755     }
 756
 757     /* Start outer loop over neighborlists */
 758     for(iidx=0; iidx<nri; iidx++)
 759     {
 760         /* Load shift vector for this list */
 761         i_shift_offset   = DIM*shiftidx[iidx];
 762
 763         /* Load limits for loop over neighbors */
 764         j_index_start    = jindex[iidx];
 765         j_index_end      = jindex[iidx+1];
 766
 767         /* Get outer coordinate index */
 768         inr              = iinr[iidx];
 769         i_coord_offset   = DIM*inr;
 770
 771         /* Load i particle coords and add shift vector */
 772         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 773                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 774
 775         fix1             = _mm_setzero_ps();
 776         fiy1             = _mm_setzero_ps();
 777         fiz1             = _mm_setzero_ps();
 778         fix2             = _mm_setzero_ps();
 779         fiy2             = _mm_setzero_ps();
 780         fiz2             = _mm_setzero_ps();
 781         fix3             = _mm_setzero_ps();
 782         fiy3             = _mm_setzero_ps();
 783         fiz3             = _mm_setzero_ps();
 784
 785         /* Start inner kernel loop */
 786         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 787         {
 788
 789             /* Get j neighbor index, and coordinate index */
 790             jnrA             = jjnr[jidx];
 791             jnrB             = jjnr[jidx+1];
 792             jnrC             = jjnr[jidx+2];
 793             jnrD             = jjnr[jidx+3];
 794             j_coord_offsetA  = DIM*jnrA;
 795             j_coord_offsetB  = DIM*jnrB;
 796             j_coord_offsetC  = DIM*jnrC;
 797             j_coord_offsetD  = DIM*jnrD;
 798
 799             /* load j atom coordinates */
 800             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 801                                               x+j_coord_offsetC,x+j_coord_offsetD,
 802                                               &jx0,&jy0,&jz0);
 803
 804             /* Calculate displacement vector */
 805             dx10             = _mm_sub_ps(ix1,jx0);
 806             dy10             = _mm_sub_ps(iy1,jy0);
 807             dz10             = _mm_sub_ps(iz1,jz0);
 808             dx20             = _mm_sub_ps(ix2,jx0);
 809             dy20             = _mm_sub_ps(iy2,jy0);
 810             dz20             = _mm_sub_ps(iz2,jz0);
 811             dx30             = _mm_sub_ps(ix3,jx0);
 812             dy30             = _mm_sub_ps(iy3,jy0);
 813             dz30             = _mm_sub_ps(iz3,jz0);
 814
 815             /* Calculate squared distance and things based on it */
 816             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 817             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 818             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 819
 820             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 821             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 822             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 823
 824             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 825             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 826             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 827
 828             /* Load parameters for j particles */
 829             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 830                                                               charge+jnrC+0,charge+jnrD+0);
 831
 832             fjx0             = _mm_setzero_ps();
 833             fjy0             = _mm_setzero_ps();
 834             fjz0             = _mm_setzero_ps();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             if (gmx_mm_any_lt(rsq10,rcutoff2))
 841             {
 842
 843             r10              = _mm_mul_ps(rsq10,rinv10);
 844
 845             /* Compute parameters for interactions between i and j atoms */
 846             qq10             = _mm_mul_ps(iq1,jq0);
 847
 848             /* EWALD ELECTROSTATICS */
 849
 850             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 851             ewrt             = _mm_mul_ps(r10,ewtabscale);
 852             ewitab           = _mm_cvttps_epi32(ewrt);
 853             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 854             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 855                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 856                                          &ewtabF,&ewtabFn);
 857             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 858             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 859
 860             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 861
 862             fscal            = felec;
 863
 864             fscal            = _mm_and_ps(fscal,cutoff_mask);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm_mul_ps(fscal,dx10);
 868             ty               = _mm_mul_ps(fscal,dy10);
 869             tz               = _mm_mul_ps(fscal,dz10);
 870
 871             /* Update vectorial force */
 872             fix1             = _mm_add_ps(fix1,tx);
 873             fiy1             = _mm_add_ps(fiy1,ty);
 874             fiz1             = _mm_add_ps(fiz1,tz);
 875
 876             fjx0             = _mm_add_ps(fjx0,tx);
 877             fjy0             = _mm_add_ps(fjy0,ty);
 878             fjz0             = _mm_add_ps(fjz0,tz);
 879
 880             }
 881
 882             /**************************
 883              * CALCULATE INTERACTIONS *
 884              **************************/
 885
 886             if (gmx_mm_any_lt(rsq20,rcutoff2))
 887             {
 888
 889             r20              = _mm_mul_ps(rsq20,rinv20);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq20             = _mm_mul_ps(iq2,jq0);
 893
 894             /* EWALD ELECTROSTATICS */
 895
 896             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 897             ewrt             = _mm_mul_ps(r20,ewtabscale);
 898             ewitab           = _mm_cvttps_epi32(ewrt);
 899             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 900             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 901                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 902                                          &ewtabF,&ewtabFn);
 903             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 904             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 905
 906             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 907
 908             fscal            = felec;
 909
 910             fscal            = _mm_and_ps(fscal,cutoff_mask);
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm_mul_ps(fscal,dx20);
 914             ty               = _mm_mul_ps(fscal,dy20);
 915             tz               = _mm_mul_ps(fscal,dz20);
 916
 917             /* Update vectorial force */
 918             fix2             = _mm_add_ps(fix2,tx);
 919             fiy2             = _mm_add_ps(fiy2,ty);
 920             fiz2             = _mm_add_ps(fiz2,tz);
 921
 922             fjx0             = _mm_add_ps(fjx0,tx);
 923             fjy0             = _mm_add_ps(fjy0,ty);
 924             fjz0             = _mm_add_ps(fjz0,tz);
 925
 926             }
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             if (gmx_mm_any_lt(rsq30,rcutoff2))
 933             {
 934
 935             r30              = _mm_mul_ps(rsq30,rinv30);
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq30             = _mm_mul_ps(iq3,jq0);
 939
 940             /* EWALD ELECTROSTATICS */
 941
 942             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 943             ewrt             = _mm_mul_ps(r30,ewtabscale);
 944             ewitab           = _mm_cvttps_epi32(ewrt);
 945             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 946             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 947                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 948                                          &ewtabF,&ewtabFn);
 949             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 950             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 951
 952             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 953
 954             fscal            = felec;
 955
 956             fscal            = _mm_and_ps(fscal,cutoff_mask);
 957
 958             /* Calculate temporary vectorial force */
 959             tx               = _mm_mul_ps(fscal,dx30);
 960             ty               = _mm_mul_ps(fscal,dy30);
 961             tz               = _mm_mul_ps(fscal,dz30);
 962
 963             /* Update vectorial force */
 964             fix3             = _mm_add_ps(fix3,tx);
 965             fiy3             = _mm_add_ps(fiy3,ty);
 966             fiz3             = _mm_add_ps(fiz3,tz);
 967
 968             fjx0             = _mm_add_ps(fjx0,tx);
 969             fjy0             = _mm_add_ps(fjy0,ty);
 970             fjz0             = _mm_add_ps(fjz0,tz);
 971
 972             }
 973
 974             fjptrA             = f+j_coord_offsetA;
 975             fjptrB             = f+j_coord_offsetB;
 976             fjptrC             = f+j_coord_offsetC;
 977             fjptrD             = f+j_coord_offsetD;
 978
 979             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 980
 981             /* Inner loop uses 117 flops */
 982         }
 983
 984         if(jidx<j_index_end)
 985         {
 986
 987             /* Get j neighbor index, and coordinate index */
 988             jnrlistA         = jjnr[jidx];
 989             jnrlistB         = jjnr[jidx+1];
 990             jnrlistC         = jjnr[jidx+2];
 991             jnrlistD         = jjnr[jidx+3];
 992             /* Sign of each element will be negative for non-real atoms.
 993              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 994              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 995              */
 996             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 997             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 998             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 999             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1000             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1001             j_coord_offsetA  = DIM*jnrA;
1002             j_coord_offsetB  = DIM*jnrB;
1003             j_coord_offsetC  = DIM*jnrC;
1004             j_coord_offsetD  = DIM*jnrD;
1005
1006             /* load j atom coordinates */
1007             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1008                                               x+j_coord_offsetC,x+j_coord_offsetD,
1009                                               &jx0,&jy0,&jz0);
1010
1011             /* Calculate displacement vector */
1012             dx10             = _mm_sub_ps(ix1,jx0);
1013             dy10             = _mm_sub_ps(iy1,jy0);
1014             dz10             = _mm_sub_ps(iz1,jz0);
1015             dx20             = _mm_sub_ps(ix2,jx0);
1016             dy20             = _mm_sub_ps(iy2,jy0);
1017             dz20             = _mm_sub_ps(iz2,jz0);
1018             dx30             = _mm_sub_ps(ix3,jx0);
1019             dy30             = _mm_sub_ps(iy3,jy0);
1020             dz30             = _mm_sub_ps(iz3,jz0);
1021
1022             /* Calculate squared distance and things based on it */
1023             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1024             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1025             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1026
1027             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1028             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1029             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1030
1031             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1032             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1033             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1034
1035             /* Load parameters for j particles */
1036             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1037                                                               charge+jnrC+0,charge+jnrD+0);
1038
1039             fjx0             = _mm_setzero_ps();
1040             fjy0             = _mm_setzero_ps();
1041             fjz0             = _mm_setzero_ps();
1042
1043             /**************************
1044              * CALCULATE INTERACTIONS *
1045              **************************/
1046
1047             if (gmx_mm_any_lt(rsq10,rcutoff2))
1048             {
1049
1050             r10              = _mm_mul_ps(rsq10,rinv10);
1051             r10              = _mm_andnot_ps(dummy_mask,r10);
1052
1053             /* Compute parameters for interactions between i and j atoms */
1054             qq10             = _mm_mul_ps(iq1,jq0);
1055
1056             /* EWALD ELECTROSTATICS */
1057
1058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059             ewrt             = _mm_mul_ps(r10,ewtabscale);
1060             ewitab           = _mm_cvttps_epi32(ewrt);
1061             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1062             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1063                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1064                                          &ewtabF,&ewtabFn);
1065             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1066             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1067
1068             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1069
1070             fscal            = felec;
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,dx10);
1078             ty               = _mm_mul_ps(fscal,dy10);
1079             tz               = _mm_mul_ps(fscal,dz10);
1080
1081             /* Update vectorial force */
1082             fix1             = _mm_add_ps(fix1,tx);
1083             fiy1             = _mm_add_ps(fiy1,ty);
1084             fiz1             = _mm_add_ps(fiz1,tz);
1085
1086             fjx0             = _mm_add_ps(fjx0,tx);
1087             fjy0             = _mm_add_ps(fjy0,ty);
1088             fjz0             = _mm_add_ps(fjz0,tz);
1089
1090             }
1091
1092             /**************************
1093              * CALCULATE INTERACTIONS *
1094              **************************/
1095
1096             if (gmx_mm_any_lt(rsq20,rcutoff2))
1097             {
1098
1099             r20              = _mm_mul_ps(rsq20,rinv20);
1100             r20              = _mm_andnot_ps(dummy_mask,r20);
1101
1102             /* Compute parameters for interactions between i and j atoms */
1103             qq20             = _mm_mul_ps(iq2,jq0);
1104
1105             /* EWALD ELECTROSTATICS */
1106
1107             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1108             ewrt             = _mm_mul_ps(r20,ewtabscale);
1109             ewitab           = _mm_cvttps_epi32(ewrt);
1110             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1111             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1112                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1113                                          &ewtabF,&ewtabFn);
1114             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1115             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1116
1117             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1118
1119             fscal            = felec;
1120
1121             fscal            = _mm_and_ps(fscal,cutoff_mask);
1122
1123             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm_mul_ps(fscal,dx20);
1127             ty               = _mm_mul_ps(fscal,dy20);
1128             tz               = _mm_mul_ps(fscal,dz20);
1129
1130             /* Update vectorial force */
1131             fix2             = _mm_add_ps(fix2,tx);
1132             fiy2             = _mm_add_ps(fiy2,ty);
1133             fiz2             = _mm_add_ps(fiz2,tz);
1134
1135             fjx0             = _mm_add_ps(fjx0,tx);
1136             fjy0             = _mm_add_ps(fjy0,ty);
1137             fjz0             = _mm_add_ps(fjz0,tz);
1138
1139             }
1140
1141             /**************************
1142              * CALCULATE INTERACTIONS *
1143              **************************/
1144
1145             if (gmx_mm_any_lt(rsq30,rcutoff2))
1146             {
1147
1148             r30              = _mm_mul_ps(rsq30,rinv30);
1149             r30              = _mm_andnot_ps(dummy_mask,r30);
1150
1151             /* Compute parameters for interactions between i and j atoms */
1152             qq30             = _mm_mul_ps(iq3,jq0);
1153
1154             /* EWALD ELECTROSTATICS */
1155
1156             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1157             ewrt             = _mm_mul_ps(r30,ewtabscale);
1158             ewitab           = _mm_cvttps_epi32(ewrt);
1159             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1160             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1161                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1162                                          &ewtabF,&ewtabFn);
1163             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1164             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1165
1166             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1167
1168             fscal            = felec;
1169
1170             fscal            = _mm_and_ps(fscal,cutoff_mask);
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm_mul_ps(fscal,dx30);
1176             ty               = _mm_mul_ps(fscal,dy30);
1177             tz               = _mm_mul_ps(fscal,dz30);
1178
1179             /* Update vectorial force */
1180             fix3             = _mm_add_ps(fix3,tx);
1181             fiy3             = _mm_add_ps(fiy3,ty);
1182             fiz3             = _mm_add_ps(fiz3,tz);
1183
1184             fjx0             = _mm_add_ps(fjx0,tx);
1185             fjy0             = _mm_add_ps(fjy0,ty);
1186             fjz0             = _mm_add_ps(fjz0,tz);
1187
1188             }
1189
1190             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1191             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1192             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1193             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1194
1195             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1196
1197             /* Inner loop uses 120 flops */
1198         }
1199
1200         /* End of innermost loop */
1201
1202         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1203                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1204
1205         /* Increment number of inner iterations */
1206         inneriter                  += j_index_end - j_index_start;
1207
1208         /* Outer loop uses 18 flops */
1209     }
1210
1211     /* Increment number of outer iterations */
1212     outeriter        += nri;
1213
1214     /* Update outer/inner flops */
1215
1216     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1217 }