src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwLJ_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_ElecCoul_VdwLJ_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_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              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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 115     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 116     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 117     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 151
 152         fix0             = _mm_setzero_ps();
 153         fiy0             = _mm_setzero_ps();
 154         fiz0             = _mm_setzero_ps();
 155         fix1             = _mm_setzero_ps();
 156         fiy1             = _mm_setzero_ps();
 157         fiz1             = _mm_setzero_ps();
 158         fix2             = _mm_setzero_ps();
 159         fiy2             = _mm_setzero_ps();
 160         fiz2             = _mm_setzero_ps();
 161         fix3             = _mm_setzero_ps();
 162         fiy3             = _mm_setzero_ps();
 163         fiz3             = _mm_setzero_ps();
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_ps();
 167         vvdwsum          = _mm_setzero_ps();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               x+j_coord_offsetC,x+j_coord_offsetD,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_ps(ix0,jx0);
 190             dy00             = _mm_sub_ps(iy0,jy0);
 191             dz00             = _mm_sub_ps(iz0,jz0);
 192             dx10             = _mm_sub_ps(ix1,jx0);
 193             dy10             = _mm_sub_ps(iy1,jy0);
 194             dz10             = _mm_sub_ps(iz1,jz0);
 195             dx20             = _mm_sub_ps(ix2,jx0);
 196             dy20             = _mm_sub_ps(iy2,jy0);
 197             dz20             = _mm_sub_ps(iz2,jz0);
 198             dx30             = _mm_sub_ps(ix3,jx0);
 199             dy30             = _mm_sub_ps(iy3,jy0);
 200             dz30             = _mm_sub_ps(iz3,jz0);
 201
 202             /* Calculate squared distance and things based on it */
 203             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 204             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 205             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 206             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 207
 208             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 209             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 210             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 211
 212             rinvsq00         = gmx_mm_inv_ps(rsq00);
 213             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 215             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 231                                          vdwparam+vdwioffset0+vdwjidx0B,
 232                                          vdwparam+vdwioffset0+vdwjidx0C,
 233                                          vdwparam+vdwioffset0+vdwjidx0D,
 234                                          &c6_00,&c12_00);
 235
 236             /* LENNARD-JONES DISPERSION/REPULSION */
 237
 238             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 239             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 240             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 241             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 242             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 246
 247             fscal            = fvdw;
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_ps(fscal,dx00);
 251             ty               = _mm_mul_ps(fscal,dy00);
 252             tz               = _mm_mul_ps(fscal,dz00);
 253
 254             /* Update vectorial force */
 255             fix0             = _mm_add_ps(fix0,tx);
 256             fiy0             = _mm_add_ps(fiy0,ty);
 257             fiz0             = _mm_add_ps(fiz0,tz);
 258
 259             fjptrA             = f+j_coord_offsetA;
 260             fjptrB             = f+j_coord_offsetB;
 261             fjptrC             = f+j_coord_offsetC;
 262             fjptrD             = f+j_coord_offsetD;
 263             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq10             = _mm_mul_ps(iq1,jq0);
 271
 272             /* COULOMB ELECTROSTATICS */
 273             velec            = _mm_mul_ps(qq10,rinv10);
 274             felec            = _mm_mul_ps(velec,rinvsq10);
 275
 276             /* Update potential sum for this i atom from the interaction with this j atom. */
 277             velecsum         = _mm_add_ps(velecsum,velec);
 278
 279             fscal            = felec;
 280
 281             /* Calculate temporary vectorial force */
 282             tx               = _mm_mul_ps(fscal,dx10);
 283             ty               = _mm_mul_ps(fscal,dy10);
 284             tz               = _mm_mul_ps(fscal,dz10);
 285
 286             /* Update vectorial force */
 287             fix1             = _mm_add_ps(fix1,tx);
 288             fiy1             = _mm_add_ps(fiy1,ty);
 289             fiz1             = _mm_add_ps(fiz1,tz);
 290
 291             fjptrA             = f+j_coord_offsetA;
 292             fjptrB             = f+j_coord_offsetB;
 293             fjptrC             = f+j_coord_offsetC;
 294             fjptrD             = f+j_coord_offsetD;
 295             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq20             = _mm_mul_ps(iq2,jq0);
 303
 304             /* COULOMB ELECTROSTATICS */
 305             velec            = _mm_mul_ps(qq20,rinv20);
 306             felec            = _mm_mul_ps(velec,rinvsq20);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_ps(fscal,dx20);
 315             ty               = _mm_mul_ps(fscal,dy20);
 316             tz               = _mm_mul_ps(fscal,dz20);
 317
 318             /* Update vectorial force */
 319             fix2             = _mm_add_ps(fix2,tx);
 320             fiy2             = _mm_add_ps(fiy2,ty);
 321             fiz2             = _mm_add_ps(fiz2,tz);
 322
 323             fjptrA             = f+j_coord_offsetA;
 324             fjptrB             = f+j_coord_offsetB;
 325             fjptrC             = f+j_coord_offsetC;
 326             fjptrD             = f+j_coord_offsetD;
 327             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             /* Compute parameters for interactions between i and j atoms */
 334             qq30             = _mm_mul_ps(iq3,jq0);
 335
 336             /* COULOMB ELECTROSTATICS */
 337             velec            = _mm_mul_ps(qq30,rinv30);
 338             felec            = _mm_mul_ps(velec,rinvsq30);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velecsum         = _mm_add_ps(velecsum,velec);
 342
 343             fscal            = felec;
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_ps(fscal,dx30);
 347             ty               = _mm_mul_ps(fscal,dy30);
 348             tz               = _mm_mul_ps(fscal,dz30);
 349
 350             /* Update vectorial force */
 351             fix3             = _mm_add_ps(fix3,tx);
 352             fiy3             = _mm_add_ps(fiy3,ty);
 353             fiz3             = _mm_add_ps(fiz3,tz);
 354
 355             fjptrA             = f+j_coord_offsetA;
 356             fjptrB             = f+j_coord_offsetB;
 357             fjptrC             = f+j_coord_offsetC;
 358             fjptrD             = f+j_coord_offsetD;
 359             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 360
 361             /* Inner loop uses 116 flops */
 362         }
 363
 364         if(jidx<j_index_end)
 365         {
 366
 367             /* Get j neighbor index, and coordinate index */
 368             jnrlistA         = jjnr[jidx];
 369             jnrlistB         = jjnr[jidx+1];
 370             jnrlistC         = jjnr[jidx+2];
 371             jnrlistD         = jjnr[jidx+3];
 372             /* Sign of each element will be negative for non-real atoms.
 373              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 374              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 375              */
 376             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 377             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 378             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 379             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 380             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 381             j_coord_offsetA  = DIM*jnrA;
 382             j_coord_offsetB  = DIM*jnrB;
 383             j_coord_offsetC  = DIM*jnrC;
 384             j_coord_offsetD  = DIM*jnrD;
 385
 386             /* load j atom coordinates */
 387             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 388                                               x+j_coord_offsetC,x+j_coord_offsetD,
 389                                               &jx0,&jy0,&jz0);
 390
 391             /* Calculate displacement vector */
 392             dx00             = _mm_sub_ps(ix0,jx0);
 393             dy00             = _mm_sub_ps(iy0,jy0);
 394             dz00             = _mm_sub_ps(iz0,jz0);
 395             dx10             = _mm_sub_ps(ix1,jx0);
 396             dy10             = _mm_sub_ps(iy1,jy0);
 397             dz10             = _mm_sub_ps(iz1,jz0);
 398             dx20             = _mm_sub_ps(ix2,jx0);
 399             dy20             = _mm_sub_ps(iy2,jy0);
 400             dz20             = _mm_sub_ps(iz2,jz0);
 401             dx30             = _mm_sub_ps(ix3,jx0);
 402             dy30             = _mm_sub_ps(iy3,jy0);
 403             dz30             = _mm_sub_ps(iz3,jz0);
 404
 405             /* Calculate squared distance and things based on it */
 406             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 407             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 408             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 409             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 410
 411             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 412             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 413             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 414
 415             rinvsq00         = gmx_mm_inv_ps(rsq00);
 416             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 417             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 418             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 419
 420             /* Load parameters for j particles */
 421             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 422                                                               charge+jnrC+0,charge+jnrD+0);
 423             vdwjidx0A        = 2*vdwtype[jnrA+0];
 424             vdwjidx0B        = 2*vdwtype[jnrB+0];
 425             vdwjidx0C        = 2*vdwtype[jnrC+0];
 426             vdwjidx0D        = 2*vdwtype[jnrD+0];
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 434                                          vdwparam+vdwioffset0+vdwjidx0B,
 435                                          vdwparam+vdwioffset0+vdwjidx0C,
 436                                          vdwparam+vdwioffset0+vdwjidx0D,
 437                                          &c6_00,&c12_00);
 438
 439             /* LENNARD-JONES DISPERSION/REPULSION */
 440
 441             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 442             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 443             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 444             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 445             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 446
 447             /* Update potential sum for this i atom from the interaction with this j atom. */
 448             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 449             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 450
 451             fscal            = fvdw;
 452
 453             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 454
 455             /* Calculate temporary vectorial force */
 456             tx               = _mm_mul_ps(fscal,dx00);
 457             ty               = _mm_mul_ps(fscal,dy00);
 458             tz               = _mm_mul_ps(fscal,dz00);
 459
 460             /* Update vectorial force */
 461             fix0             = _mm_add_ps(fix0,tx);
 462             fiy0             = _mm_add_ps(fiy0,ty);
 463             fiz0             = _mm_add_ps(fiz0,tz);
 464
 465             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 466             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 467             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 468             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 469             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             /* Compute parameters for interactions between i and j atoms */
 476             qq10             = _mm_mul_ps(iq1,jq0);
 477
 478             /* COULOMB ELECTROSTATICS */
 479             velec            = _mm_mul_ps(qq10,rinv10);
 480             felec            = _mm_mul_ps(velec,rinvsq10);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm_andnot_ps(dummy_mask,velec);
 484             velecsum         = _mm_add_ps(velecsum,velec);
 485
 486             fscal            = felec;
 487
 488             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 489
 490             /* Calculate temporary vectorial force */
 491             tx               = _mm_mul_ps(fscal,dx10);
 492             ty               = _mm_mul_ps(fscal,dy10);
 493             tz               = _mm_mul_ps(fscal,dz10);
 494
 495             /* Update vectorial force */
 496             fix1             = _mm_add_ps(fix1,tx);
 497             fiy1             = _mm_add_ps(fiy1,ty);
 498             fiz1             = _mm_add_ps(fiz1,tz);
 499
 500             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 501             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 502             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 503             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 504             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq20             = _mm_mul_ps(iq2,jq0);
 512
 513             /* COULOMB ELECTROSTATICS */
 514             velec            = _mm_mul_ps(qq20,rinv20);
 515             felec            = _mm_mul_ps(velec,rinvsq20);
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_andnot_ps(dummy_mask,velec);
 519             velecsum         = _mm_add_ps(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 524
 525             /* Calculate temporary vectorial force */
 526             tx               = _mm_mul_ps(fscal,dx20);
 527             ty               = _mm_mul_ps(fscal,dy20);
 528             tz               = _mm_mul_ps(fscal,dz20);
 529
 530             /* Update vectorial force */
 531             fix2             = _mm_add_ps(fix2,tx);
 532             fiy2             = _mm_add_ps(fiy2,ty);
 533             fiz2             = _mm_add_ps(fiz2,tz);
 534
 535             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 536             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 537             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 538             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 539             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             /* Compute parameters for interactions between i and j atoms */
 546             qq30             = _mm_mul_ps(iq3,jq0);
 547
 548             /* COULOMB ELECTROSTATICS */
 549             velec            = _mm_mul_ps(qq30,rinv30);
 550             felec            = _mm_mul_ps(velec,rinvsq30);
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_andnot_ps(dummy_mask,velec);
 554             velecsum         = _mm_add_ps(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 559
 560             /* Calculate temporary vectorial force */
 561             tx               = _mm_mul_ps(fscal,dx30);
 562             ty               = _mm_mul_ps(fscal,dy30);
 563             tz               = _mm_mul_ps(fscal,dz30);
 564
 565             /* Update vectorial force */
 566             fix3             = _mm_add_ps(fix3,tx);
 567             fiy3             = _mm_add_ps(fiy3,ty);
 568             fiz3             = _mm_add_ps(fiz3,tz);
 569
 570             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 571             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 572             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 573             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 574             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 575
 576             /* Inner loop uses 116 flops */
 577         }
 578
 579         /* End of innermost loop */
 580
 581         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 582                                               f+i_coord_offset,fshift+i_shift_offset);
 583
 584         ggid                        = gid[iidx];
 585         /* Update potential energies */
 586         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 587         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 588
 589         /* Increment number of inner iterations */
 590         inneriter                  += j_index_end - j_index_start;
 591
 592         /* Outer loop uses 26 flops */
 593     }
 594
 595     /* Increment number of outer iterations */
 596     outeriter        += nri;
 597
 598     /* Update outer/inner flops */
 599
 600     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*116);
 601 }
 602 /*
 603  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse4_1_single
 604  * Electrostatics interaction: Coulomb
 605  * VdW interaction:            LennardJones
 606  * Geometry:                   Water4-Particle
 607  * Calculate force/pot:        Force
 608  */
 609 void
 610 nb_kernel_ElecCoul_VdwLJ_GeomW4P1_F_sse4_1_single
 611                     (t_nblist * gmx_restrict                nlist,
 612                      rvec * gmx_restrict                    xx,
 613                      rvec * gmx_restrict                    ff,
 614                      t_forcerec * gmx_restrict              fr,
 615                      t_mdatoms * gmx_restrict               mdatoms,
 616                      nb_kernel_data_t * gmx_restrict        kernel_data,
 617                      t_nrnb * gmx_restrict                  nrnb)
 618 {
 619     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 620      * just 0 for non-waters.
 621      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 622      * jnr indices corresponding to data put in the four positions in the SIMD register.
 623      */
 624     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 625     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 626     int              jnrA,jnrB,jnrC,jnrD;
 627     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 628     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 629     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 630     real             rcutoff_scalar;
 631     real             *shiftvec,*fshift,*x,*f;
 632     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 633     real             scratch[4*DIM];
 634     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 635     int              vdwioffset0;
 636     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 637     int              vdwioffset1;
 638     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 639     int              vdwioffset2;
 640     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 641     int              vdwioffset3;
 642     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 643     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 644     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 645     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 646     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 647     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 648     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 649     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 650     real             *charge;
 651     int              nvdwtype;
 652     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 653     int              *vdwtype;
 654     real             *vdwparam;
 655     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 656     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 657     __m128           dummy_mask,cutoff_mask;
 658     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 659     __m128           one     = _mm_set1_ps(1.0);
 660     __m128           two     = _mm_set1_ps(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_ps(fr->epsfac);
 673     charge           = mdatoms->chargeA;
 674     nvdwtype         = fr->ntype;
 675     vdwparam         = fr->nbfp;
 676     vdwtype          = mdatoms->typeA;
 677
 678     /* Setup water-specific parameters */
 679     inr              = nlist->iinr[0];
 680     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 681     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 682     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 683     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 684
 685     /* Avoid stupid compiler warnings */
 686     jnrA = jnrB = jnrC = jnrD = 0;
 687     j_coord_offsetA = 0;
 688     j_coord_offsetB = 0;
 689     j_coord_offsetC = 0;
 690     j_coord_offsetD = 0;
 691
 692     outeriter        = 0;
 693     inneriter        = 0;
 694
 695     for(iidx=0;iidx<4*DIM;iidx++)
 696     {
 697         scratch[iidx] = 0.0;
 698     }
 699
 700     /* Start outer loop over neighborlists */
 701     for(iidx=0; iidx<nri; iidx++)
 702     {
 703         /* Load shift vector for this list */
 704         i_shift_offset   = DIM*shiftidx[iidx];
 705
 706         /* Load limits for loop over neighbors */
 707         j_index_start    = jindex[iidx];
 708         j_index_end      = jindex[iidx+1];
 709
 710         /* Get outer coordinate index */
 711         inr              = iinr[iidx];
 712         i_coord_offset   = DIM*inr;
 713
 714         /* Load i particle coords and add shift vector */
 715         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 716                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 717
 718         fix0             = _mm_setzero_ps();
 719         fiy0             = _mm_setzero_ps();
 720         fiz0             = _mm_setzero_ps();
 721         fix1             = _mm_setzero_ps();
 722         fiy1             = _mm_setzero_ps();
 723         fiz1             = _mm_setzero_ps();
 724         fix2             = _mm_setzero_ps();
 725         fiy2             = _mm_setzero_ps();
 726         fiz2             = _mm_setzero_ps();
 727         fix3             = _mm_setzero_ps();
 728         fiy3             = _mm_setzero_ps();
 729         fiz3             = _mm_setzero_ps();
 730
 731         /* Start inner kernel loop */
 732         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 733         {
 734
 735             /* Get j neighbor index, and coordinate index */
 736             jnrA             = jjnr[jidx];
 737             jnrB             = jjnr[jidx+1];
 738             jnrC             = jjnr[jidx+2];
 739             jnrD             = jjnr[jidx+3];
 740             j_coord_offsetA  = DIM*jnrA;
 741             j_coord_offsetB  = DIM*jnrB;
 742             j_coord_offsetC  = DIM*jnrC;
 743             j_coord_offsetD  = DIM*jnrD;
 744
 745             /* load j atom coordinates */
 746             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 747                                               x+j_coord_offsetC,x+j_coord_offsetD,
 748                                               &jx0,&jy0,&jz0);
 749
 750             /* Calculate displacement vector */
 751             dx00             = _mm_sub_ps(ix0,jx0);
 752             dy00             = _mm_sub_ps(iy0,jy0);
 753             dz00             = _mm_sub_ps(iz0,jz0);
 754             dx10             = _mm_sub_ps(ix1,jx0);
 755             dy10             = _mm_sub_ps(iy1,jy0);
 756             dz10             = _mm_sub_ps(iz1,jz0);
 757             dx20             = _mm_sub_ps(ix2,jx0);
 758             dy20             = _mm_sub_ps(iy2,jy0);
 759             dz20             = _mm_sub_ps(iz2,jz0);
 760             dx30             = _mm_sub_ps(ix3,jx0);
 761             dy30             = _mm_sub_ps(iy3,jy0);
 762             dz30             = _mm_sub_ps(iz3,jz0);
 763
 764             /* Calculate squared distance and things based on it */
 765             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 766             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 767             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 768             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 769
 770             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 771             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 772             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 773
 774             rinvsq00         = gmx_mm_inv_ps(rsq00);
 775             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 776             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 777             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 778
 779             /* Load parameters for j particles */
 780             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 781                                                               charge+jnrC+0,charge+jnrD+0);
 782             vdwjidx0A        = 2*vdwtype[jnrA+0];
 783             vdwjidx0B        = 2*vdwtype[jnrB+0];
 784             vdwjidx0C        = 2*vdwtype[jnrC+0];
 785             vdwjidx0D        = 2*vdwtype[jnrD+0];
 786
 787             /**************************
 788              * CALCULATE INTERACTIONS *
 789              **************************/
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 793                                          vdwparam+vdwioffset0+vdwjidx0B,
 794                                          vdwparam+vdwioffset0+vdwjidx0C,
 795                                          vdwparam+vdwioffset0+vdwjidx0D,
 796                                          &c6_00,&c12_00);
 797
 798             /* LENNARD-JONES DISPERSION/REPULSION */
 799
 800             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 801             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 802
 803             fscal            = fvdw;
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm_mul_ps(fscal,dx00);
 807             ty               = _mm_mul_ps(fscal,dy00);
 808             tz               = _mm_mul_ps(fscal,dz00);
 809
 810             /* Update vectorial force */
 811             fix0             = _mm_add_ps(fix0,tx);
 812             fiy0             = _mm_add_ps(fiy0,ty);
 813             fiz0             = _mm_add_ps(fiz0,tz);
 814
 815             fjptrA             = f+j_coord_offsetA;
 816             fjptrB             = f+j_coord_offsetB;
 817             fjptrC             = f+j_coord_offsetC;
 818             fjptrD             = f+j_coord_offsetD;
 819             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq10             = _mm_mul_ps(iq1,jq0);
 827
 828             /* COULOMB ELECTROSTATICS */
 829             velec            = _mm_mul_ps(qq10,rinv10);
 830             felec            = _mm_mul_ps(velec,rinvsq10);
 831
 832             fscal            = felec;
 833
 834             /* Calculate temporary vectorial force */
 835             tx               = _mm_mul_ps(fscal,dx10);
 836             ty               = _mm_mul_ps(fscal,dy10);
 837             tz               = _mm_mul_ps(fscal,dz10);
 838
 839             /* Update vectorial force */
 840             fix1             = _mm_add_ps(fix1,tx);
 841             fiy1             = _mm_add_ps(fiy1,ty);
 842             fiz1             = _mm_add_ps(fiz1,tz);
 843
 844             fjptrA             = f+j_coord_offsetA;
 845             fjptrB             = f+j_coord_offsetB;
 846             fjptrC             = f+j_coord_offsetC;
 847             fjptrD             = f+j_coord_offsetD;
 848             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq20             = _mm_mul_ps(iq2,jq0);
 856
 857             /* COULOMB ELECTROSTATICS */
 858             velec            = _mm_mul_ps(qq20,rinv20);
 859             felec            = _mm_mul_ps(velec,rinvsq20);
 860
 861             fscal            = felec;
 862
 863             /* Calculate temporary vectorial force */
 864             tx               = _mm_mul_ps(fscal,dx20);
 865             ty               = _mm_mul_ps(fscal,dy20);
 866             tz               = _mm_mul_ps(fscal,dz20);
 867
 868             /* Update vectorial force */
 869             fix2             = _mm_add_ps(fix2,tx);
 870             fiy2             = _mm_add_ps(fiy2,ty);
 871             fiz2             = _mm_add_ps(fiz2,tz);
 872
 873             fjptrA             = f+j_coord_offsetA;
 874             fjptrB             = f+j_coord_offsetB;
 875             fjptrC             = f+j_coord_offsetC;
 876             fjptrD             = f+j_coord_offsetD;
 877             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 878
 879             /**************************
 880              * CALCULATE INTERACTIONS *
 881              **************************/
 882
 883             /* Compute parameters for interactions between i and j atoms */
 884             qq30             = _mm_mul_ps(iq3,jq0);
 885
 886             /* COULOMB ELECTROSTATICS */
 887             velec            = _mm_mul_ps(qq30,rinv30);
 888             felec            = _mm_mul_ps(velec,rinvsq30);
 889
 890             fscal            = felec;
 891
 892             /* Calculate temporary vectorial force */
 893             tx               = _mm_mul_ps(fscal,dx30);
 894             ty               = _mm_mul_ps(fscal,dy30);
 895             tz               = _mm_mul_ps(fscal,dz30);
 896
 897             /* Update vectorial force */
 898             fix3             = _mm_add_ps(fix3,tx);
 899             fiy3             = _mm_add_ps(fiy3,ty);
 900             fiz3             = _mm_add_ps(fiz3,tz);
 901
 902             fjptrA             = f+j_coord_offsetA;
 903             fjptrB             = f+j_coord_offsetB;
 904             fjptrC             = f+j_coord_offsetC;
 905             fjptrD             = f+j_coord_offsetD;
 906             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 907
 908             /* Inner loop uses 108 flops */
 909         }
 910
 911         if(jidx<j_index_end)
 912         {
 913
 914             /* Get j neighbor index, and coordinate index */
 915             jnrlistA         = jjnr[jidx];
 916             jnrlistB         = jjnr[jidx+1];
 917             jnrlistC         = jjnr[jidx+2];
 918             jnrlistD         = jjnr[jidx+3];
 919             /* Sign of each element will be negative for non-real atoms.
 920              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 921              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 922              */
 923             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 924             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 925             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 926             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 927             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 928             j_coord_offsetA  = DIM*jnrA;
 929             j_coord_offsetB  = DIM*jnrB;
 930             j_coord_offsetC  = DIM*jnrC;
 931             j_coord_offsetD  = DIM*jnrD;
 932
 933             /* load j atom coordinates */
 934             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 935                                               x+j_coord_offsetC,x+j_coord_offsetD,
 936                                               &jx0,&jy0,&jz0);
 937
 938             /* Calculate displacement vector */
 939             dx00             = _mm_sub_ps(ix0,jx0);
 940             dy00             = _mm_sub_ps(iy0,jy0);
 941             dz00             = _mm_sub_ps(iz0,jz0);
 942             dx10             = _mm_sub_ps(ix1,jx0);
 943             dy10             = _mm_sub_ps(iy1,jy0);
 944             dz10             = _mm_sub_ps(iz1,jz0);
 945             dx20             = _mm_sub_ps(ix2,jx0);
 946             dy20             = _mm_sub_ps(iy2,jy0);
 947             dz20             = _mm_sub_ps(iz2,jz0);
 948             dx30             = _mm_sub_ps(ix3,jx0);
 949             dy30             = _mm_sub_ps(iy3,jy0);
 950             dz30             = _mm_sub_ps(iz3,jz0);
 951
 952             /* Calculate squared distance and things based on it */
 953             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 954             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 955             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 956             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 957
 958             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 959             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 960             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 961
 962             rinvsq00         = gmx_mm_inv_ps(rsq00);
 963             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 964             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 965             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 966
 967             /* Load parameters for j particles */
 968             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 969                                                               charge+jnrC+0,charge+jnrD+0);
 970             vdwjidx0A        = 2*vdwtype[jnrA+0];
 971             vdwjidx0B        = 2*vdwtype[jnrB+0];
 972             vdwjidx0C        = 2*vdwtype[jnrC+0];
 973             vdwjidx0D        = 2*vdwtype[jnrD+0];
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 981                                          vdwparam+vdwioffset0+vdwjidx0B,
 982                                          vdwparam+vdwioffset0+vdwjidx0C,
 983                                          vdwparam+vdwioffset0+vdwjidx0D,
 984                                          &c6_00,&c12_00);
 985
 986             /* LENNARD-JONES DISPERSION/REPULSION */
 987
 988             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 989             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 990
 991             fscal            = fvdw;
 992
 993             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 994
 995             /* Calculate temporary vectorial force */
 996             tx               = _mm_mul_ps(fscal,dx00);
 997             ty               = _mm_mul_ps(fscal,dy00);
 998             tz               = _mm_mul_ps(fscal,dz00);
 999
1000             /* Update vectorial force */
1001             fix0             = _mm_add_ps(fix0,tx);
1002             fiy0             = _mm_add_ps(fiy0,ty);
1003             fiz0             = _mm_add_ps(fiz0,tz);
1004
1005             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1006             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1007             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1008             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1009             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1010
1011             /**************************
1012              * CALCULATE INTERACTIONS *
1013              **************************/
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq10             = _mm_mul_ps(iq1,jq0);
1017
1018             /* COULOMB ELECTROSTATICS */
1019             velec            = _mm_mul_ps(qq10,rinv10);
1020             felec            = _mm_mul_ps(velec,rinvsq10);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1025
1026             /* Calculate temporary vectorial force */
1027             tx               = _mm_mul_ps(fscal,dx10);
1028             ty               = _mm_mul_ps(fscal,dy10);
1029             tz               = _mm_mul_ps(fscal,dz10);
1030
1031             /* Update vectorial force */
1032             fix1             = _mm_add_ps(fix1,tx);
1033             fiy1             = _mm_add_ps(fiy1,ty);
1034             fiz1             = _mm_add_ps(fiz1,tz);
1035
1036             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1037             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1038             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1039             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1040             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1041
1042             /**************************
1043              * CALCULATE INTERACTIONS *
1044              **************************/
1045
1046             /* Compute parameters for interactions between i and j atoms */
1047             qq20             = _mm_mul_ps(iq2,jq0);
1048
1049             /* COULOMB ELECTROSTATICS */
1050             velec            = _mm_mul_ps(qq20,rinv20);
1051             felec            = _mm_mul_ps(velec,rinvsq20);
1052
1053             fscal            = felec;
1054
1055             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1056
1057             /* Calculate temporary vectorial force */
1058             tx               = _mm_mul_ps(fscal,dx20);
1059             ty               = _mm_mul_ps(fscal,dy20);
1060             tz               = _mm_mul_ps(fscal,dz20);
1061
1062             /* Update vectorial force */
1063             fix2             = _mm_add_ps(fix2,tx);
1064             fiy2             = _mm_add_ps(fiy2,ty);
1065             fiz2             = _mm_add_ps(fiz2,tz);
1066
1067             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1068             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1069             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1070             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1071             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq30             = _mm_mul_ps(iq3,jq0);
1079
1080             /* COULOMB ELECTROSTATICS */
1081             velec            = _mm_mul_ps(qq30,rinv30);
1082             felec            = _mm_mul_ps(velec,rinvsq30);
1083
1084             fscal            = felec;
1085
1086             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1087
1088             /* Calculate temporary vectorial force */
1089             tx               = _mm_mul_ps(fscal,dx30);
1090             ty               = _mm_mul_ps(fscal,dy30);
1091             tz               = _mm_mul_ps(fscal,dz30);
1092
1093             /* Update vectorial force */
1094             fix3             = _mm_add_ps(fix3,tx);
1095             fiy3             = _mm_add_ps(fiy3,ty);
1096             fiz3             = _mm_add_ps(fiz3,tz);
1097
1098             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1099             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1100             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1101             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1102             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1103
1104             /* Inner loop uses 108 flops */
1105         }
1106
1107         /* End of innermost loop */
1108
1109         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1110                                               f+i_coord_offset,fshift+i_shift_offset);
1111
1112         /* Increment number of inner iterations */
1113         inneriter                  += j_index_end - j_index_start;
1114
1115         /* Outer loop uses 24 flops */
1116     }
1117
1118     /* Increment number of outer iterations */
1119     outeriter        += nri;
1120
1121     /* Update outer/inner flops */
1122
1123     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*108);
1124 }