src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwLJ_GeomW3W3_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3W3_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Water3
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomW3W3_VF_avx_128_fma_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 AVX_128, 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           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
  78     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
  79     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
  80     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
  81     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  82     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
  83     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
  84     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  85     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
  86     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
  87     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  88     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
  89     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
  90     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  97     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     krf              = _mm_set1_ps(fr->ic->k_rf);
 116     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 117     crf              = _mm_set1_ps(fr->ic->c_rf);
 118     nvdwtype         = fr->ntype;
 119     vdwparam         = fr->nbfp;
 120     vdwtype          = mdatoms->typeA;
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 125     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 126     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     jq0              = _mm_set1_ps(charge[inr+0]);
 130     jq1              = _mm_set1_ps(charge[inr+1]);
 131     jq2              = _mm_set1_ps(charge[inr+2]);
 132     vdwjidx0A        = 2*vdwtype[inr+0];
 133     qq00             = _mm_mul_ps(iq0,jq0);
 134     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 135     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 136     qq01             = _mm_mul_ps(iq0,jq1);
 137     qq02             = _mm_mul_ps(iq0,jq2);
 138     qq10             = _mm_mul_ps(iq1,jq0);
 139     qq11             = _mm_mul_ps(iq1,jq1);
 140     qq12             = _mm_mul_ps(iq1,jq2);
 141     qq20             = _mm_mul_ps(iq2,jq0);
 142     qq21             = _mm_mul_ps(iq2,jq1);
 143     qq22             = _mm_mul_ps(iq2,jq2);
 144
 145     /* Avoid stupid compiler warnings */
 146     jnrA = jnrB = jnrC = jnrD = 0;
 147     j_coord_offsetA = 0;
 148     j_coord_offsetB = 0;
 149     j_coord_offsetC = 0;
 150     j_coord_offsetD = 0;
 151
 152     outeriter        = 0;
 153     inneriter        = 0;
 154
 155     for(iidx=0;iidx<4*DIM;iidx++)
 156     {
 157         scratch[iidx] = 0.0;
 158     }
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 177
 178         fix0             = _mm_setzero_ps();
 179         fiy0             = _mm_setzero_ps();
 180         fiz0             = _mm_setzero_ps();
 181         fix1             = _mm_setzero_ps();
 182         fiy1             = _mm_setzero_ps();
 183         fiz1             = _mm_setzero_ps();
 184         fix2             = _mm_setzero_ps();
 185         fiy2             = _mm_setzero_ps();
 186         fiz2             = _mm_setzero_ps();
 187
 188         /* Reset potential sums */
 189         velecsum         = _mm_setzero_ps();
 190         vvdwsum          = _mm_setzero_ps();
 191
 192         /* Start inner kernel loop */
 193         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 194         {
 195
 196             /* Get j neighbor index, and coordinate index */
 197             jnrA             = jjnr[jidx];
 198             jnrB             = jjnr[jidx+1];
 199             jnrC             = jjnr[jidx+2];
 200             jnrD             = jjnr[jidx+3];
 201             j_coord_offsetA  = DIM*jnrA;
 202             j_coord_offsetB  = DIM*jnrB;
 203             j_coord_offsetC  = DIM*jnrC;
 204             j_coord_offsetD  = DIM*jnrD;
 205
 206             /* load j atom coordinates */
 207             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 208                                               x+j_coord_offsetC,x+j_coord_offsetD,
 209                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 210
 211             /* Calculate displacement vector */
 212             dx00             = _mm_sub_ps(ix0,jx0);
 213             dy00             = _mm_sub_ps(iy0,jy0);
 214             dz00             = _mm_sub_ps(iz0,jz0);
 215             dx01             = _mm_sub_ps(ix0,jx1);
 216             dy01             = _mm_sub_ps(iy0,jy1);
 217             dz01             = _mm_sub_ps(iz0,jz1);
 218             dx02             = _mm_sub_ps(ix0,jx2);
 219             dy02             = _mm_sub_ps(iy0,jy2);
 220             dz02             = _mm_sub_ps(iz0,jz2);
 221             dx10             = _mm_sub_ps(ix1,jx0);
 222             dy10             = _mm_sub_ps(iy1,jy0);
 223             dz10             = _mm_sub_ps(iz1,jz0);
 224             dx11             = _mm_sub_ps(ix1,jx1);
 225             dy11             = _mm_sub_ps(iy1,jy1);
 226             dz11             = _mm_sub_ps(iz1,jz1);
 227             dx12             = _mm_sub_ps(ix1,jx2);
 228             dy12             = _mm_sub_ps(iy1,jy2);
 229             dz12             = _mm_sub_ps(iz1,jz2);
 230             dx20             = _mm_sub_ps(ix2,jx0);
 231             dy20             = _mm_sub_ps(iy2,jy0);
 232             dz20             = _mm_sub_ps(iz2,jz0);
 233             dx21             = _mm_sub_ps(ix2,jx1);
 234             dy21             = _mm_sub_ps(iy2,jy1);
 235             dz21             = _mm_sub_ps(iz2,jz1);
 236             dx22             = _mm_sub_ps(ix2,jx2);
 237             dy22             = _mm_sub_ps(iy2,jy2);
 238             dz22             = _mm_sub_ps(iz2,jz2);
 239
 240             /* Calculate squared distance and things based on it */
 241             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 242             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 243             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 244             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 245             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 246             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 247             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 248             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 249             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 250
 251             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 252             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 253             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 254             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 255             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 256             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 257             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 258             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 259             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 260
 261             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 262             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 263             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 264             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 265             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 266             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 267             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 268             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 269             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 270
 271             fjx0             = _mm_setzero_ps();
 272             fjy0             = _mm_setzero_ps();
 273             fjz0             = _mm_setzero_ps();
 274             fjx1             = _mm_setzero_ps();
 275             fjy1             = _mm_setzero_ps();
 276             fjz1             = _mm_setzero_ps();
 277             fjx2             = _mm_setzero_ps();
 278             fjy2             = _mm_setzero_ps();
 279             fjz2             = _mm_setzero_ps();
 280
 281             /**************************
 282              * CALCULATE INTERACTIONS *
 283              **************************/
 284
 285             /* REACTION-FIELD ELECTROSTATICS */
 286             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 287             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 288
 289             /* LENNARD-JONES DISPERSION/REPULSION */
 290
 291             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 292             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 293             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 294             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 295             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 296
 297             /* Update potential sum for this i atom from the interaction with this j atom. */
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 300
 301             fscal            = _mm_add_ps(felec,fvdw);
 302
 303              /* Update vectorial force */
 304             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 305             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 306             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 307
 308             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 309             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 310             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             /* REACTION-FIELD ELECTROSTATICS */
 317             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 318             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm_add_ps(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325              /* Update vectorial force */
 326             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 327             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 328             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 329
 330             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 331             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 332             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 340             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347              /* Update vectorial force */
 348             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 349             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 350             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 351
 352             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 353             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 354             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             /* REACTION-FIELD ELECTROSTATICS */
 361             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 362             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 363
 364             /* Update potential sum for this i atom from the interaction with this j atom. */
 365             velecsum         = _mm_add_ps(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369              /* Update vectorial force */
 370             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 371             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 372             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 373
 374             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 375             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 376             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             /* REACTION-FIELD ELECTROSTATICS */
 383             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 384             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 385
 386             /* Update potential sum for this i atom from the interaction with this j atom. */
 387             velecsum         = _mm_add_ps(velecsum,velec);
 388
 389             fscal            = felec;
 390
 391              /* Update vectorial force */
 392             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 393             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 394             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 395
 396             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 397             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 398             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             /* REACTION-FIELD ELECTROSTATICS */
 405             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 406             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 407
 408             /* Update potential sum for this i atom from the interaction with this j atom. */
 409             velecsum         = _mm_add_ps(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413              /* Update vectorial force */
 414             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 415             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 416             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 417
 418             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 419             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 420             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             /* REACTION-FIELD ELECTROSTATICS */
 427             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 428             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 429
 430             /* Update potential sum for this i atom from the interaction with this j atom. */
 431             velecsum         = _mm_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435              /* Update vectorial force */
 436             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 437             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 438             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 439
 440             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 441             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 442             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 443
 444             /**************************
 445              * CALCULATE INTERACTIONS *
 446              **************************/
 447
 448             /* REACTION-FIELD ELECTROSTATICS */
 449             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 450             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 451
 452             /* Update potential sum for this i atom from the interaction with this j atom. */
 453             velecsum         = _mm_add_ps(velecsum,velec);
 454
 455             fscal            = felec;
 456
 457              /* Update vectorial force */
 458             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 459             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 460             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 461
 462             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 463             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 464             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             /* REACTION-FIELD ELECTROSTATICS */
 471             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 472             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velecsum         = _mm_add_ps(velecsum,velec);
 476
 477             fscal            = felec;
 478
 479              /* Update vectorial force */
 480             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 481             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 482             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 483
 484             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 485             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 486             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 487
 488             fjptrA             = f+j_coord_offsetA;
 489             fjptrB             = f+j_coord_offsetB;
 490             fjptrC             = f+j_coord_offsetC;
 491             fjptrD             = f+j_coord_offsetD;
 492
 493             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 494                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 495
 496             /* Inner loop uses 327 flops */
 497         }
 498
 499         if(jidx<j_index_end)
 500         {
 501
 502             /* Get j neighbor index, and coordinate index */
 503             jnrlistA         = jjnr[jidx];
 504             jnrlistB         = jjnr[jidx+1];
 505             jnrlistC         = jjnr[jidx+2];
 506             jnrlistD         = jjnr[jidx+3];
 507             /* Sign of each element will be negative for non-real atoms.
 508              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 509              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 510              */
 511             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 512             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 513             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 514             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 515             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 516             j_coord_offsetA  = DIM*jnrA;
 517             j_coord_offsetB  = DIM*jnrB;
 518             j_coord_offsetC  = DIM*jnrC;
 519             j_coord_offsetD  = DIM*jnrD;
 520
 521             /* load j atom coordinates */
 522             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 523                                               x+j_coord_offsetC,x+j_coord_offsetD,
 524                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
 525
 526             /* Calculate displacement vector */
 527             dx00             = _mm_sub_ps(ix0,jx0);
 528             dy00             = _mm_sub_ps(iy0,jy0);
 529             dz00             = _mm_sub_ps(iz0,jz0);
 530             dx01             = _mm_sub_ps(ix0,jx1);
 531             dy01             = _mm_sub_ps(iy0,jy1);
 532             dz01             = _mm_sub_ps(iz0,jz1);
 533             dx02             = _mm_sub_ps(ix0,jx2);
 534             dy02             = _mm_sub_ps(iy0,jy2);
 535             dz02             = _mm_sub_ps(iz0,jz2);
 536             dx10             = _mm_sub_ps(ix1,jx0);
 537             dy10             = _mm_sub_ps(iy1,jy0);
 538             dz10             = _mm_sub_ps(iz1,jz0);
 539             dx11             = _mm_sub_ps(ix1,jx1);
 540             dy11             = _mm_sub_ps(iy1,jy1);
 541             dz11             = _mm_sub_ps(iz1,jz1);
 542             dx12             = _mm_sub_ps(ix1,jx2);
 543             dy12             = _mm_sub_ps(iy1,jy2);
 544             dz12             = _mm_sub_ps(iz1,jz2);
 545             dx20             = _mm_sub_ps(ix2,jx0);
 546             dy20             = _mm_sub_ps(iy2,jy0);
 547             dz20             = _mm_sub_ps(iz2,jz0);
 548             dx21             = _mm_sub_ps(ix2,jx1);
 549             dy21             = _mm_sub_ps(iy2,jy1);
 550             dz21             = _mm_sub_ps(iz2,jz1);
 551             dx22             = _mm_sub_ps(ix2,jx2);
 552             dy22             = _mm_sub_ps(iy2,jy2);
 553             dz22             = _mm_sub_ps(iz2,jz2);
 554
 555             /* Calculate squared distance and things based on it */
 556             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 557             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
 558             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
 559             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 560             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
 561             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
 562             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 563             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
 564             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
 565
 566             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 567             rinv01           = gmx_mm_invsqrt_ps(rsq01);
 568             rinv02           = gmx_mm_invsqrt_ps(rsq02);
 569             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 570             rinv11           = gmx_mm_invsqrt_ps(rsq11);
 571             rinv12           = gmx_mm_invsqrt_ps(rsq12);
 572             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 573             rinv21           = gmx_mm_invsqrt_ps(rsq21);
 574             rinv22           = gmx_mm_invsqrt_ps(rsq22);
 575
 576             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 577             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
 578             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
 579             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 580             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
 581             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
 582             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 583             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
 584             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
 585
 586             fjx0             = _mm_setzero_ps();
 587             fjy0             = _mm_setzero_ps();
 588             fjz0             = _mm_setzero_ps();
 589             fjx1             = _mm_setzero_ps();
 590             fjy1             = _mm_setzero_ps();
 591             fjz1             = _mm_setzero_ps();
 592             fjx2             = _mm_setzero_ps();
 593             fjy2             = _mm_setzero_ps();
 594             fjz2             = _mm_setzero_ps();
 595
 596             /**************************
 597              * CALCULATE INTERACTIONS *
 598              **************************/
 599
 600             /* REACTION-FIELD ELECTROSTATICS */
 601             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 602             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 603
 604             /* LENNARD-JONES DISPERSION/REPULSION */
 605
 606             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 607             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 608             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 609             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 610             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm_andnot_ps(dummy_mask,velec);
 614             velecsum         = _mm_add_ps(velecsum,velec);
 615             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 616             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 617
 618             fscal            = _mm_add_ps(felec,fvdw);
 619
 620             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 621
 622              /* Update vectorial force */
 623             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 624             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 625             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 626
 627             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 628             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 629             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             /* REACTION-FIELD ELECTROSTATICS */
 636             velec            = _mm_mul_ps(qq01,_mm_sub_ps(_mm_macc_ps(krf,rsq01,rinv01),crf));
 637             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
 638
 639             /* Update potential sum for this i atom from the interaction with this j atom. */
 640             velec            = _mm_andnot_ps(dummy_mask,velec);
 641             velecsum         = _mm_add_ps(velecsum,velec);
 642
 643             fscal            = felec;
 644
 645             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 646
 647              /* Update vectorial force */
 648             fix0             = _mm_macc_ps(dx01,fscal,fix0);
 649             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
 650             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
 651
 652             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
 653             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
 654             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             /* REACTION-FIELD ELECTROSTATICS */
 661             velec            = _mm_mul_ps(qq02,_mm_sub_ps(_mm_macc_ps(krf,rsq02,rinv02),crf));
 662             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
 663
 664             /* Update potential sum for this i atom from the interaction with this j atom. */
 665             velec            = _mm_andnot_ps(dummy_mask,velec);
 666             velecsum         = _mm_add_ps(velecsum,velec);
 667
 668             fscal            = felec;
 669
 670             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 671
 672              /* Update vectorial force */
 673             fix0             = _mm_macc_ps(dx02,fscal,fix0);
 674             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
 675             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
 676
 677             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
 678             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
 679             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
 680
 681             /**************************
 682              * CALCULATE INTERACTIONS *
 683              **************************/
 684
 685             /* REACTION-FIELD ELECTROSTATICS */
 686             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 687             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 688
 689             /* Update potential sum for this i atom from the interaction with this j atom. */
 690             velec            = _mm_andnot_ps(dummy_mask,velec);
 691             velecsum         = _mm_add_ps(velecsum,velec);
 692
 693             fscal            = felec;
 694
 695             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 696
 697              /* Update vectorial force */
 698             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 699             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 700             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 701
 702             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 703             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 704             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 705
 706             /**************************
 707              * CALCULATE INTERACTIONS *
 708              **************************/
 709
 710             /* REACTION-FIELD ELECTROSTATICS */
 711             velec            = _mm_mul_ps(qq11,_mm_sub_ps(_mm_macc_ps(krf,rsq11,rinv11),crf));
 712             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
 713
 714             /* Update potential sum for this i atom from the interaction with this j atom. */
 715             velec            = _mm_andnot_ps(dummy_mask,velec);
 716             velecsum         = _mm_add_ps(velecsum,velec);
 717
 718             fscal            = felec;
 719
 720             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 721
 722              /* Update vectorial force */
 723             fix1             = _mm_macc_ps(dx11,fscal,fix1);
 724             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
 725             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
 726
 727             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
 728             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
 729             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
 730
 731             /**************************
 732              * CALCULATE INTERACTIONS *
 733              **************************/
 734
 735             /* REACTION-FIELD ELECTROSTATICS */
 736             velec            = _mm_mul_ps(qq12,_mm_sub_ps(_mm_macc_ps(krf,rsq12,rinv12),crf));
 737             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
 738
 739             /* Update potential sum for this i atom from the interaction with this j atom. */
 740             velec            = _mm_andnot_ps(dummy_mask,velec);
 741             velecsum         = _mm_add_ps(velecsum,velec);
 742
 743             fscal            = felec;
 744
 745             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 746
 747              /* Update vectorial force */
 748             fix1             = _mm_macc_ps(dx12,fscal,fix1);
 749             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
 750             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
 751
 752             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
 753             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
 754             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
 755
 756             /**************************
 757              * CALCULATE INTERACTIONS *
 758              **************************/
 759
 760             /* REACTION-FIELD ELECTROSTATICS */
 761             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 762             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 763
 764             /* Update potential sum for this i atom from the interaction with this j atom. */
 765             velec            = _mm_andnot_ps(dummy_mask,velec);
 766             velecsum         = _mm_add_ps(velecsum,velec);
 767
 768             fscal            = felec;
 769
 770             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 771
 772              /* Update vectorial force */
 773             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 774             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 775             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 776
 777             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 778             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 779             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             /* REACTION-FIELD ELECTROSTATICS */
 786             velec            = _mm_mul_ps(qq21,_mm_sub_ps(_mm_macc_ps(krf,rsq21,rinv21),crf));
 787             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm_andnot_ps(dummy_mask,velec);
 791             velecsum         = _mm_add_ps(velecsum,velec);
 792
 793             fscal            = felec;
 794
 795             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 796
 797              /* Update vectorial force */
 798             fix2             = _mm_macc_ps(dx21,fscal,fix2);
 799             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
 800             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
 801
 802             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
 803             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
 804             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             /* REACTION-FIELD ELECTROSTATICS */
 811             velec            = _mm_mul_ps(qq22,_mm_sub_ps(_mm_macc_ps(krf,rsq22,rinv22),crf));
 812             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
 813
 814             /* Update potential sum for this i atom from the interaction with this j atom. */
 815             velec            = _mm_andnot_ps(dummy_mask,velec);
 816             velecsum         = _mm_add_ps(velecsum,velec);
 817
 818             fscal            = felec;
 819
 820             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 821
 822              /* Update vectorial force */
 823             fix2             = _mm_macc_ps(dx22,fscal,fix2);
 824             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
 825             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
 826
 827             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
 828             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
 829             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
 830
 831             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 832             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 833             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 834             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 835
 836             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
 837                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
 838
 839             /* Inner loop uses 327 flops */
 840         }
 841
 842         /* End of innermost loop */
 843
 844         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 845                                               f+i_coord_offset,fshift+i_shift_offset);
 846
 847         ggid                        = gid[iidx];
 848         /* Update potential energies */
 849         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 850         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 851
 852         /* Increment number of inner iterations */
 853         inneriter                  += j_index_end - j_index_start;
 854
 855         /* Outer loop uses 20 flops */
 856     }
 857
 858     /* Increment number of outer iterations */
 859     outeriter        += nri;
 860
 861     /* Update outer/inner flops */
 862
 863     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_VF,outeriter*20 + inneriter*327);
 864 }
 865 /*
 866  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_128_fma_single
 867  * Electrostatics interaction: ReactionField
 868  * VdW interaction:            LennardJones
 869  * Geometry:                   Water3-Water3
 870  * Calculate force/pot:        Force
 871  */
 872 void
 873 nb_kernel_ElecRF_VdwLJ_GeomW3W3_F_avx_128_fma_single
 874                     (t_nblist * gmx_restrict                nlist,
 875                      rvec * gmx_restrict                    xx,
 876                      rvec * gmx_restrict                    ff,
 877                      t_forcerec * gmx_restrict              fr,
 878                      t_mdatoms * gmx_restrict               mdatoms,
 879                      nb_kernel_data_t * gmx_restrict        kernel_data,
 880                      t_nrnb * gmx_restrict                  nrnb)
 881 {
 882     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 883      * just 0 for non-waters.
 884      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 885      * jnr indices corresponding to data put in the four positions in the SIMD register.
 886      */
 887     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 888     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 889     int              jnrA,jnrB,jnrC,jnrD;
 890     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 891     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 892     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 893     real             rcutoff_scalar;
 894     real             *shiftvec,*fshift,*x,*f;
 895     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 896     real             scratch[4*DIM];
 897     __m128           fscal,rcutoff,rcutoff2,jidxall;
 898     int              vdwioffset0;
 899     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 900     int              vdwioffset1;
 901     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 902     int              vdwioffset2;
 903     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 904     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 905     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 906     int              vdwjidx1A,vdwjidx1B,vdwjidx1C,vdwjidx1D;
 907     __m128           jx1,jy1,jz1,fjx1,fjy1,fjz1,jq1,isaj1;
 908     int              vdwjidx2A,vdwjidx2B,vdwjidx2C,vdwjidx2D;
 909     __m128           jx2,jy2,jz2,fjx2,fjy2,fjz2,jq2,isaj2;
 910     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 911     __m128           dx01,dy01,dz01,rsq01,rinv01,rinvsq01,r01,qq01,c6_01,c12_01;
 912     __m128           dx02,dy02,dz02,rsq02,rinv02,rinvsq02,r02,qq02,c6_02,c12_02;
 913     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 914     __m128           dx11,dy11,dz11,rsq11,rinv11,rinvsq11,r11,qq11,c6_11,c12_11;
 915     __m128           dx12,dy12,dz12,rsq12,rinv12,rinvsq12,r12,qq12,c6_12,c12_12;
 916     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 917     __m128           dx21,dy21,dz21,rsq21,rinv21,rinvsq21,r21,qq21,c6_21,c12_21;
 918     __m128           dx22,dy22,dz22,rsq22,rinv22,rinvsq22,r22,qq22,c6_22,c12_22;
 919     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 920     real             *charge;
 921     int              nvdwtype;
 922     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 923     int              *vdwtype;
 924     real             *vdwparam;
 925     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 926     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 927     __m128           dummy_mask,cutoff_mask;
 928     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 929     __m128           one     = _mm_set1_ps(1.0);
 930     __m128           two     = _mm_set1_ps(2.0);
 931     x                = xx[0];
 932     f                = ff[0];
 933
 934     nri              = nlist->nri;
 935     iinr             = nlist->iinr;
 936     jindex           = nlist->jindex;
 937     jjnr             = nlist->jjnr;
 938     shiftidx         = nlist->shift;
 939     gid              = nlist->gid;
 940     shiftvec         = fr->shift_vec[0];
 941     fshift           = fr->fshift[0];
 942     facel            = _mm_set1_ps(fr->epsfac);
 943     charge           = mdatoms->chargeA;
 944     krf              = _mm_set1_ps(fr->ic->k_rf);
 945     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 946     crf              = _mm_set1_ps(fr->ic->c_rf);
 947     nvdwtype         = fr->ntype;
 948     vdwparam         = fr->nbfp;
 949     vdwtype          = mdatoms->typeA;
 950
 951     /* Setup water-specific parameters */
 952     inr              = nlist->iinr[0];
 953     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 954     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 955     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 956     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 957
 958     jq0              = _mm_set1_ps(charge[inr+0]);
 959     jq1              = _mm_set1_ps(charge[inr+1]);
 960     jq2              = _mm_set1_ps(charge[inr+2]);
 961     vdwjidx0A        = 2*vdwtype[inr+0];
 962     qq00             = _mm_mul_ps(iq0,jq0);
 963     c6_00            = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A]);
 964     c12_00           = _mm_set1_ps(vdwparam[vdwioffset0+vdwjidx0A+1]);
 965     qq01             = _mm_mul_ps(iq0,jq1);
 966     qq02             = _mm_mul_ps(iq0,jq2);
 967     qq10             = _mm_mul_ps(iq1,jq0);
 968     qq11             = _mm_mul_ps(iq1,jq1);
 969     qq12             = _mm_mul_ps(iq1,jq2);
 970     qq20             = _mm_mul_ps(iq2,jq0);
 971     qq21             = _mm_mul_ps(iq2,jq1);
 972     qq22             = _mm_mul_ps(iq2,jq2);
 973
 974     /* Avoid stupid compiler warnings */
 975     jnrA = jnrB = jnrC = jnrD = 0;
 976     j_coord_offsetA = 0;
 977     j_coord_offsetB = 0;
 978     j_coord_offsetC = 0;
 979     j_coord_offsetD = 0;
 980
 981     outeriter        = 0;
 982     inneriter        = 0;
 983
 984     for(iidx=0;iidx<4*DIM;iidx++)
 985     {
 986         scratch[iidx] = 0.0;
 987     }
 988
 989     /* Start outer loop over neighborlists */
 990     for(iidx=0; iidx<nri; iidx++)
 991     {
 992         /* Load shift vector for this list */
 993         i_shift_offset   = DIM*shiftidx[iidx];
 994
 995         /* Load limits for loop over neighbors */
 996         j_index_start    = jindex[iidx];
 997         j_index_end      = jindex[iidx+1];
 998
 999         /* Get outer coordinate index */
1000         inr              = iinr[iidx];
1001         i_coord_offset   = DIM*inr;
1002
1003         /* Load i particle coords and add shift vector */
1004         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
1005                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
1006
1007         fix0             = _mm_setzero_ps();
1008         fiy0             = _mm_setzero_ps();
1009         fiz0             = _mm_setzero_ps();
1010         fix1             = _mm_setzero_ps();
1011         fiy1             = _mm_setzero_ps();
1012         fiz1             = _mm_setzero_ps();
1013         fix2             = _mm_setzero_ps();
1014         fiy2             = _mm_setzero_ps();
1015         fiz2             = _mm_setzero_ps();
1016
1017         /* Start inner kernel loop */
1018         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
1019         {
1020
1021             /* Get j neighbor index, and coordinate index */
1022             jnrA             = jjnr[jidx];
1023             jnrB             = jjnr[jidx+1];
1024             jnrC             = jjnr[jidx+2];
1025             jnrD             = jjnr[jidx+3];
1026             j_coord_offsetA  = DIM*jnrA;
1027             j_coord_offsetB  = DIM*jnrB;
1028             j_coord_offsetC  = DIM*jnrC;
1029             j_coord_offsetD  = DIM*jnrD;
1030
1031             /* load j atom coordinates */
1032             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1033                                               x+j_coord_offsetC,x+j_coord_offsetD,
1034                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1035
1036             /* Calculate displacement vector */
1037             dx00             = _mm_sub_ps(ix0,jx0);
1038             dy00             = _mm_sub_ps(iy0,jy0);
1039             dz00             = _mm_sub_ps(iz0,jz0);
1040             dx01             = _mm_sub_ps(ix0,jx1);
1041             dy01             = _mm_sub_ps(iy0,jy1);
1042             dz01             = _mm_sub_ps(iz0,jz1);
1043             dx02             = _mm_sub_ps(ix0,jx2);
1044             dy02             = _mm_sub_ps(iy0,jy2);
1045             dz02             = _mm_sub_ps(iz0,jz2);
1046             dx10             = _mm_sub_ps(ix1,jx0);
1047             dy10             = _mm_sub_ps(iy1,jy0);
1048             dz10             = _mm_sub_ps(iz1,jz0);
1049             dx11             = _mm_sub_ps(ix1,jx1);
1050             dy11             = _mm_sub_ps(iy1,jy1);
1051             dz11             = _mm_sub_ps(iz1,jz1);
1052             dx12             = _mm_sub_ps(ix1,jx2);
1053             dy12             = _mm_sub_ps(iy1,jy2);
1054             dz12             = _mm_sub_ps(iz1,jz2);
1055             dx20             = _mm_sub_ps(ix2,jx0);
1056             dy20             = _mm_sub_ps(iy2,jy0);
1057             dz20             = _mm_sub_ps(iz2,jz0);
1058             dx21             = _mm_sub_ps(ix2,jx1);
1059             dy21             = _mm_sub_ps(iy2,jy1);
1060             dz21             = _mm_sub_ps(iz2,jz1);
1061             dx22             = _mm_sub_ps(ix2,jx2);
1062             dy22             = _mm_sub_ps(iy2,jy2);
1063             dz22             = _mm_sub_ps(iz2,jz2);
1064
1065             /* Calculate squared distance and things based on it */
1066             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1067             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1068             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1069             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1070             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1071             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1072             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1073             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1074             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1075
1076             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1077             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1078             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1079             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1080             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1081             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1082             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1083             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1084             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1085
1086             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1087             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1088             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1089             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1090             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1091             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1092             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1093             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1094             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1095
1096             fjx0             = _mm_setzero_ps();
1097             fjy0             = _mm_setzero_ps();
1098             fjz0             = _mm_setzero_ps();
1099             fjx1             = _mm_setzero_ps();
1100             fjy1             = _mm_setzero_ps();
1101             fjz1             = _mm_setzero_ps();
1102             fjx2             = _mm_setzero_ps();
1103             fjy2             = _mm_setzero_ps();
1104             fjz2             = _mm_setzero_ps();
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             /* REACTION-FIELD ELECTROSTATICS */
1111             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1112
1113             /* LENNARD-JONES DISPERSION/REPULSION */
1114
1115             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1116             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1117
1118             fscal            = _mm_add_ps(felec,fvdw);
1119
1120              /* Update vectorial force */
1121             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1122             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1123             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1124
1125             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1126             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1127             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             /* REACTION-FIELD ELECTROSTATICS */
1134             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1135
1136             fscal            = felec;
1137
1138              /* Update vectorial force */
1139             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1140             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1141             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1142
1143             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1144             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1145             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1146
1147             /**************************
1148              * CALCULATE INTERACTIONS *
1149              **************************/
1150
1151             /* REACTION-FIELD ELECTROSTATICS */
1152             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1153
1154             fscal            = felec;
1155
1156              /* Update vectorial force */
1157             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1158             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1159             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1160
1161             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1162             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1163             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1164
1165             /**************************
1166              * CALCULATE INTERACTIONS *
1167              **************************/
1168
1169             /* REACTION-FIELD ELECTROSTATICS */
1170             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1171
1172             fscal            = felec;
1173
1174              /* Update vectorial force */
1175             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1176             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1177             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1178
1179             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1180             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1181             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             /* REACTION-FIELD ELECTROSTATICS */
1188             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1189
1190             fscal            = felec;
1191
1192              /* Update vectorial force */
1193             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1194             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1195             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1196
1197             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1198             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1199             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1200
1201             /**************************
1202              * CALCULATE INTERACTIONS *
1203              **************************/
1204
1205             /* REACTION-FIELD ELECTROSTATICS */
1206             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1207
1208             fscal            = felec;
1209
1210              /* Update vectorial force */
1211             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1212             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1213             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1214
1215             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1216             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1217             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             /* REACTION-FIELD ELECTROSTATICS */
1224             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1225
1226             fscal            = felec;
1227
1228              /* Update vectorial force */
1229             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1230             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1231             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1232
1233             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1234             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1235             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             /* REACTION-FIELD ELECTROSTATICS */
1242             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1243
1244             fscal            = felec;
1245
1246              /* Update vectorial force */
1247             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1248             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1249             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1250
1251             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1252             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1253             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1254
1255             /**************************
1256              * CALCULATE INTERACTIONS *
1257              **************************/
1258
1259             /* REACTION-FIELD ELECTROSTATICS */
1260             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1261
1262             fscal            = felec;
1263
1264              /* Update vectorial force */
1265             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1266             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1267             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1268
1269             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1270             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1271             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1272
1273             fjptrA             = f+j_coord_offsetA;
1274             fjptrB             = f+j_coord_offsetB;
1275             fjptrC             = f+j_coord_offsetC;
1276             fjptrD             = f+j_coord_offsetD;
1277
1278             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1279                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1280
1281             /* Inner loop uses 277 flops */
1282         }
1283
1284         if(jidx<j_index_end)
1285         {
1286
1287             /* Get j neighbor index, and coordinate index */
1288             jnrlistA         = jjnr[jidx];
1289             jnrlistB         = jjnr[jidx+1];
1290             jnrlistC         = jjnr[jidx+2];
1291             jnrlistD         = jjnr[jidx+3];
1292             /* Sign of each element will be negative for non-real atoms.
1293              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1294              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1295              */
1296             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1297             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1298             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1299             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1300             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1301             j_coord_offsetA  = DIM*jnrA;
1302             j_coord_offsetB  = DIM*jnrB;
1303             j_coord_offsetC  = DIM*jnrC;
1304             j_coord_offsetD  = DIM*jnrD;
1305
1306             /* load j atom coordinates */
1307             gmx_mm_load_3rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1308                                               x+j_coord_offsetC,x+j_coord_offsetD,
1309                                               &jx0,&jy0,&jz0,&jx1,&jy1,&jz1,&jx2,&jy2,&jz2);
1310
1311             /* Calculate displacement vector */
1312             dx00             = _mm_sub_ps(ix0,jx0);
1313             dy00             = _mm_sub_ps(iy0,jy0);
1314             dz00             = _mm_sub_ps(iz0,jz0);
1315             dx01             = _mm_sub_ps(ix0,jx1);
1316             dy01             = _mm_sub_ps(iy0,jy1);
1317             dz01             = _mm_sub_ps(iz0,jz1);
1318             dx02             = _mm_sub_ps(ix0,jx2);
1319             dy02             = _mm_sub_ps(iy0,jy2);
1320             dz02             = _mm_sub_ps(iz0,jz2);
1321             dx10             = _mm_sub_ps(ix1,jx0);
1322             dy10             = _mm_sub_ps(iy1,jy0);
1323             dz10             = _mm_sub_ps(iz1,jz0);
1324             dx11             = _mm_sub_ps(ix1,jx1);
1325             dy11             = _mm_sub_ps(iy1,jy1);
1326             dz11             = _mm_sub_ps(iz1,jz1);
1327             dx12             = _mm_sub_ps(ix1,jx2);
1328             dy12             = _mm_sub_ps(iy1,jy2);
1329             dz12             = _mm_sub_ps(iz1,jz2);
1330             dx20             = _mm_sub_ps(ix2,jx0);
1331             dy20             = _mm_sub_ps(iy2,jy0);
1332             dz20             = _mm_sub_ps(iz2,jz0);
1333             dx21             = _mm_sub_ps(ix2,jx1);
1334             dy21             = _mm_sub_ps(iy2,jy1);
1335             dz21             = _mm_sub_ps(iz2,jz1);
1336             dx22             = _mm_sub_ps(ix2,jx2);
1337             dy22             = _mm_sub_ps(iy2,jy2);
1338             dz22             = _mm_sub_ps(iz2,jz2);
1339
1340             /* Calculate squared distance and things based on it */
1341             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1342             rsq01            = gmx_mm_calc_rsq_ps(dx01,dy01,dz01);
1343             rsq02            = gmx_mm_calc_rsq_ps(dx02,dy02,dz02);
1344             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1345             rsq11            = gmx_mm_calc_rsq_ps(dx11,dy11,dz11);
1346             rsq12            = gmx_mm_calc_rsq_ps(dx12,dy12,dz12);
1347             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1348             rsq21            = gmx_mm_calc_rsq_ps(dx21,dy21,dz21);
1349             rsq22            = gmx_mm_calc_rsq_ps(dx22,dy22,dz22);
1350
1351             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1352             rinv01           = gmx_mm_invsqrt_ps(rsq01);
1353             rinv02           = gmx_mm_invsqrt_ps(rsq02);
1354             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1355             rinv11           = gmx_mm_invsqrt_ps(rsq11);
1356             rinv12           = gmx_mm_invsqrt_ps(rsq12);
1357             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1358             rinv21           = gmx_mm_invsqrt_ps(rsq21);
1359             rinv22           = gmx_mm_invsqrt_ps(rsq22);
1360
1361             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1362             rinvsq01         = _mm_mul_ps(rinv01,rinv01);
1363             rinvsq02         = _mm_mul_ps(rinv02,rinv02);
1364             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1365             rinvsq11         = _mm_mul_ps(rinv11,rinv11);
1366             rinvsq12         = _mm_mul_ps(rinv12,rinv12);
1367             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1368             rinvsq21         = _mm_mul_ps(rinv21,rinv21);
1369             rinvsq22         = _mm_mul_ps(rinv22,rinv22);
1370
1371             fjx0             = _mm_setzero_ps();
1372             fjy0             = _mm_setzero_ps();
1373             fjz0             = _mm_setzero_ps();
1374             fjx1             = _mm_setzero_ps();
1375             fjy1             = _mm_setzero_ps();
1376             fjz1             = _mm_setzero_ps();
1377             fjx2             = _mm_setzero_ps();
1378             fjy2             = _mm_setzero_ps();
1379             fjz2             = _mm_setzero_ps();
1380
1381             /**************************
1382              * CALCULATE INTERACTIONS *
1383              **************************/
1384
1385             /* REACTION-FIELD ELECTROSTATICS */
1386             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
1387
1388             /* LENNARD-JONES DISPERSION/REPULSION */
1389
1390             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1391             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1392
1393             fscal            = _mm_add_ps(felec,fvdw);
1394
1395             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1396
1397              /* Update vectorial force */
1398             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1399             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1400             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1401
1402             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1403             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1404             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1405
1406             /**************************
1407              * CALCULATE INTERACTIONS *
1408              **************************/
1409
1410             /* REACTION-FIELD ELECTROSTATICS */
1411             felec            = _mm_mul_ps(qq01,_mm_msub_ps(rinv01,rinvsq01,krf2));
1412
1413             fscal            = felec;
1414
1415             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1416
1417              /* Update vectorial force */
1418             fix0             = _mm_macc_ps(dx01,fscal,fix0);
1419             fiy0             = _mm_macc_ps(dy01,fscal,fiy0);
1420             fiz0             = _mm_macc_ps(dz01,fscal,fiz0);
1421
1422             fjx1             = _mm_macc_ps(dx01,fscal,fjx1);
1423             fjy1             = _mm_macc_ps(dy01,fscal,fjy1);
1424             fjz1             = _mm_macc_ps(dz01,fscal,fjz1);
1425
1426             /**************************
1427              * CALCULATE INTERACTIONS *
1428              **************************/
1429
1430             /* REACTION-FIELD ELECTROSTATICS */
1431             felec            = _mm_mul_ps(qq02,_mm_msub_ps(rinv02,rinvsq02,krf2));
1432
1433             fscal            = felec;
1434
1435             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1436
1437              /* Update vectorial force */
1438             fix0             = _mm_macc_ps(dx02,fscal,fix0);
1439             fiy0             = _mm_macc_ps(dy02,fscal,fiy0);
1440             fiz0             = _mm_macc_ps(dz02,fscal,fiz0);
1441
1442             fjx2             = _mm_macc_ps(dx02,fscal,fjx2);
1443             fjy2             = _mm_macc_ps(dy02,fscal,fjy2);
1444             fjz2             = _mm_macc_ps(dz02,fscal,fjz2);
1445
1446             /**************************
1447              * CALCULATE INTERACTIONS *
1448              **************************/
1449
1450             /* REACTION-FIELD ELECTROSTATICS */
1451             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1452
1453             fscal            = felec;
1454
1455             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1456
1457              /* Update vectorial force */
1458             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1459             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1460             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1461
1462             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1463             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1464             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1465
1466             /**************************
1467              * CALCULATE INTERACTIONS *
1468              **************************/
1469
1470             /* REACTION-FIELD ELECTROSTATICS */
1471             felec            = _mm_mul_ps(qq11,_mm_msub_ps(rinv11,rinvsq11,krf2));
1472
1473             fscal            = felec;
1474
1475             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1476
1477              /* Update vectorial force */
1478             fix1             = _mm_macc_ps(dx11,fscal,fix1);
1479             fiy1             = _mm_macc_ps(dy11,fscal,fiy1);
1480             fiz1             = _mm_macc_ps(dz11,fscal,fiz1);
1481
1482             fjx1             = _mm_macc_ps(dx11,fscal,fjx1);
1483             fjy1             = _mm_macc_ps(dy11,fscal,fjy1);
1484             fjz1             = _mm_macc_ps(dz11,fscal,fjz1);
1485
1486             /**************************
1487              * CALCULATE INTERACTIONS *
1488              **************************/
1489
1490             /* REACTION-FIELD ELECTROSTATICS */
1491             felec            = _mm_mul_ps(qq12,_mm_msub_ps(rinv12,rinvsq12,krf2));
1492
1493             fscal            = felec;
1494
1495             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1496
1497              /* Update vectorial force */
1498             fix1             = _mm_macc_ps(dx12,fscal,fix1);
1499             fiy1             = _mm_macc_ps(dy12,fscal,fiy1);
1500             fiz1             = _mm_macc_ps(dz12,fscal,fiz1);
1501
1502             fjx2             = _mm_macc_ps(dx12,fscal,fjx2);
1503             fjy2             = _mm_macc_ps(dy12,fscal,fjy2);
1504             fjz2             = _mm_macc_ps(dz12,fscal,fjz2);
1505
1506             /**************************
1507              * CALCULATE INTERACTIONS *
1508              **************************/
1509
1510             /* REACTION-FIELD ELECTROSTATICS */
1511             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1512
1513             fscal            = felec;
1514
1515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1516
1517              /* Update vectorial force */
1518             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1519             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1520             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1521
1522             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1523             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1524             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1525
1526             /**************************
1527              * CALCULATE INTERACTIONS *
1528              **************************/
1529
1530             /* REACTION-FIELD ELECTROSTATICS */
1531             felec            = _mm_mul_ps(qq21,_mm_msub_ps(rinv21,rinvsq21,krf2));
1532
1533             fscal            = felec;
1534
1535             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1536
1537              /* Update vectorial force */
1538             fix2             = _mm_macc_ps(dx21,fscal,fix2);
1539             fiy2             = _mm_macc_ps(dy21,fscal,fiy2);
1540             fiz2             = _mm_macc_ps(dz21,fscal,fiz2);
1541
1542             fjx1             = _mm_macc_ps(dx21,fscal,fjx1);
1543             fjy1             = _mm_macc_ps(dy21,fscal,fjy1);
1544             fjz1             = _mm_macc_ps(dz21,fscal,fjz1);
1545
1546             /**************************
1547              * CALCULATE INTERACTIONS *
1548              **************************/
1549
1550             /* REACTION-FIELD ELECTROSTATICS */
1551             felec            = _mm_mul_ps(qq22,_mm_msub_ps(rinv22,rinvsq22,krf2));
1552
1553             fscal            = felec;
1554
1555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1556
1557              /* Update vectorial force */
1558             fix2             = _mm_macc_ps(dx22,fscal,fix2);
1559             fiy2             = _mm_macc_ps(dy22,fscal,fiy2);
1560             fiz2             = _mm_macc_ps(dz22,fscal,fiz2);
1561
1562             fjx2             = _mm_macc_ps(dx22,fscal,fjx2);
1563             fjy2             = _mm_macc_ps(dy22,fscal,fjy2);
1564             fjz2             = _mm_macc_ps(dz22,fscal,fjz2);
1565
1566             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1567             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1568             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1569             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1570
1571             gmx_mm_decrement_3rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,
1572                                                    fjx0,fjy0,fjz0,fjx1,fjy1,fjz1,fjx2,fjy2,fjz2);
1573
1574             /* Inner loop uses 277 flops */
1575         }
1576
1577         /* End of innermost loop */
1578
1579         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1580                                               f+i_coord_offset,fshift+i_shift_offset);
1581
1582         /* Increment number of inner iterations */
1583         inneriter                  += j_index_end - j_index_start;
1584
1585         /* Outer loop uses 18 flops */
1586     }
1587
1588     /* Increment number of outer iterations */
1589     outeriter        += nri;
1590
1591     /* Update outer/inner flops */
1592
1593     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3W3_F,outeriter*18 + inneriter*277);
1594 }