src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwLJ_GeomW4P1_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_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwLJ_GeomW4P1_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              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     krf              = _mm_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm_set1_ps(fr->ic->c_rf);
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 120     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 121
 122     /* Avoid stupid compiler warnings */
 123     jnrA = jnrB = jnrC = jnrD = 0;
 124     j_coord_offsetA = 0;
 125     j_coord_offsetB = 0;
 126     j_coord_offsetC = 0;
 127     j_coord_offsetD = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     for(iidx=0;iidx<4*DIM;iidx++)
 133     {
 134         scratch[iidx] = 0.0;
 135     }
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 153                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 154
 155         fix0             = _mm_setzero_ps();
 156         fiy0             = _mm_setzero_ps();
 157         fiz0             = _mm_setzero_ps();
 158         fix1             = _mm_setzero_ps();
 159         fiy1             = _mm_setzero_ps();
 160         fiz1             = _mm_setzero_ps();
 161         fix2             = _mm_setzero_ps();
 162         fiy2             = _mm_setzero_ps();
 163         fiz2             = _mm_setzero_ps();
 164         fix3             = _mm_setzero_ps();
 165         fiy3             = _mm_setzero_ps();
 166         fiz3             = _mm_setzero_ps();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_ps();
 170         vvdwsum          = _mm_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181             j_coord_offsetA  = DIM*jnrA;
 182             j_coord_offsetB  = DIM*jnrB;
 183             j_coord_offsetC  = DIM*jnrC;
 184             j_coord_offsetD  = DIM*jnrD;
 185
 186             /* load j atom coordinates */
 187             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 188                                               x+j_coord_offsetC,x+j_coord_offsetD,
 189                                               &jx0,&jy0,&jz0);
 190
 191             /* Calculate displacement vector */
 192             dx00             = _mm_sub_ps(ix0,jx0);
 193             dy00             = _mm_sub_ps(iy0,jy0);
 194             dz00             = _mm_sub_ps(iz0,jz0);
 195             dx10             = _mm_sub_ps(ix1,jx0);
 196             dy10             = _mm_sub_ps(iy1,jy0);
 197             dz10             = _mm_sub_ps(iz1,jz0);
 198             dx20             = _mm_sub_ps(ix2,jx0);
 199             dy20             = _mm_sub_ps(iy2,jy0);
 200             dz20             = _mm_sub_ps(iz2,jz0);
 201             dx30             = _mm_sub_ps(ix3,jx0);
 202             dy30             = _mm_sub_ps(iy3,jy0);
 203             dz30             = _mm_sub_ps(iz3,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 207             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 208             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 209             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 210
 211             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 212             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 213             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 214
 215             rinvsq00         = gmx_mm_inv_ps(rsq00);
 216             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 218             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                               charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             fjx0             = _mm_setzero_ps();
 229             fjy0             = _mm_setzero_ps();
 230             fjz0             = _mm_setzero_ps();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 238                                          vdwparam+vdwioffset0+vdwjidx0B,
 239                                          vdwparam+vdwioffset0+vdwjidx0C,
 240                                          vdwparam+vdwioffset0+vdwjidx0D,
 241                                          &c6_00,&c12_00);
 242
 243             /* LENNARD-JONES DISPERSION/REPULSION */
 244
 245             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 246             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 247             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 248             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 249             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 253
 254             fscal            = fvdw;
 255
 256              /* Update vectorial force */
 257             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 258             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 259             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 260
 261             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 262             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 263             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 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             /* REACTION-FIELD ELECTROSTATICS */
 273             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 274             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 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              /* Update vectorial force */
 282             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 283             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 284             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 285
 286             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 287             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 288             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 289
 290             /**************************
 291              * CALCULATE INTERACTIONS *
 292              **************************/
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq20             = _mm_mul_ps(iq2,jq0);
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 299             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306              /* Update vectorial force */
 307             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 308             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 309             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 310
 311             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 312             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 313             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq30             = _mm_mul_ps(iq3,jq0);
 321
 322             /* REACTION-FIELD ELECTROSTATICS */
 323             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 324             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331              /* Update vectorial force */
 332             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 333             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 334             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 335
 336             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 337             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 338             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 339
 340             fjptrA             = f+j_coord_offsetA;
 341             fjptrB             = f+j_coord_offsetB;
 342             fjptrC             = f+j_coord_offsetC;
 343             fjptrD             = f+j_coord_offsetD;
 344
 345             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 346
 347             /* Inner loop uses 140 flops */
 348         }
 349
 350         if(jidx<j_index_end)
 351         {
 352
 353             /* Get j neighbor index, and coordinate index */
 354             jnrlistA         = jjnr[jidx];
 355             jnrlistB         = jjnr[jidx+1];
 356             jnrlistC         = jjnr[jidx+2];
 357             jnrlistD         = jjnr[jidx+3];
 358             /* Sign of each element will be negative for non-real atoms.
 359              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 360              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 361              */
 362             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 363             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 364             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 365             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 366             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 367             j_coord_offsetA  = DIM*jnrA;
 368             j_coord_offsetB  = DIM*jnrB;
 369             j_coord_offsetC  = DIM*jnrC;
 370             j_coord_offsetD  = DIM*jnrD;
 371
 372             /* load j atom coordinates */
 373             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 374                                               x+j_coord_offsetC,x+j_coord_offsetD,
 375                                               &jx0,&jy0,&jz0);
 376
 377             /* Calculate displacement vector */
 378             dx00             = _mm_sub_ps(ix0,jx0);
 379             dy00             = _mm_sub_ps(iy0,jy0);
 380             dz00             = _mm_sub_ps(iz0,jz0);
 381             dx10             = _mm_sub_ps(ix1,jx0);
 382             dy10             = _mm_sub_ps(iy1,jy0);
 383             dz10             = _mm_sub_ps(iz1,jz0);
 384             dx20             = _mm_sub_ps(ix2,jx0);
 385             dy20             = _mm_sub_ps(iy2,jy0);
 386             dz20             = _mm_sub_ps(iz2,jz0);
 387             dx30             = _mm_sub_ps(ix3,jx0);
 388             dy30             = _mm_sub_ps(iy3,jy0);
 389             dz30             = _mm_sub_ps(iz3,jz0);
 390
 391             /* Calculate squared distance and things based on it */
 392             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 393             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 394             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 395             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 396
 397             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 398             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 399             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 400
 401             rinvsq00         = gmx_mm_inv_ps(rsq00);
 402             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 403             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 404             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 405
 406             /* Load parameters for j particles */
 407             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 408                                                               charge+jnrC+0,charge+jnrD+0);
 409             vdwjidx0A        = 2*vdwtype[jnrA+0];
 410             vdwjidx0B        = 2*vdwtype[jnrB+0];
 411             vdwjidx0C        = 2*vdwtype[jnrC+0];
 412             vdwjidx0D        = 2*vdwtype[jnrD+0];
 413
 414             fjx0             = _mm_setzero_ps();
 415             fjy0             = _mm_setzero_ps();
 416             fjz0             = _mm_setzero_ps();
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 424                                          vdwparam+vdwioffset0+vdwjidx0B,
 425                                          vdwparam+vdwioffset0+vdwjidx0C,
 426                                          vdwparam+vdwioffset0+vdwjidx0D,
 427                                          &c6_00,&c12_00);
 428
 429             /* LENNARD-JONES DISPERSION/REPULSION */
 430
 431             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 432             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 433             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 434             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 435             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 439             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 440
 441             fscal            = fvdw;
 442
 443             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 444
 445              /* Update vectorial force */
 446             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 447             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 448             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 449
 450             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 451             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 452             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             /* Compute parameters for interactions between i and j atoms */
 459             qq10             = _mm_mul_ps(iq1,jq0);
 460
 461             /* REACTION-FIELD ELECTROSTATICS */
 462             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 463             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 464
 465             /* Update potential sum for this i atom from the interaction with this j atom. */
 466             velec            = _mm_andnot_ps(dummy_mask,velec);
 467             velecsum         = _mm_add_ps(velecsum,velec);
 468
 469             fscal            = felec;
 470
 471             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 472
 473              /* Update vectorial force */
 474             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 475             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 476             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 477
 478             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 479             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 480             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 481
 482             /**************************
 483              * CALCULATE INTERACTIONS *
 484              **************************/
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq20             = _mm_mul_ps(iq2,jq0);
 488
 489             /* REACTION-FIELD ELECTROSTATICS */
 490             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 491             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 492
 493             /* Update potential sum for this i atom from the interaction with this j atom. */
 494             velec            = _mm_andnot_ps(dummy_mask,velec);
 495             velecsum         = _mm_add_ps(velecsum,velec);
 496
 497             fscal            = felec;
 498
 499             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 500
 501              /* Update vectorial force */
 502             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 503             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 504             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 505
 506             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 507             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 508             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             /* Compute parameters for interactions between i and j atoms */
 515             qq30             = _mm_mul_ps(iq3,jq0);
 516
 517             /* REACTION-FIELD ELECTROSTATICS */
 518             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 519             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_andnot_ps(dummy_mask,velec);
 523             velecsum         = _mm_add_ps(velecsum,velec);
 524
 525             fscal            = felec;
 526
 527             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 528
 529              /* Update vectorial force */
 530             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 531             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 532             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 533
 534             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 535             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 536             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 537
 538             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 539             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 540             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 541             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 542
 543             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 544
 545             /* Inner loop uses 140 flops */
 546         }
 547
 548         /* End of innermost loop */
 549
 550         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 551                                               f+i_coord_offset,fshift+i_shift_offset);
 552
 553         ggid                        = gid[iidx];
 554         /* Update potential energies */
 555         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 556         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 557
 558         /* Increment number of inner iterations */
 559         inneriter                  += j_index_end - j_index_start;
 560
 561         /* Outer loop uses 26 flops */
 562     }
 563
 564     /* Increment number of outer iterations */
 565     outeriter        += nri;
 566
 567     /* Update outer/inner flops */
 568
 569     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*140);
 570 }
 571 /*
 572  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_128_fma_single
 573  * Electrostatics interaction: ReactionField
 574  * VdW interaction:            LennardJones
 575  * Geometry:                   Water4-Particle
 576  * Calculate force/pot:        Force
 577  */
 578 void
 579 nb_kernel_ElecRF_VdwLJ_GeomW4P1_F_avx_128_fma_single
 580                     (t_nblist * gmx_restrict                nlist,
 581                      rvec * gmx_restrict                    xx,
 582                      rvec * gmx_restrict                    ff,
 583                      t_forcerec * gmx_restrict              fr,
 584                      t_mdatoms * gmx_restrict               mdatoms,
 585                      nb_kernel_data_t * gmx_restrict        kernel_data,
 586                      t_nrnb * gmx_restrict                  nrnb)
 587 {
 588     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 589      * just 0 for non-waters.
 590      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 591      * jnr indices corresponding to data put in the four positions in the SIMD register.
 592      */
 593     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 594     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 595     int              jnrA,jnrB,jnrC,jnrD;
 596     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 597     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 598     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 599     real             rcutoff_scalar;
 600     real             *shiftvec,*fshift,*x,*f;
 601     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 602     real             scratch[4*DIM];
 603     __m128           fscal,rcutoff,rcutoff2,jidxall;
 604     int              vdwioffset0;
 605     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 606     int              vdwioffset1;
 607     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 608     int              vdwioffset2;
 609     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 610     int              vdwioffset3;
 611     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 612     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 613     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 614     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 615     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 616     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 617     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 618     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 619     real             *charge;
 620     int              nvdwtype;
 621     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 622     int              *vdwtype;
 623     real             *vdwparam;
 624     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 625     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 626     __m128           dummy_mask,cutoff_mask;
 627     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 628     __m128           one     = _mm_set1_ps(1.0);
 629     __m128           two     = _mm_set1_ps(2.0);
 630     x                = xx[0];
 631     f                = ff[0];
 632
 633     nri              = nlist->nri;
 634     iinr             = nlist->iinr;
 635     jindex           = nlist->jindex;
 636     jjnr             = nlist->jjnr;
 637     shiftidx         = nlist->shift;
 638     gid              = nlist->gid;
 639     shiftvec         = fr->shift_vec[0];
 640     fshift           = fr->fshift[0];
 641     facel            = _mm_set1_ps(fr->epsfac);
 642     charge           = mdatoms->chargeA;
 643     krf              = _mm_set1_ps(fr->ic->k_rf);
 644     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 645     crf              = _mm_set1_ps(fr->ic->c_rf);
 646     nvdwtype         = fr->ntype;
 647     vdwparam         = fr->nbfp;
 648     vdwtype          = mdatoms->typeA;
 649
 650     /* Setup water-specific parameters */
 651     inr              = nlist->iinr[0];
 652     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 653     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 654     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 655     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 656
 657     /* Avoid stupid compiler warnings */
 658     jnrA = jnrB = jnrC = jnrD = 0;
 659     j_coord_offsetA = 0;
 660     j_coord_offsetB = 0;
 661     j_coord_offsetC = 0;
 662     j_coord_offsetD = 0;
 663
 664     outeriter        = 0;
 665     inneriter        = 0;
 666
 667     for(iidx=0;iidx<4*DIM;iidx++)
 668     {
 669         scratch[iidx] = 0.0;
 670     }
 671
 672     /* Start outer loop over neighborlists */
 673     for(iidx=0; iidx<nri; iidx++)
 674     {
 675         /* Load shift vector for this list */
 676         i_shift_offset   = DIM*shiftidx[iidx];
 677
 678         /* Load limits for loop over neighbors */
 679         j_index_start    = jindex[iidx];
 680         j_index_end      = jindex[iidx+1];
 681
 682         /* Get outer coordinate index */
 683         inr              = iinr[iidx];
 684         i_coord_offset   = DIM*inr;
 685
 686         /* Load i particle coords and add shift vector */
 687         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 688                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 689
 690         fix0             = _mm_setzero_ps();
 691         fiy0             = _mm_setzero_ps();
 692         fiz0             = _mm_setzero_ps();
 693         fix1             = _mm_setzero_ps();
 694         fiy1             = _mm_setzero_ps();
 695         fiz1             = _mm_setzero_ps();
 696         fix2             = _mm_setzero_ps();
 697         fiy2             = _mm_setzero_ps();
 698         fiz2             = _mm_setzero_ps();
 699         fix3             = _mm_setzero_ps();
 700         fiy3             = _mm_setzero_ps();
 701         fiz3             = _mm_setzero_ps();
 702
 703         /* Start inner kernel loop */
 704         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 705         {
 706
 707             /* Get j neighbor index, and coordinate index */
 708             jnrA             = jjnr[jidx];
 709             jnrB             = jjnr[jidx+1];
 710             jnrC             = jjnr[jidx+2];
 711             jnrD             = jjnr[jidx+3];
 712             j_coord_offsetA  = DIM*jnrA;
 713             j_coord_offsetB  = DIM*jnrB;
 714             j_coord_offsetC  = DIM*jnrC;
 715             j_coord_offsetD  = DIM*jnrD;
 716
 717             /* load j atom coordinates */
 718             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 719                                               x+j_coord_offsetC,x+j_coord_offsetD,
 720                                               &jx0,&jy0,&jz0);
 721
 722             /* Calculate displacement vector */
 723             dx00             = _mm_sub_ps(ix0,jx0);
 724             dy00             = _mm_sub_ps(iy0,jy0);
 725             dz00             = _mm_sub_ps(iz0,jz0);
 726             dx10             = _mm_sub_ps(ix1,jx0);
 727             dy10             = _mm_sub_ps(iy1,jy0);
 728             dz10             = _mm_sub_ps(iz1,jz0);
 729             dx20             = _mm_sub_ps(ix2,jx0);
 730             dy20             = _mm_sub_ps(iy2,jy0);
 731             dz20             = _mm_sub_ps(iz2,jz0);
 732             dx30             = _mm_sub_ps(ix3,jx0);
 733             dy30             = _mm_sub_ps(iy3,jy0);
 734             dz30             = _mm_sub_ps(iz3,jz0);
 735
 736             /* Calculate squared distance and things based on it */
 737             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 738             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 739             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 740             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 741
 742             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 743             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 744             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 745
 746             rinvsq00         = gmx_mm_inv_ps(rsq00);
 747             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 748             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 749             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 750
 751             /* Load parameters for j particles */
 752             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 753                                                               charge+jnrC+0,charge+jnrD+0);
 754             vdwjidx0A        = 2*vdwtype[jnrA+0];
 755             vdwjidx0B        = 2*vdwtype[jnrB+0];
 756             vdwjidx0C        = 2*vdwtype[jnrC+0];
 757             vdwjidx0D        = 2*vdwtype[jnrD+0];
 758
 759             fjx0             = _mm_setzero_ps();
 760             fjy0             = _mm_setzero_ps();
 761             fjz0             = _mm_setzero_ps();
 762
 763             /**************************
 764              * CALCULATE INTERACTIONS *
 765              **************************/
 766
 767             /* Compute parameters for interactions between i and j atoms */
 768             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 769                                          vdwparam+vdwioffset0+vdwjidx0B,
 770                                          vdwparam+vdwioffset0+vdwjidx0C,
 771                                          vdwparam+vdwioffset0+vdwjidx0D,
 772                                          &c6_00,&c12_00);
 773
 774             /* LENNARD-JONES DISPERSION/REPULSION */
 775
 776             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 777             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 778
 779             fscal            = fvdw;
 780
 781              /* Update vectorial force */
 782             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 783             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 784             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 785
 786             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 787             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 788             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             /* Compute parameters for interactions between i and j atoms */
 795             qq10             = _mm_mul_ps(iq1,jq0);
 796
 797             /* REACTION-FIELD ELECTROSTATICS */
 798             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 799
 800             fscal            = felec;
 801
 802              /* Update vectorial force */
 803             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 804             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 805             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 806
 807             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 808             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 809             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq20             = _mm_mul_ps(iq2,jq0);
 817
 818             /* REACTION-FIELD ELECTROSTATICS */
 819             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 820
 821             fscal            = felec;
 822
 823              /* Update vectorial force */
 824             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 825             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 826             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 827
 828             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 829             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 830             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 831
 832             /**************************
 833              * CALCULATE INTERACTIONS *
 834              **************************/
 835
 836             /* Compute parameters for interactions between i and j atoms */
 837             qq30             = _mm_mul_ps(iq3,jq0);
 838
 839             /* REACTION-FIELD ELECTROSTATICS */
 840             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 841
 842             fscal            = felec;
 843
 844              /* Update vectorial force */
 845             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 846             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 847             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 848
 849             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 850             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 851             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 852
 853             fjptrA             = f+j_coord_offsetA;
 854             fjptrB             = f+j_coord_offsetB;
 855             fjptrC             = f+j_coord_offsetC;
 856             fjptrD             = f+j_coord_offsetD;
 857
 858             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 859
 860             /* Inner loop uses 120 flops */
 861         }
 862
 863         if(jidx<j_index_end)
 864         {
 865
 866             /* Get j neighbor index, and coordinate index */
 867             jnrlistA         = jjnr[jidx];
 868             jnrlistB         = jjnr[jidx+1];
 869             jnrlistC         = jjnr[jidx+2];
 870             jnrlistD         = jjnr[jidx+3];
 871             /* Sign of each element will be negative for non-real atoms.
 872              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 873              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 874              */
 875             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 876             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 877             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 878             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 879             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 880             j_coord_offsetA  = DIM*jnrA;
 881             j_coord_offsetB  = DIM*jnrB;
 882             j_coord_offsetC  = DIM*jnrC;
 883             j_coord_offsetD  = DIM*jnrD;
 884
 885             /* load j atom coordinates */
 886             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 887                                               x+j_coord_offsetC,x+j_coord_offsetD,
 888                                               &jx0,&jy0,&jz0);
 889
 890             /* Calculate displacement vector */
 891             dx00             = _mm_sub_ps(ix0,jx0);
 892             dy00             = _mm_sub_ps(iy0,jy0);
 893             dz00             = _mm_sub_ps(iz0,jz0);
 894             dx10             = _mm_sub_ps(ix1,jx0);
 895             dy10             = _mm_sub_ps(iy1,jy0);
 896             dz10             = _mm_sub_ps(iz1,jz0);
 897             dx20             = _mm_sub_ps(ix2,jx0);
 898             dy20             = _mm_sub_ps(iy2,jy0);
 899             dz20             = _mm_sub_ps(iz2,jz0);
 900             dx30             = _mm_sub_ps(ix3,jx0);
 901             dy30             = _mm_sub_ps(iy3,jy0);
 902             dz30             = _mm_sub_ps(iz3,jz0);
 903
 904             /* Calculate squared distance and things based on it */
 905             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 906             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 907             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 908             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 909
 910             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 911             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 912             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 913
 914             rinvsq00         = gmx_mm_inv_ps(rsq00);
 915             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 916             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 917             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 918
 919             /* Load parameters for j particles */
 920             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 921                                                               charge+jnrC+0,charge+jnrD+0);
 922             vdwjidx0A        = 2*vdwtype[jnrA+0];
 923             vdwjidx0B        = 2*vdwtype[jnrB+0];
 924             vdwjidx0C        = 2*vdwtype[jnrC+0];
 925             vdwjidx0D        = 2*vdwtype[jnrD+0];
 926
 927             fjx0             = _mm_setzero_ps();
 928             fjy0             = _mm_setzero_ps();
 929             fjz0             = _mm_setzero_ps();
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             /* Compute parameters for interactions between i and j atoms */
 936             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 937                                          vdwparam+vdwioffset0+vdwjidx0B,
 938                                          vdwparam+vdwioffset0+vdwjidx0C,
 939                                          vdwparam+vdwioffset0+vdwjidx0D,
 940                                          &c6_00,&c12_00);
 941
 942             /* LENNARD-JONES DISPERSION/REPULSION */
 943
 944             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 945             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 946
 947             fscal            = fvdw;
 948
 949             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 950
 951              /* Update vectorial force */
 952             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 953             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 954             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 955
 956             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 957             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 958             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq10             = _mm_mul_ps(iq1,jq0);
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 973
 974              /* Update vectorial force */
 975             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 976             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 977             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 978
 979             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 980             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 981             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq20             = _mm_mul_ps(iq2,jq0);
 989
 990             /* REACTION-FIELD ELECTROSTATICS */
 991             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 992
 993             fscal            = felec;
 994
 995             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 996
 997              /* Update vectorial force */
 998             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 999             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1000             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1001
1002             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1003             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1004             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1005
1006             /**************************
1007              * CALCULATE INTERACTIONS *
1008              **************************/
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq30             = _mm_mul_ps(iq3,jq0);
1012
1013             /* REACTION-FIELD ELECTROSTATICS */
1014             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1015
1016             fscal            = felec;
1017
1018             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1019
1020              /* Update vectorial force */
1021             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1022             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1023             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1024
1025             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1026             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1027             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1028
1029             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1030             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1031             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1032             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1033
1034             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1035
1036             /* Inner loop uses 120 flops */
1037         }
1038
1039         /* End of innermost loop */
1040
1041         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1042                                               f+i_coord_offset,fshift+i_shift_offset);
1043
1044         /* Increment number of inner iterations */
1045         inneriter                  += j_index_end - j_index_start;
1046
1047         /* Outer loop uses 24 flops */
1048     }
1049
1050     /* Increment number of outer iterations */
1051     outeriter        += nri;
1052
1053     /* Update outer/inner flops */
1054
1055     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1056 }