src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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_ElecRFCut_VdwLJSh_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_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     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128           dummy_mask,cutoff_mask;
  89     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  90     __m128           one     = _mm_set1_ps(1.0);
  91     __m128           two     = _mm_set1_ps(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_ps(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     krf              = _mm_set1_ps(fr->ic->k_rf);
 106     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 107     crf              = _mm_set1_ps(fr->ic->c_rf);
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 115     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 116     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 117     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 118
 119     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 120     rcutoff_scalar   = fr->rcoulomb;
 121     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 122     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 123
 124     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 125     rvdw             = _mm_set1_ps(fr->rvdw);
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = jnrC = jnrD = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131     j_coord_offsetC = 0;
 132     j_coord_offsetD = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     for(iidx=0;iidx<4*DIM;iidx++)
 138     {
 139         scratch[iidx] = 0.0;
 140     }
 141
 142     /* Start outer loop over neighborlists */
 143     for(iidx=0; iidx<nri; iidx++)
 144     {
 145         /* Load shift vector for this list */
 146         i_shift_offset   = DIM*shiftidx[iidx];
 147
 148         /* Load limits for loop over neighbors */
 149         j_index_start    = jindex[iidx];
 150         j_index_end      = jindex[iidx+1];
 151
 152         /* Get outer coordinate index */
 153         inr              = iinr[iidx];
 154         i_coord_offset   = DIM*inr;
 155
 156         /* Load i particle coords and add shift vector */
 157         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 158                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 159
 160         fix0             = _mm_setzero_ps();
 161         fiy0             = _mm_setzero_ps();
 162         fiz0             = _mm_setzero_ps();
 163         fix1             = _mm_setzero_ps();
 164         fiy1             = _mm_setzero_ps();
 165         fiz1             = _mm_setzero_ps();
 166         fix2             = _mm_setzero_ps();
 167         fiy2             = _mm_setzero_ps();
 168         fiz2             = _mm_setzero_ps();
 169
 170         /* Reset potential sums */
 171         velecsum         = _mm_setzero_ps();
 172         vvdwsum          = _mm_setzero_ps();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               x+j_coord_offsetC,x+j_coord_offsetD,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_ps(ix0,jx0);
 195             dy00             = _mm_sub_ps(iy0,jy0);
 196             dz00             = _mm_sub_ps(iz0,jz0);
 197             dx10             = _mm_sub_ps(ix1,jx0);
 198             dy10             = _mm_sub_ps(iy1,jy0);
 199             dz10             = _mm_sub_ps(iz1,jz0);
 200             dx20             = _mm_sub_ps(ix2,jx0);
 201             dy20             = _mm_sub_ps(iy2,jy0);
 202             dz20             = _mm_sub_ps(iz2,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 207             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 208
 209             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 210             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 211             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 212
 213             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 214             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 215             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 224
 225             fjx0             = _mm_setzero_ps();
 226             fjy0             = _mm_setzero_ps();
 227             fjz0             = _mm_setzero_ps();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq00,rcutoff2))
 234             {
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm_mul_ps(iq0,jq0);
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             /* REACTION-FIELD ELECTROSTATICS */
 245             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 246             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 247
 248             /* LENNARD-JONES DISPERSION/REPULSION */
 249
 250             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 251             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 252             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 253             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 254                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 255             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 256
 257             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 258
 259             /* Update potential sum for this i atom from the interaction with this j atom. */
 260             velec            = _mm_and_ps(velec,cutoff_mask);
 261             velecsum         = _mm_add_ps(velecsum,velec);
 262             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 263             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 264
 265             fscal            = _mm_add_ps(felec,fvdw);
 266
 267             fscal            = _mm_and_ps(fscal,cutoff_mask);
 268
 269              /* Update vectorial force */
 270             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 271             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 272             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 273
 274             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 275             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 276             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 277
 278             }
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             if (gmx_mm_any_lt(rsq10,rcutoff2))
 285             {
 286
 287             /* Compute parameters for interactions between i and j atoms */
 288             qq10             = _mm_mul_ps(iq1,jq0);
 289
 290             /* REACTION-FIELD ELECTROSTATICS */
 291             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 292             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 293
 294             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_and_ps(velec,cutoff_mask);
 298             velecsum         = _mm_add_ps(velecsum,velec);
 299
 300             fscal            = felec;
 301
 302             fscal            = _mm_and_ps(fscal,cutoff_mask);
 303
 304              /* Update vectorial force */
 305             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 306             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 307             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 308
 309             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 310             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 311             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm_any_lt(rsq20,rcutoff2))
 320             {
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq20             = _mm_mul_ps(iq2,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 327             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 328
 329             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velec            = _mm_and_ps(velec,cutoff_mask);
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             fscal            = _mm_and_ps(fscal,cutoff_mask);
 338
 339              /* Update vectorial force */
 340             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 341             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 342             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 343
 344             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 345             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 346             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 347
 348             }
 349
 350             fjptrA             = f+j_coord_offsetA;
 351             fjptrB             = f+j_coord_offsetB;
 352             fjptrC             = f+j_coord_offsetC;
 353             fjptrD             = f+j_coord_offsetD;
 354
 355             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 356
 357             /* Inner loop uses 135 flops */
 358         }
 359
 360         if(jidx<j_index_end)
 361         {
 362
 363             /* Get j neighbor index, and coordinate index */
 364             jnrlistA         = jjnr[jidx];
 365             jnrlistB         = jjnr[jidx+1];
 366             jnrlistC         = jjnr[jidx+2];
 367             jnrlistD         = jjnr[jidx+3];
 368             /* Sign of each element will be negative for non-real atoms.
 369              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 370              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 371              */
 372             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 373             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 374             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 375             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 376             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 377             j_coord_offsetA  = DIM*jnrA;
 378             j_coord_offsetB  = DIM*jnrB;
 379             j_coord_offsetC  = DIM*jnrC;
 380             j_coord_offsetD  = DIM*jnrD;
 381
 382             /* load j atom coordinates */
 383             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 384                                               x+j_coord_offsetC,x+j_coord_offsetD,
 385                                               &jx0,&jy0,&jz0);
 386
 387             /* Calculate displacement vector */
 388             dx00             = _mm_sub_ps(ix0,jx0);
 389             dy00             = _mm_sub_ps(iy0,jy0);
 390             dz00             = _mm_sub_ps(iz0,jz0);
 391             dx10             = _mm_sub_ps(ix1,jx0);
 392             dy10             = _mm_sub_ps(iy1,jy0);
 393             dz10             = _mm_sub_ps(iz1,jz0);
 394             dx20             = _mm_sub_ps(ix2,jx0);
 395             dy20             = _mm_sub_ps(iy2,jy0);
 396             dz20             = _mm_sub_ps(iz2,jz0);
 397
 398             /* Calculate squared distance and things based on it */
 399             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 400             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 401             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 402
 403             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 404             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 405             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 406
 407             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 408             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 409             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 410
 411             /* Load parameters for j particles */
 412             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 413                                                               charge+jnrC+0,charge+jnrD+0);
 414             vdwjidx0A        = 2*vdwtype[jnrA+0];
 415             vdwjidx0B        = 2*vdwtype[jnrB+0];
 416             vdwjidx0C        = 2*vdwtype[jnrC+0];
 417             vdwjidx0D        = 2*vdwtype[jnrD+0];
 418
 419             fjx0             = _mm_setzero_ps();
 420             fjy0             = _mm_setzero_ps();
 421             fjz0             = _mm_setzero_ps();
 422
 423             /**************************
 424              * CALCULATE INTERACTIONS *
 425              **************************/
 426
 427             if (gmx_mm_any_lt(rsq00,rcutoff2))
 428             {
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq00             = _mm_mul_ps(iq0,jq0);
 432             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 433                                          vdwparam+vdwioffset0+vdwjidx0B,
 434                                          vdwparam+vdwioffset0+vdwjidx0C,
 435                                          vdwparam+vdwioffset0+vdwjidx0D,
 436                                          &c6_00,&c12_00);
 437
 438             /* REACTION-FIELD ELECTROSTATICS */
 439             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 440             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 441
 442             /* LENNARD-JONES DISPERSION/REPULSION */
 443
 444             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 445             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 446             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 447             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 448                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 449             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 450
 451             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 452
 453             /* Update potential sum for this i atom from the interaction with this j atom. */
 454             velec            = _mm_and_ps(velec,cutoff_mask);
 455             velec            = _mm_andnot_ps(dummy_mask,velec);
 456             velecsum         = _mm_add_ps(velecsum,velec);
 457             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 458             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 459             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 460
 461             fscal            = _mm_add_ps(felec,fvdw);
 462
 463             fscal            = _mm_and_ps(fscal,cutoff_mask);
 464
 465             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 466
 467              /* Update vectorial force */
 468             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 469             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 470             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 471
 472             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 473             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 474             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 475
 476             }
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             if (gmx_mm_any_lt(rsq10,rcutoff2))
 483             {
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq10             = _mm_mul_ps(iq1,jq0);
 487
 488             /* REACTION-FIELD ELECTROSTATICS */
 489             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 490             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 491
 492             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 493
 494             /* Update potential sum for this i atom from the interaction with this j atom. */
 495             velec            = _mm_and_ps(velec,cutoff_mask);
 496             velec            = _mm_andnot_ps(dummy_mask,velec);
 497             velecsum         = _mm_add_ps(velecsum,velec);
 498
 499             fscal            = felec;
 500
 501             fscal            = _mm_and_ps(fscal,cutoff_mask);
 502
 503             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 504
 505              /* Update vectorial force */
 506             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 507             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 508             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 509
 510             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 511             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 512             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 513
 514             }
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             if (gmx_mm_any_lt(rsq20,rcutoff2))
 521             {
 522
 523             /* Compute parameters for interactions between i and j atoms */
 524             qq20             = _mm_mul_ps(iq2,jq0);
 525
 526             /* REACTION-FIELD ELECTROSTATICS */
 527             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 528             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 529
 530             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_and_ps(velec,cutoff_mask);
 534             velec            = _mm_andnot_ps(dummy_mask,velec);
 535             velecsum         = _mm_add_ps(velecsum,velec);
 536
 537             fscal            = felec;
 538
 539             fscal            = _mm_and_ps(fscal,cutoff_mask);
 540
 541             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 542
 543              /* Update vectorial force */
 544             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 545             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 546             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 547
 548             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 549             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 550             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 551
 552             }
 553
 554             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 555             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 556             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 557             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 558
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 560
 561             /* Inner loop uses 135 flops */
 562         }
 563
 564         /* End of innermost loop */
 565
 566         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 567                                               f+i_coord_offset,fshift+i_shift_offset);
 568
 569         ggid                        = gid[iidx];
 570         /* Update potential energies */
 571         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 572         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 573
 574         /* Increment number of inner iterations */
 575         inneriter                  += j_index_end - j_index_start;
 576
 577         /* Outer loop uses 20 flops */
 578     }
 579
 580     /* Increment number of outer iterations */
 581     outeriter        += nri;
 582
 583     /* Update outer/inner flops */
 584
 585     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*135);
 586 }
 587 /*
 588  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 589  * Electrostatics interaction: ReactionField
 590  * VdW interaction:            LennardJones
 591  * Geometry:                   Water3-Particle
 592  * Calculate force/pot:        Force
 593  */
 594 void
 595 nb_kernel_ElecRFCut_VdwLJSh_GeomW3P1_F_avx_128_fma_single
 596                     (t_nblist * gmx_restrict                nlist,
 597                      rvec * gmx_restrict                    xx,
 598                      rvec * gmx_restrict                    ff,
 599                      t_forcerec * gmx_restrict              fr,
 600                      t_mdatoms * gmx_restrict               mdatoms,
 601                      nb_kernel_data_t * gmx_restrict        kernel_data,
 602                      t_nrnb * gmx_restrict                  nrnb)
 603 {
 604     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 605      * just 0 for non-waters.
 606      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 607      * jnr indices corresponding to data put in the four positions in the SIMD register.
 608      */
 609     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 610     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 611     int              jnrA,jnrB,jnrC,jnrD;
 612     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 613     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 614     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 615     real             rcutoff_scalar;
 616     real             *shiftvec,*fshift,*x,*f;
 617     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 618     real             scratch[4*DIM];
 619     __m128           fscal,rcutoff,rcutoff2,jidxall;
 620     int              vdwioffset0;
 621     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 622     int              vdwioffset1;
 623     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 624     int              vdwioffset2;
 625     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 626     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 627     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 628     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 629     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 630     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 631     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 632     real             *charge;
 633     int              nvdwtype;
 634     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 635     int              *vdwtype;
 636     real             *vdwparam;
 637     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 638     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 639     __m128           dummy_mask,cutoff_mask;
 640     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 641     __m128           one     = _mm_set1_ps(1.0);
 642     __m128           two     = _mm_set1_ps(2.0);
 643     x                = xx[0];
 644     f                = ff[0];
 645
 646     nri              = nlist->nri;
 647     iinr             = nlist->iinr;
 648     jindex           = nlist->jindex;
 649     jjnr             = nlist->jjnr;
 650     shiftidx         = nlist->shift;
 651     gid              = nlist->gid;
 652     shiftvec         = fr->shift_vec[0];
 653     fshift           = fr->fshift[0];
 654     facel            = _mm_set1_ps(fr->epsfac);
 655     charge           = mdatoms->chargeA;
 656     krf              = _mm_set1_ps(fr->ic->k_rf);
 657     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 658     crf              = _mm_set1_ps(fr->ic->c_rf);
 659     nvdwtype         = fr->ntype;
 660     vdwparam         = fr->nbfp;
 661     vdwtype          = mdatoms->typeA;
 662
 663     /* Setup water-specific parameters */
 664     inr              = nlist->iinr[0];
 665     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 666     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 667     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 668     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 669
 670     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 671     rcutoff_scalar   = fr->rcoulomb;
 672     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 673     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 674
 675     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 676     rvdw             = _mm_set1_ps(fr->rvdw);
 677
 678     /* Avoid stupid compiler warnings */
 679     jnrA = jnrB = jnrC = jnrD = 0;
 680     j_coord_offsetA = 0;
 681     j_coord_offsetB = 0;
 682     j_coord_offsetC = 0;
 683     j_coord_offsetD = 0;
 684
 685     outeriter        = 0;
 686     inneriter        = 0;
 687
 688     for(iidx=0;iidx<4*DIM;iidx++)
 689     {
 690         scratch[iidx] = 0.0;
 691     }
 692
 693     /* Start outer loop over neighborlists */
 694     for(iidx=0; iidx<nri; iidx++)
 695     {
 696         /* Load shift vector for this list */
 697         i_shift_offset   = DIM*shiftidx[iidx];
 698
 699         /* Load limits for loop over neighbors */
 700         j_index_start    = jindex[iidx];
 701         j_index_end      = jindex[iidx+1];
 702
 703         /* Get outer coordinate index */
 704         inr              = iinr[iidx];
 705         i_coord_offset   = DIM*inr;
 706
 707         /* Load i particle coords and add shift vector */
 708         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 709                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 710
 711         fix0             = _mm_setzero_ps();
 712         fiy0             = _mm_setzero_ps();
 713         fiz0             = _mm_setzero_ps();
 714         fix1             = _mm_setzero_ps();
 715         fiy1             = _mm_setzero_ps();
 716         fiz1             = _mm_setzero_ps();
 717         fix2             = _mm_setzero_ps();
 718         fiy2             = _mm_setzero_ps();
 719         fiz2             = _mm_setzero_ps();
 720
 721         /* Start inner kernel loop */
 722         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 723         {
 724
 725             /* Get j neighbor index, and coordinate index */
 726             jnrA             = jjnr[jidx];
 727             jnrB             = jjnr[jidx+1];
 728             jnrC             = jjnr[jidx+2];
 729             jnrD             = jjnr[jidx+3];
 730             j_coord_offsetA  = DIM*jnrA;
 731             j_coord_offsetB  = DIM*jnrB;
 732             j_coord_offsetC  = DIM*jnrC;
 733             j_coord_offsetD  = DIM*jnrD;
 734
 735             /* load j atom coordinates */
 736             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 737                                               x+j_coord_offsetC,x+j_coord_offsetD,
 738                                               &jx0,&jy0,&jz0);
 739
 740             /* Calculate displacement vector */
 741             dx00             = _mm_sub_ps(ix0,jx0);
 742             dy00             = _mm_sub_ps(iy0,jy0);
 743             dz00             = _mm_sub_ps(iz0,jz0);
 744             dx10             = _mm_sub_ps(ix1,jx0);
 745             dy10             = _mm_sub_ps(iy1,jy0);
 746             dz10             = _mm_sub_ps(iz1,jz0);
 747             dx20             = _mm_sub_ps(ix2,jx0);
 748             dy20             = _mm_sub_ps(iy2,jy0);
 749             dz20             = _mm_sub_ps(iz2,jz0);
 750
 751             /* Calculate squared distance and things based on it */
 752             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 753             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 754             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 755
 756             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 757             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 758             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 759
 760             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 761             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 762             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 763
 764             /* Load parameters for j particles */
 765             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 766                                                               charge+jnrC+0,charge+jnrD+0);
 767             vdwjidx0A        = 2*vdwtype[jnrA+0];
 768             vdwjidx0B        = 2*vdwtype[jnrB+0];
 769             vdwjidx0C        = 2*vdwtype[jnrC+0];
 770             vdwjidx0D        = 2*vdwtype[jnrD+0];
 771
 772             fjx0             = _mm_setzero_ps();
 773             fjy0             = _mm_setzero_ps();
 774             fjz0             = _mm_setzero_ps();
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             if (gmx_mm_any_lt(rsq00,rcutoff2))
 781             {
 782
 783             /* Compute parameters for interactions between i and j atoms */
 784             qq00             = _mm_mul_ps(iq0,jq0);
 785             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 786                                          vdwparam+vdwioffset0+vdwjidx0B,
 787                                          vdwparam+vdwioffset0+vdwjidx0C,
 788                                          vdwparam+vdwioffset0+vdwjidx0D,
 789                                          &c6_00,&c12_00);
 790
 791             /* REACTION-FIELD ELECTROSTATICS */
 792             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 793
 794             /* LENNARD-JONES DISPERSION/REPULSION */
 795
 796             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 797             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 798
 799             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 800
 801             fscal            = _mm_add_ps(felec,fvdw);
 802
 803             fscal            = _mm_and_ps(fscal,cutoff_mask);
 804
 805              /* Update vectorial force */
 806             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 807             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 808             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 809
 810             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 811             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 812             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 813
 814             }
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             if (gmx_mm_any_lt(rsq10,rcutoff2))
 821             {
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             qq10             = _mm_mul_ps(iq1,jq0);
 825
 826             /* REACTION-FIELD ELECTROSTATICS */
 827             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 828
 829             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm_and_ps(fscal,cutoff_mask);
 834
 835              /* Update vectorial force */
 836             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 837             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 838             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 839
 840             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 841             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 842             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 843
 844             }
 845
 846             /**************************
 847              * CALCULATE INTERACTIONS *
 848              **************************/
 849
 850             if (gmx_mm_any_lt(rsq20,rcutoff2))
 851             {
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             qq20             = _mm_mul_ps(iq2,jq0);
 855
 856             /* REACTION-FIELD ELECTROSTATICS */
 857             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 858
 859             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 860
 861             fscal            = felec;
 862
 863             fscal            = _mm_and_ps(fscal,cutoff_mask);
 864
 865              /* Update vectorial force */
 866             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 867             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 868             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 869
 870             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 871             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 872             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 873
 874             }
 875
 876             fjptrA             = f+j_coord_offsetA;
 877             fjptrB             = f+j_coord_offsetB;
 878             fjptrC             = f+j_coord_offsetC;
 879             fjptrD             = f+j_coord_offsetD;
 880
 881             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 882
 883             /* Inner loop uses 106 flops */
 884         }
 885
 886         if(jidx<j_index_end)
 887         {
 888
 889             /* Get j neighbor index, and coordinate index */
 890             jnrlistA         = jjnr[jidx];
 891             jnrlistB         = jjnr[jidx+1];
 892             jnrlistC         = jjnr[jidx+2];
 893             jnrlistD         = jjnr[jidx+3];
 894             /* Sign of each element will be negative for non-real atoms.
 895              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 896              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 897              */
 898             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 899             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 900             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 901             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 902             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 903             j_coord_offsetA  = DIM*jnrA;
 904             j_coord_offsetB  = DIM*jnrB;
 905             j_coord_offsetC  = DIM*jnrC;
 906             j_coord_offsetD  = DIM*jnrD;
 907
 908             /* load j atom coordinates */
 909             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 910                                               x+j_coord_offsetC,x+j_coord_offsetD,
 911                                               &jx0,&jy0,&jz0);
 912
 913             /* Calculate displacement vector */
 914             dx00             = _mm_sub_ps(ix0,jx0);
 915             dy00             = _mm_sub_ps(iy0,jy0);
 916             dz00             = _mm_sub_ps(iz0,jz0);
 917             dx10             = _mm_sub_ps(ix1,jx0);
 918             dy10             = _mm_sub_ps(iy1,jy0);
 919             dz10             = _mm_sub_ps(iz1,jz0);
 920             dx20             = _mm_sub_ps(ix2,jx0);
 921             dy20             = _mm_sub_ps(iy2,jy0);
 922             dz20             = _mm_sub_ps(iz2,jz0);
 923
 924             /* Calculate squared distance and things based on it */
 925             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 926             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 927             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 928
 929             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 930             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 931             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 932
 933             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 934             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 935             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 936
 937             /* Load parameters for j particles */
 938             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 939                                                               charge+jnrC+0,charge+jnrD+0);
 940             vdwjidx0A        = 2*vdwtype[jnrA+0];
 941             vdwjidx0B        = 2*vdwtype[jnrB+0];
 942             vdwjidx0C        = 2*vdwtype[jnrC+0];
 943             vdwjidx0D        = 2*vdwtype[jnrD+0];
 944
 945             fjx0             = _mm_setzero_ps();
 946             fjy0             = _mm_setzero_ps();
 947             fjz0             = _mm_setzero_ps();
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             if (gmx_mm_any_lt(rsq00,rcutoff2))
 954             {
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq00             = _mm_mul_ps(iq0,jq0);
 958             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 959                                          vdwparam+vdwioffset0+vdwjidx0B,
 960                                          vdwparam+vdwioffset0+vdwjidx0C,
 961                                          vdwparam+vdwioffset0+vdwjidx0D,
 962                                          &c6_00,&c12_00);
 963
 964             /* REACTION-FIELD ELECTROSTATICS */
 965             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 966
 967             /* LENNARD-JONES DISPERSION/REPULSION */
 968
 969             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 970             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 971
 972             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 973
 974             fscal            = _mm_add_ps(felec,fvdw);
 975
 976             fscal            = _mm_and_ps(fscal,cutoff_mask);
 977
 978             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 979
 980              /* Update vectorial force */
 981             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 982             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 983             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 984
 985             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 986             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 987             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 988
 989             }
 990
 991             /**************************
 992              * CALCULATE INTERACTIONS *
 993              **************************/
 994
 995             if (gmx_mm_any_lt(rsq10,rcutoff2))
 996             {
 997
 998             /* Compute parameters for interactions between i and j atoms */
 999             qq10             = _mm_mul_ps(iq1,jq0);
1000
1001             /* REACTION-FIELD ELECTROSTATICS */
1002             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1003
1004             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1005
1006             fscal            = felec;
1007
1008             fscal            = _mm_and_ps(fscal,cutoff_mask);
1009
1010             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1011
1012              /* Update vectorial force */
1013             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1014             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1015             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1016
1017             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1018             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1019             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1020
1021             }
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             if (gmx_mm_any_lt(rsq20,rcutoff2))
1028             {
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq20             = _mm_mul_ps(iq2,jq0);
1032
1033             /* REACTION-FIELD ELECTROSTATICS */
1034             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1035
1036             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1037
1038             fscal            = felec;
1039
1040             fscal            = _mm_and_ps(fscal,cutoff_mask);
1041
1042             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1043
1044              /* Update vectorial force */
1045             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1046             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1047             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1048
1049             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1050             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1051             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1052
1053             }
1054
1055             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1056             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1057             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1058             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1059
1060             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1061
1062             /* Inner loop uses 106 flops */
1063         }
1064
1065         /* End of innermost loop */
1066
1067         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1068                                               f+i_coord_offset,fshift+i_shift_offset);
1069
1070         /* Increment number of inner iterations */
1071         inneriter                  += j_index_end - j_index_start;
1072
1073         /* Outer loop uses 18 flops */
1074     }
1075
1076     /* Increment number of outer iterations */
1077     outeriter        += nri;
1078
1079     /* Update outer/inner flops */
1080
1081     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*106);
1082 }