src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwLJ_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_ElecEw_VdwLJ_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_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     __m128i          ewitab;
  89     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  91     real             *ewtab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 114     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 115     beta2            = _mm_mul_ps(beta,beta);
 116     beta3            = _mm_mul_ps(beta,beta2);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164         fix1             = _mm_setzero_ps();
 165         fiy1             = _mm_setzero_ps();
 166         fiz1             = _mm_setzero_ps();
 167         fix2             = _mm_setzero_ps();
 168         fiy2             = _mm_setzero_ps();
 169         fiz2             = _mm_setzero_ps();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173         vvdwsum          = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184             j_coord_offsetA  = DIM*jnrA;
 185             j_coord_offsetB  = DIM*jnrB;
 186             j_coord_offsetC  = DIM*jnrC;
 187             j_coord_offsetD  = DIM*jnrD;
 188
 189             /* load j atom coordinates */
 190             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 191                                               x+j_coord_offsetC,x+j_coord_offsetD,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_ps(ix0,jx0);
 196             dy00             = _mm_sub_ps(iy0,jy0);
 197             dz00             = _mm_sub_ps(iz0,jz0);
 198             dx10             = _mm_sub_ps(ix1,jx0);
 199             dy10             = _mm_sub_ps(iy1,jy0);
 200             dz10             = _mm_sub_ps(iz1,jz0);
 201             dx20             = _mm_sub_ps(ix2,jx0);
 202             dy20             = _mm_sub_ps(iy2,jy0);
 203             dz20             = _mm_sub_ps(iz2,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
 210             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 211             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 212             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 213
 214             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 215             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 216             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 217
 218             /* Load parameters for j particles */
 219             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 220                                                               charge+jnrC+0,charge+jnrD+0);
 221             vdwjidx0A        = 2*vdwtype[jnrA+0];
 222             vdwjidx0B        = 2*vdwtype[jnrB+0];
 223             vdwjidx0C        = 2*vdwtype[jnrC+0];
 224             vdwjidx0D        = 2*vdwtype[jnrD+0];
 225
 226             fjx0             = _mm_setzero_ps();
 227             fjy0             = _mm_setzero_ps();
 228             fjz0             = _mm_setzero_ps();
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             r00              = _mm_mul_ps(rsq00,rinv00);
 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             /* EWALD ELECTROSTATICS */
 245
 246             /* Analytical PME correction */
 247             zeta2            = _mm_mul_ps(beta2,rsq00);
 248             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 249             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 250             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 251             felec            = _mm_mul_ps(qq00,felec);
 252             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 253             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 254             velec            = _mm_mul_ps(qq00,velec);
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 261             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 262             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 267
 268             fscal            = _mm_add_ps(felec,fvdw);
 269
 270              /* Update vectorial force */
 271             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 272             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 273             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 274
 275             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 276             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 277             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 278
 279             /**************************
 280              * CALCULATE INTERACTIONS *
 281              **************************/
 282
 283             r10              = _mm_mul_ps(rsq10,rinv10);
 284
 285             /* Compute parameters for interactions between i and j atoms */
 286             qq10             = _mm_mul_ps(iq1,jq0);
 287
 288             /* EWALD ELECTROSTATICS */
 289
 290             /* Analytical PME correction */
 291             zeta2            = _mm_mul_ps(beta2,rsq10);
 292             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 293             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 294             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 295             felec            = _mm_mul_ps(qq10,felec);
 296             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 297             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 298             velec            = _mm_mul_ps(qq10,velec);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velecsum         = _mm_add_ps(velecsum,velec);
 302
 303             fscal            = felec;
 304
 305              /* Update vectorial force */
 306             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 307             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 308             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 309
 310             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 311             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 312             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r20              = _mm_mul_ps(rsq20,rinv20);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq20             = _mm_mul_ps(iq2,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Analytical PME correction */
 326             zeta2            = _mm_mul_ps(beta2,rsq20);
 327             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 328             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 329             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 330             felec            = _mm_mul_ps(qq20,felec);
 331             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 332             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 333             velec            = _mm_mul_ps(qq20,velec);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340              /* Update vectorial force */
 341             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 342             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 343             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 344
 345             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 346             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 347             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 348
 349             fjptrA             = f+j_coord_offsetA;
 350             fjptrB             = f+j_coord_offsetB;
 351             fjptrC             = f+j_coord_offsetC;
 352             fjptrD             = f+j_coord_offsetD;
 353
 354             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 355
 356             /* Inner loop uses 99 flops */
 357         }
 358
 359         if(jidx<j_index_end)
 360         {
 361
 362             /* Get j neighbor index, and coordinate index */
 363             jnrlistA         = jjnr[jidx];
 364             jnrlistB         = jjnr[jidx+1];
 365             jnrlistC         = jjnr[jidx+2];
 366             jnrlistD         = jjnr[jidx+3];
 367             /* Sign of each element will be negative for non-real atoms.
 368              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 369              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 370              */
 371             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 372             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 373             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 374             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 375             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 376             j_coord_offsetA  = DIM*jnrA;
 377             j_coord_offsetB  = DIM*jnrB;
 378             j_coord_offsetC  = DIM*jnrC;
 379             j_coord_offsetD  = DIM*jnrD;
 380
 381             /* load j atom coordinates */
 382             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 383                                               x+j_coord_offsetC,x+j_coord_offsetD,
 384                                               &jx0,&jy0,&jz0);
 385
 386             /* Calculate displacement vector */
 387             dx00             = _mm_sub_ps(ix0,jx0);
 388             dy00             = _mm_sub_ps(iy0,jy0);
 389             dz00             = _mm_sub_ps(iz0,jz0);
 390             dx10             = _mm_sub_ps(ix1,jx0);
 391             dy10             = _mm_sub_ps(iy1,jy0);
 392             dz10             = _mm_sub_ps(iz1,jz0);
 393             dx20             = _mm_sub_ps(ix2,jx0);
 394             dy20             = _mm_sub_ps(iy2,jy0);
 395             dz20             = _mm_sub_ps(iz2,jz0);
 396
 397             /* Calculate squared distance and things based on it */
 398             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 399             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 400             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 401
 402             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 403             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 404             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 405
 406             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 407             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 408             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 409
 410             /* Load parameters for j particles */
 411             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 412                                                               charge+jnrC+0,charge+jnrD+0);
 413             vdwjidx0A        = 2*vdwtype[jnrA+0];
 414             vdwjidx0B        = 2*vdwtype[jnrB+0];
 415             vdwjidx0C        = 2*vdwtype[jnrC+0];
 416             vdwjidx0D        = 2*vdwtype[jnrD+0];
 417
 418             fjx0             = _mm_setzero_ps();
 419             fjy0             = _mm_setzero_ps();
 420             fjz0             = _mm_setzero_ps();
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r00              = _mm_mul_ps(rsq00,rinv00);
 427             r00              = _mm_andnot_ps(dummy_mask,r00);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq00             = _mm_mul_ps(iq0,jq0);
 431             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 432                                          vdwparam+vdwioffset0+vdwjidx0B,
 433                                          vdwparam+vdwioffset0+vdwjidx0C,
 434                                          vdwparam+vdwioffset0+vdwjidx0D,
 435                                          &c6_00,&c12_00);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Analytical PME correction */
 440             zeta2            = _mm_mul_ps(beta2,rsq00);
 441             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 442             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 443             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 444             felec            = _mm_mul_ps(qq00,felec);
 445             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 446             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 447             velec            = _mm_mul_ps(qq00,velec);
 448
 449             /* LENNARD-JONES DISPERSION/REPULSION */
 450
 451             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 452             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 453             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 454             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 455             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 456
 457             /* Update potential sum for this i atom from the interaction with this j atom. */
 458             velec            = _mm_andnot_ps(dummy_mask,velec);
 459             velecsum         = _mm_add_ps(velecsum,velec);
 460             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 461             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 462
 463             fscal            = _mm_add_ps(felec,fvdw);
 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              * CALCULATE INTERACTIONS *
 478              **************************/
 479
 480             r10              = _mm_mul_ps(rsq10,rinv10);
 481             r10              = _mm_andnot_ps(dummy_mask,r10);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq10             = _mm_mul_ps(iq1,jq0);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Analytical PME correction */
 489             zeta2            = _mm_mul_ps(beta2,rsq10);
 490             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 491             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 492             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 493             felec            = _mm_mul_ps(qq10,felec);
 494             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 495             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 496             velec            = _mm_mul_ps(qq10,velec);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm_andnot_ps(dummy_mask,velec);
 500             velecsum         = _mm_add_ps(velecsum,velec);
 501
 502             fscal            = felec;
 503
 504             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 505
 506              /* Update vectorial force */
 507             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 508             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 509             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 510
 511             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 512             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 513             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r20              = _mm_mul_ps(rsq20,rinv20);
 520             r20              = _mm_andnot_ps(dummy_mask,r20);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq20             = _mm_mul_ps(iq2,jq0);
 524
 525             /* EWALD ELECTROSTATICS */
 526
 527             /* Analytical PME correction */
 528             zeta2            = _mm_mul_ps(beta2,rsq20);
 529             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 530             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 531             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 532             felec            = _mm_mul_ps(qq20,felec);
 533             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 534             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 535             velec            = _mm_mul_ps(qq20,velec);
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545              /* Update vectorial force */
 546             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 547             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 548             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 549
 550             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 551             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 552             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 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 102 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*102);
 586 }
 587 /*
 588  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_avx_128_fma_single
 589  * Electrostatics interaction: Ewald
 590  * VdW interaction:            LennardJones
 591  * Geometry:                   Water3-Particle
 592  * Calculate force/pot:        Force
 593  */
 594 void
 595 nb_kernel_ElecEw_VdwLJ_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     __m128i          ewitab;
 640     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 641     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 642     real             *ewtab;
 643     __m128           dummy_mask,cutoff_mask;
 644     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 645     __m128           one     = _mm_set1_ps(1.0);
 646     __m128           two     = _mm_set1_ps(2.0);
 647     x                = xx[0];
 648     f                = ff[0];
 649
 650     nri              = nlist->nri;
 651     iinr             = nlist->iinr;
 652     jindex           = nlist->jindex;
 653     jjnr             = nlist->jjnr;
 654     shiftidx         = nlist->shift;
 655     gid              = nlist->gid;
 656     shiftvec         = fr->shift_vec[0];
 657     fshift           = fr->fshift[0];
 658     facel            = _mm_set1_ps(fr->epsfac);
 659     charge           = mdatoms->chargeA;
 660     nvdwtype         = fr->ntype;
 661     vdwparam         = fr->nbfp;
 662     vdwtype          = mdatoms->typeA;
 663
 664     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 665     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 666     beta2            = _mm_mul_ps(beta,beta);
 667     beta3            = _mm_mul_ps(beta,beta2);
 668     ewtab            = fr->ic->tabq_coul_F;
 669     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 670     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 671
 672     /* Setup water-specific parameters */
 673     inr              = nlist->iinr[0];
 674     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 675     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 676     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 677     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 678
 679     /* Avoid stupid compiler warnings */
 680     jnrA = jnrB = jnrC = jnrD = 0;
 681     j_coord_offsetA = 0;
 682     j_coord_offsetB = 0;
 683     j_coord_offsetC = 0;
 684     j_coord_offsetD = 0;
 685
 686     outeriter        = 0;
 687     inneriter        = 0;
 688
 689     for(iidx=0;iidx<4*DIM;iidx++)
 690     {
 691         scratch[iidx] = 0.0;
 692     }
 693
 694     /* Start outer loop over neighborlists */
 695     for(iidx=0; iidx<nri; iidx++)
 696     {
 697         /* Load shift vector for this list */
 698         i_shift_offset   = DIM*shiftidx[iidx];
 699
 700         /* Load limits for loop over neighbors */
 701         j_index_start    = jindex[iidx];
 702         j_index_end      = jindex[iidx+1];
 703
 704         /* Get outer coordinate index */
 705         inr              = iinr[iidx];
 706         i_coord_offset   = DIM*inr;
 707
 708         /* Load i particle coords and add shift vector */
 709         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 710                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 711
 712         fix0             = _mm_setzero_ps();
 713         fiy0             = _mm_setzero_ps();
 714         fiz0             = _mm_setzero_ps();
 715         fix1             = _mm_setzero_ps();
 716         fiy1             = _mm_setzero_ps();
 717         fiz1             = _mm_setzero_ps();
 718         fix2             = _mm_setzero_ps();
 719         fiy2             = _mm_setzero_ps();
 720         fiz2             = _mm_setzero_ps();
 721
 722         /* Start inner kernel loop */
 723         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 724         {
 725
 726             /* Get j neighbor index, and coordinate index */
 727             jnrA             = jjnr[jidx];
 728             jnrB             = jjnr[jidx+1];
 729             jnrC             = jjnr[jidx+2];
 730             jnrD             = jjnr[jidx+3];
 731             j_coord_offsetA  = DIM*jnrA;
 732             j_coord_offsetB  = DIM*jnrB;
 733             j_coord_offsetC  = DIM*jnrC;
 734             j_coord_offsetD  = DIM*jnrD;
 735
 736             /* load j atom coordinates */
 737             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 738                                               x+j_coord_offsetC,x+j_coord_offsetD,
 739                                               &jx0,&jy0,&jz0);
 740
 741             /* Calculate displacement vector */
 742             dx00             = _mm_sub_ps(ix0,jx0);
 743             dy00             = _mm_sub_ps(iy0,jy0);
 744             dz00             = _mm_sub_ps(iz0,jz0);
 745             dx10             = _mm_sub_ps(ix1,jx0);
 746             dy10             = _mm_sub_ps(iy1,jy0);
 747             dz10             = _mm_sub_ps(iz1,jz0);
 748             dx20             = _mm_sub_ps(ix2,jx0);
 749             dy20             = _mm_sub_ps(iy2,jy0);
 750             dz20             = _mm_sub_ps(iz2,jz0);
 751
 752             /* Calculate squared distance and things based on it */
 753             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 754             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 755             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 756
 757             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 758             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 759             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 760
 761             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 762             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 763             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 764
 765             /* Load parameters for j particles */
 766             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 767                                                               charge+jnrC+0,charge+jnrD+0);
 768             vdwjidx0A        = 2*vdwtype[jnrA+0];
 769             vdwjidx0B        = 2*vdwtype[jnrB+0];
 770             vdwjidx0C        = 2*vdwtype[jnrC+0];
 771             vdwjidx0D        = 2*vdwtype[jnrD+0];
 772
 773             fjx0             = _mm_setzero_ps();
 774             fjy0             = _mm_setzero_ps();
 775             fjz0             = _mm_setzero_ps();
 776
 777             /**************************
 778              * CALCULATE INTERACTIONS *
 779              **************************/
 780
 781             r00              = _mm_mul_ps(rsq00,rinv00);
 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             /* EWALD ELECTROSTATICS */
 792
 793             /* Analytical PME correction */
 794             zeta2            = _mm_mul_ps(beta2,rsq00);
 795             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 796             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 797             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 798             felec            = _mm_mul_ps(qq00,felec);
 799
 800             /* LENNARD-JONES DISPERSION/REPULSION */
 801
 802             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 803             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 804
 805             fscal            = _mm_add_ps(felec,fvdw);
 806
 807              /* Update vectorial force */
 808             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 809             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 810             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 811
 812             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 813             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 814             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 815
 816             /**************************
 817              * CALCULATE INTERACTIONS *
 818              **************************/
 819
 820             r10              = _mm_mul_ps(rsq10,rinv10);
 821
 822             /* Compute parameters for interactions between i and j atoms */
 823             qq10             = _mm_mul_ps(iq1,jq0);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Analytical PME correction */
 828             zeta2            = _mm_mul_ps(beta2,rsq10);
 829             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 830             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 831             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 832             felec            = _mm_mul_ps(qq10,felec);
 833
 834             fscal            = felec;
 835
 836              /* Update vectorial force */
 837             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 838             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 839             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 840
 841             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 842             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 843             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             r20              = _mm_mul_ps(rsq20,rinv20);
 850
 851             /* Compute parameters for interactions between i and j atoms */
 852             qq20             = _mm_mul_ps(iq2,jq0);
 853
 854             /* EWALD ELECTROSTATICS */
 855
 856             /* Analytical PME correction */
 857             zeta2            = _mm_mul_ps(beta2,rsq20);
 858             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 859             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 860             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 861             felec            = _mm_mul_ps(qq20,felec);
 862
 863             fscal            = felec;
 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             fjptrA             = f+j_coord_offsetA;
 875             fjptrB             = f+j_coord_offsetB;
 876             fjptrC             = f+j_coord_offsetC;
 877             fjptrD             = f+j_coord_offsetD;
 878
 879             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 880
 881             /* Inner loop uses 91 flops */
 882         }
 883
 884         if(jidx<j_index_end)
 885         {
 886
 887             /* Get j neighbor index, and coordinate index */
 888             jnrlistA         = jjnr[jidx];
 889             jnrlistB         = jjnr[jidx+1];
 890             jnrlistC         = jjnr[jidx+2];
 891             jnrlistD         = jjnr[jidx+3];
 892             /* Sign of each element will be negative for non-real atoms.
 893              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 894              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 895              */
 896             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 897             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 898             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 899             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 900             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 901             j_coord_offsetA  = DIM*jnrA;
 902             j_coord_offsetB  = DIM*jnrB;
 903             j_coord_offsetC  = DIM*jnrC;
 904             j_coord_offsetD  = DIM*jnrD;
 905
 906             /* load j atom coordinates */
 907             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 908                                               x+j_coord_offsetC,x+j_coord_offsetD,
 909                                               &jx0,&jy0,&jz0);
 910
 911             /* Calculate displacement vector */
 912             dx00             = _mm_sub_ps(ix0,jx0);
 913             dy00             = _mm_sub_ps(iy0,jy0);
 914             dz00             = _mm_sub_ps(iz0,jz0);
 915             dx10             = _mm_sub_ps(ix1,jx0);
 916             dy10             = _mm_sub_ps(iy1,jy0);
 917             dz10             = _mm_sub_ps(iz1,jz0);
 918             dx20             = _mm_sub_ps(ix2,jx0);
 919             dy20             = _mm_sub_ps(iy2,jy0);
 920             dz20             = _mm_sub_ps(iz2,jz0);
 921
 922             /* Calculate squared distance and things based on it */
 923             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 924             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 925             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 926
 927             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 928             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 929             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 930
 931             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 932             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 933             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 934
 935             /* Load parameters for j particles */
 936             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 937                                                               charge+jnrC+0,charge+jnrD+0);
 938             vdwjidx0A        = 2*vdwtype[jnrA+0];
 939             vdwjidx0B        = 2*vdwtype[jnrB+0];
 940             vdwjidx0C        = 2*vdwtype[jnrC+0];
 941             vdwjidx0D        = 2*vdwtype[jnrD+0];
 942
 943             fjx0             = _mm_setzero_ps();
 944             fjy0             = _mm_setzero_ps();
 945             fjz0             = _mm_setzero_ps();
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r00              = _mm_mul_ps(rsq00,rinv00);
 952             r00              = _mm_andnot_ps(dummy_mask,r00);
 953
 954             /* Compute parameters for interactions between i and j atoms */
 955             qq00             = _mm_mul_ps(iq0,jq0);
 956             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 957                                          vdwparam+vdwioffset0+vdwjidx0B,
 958                                          vdwparam+vdwioffset0+vdwjidx0C,
 959                                          vdwparam+vdwioffset0+vdwjidx0D,
 960                                          &c6_00,&c12_00);
 961
 962             /* EWALD ELECTROSTATICS */
 963
 964             /* Analytical PME correction */
 965             zeta2            = _mm_mul_ps(beta2,rsq00);
 966             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 967             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 968             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 969             felec            = _mm_mul_ps(qq00,felec);
 970
 971             /* LENNARD-JONES DISPERSION/REPULSION */
 972
 973             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 974             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 975
 976             fscal            = _mm_add_ps(felec,fvdw);
 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              * CALCULATE INTERACTIONS *
 991              **************************/
 992
 993             r10              = _mm_mul_ps(rsq10,rinv10);
 994             r10              = _mm_andnot_ps(dummy_mask,r10);
 995
 996             /* Compute parameters for interactions between i and j atoms */
 997             qq10             = _mm_mul_ps(iq1,jq0);
 998
 999             /* EWALD ELECTROSTATICS */
1000
1001             /* Analytical PME correction */
1002             zeta2            = _mm_mul_ps(beta2,rsq10);
1003             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1004             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1005             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1006             felec            = _mm_mul_ps(qq10,felec);
1007
1008             fscal            = felec;
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              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r20              = _mm_mul_ps(rsq20,rinv20);
1026             r20              = _mm_andnot_ps(dummy_mask,r20);
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq20             = _mm_mul_ps(iq2,jq0);
1030
1031             /* EWALD ELECTROSTATICS */
1032
1033             /* Analytical PME correction */
1034             zeta2            = _mm_mul_ps(beta2,rsq20);
1035             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1036             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1037             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1038             felec            = _mm_mul_ps(qq20,felec);
1039
1040             fscal            = felec;
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             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1054             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1055             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1056             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1057
1058             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1059
1060             /* Inner loop uses 94 flops */
1061         }
1062
1063         /* End of innermost loop */
1064
1065         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1066                                               f+i_coord_offset,fshift+i_shift_offset);
1067
1068         /* Increment number of inner iterations */
1069         inneriter                  += j_index_end - j_index_start;
1070
1071         /* Outer loop uses 18 flops */
1072     }
1073
1074     /* Increment number of outer iterations */
1075     outeriter        += nri;
1076
1077     /* Update outer/inner flops */
1078
1079     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*94);
1080 }