src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset1;
  70     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  71     int              vdwioffset2;
  72     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  73     int              vdwioffset3;
  74     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     __m128i          ewitab;
  83     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  84     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  85     real             *ewtab;
  86     __m128           dummy_mask,cutoff_mask;
  87     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  88     __m128           one     = _mm_set1_ps(1.0);
  89     __m128           two     = _mm_set1_ps(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_ps(fr->epsfac);
 102     charge           = mdatoms->chargeA;
 103
 104     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 105     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 106     beta2            = _mm_mul_ps(beta,beta);
 107     beta3            = _mm_mul_ps(beta,beta2);
 108     ewtab            = fr->ic->tabq_coul_FDV0;
 109     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 110     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 111
 112     /* Setup water-specific parameters */
 113     inr              = nlist->iinr[0];
 114     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 115     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 116     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 117
 118     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 119     rcutoff_scalar   = fr->rcoulomb;
 120     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 121     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 154                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 155
 156         fix1             = _mm_setzero_ps();
 157         fiy1             = _mm_setzero_ps();
 158         fiz1             = _mm_setzero_ps();
 159         fix2             = _mm_setzero_ps();
 160         fiy2             = _mm_setzero_ps();
 161         fiz2             = _mm_setzero_ps();
 162         fix3             = _mm_setzero_ps();
 163         fiy3             = _mm_setzero_ps();
 164         fiz3             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               x+j_coord_offsetC,x+j_coord_offsetD,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx10             = _mm_sub_ps(ix1,jx0);
 190             dy10             = _mm_sub_ps(iy1,jy0);
 191             dz10             = _mm_sub_ps(iz1,jz0);
 192             dx20             = _mm_sub_ps(ix2,jx0);
 193             dy20             = _mm_sub_ps(iy2,jy0);
 194             dz20             = _mm_sub_ps(iz2,jz0);
 195             dx30             = _mm_sub_ps(ix3,jx0);
 196             dy30             = _mm_sub_ps(iy3,jy0);
 197             dz30             = _mm_sub_ps(iz3,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 201             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 202             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 203
 204             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 205             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 206             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 207
 208             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 209             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 210             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                               charge+jnrC+0,charge+jnrD+0);
 215
 216             fjx0             = _mm_setzero_ps();
 217             fjy0             = _mm_setzero_ps();
 218             fjz0             = _mm_setzero_ps();
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             if (gmx_mm_any_lt(rsq10,rcutoff2))
 225             {
 226
 227             r10              = _mm_mul_ps(rsq10,rinv10);
 228
 229             /* Compute parameters for interactions between i and j atoms */
 230             qq10             = _mm_mul_ps(iq1,jq0);
 231
 232             /* EWALD ELECTROSTATICS */
 233
 234             /* Analytical PME correction */
 235             zeta2            = _mm_mul_ps(beta2,rsq10);
 236             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 237             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 238             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 239             felec            = _mm_mul_ps(qq10,felec);
 240             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 241             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 242             velec            = _mm_mul_ps(qq10,velec);
 243
 244             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _mm_and_ps(velec,cutoff_mask);
 248             velecsum         = _mm_add_ps(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             fscal            = _mm_and_ps(fscal,cutoff_mask);
 253
 254              /* Update vectorial force */
 255             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 256             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 257             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 258
 259             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 260             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 261             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 262
 263             }
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             if (gmx_mm_any_lt(rsq20,rcutoff2))
 270             {
 271
 272             r20              = _mm_mul_ps(rsq20,rinv20);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq20             = _mm_mul_ps(iq2,jq0);
 276
 277             /* EWALD ELECTROSTATICS */
 278
 279             /* Analytical PME correction */
 280             zeta2            = _mm_mul_ps(beta2,rsq20);
 281             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 282             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 283             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 284             felec            = _mm_mul_ps(qq20,felec);
 285             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 286             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 287             velec            = _mm_mul_ps(qq20,velec);
 288
 289             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             velec            = _mm_and_ps(velec,cutoff_mask);
 293             velecsum         = _mm_add_ps(velecsum,velec);
 294
 295             fscal            = felec;
 296
 297             fscal            = _mm_and_ps(fscal,cutoff_mask);
 298
 299              /* Update vectorial force */
 300             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 301             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 302             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 303
 304             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 305             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 306             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 307
 308             }
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             if (gmx_mm_any_lt(rsq30,rcutoff2))
 315             {
 316
 317             r30              = _mm_mul_ps(rsq30,rinv30);
 318
 319             /* Compute parameters for interactions between i and j atoms */
 320             qq30             = _mm_mul_ps(iq3,jq0);
 321
 322             /* EWALD ELECTROSTATICS */
 323
 324             /* Analytical PME correction */
 325             zeta2            = _mm_mul_ps(beta2,rsq30);
 326             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 327             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 328             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 329             felec            = _mm_mul_ps(qq30,felec);
 330             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 331             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 332             velec            = _mm_mul_ps(qq30,velec);
 333
 334             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velec            = _mm_and_ps(velec,cutoff_mask);
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342             fscal            = _mm_and_ps(fscal,cutoff_mask);
 343
 344              /* Update vectorial force */
 345             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 346             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 347             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 348
 349             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 350             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 351             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 352
 353             }
 354
 355             fjptrA             = f+j_coord_offsetA;
 356             fjptrB             = f+j_coord_offsetB;
 357             fjptrC             = f+j_coord_offsetC;
 358             fjptrD             = f+j_coord_offsetD;
 359
 360             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 361
 362             /* Inner loop uses 99 flops */
 363         }
 364
 365         if(jidx<j_index_end)
 366         {
 367
 368             /* Get j neighbor index, and coordinate index */
 369             jnrlistA         = jjnr[jidx];
 370             jnrlistB         = jjnr[jidx+1];
 371             jnrlistC         = jjnr[jidx+2];
 372             jnrlistD         = jjnr[jidx+3];
 373             /* Sign of each element will be negative for non-real atoms.
 374              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 375              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 376              */
 377             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 378             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 379             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 380             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 381             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 382             j_coord_offsetA  = DIM*jnrA;
 383             j_coord_offsetB  = DIM*jnrB;
 384             j_coord_offsetC  = DIM*jnrC;
 385             j_coord_offsetD  = DIM*jnrD;
 386
 387             /* load j atom coordinates */
 388             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 389                                               x+j_coord_offsetC,x+j_coord_offsetD,
 390                                               &jx0,&jy0,&jz0);
 391
 392             /* Calculate displacement vector */
 393             dx10             = _mm_sub_ps(ix1,jx0);
 394             dy10             = _mm_sub_ps(iy1,jy0);
 395             dz10             = _mm_sub_ps(iz1,jz0);
 396             dx20             = _mm_sub_ps(ix2,jx0);
 397             dy20             = _mm_sub_ps(iy2,jy0);
 398             dz20             = _mm_sub_ps(iz2,jz0);
 399             dx30             = _mm_sub_ps(ix3,jx0);
 400             dy30             = _mm_sub_ps(iy3,jy0);
 401             dz30             = _mm_sub_ps(iz3,jz0);
 402
 403             /* Calculate squared distance and things based on it */
 404             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 405             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 406             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 407
 408             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 409             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 410             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 411
 412             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 413             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 414             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 415
 416             /* Load parameters for j particles */
 417             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 418                                                               charge+jnrC+0,charge+jnrD+0);
 419
 420             fjx0             = _mm_setzero_ps();
 421             fjy0             = _mm_setzero_ps();
 422             fjz0             = _mm_setzero_ps();
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             if (gmx_mm_any_lt(rsq10,rcutoff2))
 429             {
 430
 431             r10              = _mm_mul_ps(rsq10,rinv10);
 432             r10              = _mm_andnot_ps(dummy_mask,r10);
 433
 434             /* Compute parameters for interactions between i and j atoms */
 435             qq10             = _mm_mul_ps(iq1,jq0);
 436
 437             /* EWALD ELECTROSTATICS */
 438
 439             /* Analytical PME correction */
 440             zeta2            = _mm_mul_ps(beta2,rsq10);
 441             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 442             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 443             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 444             felec            = _mm_mul_ps(qq10,felec);
 445             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 446             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv10,sh_ewald));
 447             velec            = _mm_mul_ps(qq10,velec);
 448
 449             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 450
 451             /* Update potential sum for this i atom from the interaction with this j atom. */
 452             velec            = _mm_and_ps(velec,cutoff_mask);
 453             velec            = _mm_andnot_ps(dummy_mask,velec);
 454             velecsum         = _mm_add_ps(velecsum,velec);
 455
 456             fscal            = felec;
 457
 458             fscal            = _mm_and_ps(fscal,cutoff_mask);
 459
 460             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 461
 462              /* Update vectorial force */
 463             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 464             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 465             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 466
 467             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 468             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 469             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 470
 471             }
 472
 473             /**************************
 474              * CALCULATE INTERACTIONS *
 475              **************************/
 476
 477             if (gmx_mm_any_lt(rsq20,rcutoff2))
 478             {
 479
 480             r20              = _mm_mul_ps(rsq20,rinv20);
 481             r20              = _mm_andnot_ps(dummy_mask,r20);
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             qq20             = _mm_mul_ps(iq2,jq0);
 485
 486             /* EWALD ELECTROSTATICS */
 487
 488             /* Analytical PME correction */
 489             zeta2            = _mm_mul_ps(beta2,rsq20);
 490             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 491             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 492             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 493             felec            = _mm_mul_ps(qq20,felec);
 494             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 495             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv20,sh_ewald));
 496             velec            = _mm_mul_ps(qq20,velec);
 497
 498             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             velec            = _mm_and_ps(velec,cutoff_mask);
 502             velec            = _mm_andnot_ps(dummy_mask,velec);
 503             velecsum         = _mm_add_ps(velecsum,velec);
 504
 505             fscal            = felec;
 506
 507             fscal            = _mm_and_ps(fscal,cutoff_mask);
 508
 509             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 510
 511              /* Update vectorial force */
 512             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 513             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 514             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 515
 516             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 517             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 518             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 519
 520             }
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             if (gmx_mm_any_lt(rsq30,rcutoff2))
 527             {
 528
 529             r30              = _mm_mul_ps(rsq30,rinv30);
 530             r30              = _mm_andnot_ps(dummy_mask,r30);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq30             = _mm_mul_ps(iq3,jq0);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Analytical PME correction */
 538             zeta2            = _mm_mul_ps(beta2,rsq30);
 539             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 540             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 541             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 542             felec            = _mm_mul_ps(qq30,felec);
 543             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 544             velec            = _mm_nmacc_ps(pmecorrV,beta,_mm_sub_ps(rinv30,sh_ewald));
 545             velec            = _mm_mul_ps(qq30,velec);
 546
 547             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_and_ps(velec,cutoff_mask);
 551             velec            = _mm_andnot_ps(dummy_mask,velec);
 552             velecsum         = _mm_add_ps(velecsum,velec);
 553
 554             fscal            = felec;
 555
 556             fscal            = _mm_and_ps(fscal,cutoff_mask);
 557
 558             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 559
 560              /* Update vectorial force */
 561             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 562             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 563             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 564
 565             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 566             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 567             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 568
 569             }
 570
 571             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 572             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 573             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 574             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 575
 576             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 577
 578             /* Inner loop uses 102 flops */
 579         }
 580
 581         /* End of innermost loop */
 582
 583         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 584                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 585
 586         ggid                        = gid[iidx];
 587         /* Update potential energies */
 588         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 589
 590         /* Increment number of inner iterations */
 591         inneriter                  += j_index_end - j_index_start;
 592
 593         /* Outer loop uses 19 flops */
 594     }
 595
 596     /* Increment number of outer iterations */
 597     outeriter        += nri;
 598
 599     /* Update outer/inner flops */
 600
 601     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*102);
 602 }
 603 /*
 604  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_128_fma_single
 605  * Electrostatics interaction: Ewald
 606  * VdW interaction:            None
 607  * Geometry:                   Water4-Particle
 608  * Calculate force/pot:        Force
 609  */
 610 void
 611 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_avx_128_fma_single
 612                     (t_nblist * gmx_restrict                nlist,
 613                      rvec * gmx_restrict                    xx,
 614                      rvec * gmx_restrict                    ff,
 615                      t_forcerec * gmx_restrict              fr,
 616                      t_mdatoms * gmx_restrict               mdatoms,
 617                      nb_kernel_data_t * gmx_restrict        kernel_data,
 618                      t_nrnb * gmx_restrict                  nrnb)
 619 {
 620     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 621      * just 0 for non-waters.
 622      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 623      * jnr indices corresponding to data put in the four positions in the SIMD register.
 624      */
 625     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 626     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 627     int              jnrA,jnrB,jnrC,jnrD;
 628     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 629     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 630     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 631     real             rcutoff_scalar;
 632     real             *shiftvec,*fshift,*x,*f;
 633     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 634     real             scratch[4*DIM];
 635     __m128           fscal,rcutoff,rcutoff2,jidxall;
 636     int              vdwioffset1;
 637     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwioffset3;
 641     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 642     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 643     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 644     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 645     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 646     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 647     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 648     real             *charge;
 649     __m128i          ewitab;
 650     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 651     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 652     real             *ewtab;
 653     __m128           dummy_mask,cutoff_mask;
 654     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 655     __m128           one     = _mm_set1_ps(1.0);
 656     __m128           two     = _mm_set1_ps(2.0);
 657     x                = xx[0];
 658     f                = ff[0];
 659
 660     nri              = nlist->nri;
 661     iinr             = nlist->iinr;
 662     jindex           = nlist->jindex;
 663     jjnr             = nlist->jjnr;
 664     shiftidx         = nlist->shift;
 665     gid              = nlist->gid;
 666     shiftvec         = fr->shift_vec[0];
 667     fshift           = fr->fshift[0];
 668     facel            = _mm_set1_ps(fr->epsfac);
 669     charge           = mdatoms->chargeA;
 670
 671     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 672     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 673     beta2            = _mm_mul_ps(beta,beta);
 674     beta3            = _mm_mul_ps(beta,beta2);
 675     ewtab            = fr->ic->tabq_coul_F;
 676     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 677     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 678
 679     /* Setup water-specific parameters */
 680     inr              = nlist->iinr[0];
 681     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 682     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 683     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 684
 685     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 686     rcutoff_scalar   = fr->rcoulomb;
 687     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 688     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 689
 690     /* Avoid stupid compiler warnings */
 691     jnrA = jnrB = jnrC = jnrD = 0;
 692     j_coord_offsetA = 0;
 693     j_coord_offsetB = 0;
 694     j_coord_offsetC = 0;
 695     j_coord_offsetD = 0;
 696
 697     outeriter        = 0;
 698     inneriter        = 0;
 699
 700     for(iidx=0;iidx<4*DIM;iidx++)
 701     {
 702         scratch[iidx] = 0.0;
 703     }
 704
 705     /* Start outer loop over neighborlists */
 706     for(iidx=0; iidx<nri; iidx++)
 707     {
 708         /* Load shift vector for this list */
 709         i_shift_offset   = DIM*shiftidx[iidx];
 710
 711         /* Load limits for loop over neighbors */
 712         j_index_start    = jindex[iidx];
 713         j_index_end      = jindex[iidx+1];
 714
 715         /* Get outer coordinate index */
 716         inr              = iinr[iidx];
 717         i_coord_offset   = DIM*inr;
 718
 719         /* Load i particle coords and add shift vector */
 720         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 721                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 722
 723         fix1             = _mm_setzero_ps();
 724         fiy1             = _mm_setzero_ps();
 725         fiz1             = _mm_setzero_ps();
 726         fix2             = _mm_setzero_ps();
 727         fiy2             = _mm_setzero_ps();
 728         fiz2             = _mm_setzero_ps();
 729         fix3             = _mm_setzero_ps();
 730         fiy3             = _mm_setzero_ps();
 731         fiz3             = _mm_setzero_ps();
 732
 733         /* Start inner kernel loop */
 734         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 735         {
 736
 737             /* Get j neighbor index, and coordinate index */
 738             jnrA             = jjnr[jidx];
 739             jnrB             = jjnr[jidx+1];
 740             jnrC             = jjnr[jidx+2];
 741             jnrD             = jjnr[jidx+3];
 742             j_coord_offsetA  = DIM*jnrA;
 743             j_coord_offsetB  = DIM*jnrB;
 744             j_coord_offsetC  = DIM*jnrC;
 745             j_coord_offsetD  = DIM*jnrD;
 746
 747             /* load j atom coordinates */
 748             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 749                                               x+j_coord_offsetC,x+j_coord_offsetD,
 750                                               &jx0,&jy0,&jz0);
 751
 752             /* Calculate displacement vector */
 753             dx10             = _mm_sub_ps(ix1,jx0);
 754             dy10             = _mm_sub_ps(iy1,jy0);
 755             dz10             = _mm_sub_ps(iz1,jz0);
 756             dx20             = _mm_sub_ps(ix2,jx0);
 757             dy20             = _mm_sub_ps(iy2,jy0);
 758             dz20             = _mm_sub_ps(iz2,jz0);
 759             dx30             = _mm_sub_ps(ix3,jx0);
 760             dy30             = _mm_sub_ps(iy3,jy0);
 761             dz30             = _mm_sub_ps(iz3,jz0);
 762
 763             /* Calculate squared distance and things based on it */
 764             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 765             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 766             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 767
 768             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 769             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 770             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 771
 772             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 773             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 774             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 775
 776             /* Load parameters for j particles */
 777             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 778                                                               charge+jnrC+0,charge+jnrD+0);
 779
 780             fjx0             = _mm_setzero_ps();
 781             fjy0             = _mm_setzero_ps();
 782             fjz0             = _mm_setzero_ps();
 783
 784             /**************************
 785              * CALCULATE INTERACTIONS *
 786              **************************/
 787
 788             if (gmx_mm_any_lt(rsq10,rcutoff2))
 789             {
 790
 791             r10              = _mm_mul_ps(rsq10,rinv10);
 792
 793             /* Compute parameters for interactions between i and j atoms */
 794             qq10             = _mm_mul_ps(iq1,jq0);
 795
 796             /* EWALD ELECTROSTATICS */
 797
 798             /* Analytical PME correction */
 799             zeta2            = _mm_mul_ps(beta2,rsq10);
 800             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 801             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 802             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 803             felec            = _mm_mul_ps(qq10,felec);
 804
 805             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 806
 807             fscal            = felec;
 808
 809             fscal            = _mm_and_ps(fscal,cutoff_mask);
 810
 811              /* Update vectorial force */
 812             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 813             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 814             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 815
 816             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 817             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 818             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 819
 820             }
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             if (gmx_mm_any_lt(rsq20,rcutoff2))
 827             {
 828
 829             r20              = _mm_mul_ps(rsq20,rinv20);
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq20             = _mm_mul_ps(iq2,jq0);
 833
 834             /* EWALD ELECTROSTATICS */
 835
 836             /* Analytical PME correction */
 837             zeta2            = _mm_mul_ps(beta2,rsq20);
 838             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 839             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 840             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 841             felec            = _mm_mul_ps(qq20,felec);
 842
 843             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 844
 845             fscal            = felec;
 846
 847             fscal            = _mm_and_ps(fscal,cutoff_mask);
 848
 849              /* Update vectorial force */
 850             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 851             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 852             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 853
 854             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 855             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 856             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 857
 858             }
 859
 860             /**************************
 861              * CALCULATE INTERACTIONS *
 862              **************************/
 863
 864             if (gmx_mm_any_lt(rsq30,rcutoff2))
 865             {
 866
 867             r30              = _mm_mul_ps(rsq30,rinv30);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq30             = _mm_mul_ps(iq3,jq0);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Analytical PME correction */
 875             zeta2            = _mm_mul_ps(beta2,rsq30);
 876             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
 877             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 878             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 879             felec            = _mm_mul_ps(qq30,felec);
 880
 881             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 882
 883             fscal            = felec;
 884
 885             fscal            = _mm_and_ps(fscal,cutoff_mask);
 886
 887              /* Update vectorial force */
 888             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 889             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 890             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 891
 892             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 893             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 894             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 895
 896             }
 897
 898             fjptrA             = f+j_coord_offsetA;
 899             fjptrB             = f+j_coord_offsetB;
 900             fjptrC             = f+j_coord_offsetC;
 901             fjptrD             = f+j_coord_offsetD;
 902
 903             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 904
 905             /* Inner loop uses 93 flops */
 906         }
 907
 908         if(jidx<j_index_end)
 909         {
 910
 911             /* Get j neighbor index, and coordinate index */
 912             jnrlistA         = jjnr[jidx];
 913             jnrlistB         = jjnr[jidx+1];
 914             jnrlistC         = jjnr[jidx+2];
 915             jnrlistD         = jjnr[jidx+3];
 916             /* Sign of each element will be negative for non-real atoms.
 917              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 918              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 919              */
 920             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 921             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 922             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 923             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 924             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 925             j_coord_offsetA  = DIM*jnrA;
 926             j_coord_offsetB  = DIM*jnrB;
 927             j_coord_offsetC  = DIM*jnrC;
 928             j_coord_offsetD  = DIM*jnrD;
 929
 930             /* load j atom coordinates */
 931             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 932                                               x+j_coord_offsetC,x+j_coord_offsetD,
 933                                               &jx0,&jy0,&jz0);
 934
 935             /* Calculate displacement vector */
 936             dx10             = _mm_sub_ps(ix1,jx0);
 937             dy10             = _mm_sub_ps(iy1,jy0);
 938             dz10             = _mm_sub_ps(iz1,jz0);
 939             dx20             = _mm_sub_ps(ix2,jx0);
 940             dy20             = _mm_sub_ps(iy2,jy0);
 941             dz20             = _mm_sub_ps(iz2,jz0);
 942             dx30             = _mm_sub_ps(ix3,jx0);
 943             dy30             = _mm_sub_ps(iy3,jy0);
 944             dz30             = _mm_sub_ps(iz3,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 948             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 949             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 950
 951             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 952             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 953             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 954
 955             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 956             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 957             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 958
 959             /* Load parameters for j particles */
 960             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 961                                                               charge+jnrC+0,charge+jnrD+0);
 962
 963             fjx0             = _mm_setzero_ps();
 964             fjy0             = _mm_setzero_ps();
 965             fjz0             = _mm_setzero_ps();
 966
 967             /**************************
 968              * CALCULATE INTERACTIONS *
 969              **************************/
 970
 971             if (gmx_mm_any_lt(rsq10,rcutoff2))
 972             {
 973
 974             r10              = _mm_mul_ps(rsq10,rinv10);
 975             r10              = _mm_andnot_ps(dummy_mask,r10);
 976
 977             /* Compute parameters for interactions between i and j atoms */
 978             qq10             = _mm_mul_ps(iq1,jq0);
 979
 980             /* EWALD ELECTROSTATICS */
 981
 982             /* Analytical PME correction */
 983             zeta2            = _mm_mul_ps(beta2,rsq10);
 984             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 985             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 986             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 987             felec            = _mm_mul_ps(qq10,felec);
 988
 989             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 990
 991             fscal            = felec;
 992
 993             fscal            = _mm_and_ps(fscal,cutoff_mask);
 994
 995             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 996
 997              /* Update vectorial force */
 998             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 999             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1000             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1001
1002             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1003             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1004             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1005
1006             }
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm_any_lt(rsq20,rcutoff2))
1013             {
1014
1015             r20              = _mm_mul_ps(rsq20,rinv20);
1016             r20              = _mm_andnot_ps(dummy_mask,r20);
1017
1018             /* Compute parameters for interactions between i and j atoms */
1019             qq20             = _mm_mul_ps(iq2,jq0);
1020
1021             /* EWALD ELECTROSTATICS */
1022
1023             /* Analytical PME correction */
1024             zeta2            = _mm_mul_ps(beta2,rsq20);
1025             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1026             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1027             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1028             felec            = _mm_mul_ps(qq20,felec);
1029
1030             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm_and_ps(fscal,cutoff_mask);
1035
1036             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1037
1038              /* Update vectorial force */
1039             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1040             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1041             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1042
1043             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1044             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1045             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1046
1047             }
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             if (gmx_mm_any_lt(rsq30,rcutoff2))
1054             {
1055
1056             r30              = _mm_mul_ps(rsq30,rinv30);
1057             r30              = _mm_andnot_ps(dummy_mask,r30);
1058
1059             /* Compute parameters for interactions between i and j atoms */
1060             qq30             = _mm_mul_ps(iq3,jq0);
1061
1062             /* EWALD ELECTROSTATICS */
1063
1064             /* Analytical PME correction */
1065             zeta2            = _mm_mul_ps(beta2,rsq30);
1066             rinv3            = _mm_mul_ps(rinvsq30,rinv30);
1067             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1068             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1069             felec            = _mm_mul_ps(qq30,felec);
1070
1071             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1072
1073             fscal            = felec;
1074
1075             fscal            = _mm_and_ps(fscal,cutoff_mask);
1076
1077             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1078
1079              /* Update vectorial force */
1080             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1081             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1082             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1083
1084             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1085             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1086             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1087
1088             }
1089
1090             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1091             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1092             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1093             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1094
1095             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1096
1097             /* Inner loop uses 96 flops */
1098         }
1099
1100         /* End of innermost loop */
1101
1102         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1103                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1104
1105         /* Increment number of inner iterations */
1106         inneriter                  += j_index_end - j_index_start;
1107
1108         /* Outer loop uses 18 flops */
1109     }
1110
1111     /* Increment number of outer iterations */
1112     outeriter        += nri;
1113
1114     /* Update outer/inner flops */
1115
1116     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*96);
1117 }