src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_VF_avx_128_fma_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     krf              = _mm_set1_ps(fr->ic->k_rf);
 109     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 110     crf              = _mm_set1_ps(fr->ic->c_rf);
 111     nvdwtype         = fr->ntype;
 112     vdwparam         = fr->nbfp;
 113     vdwtype          = mdatoms->typeA;
 114
 115     /* Setup water-specific parameters */
 116     inr              = nlist->iinr[0];
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 120     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 121
 122     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 123     rcutoff_scalar   = fr->rcoulomb;
 124     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 125     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 126
 127     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 128     rvdw             = _mm_set1_ps(fr->rvdw);
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     for(iidx=0;iidx<4*DIM;iidx++)
 141     {
 142         scratch[iidx] = 0.0;
 143     }
 144
 145     /* Start outer loop over neighborlists */
 146     for(iidx=0; iidx<nri; iidx++)
 147     {
 148         /* Load shift vector for this list */
 149         i_shift_offset   = DIM*shiftidx[iidx];
 150
 151         /* Load limits for loop over neighbors */
 152         j_index_start    = jindex[iidx];
 153         j_index_end      = jindex[iidx+1];
 154
 155         /* Get outer coordinate index */
 156         inr              = iinr[iidx];
 157         i_coord_offset   = DIM*inr;
 158
 159         /* Load i particle coords and add shift vector */
 160         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 161                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 162
 163         fix0             = _mm_setzero_ps();
 164         fiy0             = _mm_setzero_ps();
 165         fiz0             = _mm_setzero_ps();
 166         fix1             = _mm_setzero_ps();
 167         fiy1             = _mm_setzero_ps();
 168         fiz1             = _mm_setzero_ps();
 169         fix2             = _mm_setzero_ps();
 170         fiy2             = _mm_setzero_ps();
 171         fiz2             = _mm_setzero_ps();
 172         fix3             = _mm_setzero_ps();
 173         fiy3             = _mm_setzero_ps();
 174         fiz3             = _mm_setzero_ps();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189             j_coord_offsetA  = DIM*jnrA;
 190             j_coord_offsetB  = DIM*jnrB;
 191             j_coord_offsetC  = DIM*jnrC;
 192             j_coord_offsetD  = DIM*jnrD;
 193
 194             /* load j atom coordinates */
 195             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 196                                               x+j_coord_offsetC,x+j_coord_offsetD,
 197                                               &jx0,&jy0,&jz0);
 198
 199             /* Calculate displacement vector */
 200             dx00             = _mm_sub_ps(ix0,jx0);
 201             dy00             = _mm_sub_ps(iy0,jy0);
 202             dz00             = _mm_sub_ps(iz0,jz0);
 203             dx10             = _mm_sub_ps(ix1,jx0);
 204             dy10             = _mm_sub_ps(iy1,jy0);
 205             dz10             = _mm_sub_ps(iz1,jz0);
 206             dx20             = _mm_sub_ps(ix2,jx0);
 207             dy20             = _mm_sub_ps(iy2,jy0);
 208             dz20             = _mm_sub_ps(iz2,jz0);
 209             dx30             = _mm_sub_ps(ix3,jx0);
 210             dy30             = _mm_sub_ps(iy3,jy0);
 211             dz30             = _mm_sub_ps(iz3,jz0);
 212
 213             /* Calculate squared distance and things based on it */
 214             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 215             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 216             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 217             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 218
 219             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 221             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 222
 223             rinvsq00         = gmx_mm_inv_ps(rsq00);
 224             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 225             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 226             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 227
 228             /* Load parameters for j particles */
 229             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 230                                                               charge+jnrC+0,charge+jnrD+0);
 231             vdwjidx0A        = 2*vdwtype[jnrA+0];
 232             vdwjidx0B        = 2*vdwtype[jnrB+0];
 233             vdwjidx0C        = 2*vdwtype[jnrC+0];
 234             vdwjidx0D        = 2*vdwtype[jnrD+0];
 235
 236             fjx0             = _mm_setzero_ps();
 237             fjy0             = _mm_setzero_ps();
 238             fjz0             = _mm_setzero_ps();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,
 250                                          vdwparam+vdwioffset0+vdwjidx0C,
 251                                          vdwparam+vdwioffset0+vdwjidx0D,
 252                                          &c6_00,&c12_00);
 253
 254             /* LENNARD-JONES DISPERSION/REPULSION */
 255
 256             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 257             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 258             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 259             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 260                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 261             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 262
 263             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = fvdw;
 270
 271             fscal            = _mm_and_ps(fscal,cutoff_mask);
 272
 273              /* Update vectorial force */
 274             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 275             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 276             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 277
 278             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 279             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 280             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 281
 282             }
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm_any_lt(rsq10,rcutoff2))
 289             {
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_ps(iq1,jq0);
 293
 294             /* REACTION-FIELD ELECTROSTATICS */
 295             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 296             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 297
 298             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velec            = _mm_and_ps(velec,cutoff_mask);
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303
 304             fscal            = felec;
 305
 306             fscal            = _mm_and_ps(fscal,cutoff_mask);
 307
 308              /* Update vectorial force */
 309             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 310             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 311             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 312
 313             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 314             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 315             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 316
 317             }
 318
 319             /**************************
 320              * CALCULATE INTERACTIONS *
 321              **************************/
 322
 323             if (gmx_mm_any_lt(rsq20,rcutoff2))
 324             {
 325
 326             /* Compute parameters for interactions between i and j atoms */
 327             qq20             = _mm_mul_ps(iq2,jq0);
 328
 329             /* REACTION-FIELD ELECTROSTATICS */
 330             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 331             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 332
 333             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velec            = _mm_and_ps(velec,cutoff_mask);
 337             velecsum         = _mm_add_ps(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             fscal            = _mm_and_ps(fscal,cutoff_mask);
 342
 343              /* Update vectorial force */
 344             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 345             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 346             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 347
 348             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 349             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 350             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 351
 352             }
 353
 354             /**************************
 355              * CALCULATE INTERACTIONS *
 356              **************************/
 357
 358             if (gmx_mm_any_lt(rsq30,rcutoff2))
 359             {
 360
 361             /* Compute parameters for interactions between i and j atoms */
 362             qq30             = _mm_mul_ps(iq3,jq0);
 363
 364             /* REACTION-FIELD ELECTROSTATICS */
 365             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 366             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 367
 368             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm_and_ps(fscal,cutoff_mask);
 377
 378              /* Update vectorial force */
 379             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 380             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 381             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 382
 383             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 384             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 385             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 386
 387             }
 388
 389             fjptrA             = f+j_coord_offsetA;
 390             fjptrB             = f+j_coord_offsetB;
 391             fjptrC             = f+j_coord_offsetC;
 392             fjptrD             = f+j_coord_offsetD;
 393
 394             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 395
 396             /* Inner loop uses 161 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             /* Get j neighbor index, and coordinate index */
 403             jnrlistA         = jjnr[jidx];
 404             jnrlistB         = jjnr[jidx+1];
 405             jnrlistC         = jjnr[jidx+2];
 406             jnrlistD         = jjnr[jidx+3];
 407             /* Sign of each element will be negative for non-real atoms.
 408              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 409              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 410              */
 411             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 412             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 413             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 414             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 415             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 416             j_coord_offsetA  = DIM*jnrA;
 417             j_coord_offsetB  = DIM*jnrB;
 418             j_coord_offsetC  = DIM*jnrC;
 419             j_coord_offsetD  = DIM*jnrD;
 420
 421             /* load j atom coordinates */
 422             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 423                                               x+j_coord_offsetC,x+j_coord_offsetD,
 424                                               &jx0,&jy0,&jz0);
 425
 426             /* Calculate displacement vector */
 427             dx00             = _mm_sub_ps(ix0,jx0);
 428             dy00             = _mm_sub_ps(iy0,jy0);
 429             dz00             = _mm_sub_ps(iz0,jz0);
 430             dx10             = _mm_sub_ps(ix1,jx0);
 431             dy10             = _mm_sub_ps(iy1,jy0);
 432             dz10             = _mm_sub_ps(iz1,jz0);
 433             dx20             = _mm_sub_ps(ix2,jx0);
 434             dy20             = _mm_sub_ps(iy2,jy0);
 435             dz20             = _mm_sub_ps(iz2,jz0);
 436             dx30             = _mm_sub_ps(ix3,jx0);
 437             dy30             = _mm_sub_ps(iy3,jy0);
 438             dz30             = _mm_sub_ps(iz3,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 442             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 443             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 444             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 445
 446             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 447             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 448             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 449
 450             rinvsq00         = gmx_mm_inv_ps(rsq00);
 451             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 452             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 453             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 454
 455             /* Load parameters for j particles */
 456             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 457                                                               charge+jnrC+0,charge+jnrD+0);
 458             vdwjidx0A        = 2*vdwtype[jnrA+0];
 459             vdwjidx0B        = 2*vdwtype[jnrB+0];
 460             vdwjidx0C        = 2*vdwtype[jnrC+0];
 461             vdwjidx0D        = 2*vdwtype[jnrD+0];
 462
 463             fjx0             = _mm_setzero_ps();
 464             fjy0             = _mm_setzero_ps();
 465             fjz0             = _mm_setzero_ps();
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             if (gmx_mm_any_lt(rsq00,rcutoff2))
 472             {
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 476                                          vdwparam+vdwioffset0+vdwjidx0B,
 477                                          vdwparam+vdwioffset0+vdwjidx0C,
 478                                          vdwparam+vdwioffset0+vdwjidx0D,
 479                                          &c6_00,&c12_00);
 480
 481             /* LENNARD-JONES DISPERSION/REPULSION */
 482
 483             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 484             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 485             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 486             vvdw             = _mm_msub_ps(_mm_nmacc_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 487                                           _mm_mul_ps( _mm_nmacc_ps(c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 488             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 489
 490             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 494             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 495             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 496
 497             fscal            = fvdw;
 498
 499             fscal            = _mm_and_ps(fscal,cutoff_mask);
 500
 501             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 502
 503              /* Update vectorial force */
 504             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 505             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 506             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 507
 508             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 509             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 510             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 511
 512             }
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             if (gmx_mm_any_lt(rsq10,rcutoff2))
 519             {
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq10             = _mm_mul_ps(iq1,jq0);
 523
 524             /* REACTION-FIELD ELECTROSTATICS */
 525             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 526             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 527
 528             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_and_ps(velec,cutoff_mask);
 532             velec            = _mm_andnot_ps(dummy_mask,velec);
 533             velecsum         = _mm_add_ps(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_and_ps(fscal,cutoff_mask);
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541              /* Update vectorial force */
 542             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 543             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 544             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 545
 546             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 547             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 548             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 549
 550             }
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             if (gmx_mm_any_lt(rsq20,rcutoff2))
 557             {
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq20             = _mm_mul_ps(iq2,jq0);
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 564             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 565
 566             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_and_ps(velec,cutoff_mask);
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_and_ps(fscal,cutoff_mask);
 576
 577             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 578
 579              /* Update vectorial force */
 580             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 581             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 582             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 583
 584             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 585             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 586             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 587
 588             }
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             if (gmx_mm_any_lt(rsq30,rcutoff2))
 595             {
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq30             = _mm_mul_ps(iq3,jq0);
 599
 600             /* REACTION-FIELD ELECTROSTATICS */
 601             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_macc_ps(krf,rsq30,rinv30),crf));
 602             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 603
 604             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 605
 606             /* Update potential sum for this i atom from the interaction with this j atom. */
 607             velec            = _mm_and_ps(velec,cutoff_mask);
 608             velec            = _mm_andnot_ps(dummy_mask,velec);
 609             velecsum         = _mm_add_ps(velecsum,velec);
 610
 611             fscal            = felec;
 612
 613             fscal            = _mm_and_ps(fscal,cutoff_mask);
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617              /* Update vectorial force */
 618             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 619             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 620             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 621
 622             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 623             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 624             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 625
 626             }
 627
 628             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 629             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 630             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 631             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 632
 633             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 634
 635             /* Inner loop uses 161 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         ggid                        = gid[iidx];
 644         /* Update potential energies */
 645         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 646         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 26 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*161);
 660 }
 661 /*
 662  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 663  * Electrostatics interaction: ReactionField
 664  * VdW interaction:            LennardJones
 665  * Geometry:                   Water4-Particle
 666  * Calculate force/pot:        Force
 667  */
 668 void
 669 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_avx_128_fma_single
 670                     (t_nblist * gmx_restrict                nlist,
 671                      rvec * gmx_restrict                    xx,
 672                      rvec * gmx_restrict                    ff,
 673                      t_forcerec * gmx_restrict              fr,
 674                      t_mdatoms * gmx_restrict               mdatoms,
 675                      nb_kernel_data_t * gmx_restrict        kernel_data,
 676                      t_nrnb * gmx_restrict                  nrnb)
 677 {
 678     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 679      * just 0 for non-waters.
 680      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 681      * jnr indices corresponding to data put in the four positions in the SIMD register.
 682      */
 683     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 684     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 685     int              jnrA,jnrB,jnrC,jnrD;
 686     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 687     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 688     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 689     real             rcutoff_scalar;
 690     real             *shiftvec,*fshift,*x,*f;
 691     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 692     real             scratch[4*DIM];
 693     __m128           fscal,rcutoff,rcutoff2,jidxall;
 694     int              vdwioffset0;
 695     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 696     int              vdwioffset1;
 697     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 698     int              vdwioffset2;
 699     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 700     int              vdwioffset3;
 701     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 702     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 703     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 704     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 705     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 706     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 707     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 708     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 709     real             *charge;
 710     int              nvdwtype;
 711     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 712     int              *vdwtype;
 713     real             *vdwparam;
 714     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 715     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 716     __m128           dummy_mask,cutoff_mask;
 717     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 718     __m128           one     = _mm_set1_ps(1.0);
 719     __m128           two     = _mm_set1_ps(2.0);
 720     x                = xx[0];
 721     f                = ff[0];
 722
 723     nri              = nlist->nri;
 724     iinr             = nlist->iinr;
 725     jindex           = nlist->jindex;
 726     jjnr             = nlist->jjnr;
 727     shiftidx         = nlist->shift;
 728     gid              = nlist->gid;
 729     shiftvec         = fr->shift_vec[0];
 730     fshift           = fr->fshift[0];
 731     facel            = _mm_set1_ps(fr->epsfac);
 732     charge           = mdatoms->chargeA;
 733     krf              = _mm_set1_ps(fr->ic->k_rf);
 734     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 735     crf              = _mm_set1_ps(fr->ic->c_rf);
 736     nvdwtype         = fr->ntype;
 737     vdwparam         = fr->nbfp;
 738     vdwtype          = mdatoms->typeA;
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 743     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 744     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 745     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 746
 747     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 748     rcutoff_scalar   = fr->rcoulomb;
 749     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 750     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 751
 752     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 753     rvdw             = _mm_set1_ps(fr->rvdw);
 754
 755     /* Avoid stupid compiler warnings */
 756     jnrA = jnrB = jnrC = jnrD = 0;
 757     j_coord_offsetA = 0;
 758     j_coord_offsetB = 0;
 759     j_coord_offsetC = 0;
 760     j_coord_offsetD = 0;
 761
 762     outeriter        = 0;
 763     inneriter        = 0;
 764
 765     for(iidx=0;iidx<4*DIM;iidx++)
 766     {
 767         scratch[iidx] = 0.0;
 768     }
 769
 770     /* Start outer loop over neighborlists */
 771     for(iidx=0; iidx<nri; iidx++)
 772     {
 773         /* Load shift vector for this list */
 774         i_shift_offset   = DIM*shiftidx[iidx];
 775
 776         /* Load limits for loop over neighbors */
 777         j_index_start    = jindex[iidx];
 778         j_index_end      = jindex[iidx+1];
 779
 780         /* Get outer coordinate index */
 781         inr              = iinr[iidx];
 782         i_coord_offset   = DIM*inr;
 783
 784         /* Load i particle coords and add shift vector */
 785         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 786                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 787
 788         fix0             = _mm_setzero_ps();
 789         fiy0             = _mm_setzero_ps();
 790         fiz0             = _mm_setzero_ps();
 791         fix1             = _mm_setzero_ps();
 792         fiy1             = _mm_setzero_ps();
 793         fiz1             = _mm_setzero_ps();
 794         fix2             = _mm_setzero_ps();
 795         fiy2             = _mm_setzero_ps();
 796         fiz2             = _mm_setzero_ps();
 797         fix3             = _mm_setzero_ps();
 798         fiy3             = _mm_setzero_ps();
 799         fiz3             = _mm_setzero_ps();
 800
 801         /* Start inner kernel loop */
 802         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 803         {
 804
 805             /* Get j neighbor index, and coordinate index */
 806             jnrA             = jjnr[jidx];
 807             jnrB             = jjnr[jidx+1];
 808             jnrC             = jjnr[jidx+2];
 809             jnrD             = jjnr[jidx+3];
 810             j_coord_offsetA  = DIM*jnrA;
 811             j_coord_offsetB  = DIM*jnrB;
 812             j_coord_offsetC  = DIM*jnrC;
 813             j_coord_offsetD  = DIM*jnrD;
 814
 815             /* load j atom coordinates */
 816             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 817                                               x+j_coord_offsetC,x+j_coord_offsetD,
 818                                               &jx0,&jy0,&jz0);
 819
 820             /* Calculate displacement vector */
 821             dx00             = _mm_sub_ps(ix0,jx0);
 822             dy00             = _mm_sub_ps(iy0,jy0);
 823             dz00             = _mm_sub_ps(iz0,jz0);
 824             dx10             = _mm_sub_ps(ix1,jx0);
 825             dy10             = _mm_sub_ps(iy1,jy0);
 826             dz10             = _mm_sub_ps(iz1,jz0);
 827             dx20             = _mm_sub_ps(ix2,jx0);
 828             dy20             = _mm_sub_ps(iy2,jy0);
 829             dz20             = _mm_sub_ps(iz2,jz0);
 830             dx30             = _mm_sub_ps(ix3,jx0);
 831             dy30             = _mm_sub_ps(iy3,jy0);
 832             dz30             = _mm_sub_ps(iz3,jz0);
 833
 834             /* Calculate squared distance and things based on it */
 835             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 836             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 837             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 838             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 839
 840             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 841             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 842             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 843
 844             rinvsq00         = gmx_mm_inv_ps(rsq00);
 845             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 846             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 847             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 848
 849             /* Load parameters for j particles */
 850             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 851                                                               charge+jnrC+0,charge+jnrD+0);
 852             vdwjidx0A        = 2*vdwtype[jnrA+0];
 853             vdwjidx0B        = 2*vdwtype[jnrB+0];
 854             vdwjidx0C        = 2*vdwtype[jnrC+0];
 855             vdwjidx0D        = 2*vdwtype[jnrD+0];
 856
 857             fjx0             = _mm_setzero_ps();
 858             fjy0             = _mm_setzero_ps();
 859             fjz0             = _mm_setzero_ps();
 860
 861             /**************************
 862              * CALCULATE INTERACTIONS *
 863              **************************/
 864
 865             if (gmx_mm_any_lt(rsq00,rcutoff2))
 866             {
 867
 868             /* Compute parameters for interactions between i and j atoms */
 869             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 870                                          vdwparam+vdwioffset0+vdwjidx0B,
 871                                          vdwparam+vdwioffset0+vdwjidx0C,
 872                                          vdwparam+vdwioffset0+vdwjidx0D,
 873                                          &c6_00,&c12_00);
 874
 875             /* LENNARD-JONES DISPERSION/REPULSION */
 876
 877             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 878             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 879
 880             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 881
 882             fscal            = fvdw;
 883
 884             fscal            = _mm_and_ps(fscal,cutoff_mask);
 885
 886              /* Update vectorial force */
 887             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 888             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 889             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 890
 891             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 892             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 893             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 894
 895             }
 896
 897             /**************************
 898              * CALCULATE INTERACTIONS *
 899              **************************/
 900
 901             if (gmx_mm_any_lt(rsq10,rcutoff2))
 902             {
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq10             = _mm_mul_ps(iq1,jq0);
 906
 907             /* REACTION-FIELD ELECTROSTATICS */
 908             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 909
 910             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 911
 912             fscal            = felec;
 913
 914             fscal            = _mm_and_ps(fscal,cutoff_mask);
 915
 916              /* Update vectorial force */
 917             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 918             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 919             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 920
 921             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 922             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 923             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 924
 925             }
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             if (gmx_mm_any_lt(rsq20,rcutoff2))
 932             {
 933
 934             /* Compute parameters for interactions between i and j atoms */
 935             qq20             = _mm_mul_ps(iq2,jq0);
 936
 937             /* REACTION-FIELD ELECTROSTATICS */
 938             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 939
 940             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 941
 942             fscal            = felec;
 943
 944             fscal            = _mm_and_ps(fscal,cutoff_mask);
 945
 946              /* Update vectorial force */
 947             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 948             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 949             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 950
 951             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 952             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 953             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 954
 955             }
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             if (gmx_mm_any_lt(rsq30,rcutoff2))
 962             {
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq30             = _mm_mul_ps(iq3,jq0);
 966
 967             /* REACTION-FIELD ELECTROSTATICS */
 968             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
 969
 970             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 971
 972             fscal            = felec;
 973
 974             fscal            = _mm_and_ps(fscal,cutoff_mask);
 975
 976              /* Update vectorial force */
 977             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 978             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 979             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 980
 981             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 982             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 983             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 984
 985             }
 986
 987             fjptrA             = f+j_coord_offsetA;
 988             fjptrB             = f+j_coord_offsetB;
 989             fjptrC             = f+j_coord_offsetC;
 990             fjptrD             = f+j_coord_offsetD;
 991
 992             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 993
 994             /* Inner loop uses 132 flops */
 995         }
 996
 997         if(jidx<j_index_end)
 998         {
 999
1000             /* Get j neighbor index, and coordinate index */
1001             jnrlistA         = jjnr[jidx];
1002             jnrlistB         = jjnr[jidx+1];
1003             jnrlistC         = jjnr[jidx+2];
1004             jnrlistD         = jjnr[jidx+3];
1005             /* Sign of each element will be negative for non-real atoms.
1006              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1007              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1008              */
1009             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1010             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1011             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1012             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1013             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1014             j_coord_offsetA  = DIM*jnrA;
1015             j_coord_offsetB  = DIM*jnrB;
1016             j_coord_offsetC  = DIM*jnrC;
1017             j_coord_offsetD  = DIM*jnrD;
1018
1019             /* load j atom coordinates */
1020             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1021                                               x+j_coord_offsetC,x+j_coord_offsetD,
1022                                               &jx0,&jy0,&jz0);
1023
1024             /* Calculate displacement vector */
1025             dx00             = _mm_sub_ps(ix0,jx0);
1026             dy00             = _mm_sub_ps(iy0,jy0);
1027             dz00             = _mm_sub_ps(iz0,jz0);
1028             dx10             = _mm_sub_ps(ix1,jx0);
1029             dy10             = _mm_sub_ps(iy1,jy0);
1030             dz10             = _mm_sub_ps(iz1,jz0);
1031             dx20             = _mm_sub_ps(ix2,jx0);
1032             dy20             = _mm_sub_ps(iy2,jy0);
1033             dz20             = _mm_sub_ps(iz2,jz0);
1034             dx30             = _mm_sub_ps(ix3,jx0);
1035             dy30             = _mm_sub_ps(iy3,jy0);
1036             dz30             = _mm_sub_ps(iz3,jz0);
1037
1038             /* Calculate squared distance and things based on it */
1039             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1040             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1041             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1042             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1043
1044             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1045             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1046             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1047
1048             rinvsq00         = gmx_mm_inv_ps(rsq00);
1049             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1050             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1051             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1052
1053             /* Load parameters for j particles */
1054             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1055                                                               charge+jnrC+0,charge+jnrD+0);
1056             vdwjidx0A        = 2*vdwtype[jnrA+0];
1057             vdwjidx0B        = 2*vdwtype[jnrB+0];
1058             vdwjidx0C        = 2*vdwtype[jnrC+0];
1059             vdwjidx0D        = 2*vdwtype[jnrD+0];
1060
1061             fjx0             = _mm_setzero_ps();
1062             fjy0             = _mm_setzero_ps();
1063             fjz0             = _mm_setzero_ps();
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm_any_lt(rsq00,rcutoff2))
1070             {
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1074                                          vdwparam+vdwioffset0+vdwjidx0B,
1075                                          vdwparam+vdwioffset0+vdwjidx0C,
1076                                          vdwparam+vdwioffset0+vdwjidx0D,
1077                                          &c6_00,&c12_00);
1078
1079             /* LENNARD-JONES DISPERSION/REPULSION */
1080
1081             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1082             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1083
1084             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1085
1086             fscal            = fvdw;
1087
1088             fscal            = _mm_and_ps(fscal,cutoff_mask);
1089
1090             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1091
1092              /* Update vectorial force */
1093             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1094             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1095             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1096
1097             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1098             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1099             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1100
1101             }
1102
1103             /**************************
1104              * CALCULATE INTERACTIONS *
1105              **************************/
1106
1107             if (gmx_mm_any_lt(rsq10,rcutoff2))
1108             {
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             qq10             = _mm_mul_ps(iq1,jq0);
1112
1113             /* REACTION-FIELD ELECTROSTATICS */
1114             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1115
1116             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1117
1118             fscal            = felec;
1119
1120             fscal            = _mm_and_ps(fscal,cutoff_mask);
1121
1122             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1123
1124              /* Update vectorial force */
1125             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1126             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1127             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1128
1129             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1130             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1131             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1132
1133             }
1134
1135             /**************************
1136              * CALCULATE INTERACTIONS *
1137              **************************/
1138
1139             if (gmx_mm_any_lt(rsq20,rcutoff2))
1140             {
1141
1142             /* Compute parameters for interactions between i and j atoms */
1143             qq20             = _mm_mul_ps(iq2,jq0);
1144
1145             /* REACTION-FIELD ELECTROSTATICS */
1146             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1147
1148             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1149
1150             fscal            = felec;
1151
1152             fscal            = _mm_and_ps(fscal,cutoff_mask);
1153
1154             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1155
1156              /* Update vectorial force */
1157             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1158             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1159             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1160
1161             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1162             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1163             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1164
1165             }
1166
1167             /**************************
1168              * CALCULATE INTERACTIONS *
1169              **************************/
1170
1171             if (gmx_mm_any_lt(rsq30,rcutoff2))
1172             {
1173
1174             /* Compute parameters for interactions between i and j atoms */
1175             qq30             = _mm_mul_ps(iq3,jq0);
1176
1177             /* REACTION-FIELD ELECTROSTATICS */
1178             felec            = _mm_mul_ps(qq30,_mm_msub_ps(rinv30,rinvsq30,krf2));
1179
1180             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1181
1182             fscal            = felec;
1183
1184             fscal            = _mm_and_ps(fscal,cutoff_mask);
1185
1186             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1187
1188              /* Update vectorial force */
1189             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1190             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1191             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1192
1193             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1194             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1195             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1196
1197             }
1198
1199             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1200             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1201             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1202             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1203
1204             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1205
1206             /* Inner loop uses 132 flops */
1207         }
1208
1209         /* End of innermost loop */
1210
1211         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1212                                               f+i_coord_offset,fshift+i_shift_offset);
1213
1214         /* Increment number of inner iterations */
1215         inneriter                  += j_index_end - j_index_start;
1216
1217         /* Outer loop uses 24 flops */
1218     }
1219
1220     /* Increment number of outer iterations */
1221     outeriter        += nri;
1222
1223     /* Update outer/inner flops */
1224
1225     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*132);
1226 }