src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwNone_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double 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_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128d          dummy_mask,cutoff_mask;
  80     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  81     __m128d          one     = _mm_set1_pd(1.0);
  82     __m128d          two     = _mm_set1_pd(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_pd(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm_set1_pd(fr->ic->k_rf);
  97     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_pd(fr->ic->c_rf);
  99
 100     /* Setup water-specific parameters */
 101     inr              = nlist->iinr[0];
 102     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 103     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 104     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 105
 106     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 107     rcutoff_scalar   = fr->rcoulomb;
 108     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 109     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 135                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 136
 137         fix1             = _mm_setzero_pd();
 138         fiy1             = _mm_setzero_pd();
 139         fiz1             = _mm_setzero_pd();
 140         fix2             = _mm_setzero_pd();
 141         fiy2             = _mm_setzero_pd();
 142         fiz2             = _mm_setzero_pd();
 143         fix3             = _mm_setzero_pd();
 144         fiy3             = _mm_setzero_pd();
 145         fiz3             = _mm_setzero_pd();
 146
 147         /* Reset potential sums */
 148         velecsum         = _mm_setzero_pd();
 149
 150         /* Start inner kernel loop */
 151         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 152         {
 153
 154             /* Get j neighbor index, and coordinate index */
 155             jnrA             = jjnr[jidx];
 156             jnrB             = jjnr[jidx+1];
 157             j_coord_offsetA  = DIM*jnrA;
 158             j_coord_offsetB  = DIM*jnrB;
 159
 160             /* load j atom coordinates */
 161             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 162                                               &jx0,&jy0,&jz0);
 163
 164             /* Calculate displacement vector */
 165             dx10             = _mm_sub_pd(ix1,jx0);
 166             dy10             = _mm_sub_pd(iy1,jy0);
 167             dz10             = _mm_sub_pd(iz1,jz0);
 168             dx20             = _mm_sub_pd(ix2,jx0);
 169             dy20             = _mm_sub_pd(iy2,jy0);
 170             dz20             = _mm_sub_pd(iz2,jz0);
 171             dx30             = _mm_sub_pd(ix3,jx0);
 172             dy30             = _mm_sub_pd(iy3,jy0);
 173             dz30             = _mm_sub_pd(iz3,jz0);
 174
 175             /* Calculate squared distance and things based on it */
 176             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 177             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 178             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 179
 180             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 181             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 182             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 183
 184             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 185             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 186             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 187
 188             /* Load parameters for j particles */
 189             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 190
 191             fjx0             = _mm_setzero_pd();
 192             fjy0             = _mm_setzero_pd();
 193             fjz0             = _mm_setzero_pd();
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq10,rcutoff2))
 200             {
 201
 202             /* Compute parameters for interactions between i and j atoms */
 203             qq10             = _mm_mul_pd(iq1,jq0);
 204
 205             /* REACTION-FIELD ELECTROSTATICS */
 206             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 207             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 208
 209             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 210
 211             /* Update potential sum for this i atom from the interaction with this j atom. */
 212             velec            = _mm_and_pd(velec,cutoff_mask);
 213             velecsum         = _mm_add_pd(velecsum,velec);
 214
 215             fscal            = felec;
 216
 217             fscal            = _mm_and_pd(fscal,cutoff_mask);
 218
 219             /* Update vectorial force */
 220             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 221             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 222             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 223
 224             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 225             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 226             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 227
 228             }
 229
 230             /**************************
 231              * CALCULATE INTERACTIONS *
 232              **************************/
 233
 234             if (gmx_mm_any_lt(rsq20,rcutoff2))
 235             {
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq20             = _mm_mul_pd(iq2,jq0);
 239
 240             /* REACTION-FIELD ELECTROSTATICS */
 241             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 242             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 243
 244             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 245
 246             /* Update potential sum for this i atom from the interaction with this j atom. */
 247             velec            = _mm_and_pd(velec,cutoff_mask);
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             fscal            = _mm_and_pd(fscal,cutoff_mask);
 253
 254             /* Update vectorial force */
 255             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 256             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 257             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 258
 259             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 260             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 261             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 262
 263             }
 264
 265             /**************************
 266              * CALCULATE INTERACTIONS *
 267              **************************/
 268
 269             if (gmx_mm_any_lt(rsq30,rcutoff2))
 270             {
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq30             = _mm_mul_pd(iq3,jq0);
 274
 275             /* REACTION-FIELD ELECTROSTATICS */
 276             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 277             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 278
 279             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velec            = _mm_and_pd(velec,cutoff_mask);
 283             velecsum         = _mm_add_pd(velecsum,velec);
 284
 285             fscal            = felec;
 286
 287             fscal            = _mm_and_pd(fscal,cutoff_mask);
 288
 289             /* Update vectorial force */
 290             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 291             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 292             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 293
 294             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 295             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 296             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 297
 298             }
 299
 300             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 301
 302             /* Inner loop uses 120 flops */
 303         }
 304
 305         if(jidx<j_index_end)
 306         {
 307
 308             jnrA             = jjnr[jidx];
 309             j_coord_offsetA  = DIM*jnrA;
 310
 311             /* load j atom coordinates */
 312             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 313                                               &jx0,&jy0,&jz0);
 314
 315             /* Calculate displacement vector */
 316             dx10             = _mm_sub_pd(ix1,jx0);
 317             dy10             = _mm_sub_pd(iy1,jy0);
 318             dz10             = _mm_sub_pd(iz1,jz0);
 319             dx20             = _mm_sub_pd(ix2,jx0);
 320             dy20             = _mm_sub_pd(iy2,jy0);
 321             dz20             = _mm_sub_pd(iz2,jz0);
 322             dx30             = _mm_sub_pd(ix3,jx0);
 323             dy30             = _mm_sub_pd(iy3,jy0);
 324             dz30             = _mm_sub_pd(iz3,jz0);
 325
 326             /* Calculate squared distance and things based on it */
 327             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 328             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 329             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 330
 331             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 332             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 333             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 334
 335             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 336             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 337             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 338
 339             /* Load parameters for j particles */
 340             jq0              = _mm_load_sd(charge+jnrA+0);
 341
 342             fjx0             = _mm_setzero_pd();
 343             fjy0             = _mm_setzero_pd();
 344             fjz0             = _mm_setzero_pd();
 345
 346             /**************************
 347              * CALCULATE INTERACTIONS *
 348              **************************/
 349
 350             if (gmx_mm_any_lt(rsq10,rcutoff2))
 351             {
 352
 353             /* Compute parameters for interactions between i and j atoms */
 354             qq10             = _mm_mul_pd(iq1,jq0);
 355
 356             /* REACTION-FIELD ELECTROSTATICS */
 357             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 358             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 359
 360             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 361
 362             /* Update potential sum for this i atom from the interaction with this j atom. */
 363             velec            = _mm_and_pd(velec,cutoff_mask);
 364             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 365             velecsum         = _mm_add_pd(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm_and_pd(fscal,cutoff_mask);
 370
 371             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 372
 373             /* Update vectorial force */
 374             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 375             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 376             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 377
 378             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 379             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 380             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 381
 382             }
 383
 384             /**************************
 385              * CALCULATE INTERACTIONS *
 386              **************************/
 387
 388             if (gmx_mm_any_lt(rsq20,rcutoff2))
 389             {
 390
 391             /* Compute parameters for interactions between i and j atoms */
 392             qq20             = _mm_mul_pd(iq2,jq0);
 393
 394             /* REACTION-FIELD ELECTROSTATICS */
 395             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 396             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 397
 398             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 399
 400             /* Update potential sum for this i atom from the interaction with this j atom. */
 401             velec            = _mm_and_pd(velec,cutoff_mask);
 402             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 403             velecsum         = _mm_add_pd(velecsum,velec);
 404
 405             fscal            = felec;
 406
 407             fscal            = _mm_and_pd(fscal,cutoff_mask);
 408
 409             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 410
 411             /* Update vectorial force */
 412             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 413             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 414             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 415
 416             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 417             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 418             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 419
 420             }
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             if (gmx_mm_any_lt(rsq30,rcutoff2))
 427             {
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq30             = _mm_mul_pd(iq3,jq0);
 431
 432             /* REACTION-FIELD ELECTROSTATICS */
 433             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 434             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 435
 436             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 437
 438             /* Update potential sum for this i atom from the interaction with this j atom. */
 439             velec            = _mm_and_pd(velec,cutoff_mask);
 440             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 441             velecsum         = _mm_add_pd(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             fscal            = _mm_and_pd(fscal,cutoff_mask);
 446
 447             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 448
 449             /* Update vectorial force */
 450             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 451             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 452             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 453
 454             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 455             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 456             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 457
 458             }
 459
 460             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 461
 462             /* Inner loop uses 120 flops */
 463         }
 464
 465         /* End of innermost loop */
 466
 467         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 468                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 469
 470         ggid                        = gid[iidx];
 471         /* Update potential energies */
 472         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 473
 474         /* Increment number of inner iterations */
 475         inneriter                  += j_index_end - j_index_start;
 476
 477         /* Outer loop uses 19 flops */
 478     }
 479
 480     /* Increment number of outer iterations */
 481     outeriter        += nri;
 482
 483     /* Update outer/inner flops */
 484
 485     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*120);
 486 }
 487 /*
 488  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 489  * Electrostatics interaction: ReactionField
 490  * VdW interaction:            None
 491  * Geometry:                   Water4-Particle
 492  * Calculate force/pot:        Force
 493  */
 494 void
 495 nb_kernel_ElecRFCut_VdwNone_GeomW4P1_F_avx_128_fma_double
 496                     (t_nblist * gmx_restrict                nlist,
 497                      rvec * gmx_restrict                    xx,
 498                      rvec * gmx_restrict                    ff,
 499                      t_forcerec * gmx_restrict              fr,
 500                      t_mdatoms * gmx_restrict               mdatoms,
 501                      nb_kernel_data_t * gmx_restrict        kernel_data,
 502                      t_nrnb * gmx_restrict                  nrnb)
 503 {
 504     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 505      * just 0 for non-waters.
 506      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 507      * jnr indices corresponding to data put in the four positions in the SIMD register.
 508      */
 509     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 510     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 511     int              jnrA,jnrB;
 512     int              j_coord_offsetA,j_coord_offsetB;
 513     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 514     real             rcutoff_scalar;
 515     real             *shiftvec,*fshift,*x,*f;
 516     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 517     int              vdwioffset1;
 518     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 519     int              vdwioffset2;
 520     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 521     int              vdwioffset3;
 522     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 523     int              vdwjidx0A,vdwjidx0B;
 524     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 525     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 526     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 527     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 528     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 529     real             *charge;
 530     __m128d          dummy_mask,cutoff_mask;
 531     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 532     __m128d          one     = _mm_set1_pd(1.0);
 533     __m128d          two     = _mm_set1_pd(2.0);
 534     x                = xx[0];
 535     f                = ff[0];
 536
 537     nri              = nlist->nri;
 538     iinr             = nlist->iinr;
 539     jindex           = nlist->jindex;
 540     jjnr             = nlist->jjnr;
 541     shiftidx         = nlist->shift;
 542     gid              = nlist->gid;
 543     shiftvec         = fr->shift_vec[0];
 544     fshift           = fr->fshift[0];
 545     facel            = _mm_set1_pd(fr->epsfac);
 546     charge           = mdatoms->chargeA;
 547     krf              = _mm_set1_pd(fr->ic->k_rf);
 548     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 549     crf              = _mm_set1_pd(fr->ic->c_rf);
 550
 551     /* Setup water-specific parameters */
 552     inr              = nlist->iinr[0];
 553     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 554     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 555     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 556
 557     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 558     rcutoff_scalar   = fr->rcoulomb;
 559     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 560     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 561
 562     /* Avoid stupid compiler warnings */
 563     jnrA = jnrB = 0;
 564     j_coord_offsetA = 0;
 565     j_coord_offsetB = 0;
 566
 567     outeriter        = 0;
 568     inneriter        = 0;
 569
 570     /* Start outer loop over neighborlists */
 571     for(iidx=0; iidx<nri; iidx++)
 572     {
 573         /* Load shift vector for this list */
 574         i_shift_offset   = DIM*shiftidx[iidx];
 575
 576         /* Load limits for loop over neighbors */
 577         j_index_start    = jindex[iidx];
 578         j_index_end      = jindex[iidx+1];
 579
 580         /* Get outer coordinate index */
 581         inr              = iinr[iidx];
 582         i_coord_offset   = DIM*inr;
 583
 584         /* Load i particle coords and add shift vector */
 585         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 586                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 587
 588         fix1             = _mm_setzero_pd();
 589         fiy1             = _mm_setzero_pd();
 590         fiz1             = _mm_setzero_pd();
 591         fix2             = _mm_setzero_pd();
 592         fiy2             = _mm_setzero_pd();
 593         fiz2             = _mm_setzero_pd();
 594         fix3             = _mm_setzero_pd();
 595         fiy3             = _mm_setzero_pd();
 596         fiz3             = _mm_setzero_pd();
 597
 598         /* Start inner kernel loop */
 599         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 600         {
 601
 602             /* Get j neighbor index, and coordinate index */
 603             jnrA             = jjnr[jidx];
 604             jnrB             = jjnr[jidx+1];
 605             j_coord_offsetA  = DIM*jnrA;
 606             j_coord_offsetB  = DIM*jnrB;
 607
 608             /* load j atom coordinates */
 609             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 610                                               &jx0,&jy0,&jz0);
 611
 612             /* Calculate displacement vector */
 613             dx10             = _mm_sub_pd(ix1,jx0);
 614             dy10             = _mm_sub_pd(iy1,jy0);
 615             dz10             = _mm_sub_pd(iz1,jz0);
 616             dx20             = _mm_sub_pd(ix2,jx0);
 617             dy20             = _mm_sub_pd(iy2,jy0);
 618             dz20             = _mm_sub_pd(iz2,jz0);
 619             dx30             = _mm_sub_pd(ix3,jx0);
 620             dy30             = _mm_sub_pd(iy3,jy0);
 621             dz30             = _mm_sub_pd(iz3,jz0);
 622
 623             /* Calculate squared distance and things based on it */
 624             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 625             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 626             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 627
 628             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 629             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 630             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 631
 632             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 633             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 634             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 635
 636             /* Load parameters for j particles */
 637             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 638
 639             fjx0             = _mm_setzero_pd();
 640             fjy0             = _mm_setzero_pd();
 641             fjz0             = _mm_setzero_pd();
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             if (gmx_mm_any_lt(rsq10,rcutoff2))
 648             {
 649
 650             /* Compute parameters for interactions between i and j atoms */
 651             qq10             = _mm_mul_pd(iq1,jq0);
 652
 653             /* REACTION-FIELD ELECTROSTATICS */
 654             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 655
 656             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 657
 658             fscal            = felec;
 659
 660             fscal            = _mm_and_pd(fscal,cutoff_mask);
 661
 662             /* Update vectorial force */
 663             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 664             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 665             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 666
 667             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 668             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 669             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 670
 671             }
 672
 673             /**************************
 674              * CALCULATE INTERACTIONS *
 675              **************************/
 676
 677             if (gmx_mm_any_lt(rsq20,rcutoff2))
 678             {
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq20             = _mm_mul_pd(iq2,jq0);
 682
 683             /* REACTION-FIELD ELECTROSTATICS */
 684             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 685
 686             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 687
 688             fscal            = felec;
 689
 690             fscal            = _mm_and_pd(fscal,cutoff_mask);
 691
 692             /* Update vectorial force */
 693             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 694             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 695             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 696
 697             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 698             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 699             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 700
 701             }
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             if (gmx_mm_any_lt(rsq30,rcutoff2))
 708             {
 709
 710             /* Compute parameters for interactions between i and j atoms */
 711             qq30             = _mm_mul_pd(iq3,jq0);
 712
 713             /* REACTION-FIELD ELECTROSTATICS */
 714             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 715
 716             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 717
 718             fscal            = felec;
 719
 720             fscal            = _mm_and_pd(fscal,cutoff_mask);
 721
 722             /* Update vectorial force */
 723             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 724             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 725             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 726
 727             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 728             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 729             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 730
 731             }
 732
 733             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 734
 735             /* Inner loop uses 102 flops */
 736         }
 737
 738         if(jidx<j_index_end)
 739         {
 740
 741             jnrA             = jjnr[jidx];
 742             j_coord_offsetA  = DIM*jnrA;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 746                                               &jx0,&jy0,&jz0);
 747
 748             /* Calculate displacement vector */
 749             dx10             = _mm_sub_pd(ix1,jx0);
 750             dy10             = _mm_sub_pd(iy1,jy0);
 751             dz10             = _mm_sub_pd(iz1,jz0);
 752             dx20             = _mm_sub_pd(ix2,jx0);
 753             dy20             = _mm_sub_pd(iy2,jy0);
 754             dz20             = _mm_sub_pd(iz2,jz0);
 755             dx30             = _mm_sub_pd(ix3,jx0);
 756             dy30             = _mm_sub_pd(iy3,jy0);
 757             dz30             = _mm_sub_pd(iz3,jz0);
 758
 759             /* Calculate squared distance and things based on it */
 760             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 761             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 762             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 763
 764             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 765             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 766             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 767
 768             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 769             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 770             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 771
 772             /* Load parameters for j particles */
 773             jq0              = _mm_load_sd(charge+jnrA+0);
 774
 775             fjx0             = _mm_setzero_pd();
 776             fjy0             = _mm_setzero_pd();
 777             fjz0             = _mm_setzero_pd();
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             if (gmx_mm_any_lt(rsq10,rcutoff2))
 784             {
 785
 786             /* Compute parameters for interactions between i and j atoms */
 787             qq10             = _mm_mul_pd(iq1,jq0);
 788
 789             /* REACTION-FIELD ELECTROSTATICS */
 790             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 791
 792             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 793
 794             fscal            = felec;
 795
 796             fscal            = _mm_and_pd(fscal,cutoff_mask);
 797
 798             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 799
 800             /* Update vectorial force */
 801             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 802             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 803             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 804
 805             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 806             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 807             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 808
 809             }
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             if (gmx_mm_any_lt(rsq20,rcutoff2))
 816             {
 817
 818             /* Compute parameters for interactions between i and j atoms */
 819             qq20             = _mm_mul_pd(iq2,jq0);
 820
 821             /* REACTION-FIELD ELECTROSTATICS */
 822             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 823
 824             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 825
 826             fscal            = felec;
 827
 828             fscal            = _mm_and_pd(fscal,cutoff_mask);
 829
 830             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 831
 832             /* Update vectorial force */
 833             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 834             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 835             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 836
 837             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 838             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 839             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 840
 841             }
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             if (gmx_mm_any_lt(rsq30,rcutoff2))
 848             {
 849
 850             /* Compute parameters for interactions between i and j atoms */
 851             qq30             = _mm_mul_pd(iq3,jq0);
 852
 853             /* REACTION-FIELD ELECTROSTATICS */
 854             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 855
 856             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 857
 858             fscal            = felec;
 859
 860             fscal            = _mm_and_pd(fscal,cutoff_mask);
 861
 862             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 863
 864             /* Update vectorial force */
 865             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 866             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 867             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 868
 869             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 870             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 871             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 872
 873             }
 874
 875             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 876
 877             /* Inner loop uses 102 flops */
 878         }
 879
 880         /* End of innermost loop */
 881
 882         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 883                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 884
 885         /* Increment number of inner iterations */
 886         inneriter                  += j_index_end - j_index_start;
 887
 888         /* Outer loop uses 18 flops */
 889     }
 890
 891     /* Increment number of outer iterations */
 892     outeriter        += nri;
 893
 894     /* Update outer/inner flops */
 895
 896     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*102);
 897 }