src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwNone_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomW4P1_VF_sse2_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 SSE, 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              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128           dummy_mask,cutoff_mask;
  80     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  81     __m128           one     = _mm_set1_ps(1.0);
  82     __m128           two     = _mm_set1_ps(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_ps(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96
  97     /* Setup water-specific parameters */
  98     inr              = nlist->iinr[0];
  99     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 100     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 101     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = jnrC = jnrD = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107     j_coord_offsetC = 0;
 108     j_coord_offsetD = 0;
 109
 110     outeriter        = 0;
 111     inneriter        = 0;
 112
 113     /* Start outer loop over neighborlists */
 114     for(iidx=0; iidx<nri; iidx++)
 115     {
 116         /* Load shift vector for this list */
 117         i_shift_offset   = DIM*shiftidx[iidx];
 118         shX              = shiftvec[i_shift_offset+XX];
 119         shY              = shiftvec[i_shift_offset+YY];
 120         shZ              = shiftvec[i_shift_offset+ZZ];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 132         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 133         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 134         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 135         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 136         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 137         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 138         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 139         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 140
 141         fix1             = _mm_setzero_ps();
 142         fiy1             = _mm_setzero_ps();
 143         fiz1             = _mm_setzero_ps();
 144         fix2             = _mm_setzero_ps();
 145         fiy2             = _mm_setzero_ps();
 146         fiz2             = _mm_setzero_ps();
 147         fix3             = _mm_setzero_ps();
 148         fiy3             = _mm_setzero_ps();
 149         fiz3             = _mm_setzero_ps();
 150
 151         /* Reset potential sums */
 152         velecsum         = _mm_setzero_ps();
 153
 154         /* Start inner kernel loop */
 155         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 156         {
 157
 158             /* Get j neighbor index, and coordinate index */
 159             jnrA             = jjnr[jidx];
 160             jnrB             = jjnr[jidx+1];
 161             jnrC             = jjnr[jidx+2];
 162             jnrD             = jjnr[jidx+3];
 163
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx10             = _mm_sub_ps(ix1,jx0);
 176             dy10             = _mm_sub_ps(iy1,jy0);
 177             dz10             = _mm_sub_ps(iz1,jz0);
 178             dx20             = _mm_sub_ps(ix2,jx0);
 179             dy20             = _mm_sub_ps(iy2,jy0);
 180             dz20             = _mm_sub_ps(iz2,jz0);
 181             dx30             = _mm_sub_ps(ix3,jx0);
 182             dy30             = _mm_sub_ps(iy3,jy0);
 183             dz30             = _mm_sub_ps(iz3,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 187             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 188             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 189
 190             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 191             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 192             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 193
 194             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 195             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 196             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201
 202             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq10             = _mm_mul_ps(iq1,jq0);
 208
 209             /* COULOMB ELECTROSTATICS */
 210             velec            = _mm_mul_ps(qq10,rinv10);
 211             felec            = _mm_mul_ps(velec,rinvsq10);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velecsum         = _mm_add_ps(velecsum,velec);
 215
 216             fscal            = felec;
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm_mul_ps(fscal,dx10);
 220             ty               = _mm_mul_ps(fscal,dy10);
 221             tz               = _mm_mul_ps(fscal,dz10);
 222
 223             /* Update vectorial force */
 224             fix1             = _mm_add_ps(fix1,tx);
 225             fiy1             = _mm_add_ps(fiy1,ty);
 226             fiz1             = _mm_add_ps(fiz1,tz);
 227
 228             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 229                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 230                                                    tx,ty,tz);
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq20             = _mm_mul_ps(iq2,jq0);
 238
 239             /* COULOMB ELECTROSTATICS */
 240             velec            = _mm_mul_ps(qq20,rinv20);
 241             felec            = _mm_mul_ps(velec,rinvsq20);
 242
 243             /* Update potential sum for this i atom from the interaction with this j atom. */
 244             velecsum         = _mm_add_ps(velecsum,velec);
 245
 246             fscal            = felec;
 247
 248             /* Calculate temporary vectorial force */
 249             tx               = _mm_mul_ps(fscal,dx20);
 250             ty               = _mm_mul_ps(fscal,dy20);
 251             tz               = _mm_mul_ps(fscal,dz20);
 252
 253             /* Update vectorial force */
 254             fix2             = _mm_add_ps(fix2,tx);
 255             fiy2             = _mm_add_ps(fiy2,ty);
 256             fiz2             = _mm_add_ps(fiz2,tz);
 257
 258             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 259                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 260                                                    tx,ty,tz);
 261
 262             /**************************
 263              * CALCULATE INTERACTIONS *
 264              **************************/
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq30             = _mm_mul_ps(iq3,jq0);
 268
 269             /* COULOMB ELECTROSTATICS */
 270             velec            = _mm_mul_ps(qq30,rinv30);
 271             felec            = _mm_mul_ps(velec,rinvsq30);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             velecsum         = _mm_add_ps(velecsum,velec);
 275
 276             fscal            = felec;
 277
 278             /* Calculate temporary vectorial force */
 279             tx               = _mm_mul_ps(fscal,dx30);
 280             ty               = _mm_mul_ps(fscal,dy30);
 281             tz               = _mm_mul_ps(fscal,dz30);
 282
 283             /* Update vectorial force */
 284             fix3             = _mm_add_ps(fix3,tx);
 285             fiy3             = _mm_add_ps(fiy3,ty);
 286             fiz3             = _mm_add_ps(fiz3,tz);
 287
 288             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 289                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 290                                                    tx,ty,tz);
 291
 292             /* Inner loop uses 84 flops */
 293         }
 294
 295         if(jidx<j_index_end)
 296         {
 297
 298             /* Get j neighbor index, and coordinate index */
 299             jnrA             = jjnr[jidx];
 300             jnrB             = jjnr[jidx+1];
 301             jnrC             = jjnr[jidx+2];
 302             jnrD             = jjnr[jidx+3];
 303
 304             /* Sign of each element will be negative for non-real atoms.
 305              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 306              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 307              */
 308             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 309             jnrA       = (jnrA>=0) ? jnrA : 0;
 310             jnrB       = (jnrB>=0) ? jnrB : 0;
 311             jnrC       = (jnrC>=0) ? jnrC : 0;
 312             jnrD       = (jnrD>=0) ? jnrD : 0;
 313
 314             j_coord_offsetA  = DIM*jnrA;
 315             j_coord_offsetB  = DIM*jnrB;
 316             j_coord_offsetC  = DIM*jnrC;
 317             j_coord_offsetD  = DIM*jnrD;
 318
 319             /* load j atom coordinates */
 320             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 321                                               x+j_coord_offsetC,x+j_coord_offsetD,
 322                                               &jx0,&jy0,&jz0);
 323
 324             /* Calculate displacement vector */
 325             dx10             = _mm_sub_ps(ix1,jx0);
 326             dy10             = _mm_sub_ps(iy1,jy0);
 327             dz10             = _mm_sub_ps(iz1,jz0);
 328             dx20             = _mm_sub_ps(ix2,jx0);
 329             dy20             = _mm_sub_ps(iy2,jy0);
 330             dz20             = _mm_sub_ps(iz2,jz0);
 331             dx30             = _mm_sub_ps(ix3,jx0);
 332             dy30             = _mm_sub_ps(iy3,jy0);
 333             dz30             = _mm_sub_ps(iz3,jz0);
 334
 335             /* Calculate squared distance and things based on it */
 336             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 337             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 338             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 339
 340             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 341             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 342             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 343
 344             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 345             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 346             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 347
 348             /* Load parameters for j particles */
 349             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 350                                                               charge+jnrC+0,charge+jnrD+0);
 351
 352             /**************************
 353              * CALCULATE INTERACTIONS *
 354              **************************/
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq10             = _mm_mul_ps(iq1,jq0);
 358
 359             /* COULOMB ELECTROSTATICS */
 360             velec            = _mm_mul_ps(qq10,rinv10);
 361             felec            = _mm_mul_ps(velec,rinvsq10);
 362
 363             /* Update potential sum for this i atom from the interaction with this j atom. */
 364             velec            = _mm_andnot_ps(dummy_mask,velec);
 365             velecsum         = _mm_add_ps(velecsum,velec);
 366
 367             fscal            = felec;
 368
 369             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_ps(fscal,dx10);
 373             ty               = _mm_mul_ps(fscal,dy10);
 374             tz               = _mm_mul_ps(fscal,dz10);
 375
 376             /* Update vectorial force */
 377             fix1             = _mm_add_ps(fix1,tx);
 378             fiy1             = _mm_add_ps(fiy1,ty);
 379             fiz1             = _mm_add_ps(fiz1,tz);
 380
 381             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 382                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 383                                                    tx,ty,tz);
 384
 385             /**************************
 386              * CALCULATE INTERACTIONS *
 387              **************************/
 388
 389             /* Compute parameters for interactions between i and j atoms */
 390             qq20             = _mm_mul_ps(iq2,jq0);
 391
 392             /* COULOMB ELECTROSTATICS */
 393             velec            = _mm_mul_ps(qq20,rinv20);
 394             felec            = _mm_mul_ps(velec,rinvsq20);
 395
 396             /* Update potential sum for this i atom from the interaction with this j atom. */
 397             velec            = _mm_andnot_ps(dummy_mask,velec);
 398             velecsum         = _mm_add_ps(velecsum,velec);
 399
 400             fscal            = felec;
 401
 402             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 403
 404             /* Calculate temporary vectorial force */
 405             tx               = _mm_mul_ps(fscal,dx20);
 406             ty               = _mm_mul_ps(fscal,dy20);
 407             tz               = _mm_mul_ps(fscal,dz20);
 408
 409             /* Update vectorial force */
 410             fix2             = _mm_add_ps(fix2,tx);
 411             fiy2             = _mm_add_ps(fiy2,ty);
 412             fiz2             = _mm_add_ps(fiz2,tz);
 413
 414             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 415                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 416                                                    tx,ty,tz);
 417
 418             /**************************
 419              * CALCULATE INTERACTIONS *
 420              **************************/
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             qq30             = _mm_mul_ps(iq3,jq0);
 424
 425             /* COULOMB ELECTROSTATICS */
 426             velec            = _mm_mul_ps(qq30,rinv30);
 427             felec            = _mm_mul_ps(velec,rinvsq30);
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velec            = _mm_andnot_ps(dummy_mask,velec);
 431             velecsum         = _mm_add_ps(velecsum,velec);
 432
 433             fscal            = felec;
 434
 435             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 436
 437             /* Calculate temporary vectorial force */
 438             tx               = _mm_mul_ps(fscal,dx30);
 439             ty               = _mm_mul_ps(fscal,dy30);
 440             tz               = _mm_mul_ps(fscal,dz30);
 441
 442             /* Update vectorial force */
 443             fix3             = _mm_add_ps(fix3,tx);
 444             fiy3             = _mm_add_ps(fiy3,ty);
 445             fiz3             = _mm_add_ps(fiz3,tz);
 446
 447             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 448                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 449                                                    tx,ty,tz);
 450
 451             /* Inner loop uses 84 flops */
 452         }
 453
 454         /* End of innermost loop */
 455
 456         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 457                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 458
 459         ggid                        = gid[iidx];
 460         /* Update potential energies */
 461         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 462
 463         /* Increment number of inner iterations */
 464         inneriter                  += j_index_end - j_index_start;
 465
 466         /* Outer loop uses 28 flops */
 467     }
 468
 469     /* Increment number of outer iterations */
 470     outeriter        += nri;
 471
 472     /* Update outer/inner flops */
 473
 474     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*28 + inneriter*84);
 475 }
 476 /*
 477  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_sse2_single
 478  * Electrostatics interaction: Coulomb
 479  * VdW interaction:            None
 480  * Geometry:                   Water4-Particle
 481  * Calculate force/pot:        Force
 482  */
 483 void
 484 nb_kernel_ElecCoul_VdwNone_GeomW4P1_F_sse2_single
 485                     (t_nblist * gmx_restrict                nlist,
 486                      rvec * gmx_restrict                    xx,
 487                      rvec * gmx_restrict                    ff,
 488                      t_forcerec * gmx_restrict              fr,
 489                      t_mdatoms * gmx_restrict               mdatoms,
 490                      nb_kernel_data_t * gmx_restrict        kernel_data,
 491                      t_nrnb * gmx_restrict                  nrnb)
 492 {
 493     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 494      * just 0 for non-waters.
 495      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 496      * jnr indices corresponding to data put in the four positions in the SIMD register.
 497      */
 498     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 499     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 500     int              jnrA,jnrB,jnrC,jnrD;
 501     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 502     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 503     real             shX,shY,shZ,rcutoff_scalar;
 504     real             *shiftvec,*fshift,*x,*f;
 505     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 506     int              vdwioffset1;
 507     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 508     int              vdwioffset2;
 509     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 510     int              vdwioffset3;
 511     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 512     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 513     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 514     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 515     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 516     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 517     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 518     real             *charge;
 519     __m128           dummy_mask,cutoff_mask;
 520     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 521     __m128           one     = _mm_set1_ps(1.0);
 522     __m128           two     = _mm_set1_ps(2.0);
 523     x                = xx[0];
 524     f                = ff[0];
 525
 526     nri              = nlist->nri;
 527     iinr             = nlist->iinr;
 528     jindex           = nlist->jindex;
 529     jjnr             = nlist->jjnr;
 530     shiftidx         = nlist->shift;
 531     gid              = nlist->gid;
 532     shiftvec         = fr->shift_vec[0];
 533     fshift           = fr->fshift[0];
 534     facel            = _mm_set1_ps(fr->epsfac);
 535     charge           = mdatoms->chargeA;
 536
 537     /* Setup water-specific parameters */
 538     inr              = nlist->iinr[0];
 539     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 540     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 541     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 542
 543     /* Avoid stupid compiler warnings */
 544     jnrA = jnrB = jnrC = jnrD = 0;
 545     j_coord_offsetA = 0;
 546     j_coord_offsetB = 0;
 547     j_coord_offsetC = 0;
 548     j_coord_offsetD = 0;
 549
 550     outeriter        = 0;
 551     inneriter        = 0;
 552
 553     /* Start outer loop over neighborlists */
 554     for(iidx=0; iidx<nri; iidx++)
 555     {
 556         /* Load shift vector for this list */
 557         i_shift_offset   = DIM*shiftidx[iidx];
 558         shX              = shiftvec[i_shift_offset+XX];
 559         shY              = shiftvec[i_shift_offset+YY];
 560         shZ              = shiftvec[i_shift_offset+ZZ];
 561
 562         /* Load limits for loop over neighbors */
 563         j_index_start    = jindex[iidx];
 564         j_index_end      = jindex[iidx+1];
 565
 566         /* Get outer coordinate index */
 567         inr              = iinr[iidx];
 568         i_coord_offset   = DIM*inr;
 569
 570         /* Load i particle coords and add shift vector */
 571         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 572         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 573         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 574         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 575         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 576         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 577         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 578         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 579         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 580
 581         fix1             = _mm_setzero_ps();
 582         fiy1             = _mm_setzero_ps();
 583         fiz1             = _mm_setzero_ps();
 584         fix2             = _mm_setzero_ps();
 585         fiy2             = _mm_setzero_ps();
 586         fiz2             = _mm_setzero_ps();
 587         fix3             = _mm_setzero_ps();
 588         fiy3             = _mm_setzero_ps();
 589         fiz3             = _mm_setzero_ps();
 590
 591         /* Start inner kernel loop */
 592         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 593         {
 594
 595             /* Get j neighbor index, and coordinate index */
 596             jnrA             = jjnr[jidx];
 597             jnrB             = jjnr[jidx+1];
 598             jnrC             = jjnr[jidx+2];
 599             jnrD             = jjnr[jidx+3];
 600
 601             j_coord_offsetA  = DIM*jnrA;
 602             j_coord_offsetB  = DIM*jnrB;
 603             j_coord_offsetC  = DIM*jnrC;
 604             j_coord_offsetD  = DIM*jnrD;
 605
 606             /* load j atom coordinates */
 607             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 608                                               x+j_coord_offsetC,x+j_coord_offsetD,
 609                                               &jx0,&jy0,&jz0);
 610
 611             /* Calculate displacement vector */
 612             dx10             = _mm_sub_ps(ix1,jx0);
 613             dy10             = _mm_sub_ps(iy1,jy0);
 614             dz10             = _mm_sub_ps(iz1,jz0);
 615             dx20             = _mm_sub_ps(ix2,jx0);
 616             dy20             = _mm_sub_ps(iy2,jy0);
 617             dz20             = _mm_sub_ps(iz2,jz0);
 618             dx30             = _mm_sub_ps(ix3,jx0);
 619             dy30             = _mm_sub_ps(iy3,jy0);
 620             dz30             = _mm_sub_ps(iz3,jz0);
 621
 622             /* Calculate squared distance and things based on it */
 623             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 624             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 625             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 626
 627             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 628             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 629             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 630
 631             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 632             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 633             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 634
 635             /* Load parameters for j particles */
 636             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 637                                                               charge+jnrC+0,charge+jnrD+0);
 638
 639             /**************************
 640              * CALCULATE INTERACTIONS *
 641              **************************/
 642
 643             /* Compute parameters for interactions between i and j atoms */
 644             qq10             = _mm_mul_ps(iq1,jq0);
 645
 646             /* COULOMB ELECTROSTATICS */
 647             velec            = _mm_mul_ps(qq10,rinv10);
 648             felec            = _mm_mul_ps(velec,rinvsq10);
 649
 650             fscal            = felec;
 651
 652             /* Calculate temporary vectorial force */
 653             tx               = _mm_mul_ps(fscal,dx10);
 654             ty               = _mm_mul_ps(fscal,dy10);
 655             tz               = _mm_mul_ps(fscal,dz10);
 656
 657             /* Update vectorial force */
 658             fix1             = _mm_add_ps(fix1,tx);
 659             fiy1             = _mm_add_ps(fiy1,ty);
 660             fiz1             = _mm_add_ps(fiz1,tz);
 661
 662             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 663                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 664                                                    tx,ty,tz);
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq20             = _mm_mul_ps(iq2,jq0);
 672
 673             /* COULOMB ELECTROSTATICS */
 674             velec            = _mm_mul_ps(qq20,rinv20);
 675             felec            = _mm_mul_ps(velec,rinvsq20);
 676
 677             fscal            = felec;
 678
 679             /* Calculate temporary vectorial force */
 680             tx               = _mm_mul_ps(fscal,dx20);
 681             ty               = _mm_mul_ps(fscal,dy20);
 682             tz               = _mm_mul_ps(fscal,dz20);
 683
 684             /* Update vectorial force */
 685             fix2             = _mm_add_ps(fix2,tx);
 686             fiy2             = _mm_add_ps(fiy2,ty);
 687             fiz2             = _mm_add_ps(fiz2,tz);
 688
 689             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 690                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 691                                                    tx,ty,tz);
 692
 693             /**************************
 694              * CALCULATE INTERACTIONS *
 695              **************************/
 696
 697             /* Compute parameters for interactions between i and j atoms */
 698             qq30             = _mm_mul_ps(iq3,jq0);
 699
 700             /* COULOMB ELECTROSTATICS */
 701             velec            = _mm_mul_ps(qq30,rinv30);
 702             felec            = _mm_mul_ps(velec,rinvsq30);
 703
 704             fscal            = felec;
 705
 706             /* Calculate temporary vectorial force */
 707             tx               = _mm_mul_ps(fscal,dx30);
 708             ty               = _mm_mul_ps(fscal,dy30);
 709             tz               = _mm_mul_ps(fscal,dz30);
 710
 711             /* Update vectorial force */
 712             fix3             = _mm_add_ps(fix3,tx);
 713             fiy3             = _mm_add_ps(fiy3,ty);
 714             fiz3             = _mm_add_ps(fiz3,tz);
 715
 716             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 717                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 718                                                    tx,ty,tz);
 719
 720             /* Inner loop uses 81 flops */
 721         }
 722
 723         if(jidx<j_index_end)
 724         {
 725
 726             /* Get j neighbor index, and coordinate index */
 727             jnrA             = jjnr[jidx];
 728             jnrB             = jjnr[jidx+1];
 729             jnrC             = jjnr[jidx+2];
 730             jnrD             = jjnr[jidx+3];
 731
 732             /* Sign of each element will be negative for non-real atoms.
 733              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 734              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 735              */
 736             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 737             jnrA       = (jnrA>=0) ? jnrA : 0;
 738             jnrB       = (jnrB>=0) ? jnrB : 0;
 739             jnrC       = (jnrC>=0) ? jnrC : 0;
 740             jnrD       = (jnrD>=0) ? jnrD : 0;
 741
 742             j_coord_offsetA  = DIM*jnrA;
 743             j_coord_offsetB  = DIM*jnrB;
 744             j_coord_offsetC  = DIM*jnrC;
 745             j_coord_offsetD  = DIM*jnrD;
 746
 747             /* load j atom coordinates */
 748             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 749                                               x+j_coord_offsetC,x+j_coord_offsetD,
 750                                               &jx0,&jy0,&jz0);
 751
 752             /* Calculate displacement vector */
 753             dx10             = _mm_sub_ps(ix1,jx0);
 754             dy10             = _mm_sub_ps(iy1,jy0);
 755             dz10             = _mm_sub_ps(iz1,jz0);
 756             dx20             = _mm_sub_ps(ix2,jx0);
 757             dy20             = _mm_sub_ps(iy2,jy0);
 758             dz20             = _mm_sub_ps(iz2,jz0);
 759             dx30             = _mm_sub_ps(ix3,jx0);
 760             dy30             = _mm_sub_ps(iy3,jy0);
 761             dz30             = _mm_sub_ps(iz3,jz0);
 762
 763             /* Calculate squared distance and things based on it */
 764             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 765             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 766             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 767
 768             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 769             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 770             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 771
 772             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 773             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 774             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 775
 776             /* Load parameters for j particles */
 777             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 778                                                               charge+jnrC+0,charge+jnrD+0);
 779
 780             /**************************
 781              * CALCULATE INTERACTIONS *
 782              **************************/
 783
 784             /* Compute parameters for interactions between i and j atoms */
 785             qq10             = _mm_mul_ps(iq1,jq0);
 786
 787             /* COULOMB ELECTROSTATICS */
 788             velec            = _mm_mul_ps(qq10,rinv10);
 789             felec            = _mm_mul_ps(velec,rinvsq10);
 790
 791             fscal            = felec;
 792
 793             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 794
 795             /* Calculate temporary vectorial force */
 796             tx               = _mm_mul_ps(fscal,dx10);
 797             ty               = _mm_mul_ps(fscal,dy10);
 798             tz               = _mm_mul_ps(fscal,dz10);
 799
 800             /* Update vectorial force */
 801             fix1             = _mm_add_ps(fix1,tx);
 802             fiy1             = _mm_add_ps(fiy1,ty);
 803             fiz1             = _mm_add_ps(fiz1,tz);
 804
 805             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 806                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 807                                                    tx,ty,tz);
 808
 809             /**************************
 810              * CALCULATE INTERACTIONS *
 811              **************************/
 812
 813             /* Compute parameters for interactions between i and j atoms */
 814             qq20             = _mm_mul_ps(iq2,jq0);
 815
 816             /* COULOMB ELECTROSTATICS */
 817             velec            = _mm_mul_ps(qq20,rinv20);
 818             felec            = _mm_mul_ps(velec,rinvsq20);
 819
 820             fscal            = felec;
 821
 822             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm_mul_ps(fscal,dx20);
 826             ty               = _mm_mul_ps(fscal,dy20);
 827             tz               = _mm_mul_ps(fscal,dz20);
 828
 829             /* Update vectorial force */
 830             fix2             = _mm_add_ps(fix2,tx);
 831             fiy2             = _mm_add_ps(fiy2,ty);
 832             fiz2             = _mm_add_ps(fiz2,tz);
 833
 834             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 835                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 836                                                    tx,ty,tz);
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             qq30             = _mm_mul_ps(iq3,jq0);
 844
 845             /* COULOMB ELECTROSTATICS */
 846             velec            = _mm_mul_ps(qq30,rinv30);
 847             felec            = _mm_mul_ps(velec,rinvsq30);
 848
 849             fscal            = felec;
 850
 851             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 852
 853             /* Calculate temporary vectorial force */
 854             tx               = _mm_mul_ps(fscal,dx30);
 855             ty               = _mm_mul_ps(fscal,dy30);
 856             tz               = _mm_mul_ps(fscal,dz30);
 857
 858             /* Update vectorial force */
 859             fix3             = _mm_add_ps(fix3,tx);
 860             fiy3             = _mm_add_ps(fiy3,ty);
 861             fiz3             = _mm_add_ps(fiz3,tz);
 862
 863             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 864                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 865                                                    tx,ty,tz);
 866
 867             /* Inner loop uses 81 flops */
 868         }
 869
 870         /* End of innermost loop */
 871
 872         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 873                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 874
 875         /* Increment number of inner iterations */
 876         inneriter                  += j_index_end - j_index_start;
 877
 878         /* Outer loop uses 27 flops */
 879     }
 880
 881     /* Increment number of outer iterations */
 882     outeriter        += nri;
 883
 884     /* Update outer/inner flops */
 885
 886     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*27 + inneriter*81);
 887 }