src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSh_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_ElecRFCut_VdwLJSh_GeomW4P1_VF_sse2_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_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              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           tx,ty,tz,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             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             if (gmx_mm_any_lt(rsq00,rcutoff2))
 241             {
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,
 246                                          vdwparam+vdwioffset0+vdwjidx0C,
 247                                          vdwparam+vdwioffset0+vdwjidx0D,
 248                                          &c6_00,&c12_00);
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 254             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 255             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 256                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 257             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 258
 259             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 263             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 264
 265             fscal            = fvdw;
 266
 267             fscal            = _mm_and_ps(fscal,cutoff_mask);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_ps(fscal,dx00);
 271             ty               = _mm_mul_ps(fscal,dy00);
 272             tz               = _mm_mul_ps(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_ps(fix0,tx);
 276             fiy0             = _mm_add_ps(fiy0,ty);
 277             fiz0             = _mm_add_ps(fiz0,tz);
 278
 279             fjptrA             = f+j_coord_offsetA;
 280             fjptrB             = f+j_coord_offsetB;
 281             fjptrC             = f+j_coord_offsetC;
 282             fjptrD             = f+j_coord_offsetD;
 283             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 284
 285             }
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm_any_lt(rsq10,rcutoff2))
 292             {
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm_mul_ps(iq1,jq0);
 296
 297             /* REACTION-FIELD ELECTROSTATICS */
 298             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 299             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 300
 301             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velec            = _mm_and_ps(velec,cutoff_mask);
 305             velecsum         = _mm_add_ps(velecsum,velec);
 306
 307             fscal            = felec;
 308
 309             fscal            = _mm_and_ps(fscal,cutoff_mask);
 310
 311             /* Calculate temporary vectorial force */
 312             tx               = _mm_mul_ps(fscal,dx10);
 313             ty               = _mm_mul_ps(fscal,dy10);
 314             tz               = _mm_mul_ps(fscal,dz10);
 315
 316             /* Update vectorial force */
 317             fix1             = _mm_add_ps(fix1,tx);
 318             fiy1             = _mm_add_ps(fiy1,ty);
 319             fiz1             = _mm_add_ps(fiz1,tz);
 320
 321             fjptrA             = f+j_coord_offsetA;
 322             fjptrB             = f+j_coord_offsetB;
 323             fjptrC             = f+j_coord_offsetC;
 324             fjptrD             = f+j_coord_offsetD;
 325             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 326
 327             }
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             if (gmx_mm_any_lt(rsq20,rcutoff2))
 334             {
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* REACTION-FIELD ELECTROSTATICS */
 340             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 341             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 342
 343             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 344
 345             /* Update potential sum for this i atom from the interaction with this j atom. */
 346             velec            = _mm_and_ps(velec,cutoff_mask);
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             fscal            = _mm_and_ps(fscal,cutoff_mask);
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx20);
 355             ty               = _mm_mul_ps(fscal,dy20);
 356             tz               = _mm_mul_ps(fscal,dz20);
 357
 358             /* Update vectorial force */
 359             fix2             = _mm_add_ps(fix2,tx);
 360             fiy2             = _mm_add_ps(fiy2,ty);
 361             fiz2             = _mm_add_ps(fiz2,tz);
 362
 363             fjptrA             = f+j_coord_offsetA;
 364             fjptrB             = f+j_coord_offsetB;
 365             fjptrC             = f+j_coord_offsetC;
 366             fjptrD             = f+j_coord_offsetD;
 367             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 368
 369             }
 370
 371             /**************************
 372              * CALCULATE INTERACTIONS *
 373              **************************/
 374
 375             if (gmx_mm_any_lt(rsq30,rcutoff2))
 376             {
 377
 378             /* Compute parameters for interactions between i and j atoms */
 379             qq30             = _mm_mul_ps(iq3,jq0);
 380
 381             /* REACTION-FIELD ELECTROSTATICS */
 382             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 383             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 384
 385             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velec            = _mm_and_ps(velec,cutoff_mask);
 389             velecsum         = _mm_add_ps(velecsum,velec);
 390
 391             fscal            = felec;
 392
 393             fscal            = _mm_and_ps(fscal,cutoff_mask);
 394
 395             /* Calculate temporary vectorial force */
 396             tx               = _mm_mul_ps(fscal,dx30);
 397             ty               = _mm_mul_ps(fscal,dy30);
 398             tz               = _mm_mul_ps(fscal,dz30);
 399
 400             /* Update vectorial force */
 401             fix3             = _mm_add_ps(fix3,tx);
 402             fiy3             = _mm_add_ps(fiy3,ty);
 403             fiz3             = _mm_add_ps(fiz3,tz);
 404
 405             fjptrA             = f+j_coord_offsetA;
 406             fjptrB             = f+j_coord_offsetB;
 407             fjptrC             = f+j_coord_offsetC;
 408             fjptrD             = f+j_coord_offsetD;
 409             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 410
 411             }
 412
 413             /* Inner loop uses 149 flops */
 414         }
 415
 416         if(jidx<j_index_end)
 417         {
 418
 419             /* Get j neighbor index, and coordinate index */
 420             jnrlistA         = jjnr[jidx];
 421             jnrlistB         = jjnr[jidx+1];
 422             jnrlistC         = jjnr[jidx+2];
 423             jnrlistD         = jjnr[jidx+3];
 424             /* Sign of each element will be negative for non-real atoms.
 425              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 426              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 427              */
 428             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 429             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 430             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 431             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 432             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 433             j_coord_offsetA  = DIM*jnrA;
 434             j_coord_offsetB  = DIM*jnrB;
 435             j_coord_offsetC  = DIM*jnrC;
 436             j_coord_offsetD  = DIM*jnrD;
 437
 438             /* load j atom coordinates */
 439             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 440                                               x+j_coord_offsetC,x+j_coord_offsetD,
 441                                               &jx0,&jy0,&jz0);
 442
 443             /* Calculate displacement vector */
 444             dx00             = _mm_sub_ps(ix0,jx0);
 445             dy00             = _mm_sub_ps(iy0,jy0);
 446             dz00             = _mm_sub_ps(iz0,jz0);
 447             dx10             = _mm_sub_ps(ix1,jx0);
 448             dy10             = _mm_sub_ps(iy1,jy0);
 449             dz10             = _mm_sub_ps(iz1,jz0);
 450             dx20             = _mm_sub_ps(ix2,jx0);
 451             dy20             = _mm_sub_ps(iy2,jy0);
 452             dz20             = _mm_sub_ps(iz2,jz0);
 453             dx30             = _mm_sub_ps(ix3,jx0);
 454             dy30             = _mm_sub_ps(iy3,jy0);
 455             dz30             = _mm_sub_ps(iz3,jz0);
 456
 457             /* Calculate squared distance and things based on it */
 458             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 459             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 460             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 461             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 462
 463             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 464             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 465             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 466
 467             rinvsq00         = gmx_mm_inv_ps(rsq00);
 468             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 469             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 470             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 471
 472             /* Load parameters for j particles */
 473             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 474                                                               charge+jnrC+0,charge+jnrD+0);
 475             vdwjidx0A        = 2*vdwtype[jnrA+0];
 476             vdwjidx0B        = 2*vdwtype[jnrB+0];
 477             vdwjidx0C        = 2*vdwtype[jnrC+0];
 478             vdwjidx0D        = 2*vdwtype[jnrD+0];
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             if (gmx_mm_any_lt(rsq00,rcutoff2))
 485             {
 486
 487             /* Compute parameters for interactions between i and j atoms */
 488             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 489                                          vdwparam+vdwioffset0+vdwjidx0B,
 490                                          vdwparam+vdwioffset0+vdwjidx0C,
 491                                          vdwparam+vdwioffset0+vdwjidx0D,
 492                                          &c6_00,&c12_00);
 493
 494             /* LENNARD-JONES DISPERSION/REPULSION */
 495
 496             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 497             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 498             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 499             vvdw             = _mm_sub_ps(_mm_mul_ps( _mm_sub_ps(vvdw12 , _mm_mul_ps(c12_00,_mm_mul_ps(sh_vdw_invrcut6,sh_vdw_invrcut6))), one_twelfth) ,
 500                                           _mm_mul_ps( _mm_sub_ps(vvdw6,_mm_mul_ps(c6_00,sh_vdw_invrcut6)),one_sixth));
 501             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 502
 503             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 507             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 508             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 509
 510             fscal            = fvdw;
 511
 512             fscal            = _mm_and_ps(fscal,cutoff_mask);
 513
 514             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 515
 516             /* Calculate temporary vectorial force */
 517             tx               = _mm_mul_ps(fscal,dx00);
 518             ty               = _mm_mul_ps(fscal,dy00);
 519             tz               = _mm_mul_ps(fscal,dz00);
 520
 521             /* Update vectorial force */
 522             fix0             = _mm_add_ps(fix0,tx);
 523             fiy0             = _mm_add_ps(fiy0,ty);
 524             fiz0             = _mm_add_ps(fiz0,tz);
 525
 526             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 527             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 528             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 529             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 530             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 531
 532             }
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             if (gmx_mm_any_lt(rsq10,rcutoff2))
 539             {
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_ps(iq1,jq0);
 543
 544             /* REACTION-FIELD ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 546             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 547
 548             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             velec            = _mm_and_ps(velec,cutoff_mask);
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_and_ps(fscal,cutoff_mask);
 558
 559             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 560
 561             /* Calculate temporary vectorial force */
 562             tx               = _mm_mul_ps(fscal,dx10);
 563             ty               = _mm_mul_ps(fscal,dy10);
 564             tz               = _mm_mul_ps(fscal,dz10);
 565
 566             /* Update vectorial force */
 567             fix1             = _mm_add_ps(fix1,tx);
 568             fiy1             = _mm_add_ps(fiy1,ty);
 569             fiz1             = _mm_add_ps(fiz1,tz);
 570
 571             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 572             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 573             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 574             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 575             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 576
 577             }
 578
 579             /**************************
 580              * CALCULATE INTERACTIONS *
 581              **************************/
 582
 583             if (gmx_mm_any_lt(rsq20,rcutoff2))
 584             {
 585
 586             /* Compute parameters for interactions between i and j atoms */
 587             qq20             = _mm_mul_ps(iq2,jq0);
 588
 589             /* REACTION-FIELD ELECTROSTATICS */
 590             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 591             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 592
 593             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 594
 595             /* Update potential sum for this i atom from the interaction with this j atom. */
 596             velec            = _mm_and_ps(velec,cutoff_mask);
 597             velec            = _mm_andnot_ps(dummy_mask,velec);
 598             velecsum         = _mm_add_ps(velecsum,velec);
 599
 600             fscal            = felec;
 601
 602             fscal            = _mm_and_ps(fscal,cutoff_mask);
 603
 604             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 605
 606             /* Calculate temporary vectorial force */
 607             tx               = _mm_mul_ps(fscal,dx20);
 608             ty               = _mm_mul_ps(fscal,dy20);
 609             tz               = _mm_mul_ps(fscal,dz20);
 610
 611             /* Update vectorial force */
 612             fix2             = _mm_add_ps(fix2,tx);
 613             fiy2             = _mm_add_ps(fiy2,ty);
 614             fiz2             = _mm_add_ps(fiz2,tz);
 615
 616             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 617             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 618             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 619             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 620             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 621
 622             }
 623
 624             /**************************
 625              * CALCULATE INTERACTIONS *
 626              **************************/
 627
 628             if (gmx_mm_any_lt(rsq30,rcutoff2))
 629             {
 630
 631             /* Compute parameters for interactions between i and j atoms */
 632             qq30             = _mm_mul_ps(iq3,jq0);
 633
 634             /* REACTION-FIELD ELECTROSTATICS */
 635             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 636             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 637
 638             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 639
 640             /* Update potential sum for this i atom from the interaction with this j atom. */
 641             velec            = _mm_and_ps(velec,cutoff_mask);
 642             velec            = _mm_andnot_ps(dummy_mask,velec);
 643             velecsum         = _mm_add_ps(velecsum,velec);
 644
 645             fscal            = felec;
 646
 647             fscal            = _mm_and_ps(fscal,cutoff_mask);
 648
 649             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 650
 651             /* Calculate temporary vectorial force */
 652             tx               = _mm_mul_ps(fscal,dx30);
 653             ty               = _mm_mul_ps(fscal,dy30);
 654             tz               = _mm_mul_ps(fscal,dz30);
 655
 656             /* Update vectorial force */
 657             fix3             = _mm_add_ps(fix3,tx);
 658             fiy3             = _mm_add_ps(fiy3,ty);
 659             fiz3             = _mm_add_ps(fiz3,tz);
 660
 661             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 662             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 663             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 664             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 665             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 666
 667             }
 668
 669             /* Inner loop uses 149 flops */
 670         }
 671
 672         /* End of innermost loop */
 673
 674         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 675                                               f+i_coord_offset,fshift+i_shift_offset);
 676
 677         ggid                        = gid[iidx];
 678         /* Update potential energies */
 679         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 680         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 681
 682         /* Increment number of inner iterations */
 683         inneriter                  += j_index_end - j_index_start;
 684
 685         /* Outer loop uses 26 flops */
 686     }
 687
 688     /* Increment number of outer iterations */
 689     outeriter        += nri;
 690
 691     /* Update outer/inner flops */
 692
 693     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*149);
 694 }
 695 /*
 696  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse2_single
 697  * Electrostatics interaction: ReactionField
 698  * VdW interaction:            LennardJones
 699  * Geometry:                   Water4-Particle
 700  * Calculate force/pot:        Force
 701  */
 702 void
 703 nb_kernel_ElecRFCut_VdwLJSh_GeomW4P1_F_sse2_single
 704                     (t_nblist * gmx_restrict                nlist,
 705                      rvec * gmx_restrict                    xx,
 706                      rvec * gmx_restrict                    ff,
 707                      t_forcerec * gmx_restrict              fr,
 708                      t_mdatoms * gmx_restrict               mdatoms,
 709                      nb_kernel_data_t * gmx_restrict        kernel_data,
 710                      t_nrnb * gmx_restrict                  nrnb)
 711 {
 712     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 713      * just 0 for non-waters.
 714      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 715      * jnr indices corresponding to data put in the four positions in the SIMD register.
 716      */
 717     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 718     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 719     int              jnrA,jnrB,jnrC,jnrD;
 720     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 721     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 722     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 723     real             rcutoff_scalar;
 724     real             *shiftvec,*fshift,*x,*f;
 725     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 726     real             scratch[4*DIM];
 727     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 728     int              vdwioffset0;
 729     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 730     int              vdwioffset1;
 731     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 732     int              vdwioffset2;
 733     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 734     int              vdwioffset3;
 735     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 736     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 737     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 738     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 739     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 740     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 741     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 742     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 743     real             *charge;
 744     int              nvdwtype;
 745     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 746     int              *vdwtype;
 747     real             *vdwparam;
 748     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 749     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 750     __m128           dummy_mask,cutoff_mask;
 751     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 752     __m128           one     = _mm_set1_ps(1.0);
 753     __m128           two     = _mm_set1_ps(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_ps(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     krf              = _mm_set1_ps(fr->ic->k_rf);
 768     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 769     crf              = _mm_set1_ps(fr->ic->c_rf);
 770     nvdwtype         = fr->ntype;
 771     vdwparam         = fr->nbfp;
 772     vdwtype          = mdatoms->typeA;
 773
 774     /* Setup water-specific parameters */
 775     inr              = nlist->iinr[0];
 776     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 777     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 778     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 779     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 780
 781     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 782     rcutoff_scalar   = fr->rcoulomb;
 783     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 784     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 785
 786     sh_vdw_invrcut6  = _mm_set1_ps(fr->ic->sh_invrc6);
 787     rvdw             = _mm_set1_ps(fr->rvdw);
 788
 789     /* Avoid stupid compiler warnings */
 790     jnrA = jnrB = jnrC = jnrD = 0;
 791     j_coord_offsetA = 0;
 792     j_coord_offsetB = 0;
 793     j_coord_offsetC = 0;
 794     j_coord_offsetD = 0;
 795
 796     outeriter        = 0;
 797     inneriter        = 0;
 798
 799     for(iidx=0;iidx<4*DIM;iidx++)
 800     {
 801         scratch[iidx] = 0.0;
 802     }
 803
 804     /* Start outer loop over neighborlists */
 805     for(iidx=0; iidx<nri; iidx++)
 806     {
 807         /* Load shift vector for this list */
 808         i_shift_offset   = DIM*shiftidx[iidx];
 809
 810         /* Load limits for loop over neighbors */
 811         j_index_start    = jindex[iidx];
 812         j_index_end      = jindex[iidx+1];
 813
 814         /* Get outer coordinate index */
 815         inr              = iinr[iidx];
 816         i_coord_offset   = DIM*inr;
 817
 818         /* Load i particle coords and add shift vector */
 819         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 820                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 821
 822         fix0             = _mm_setzero_ps();
 823         fiy0             = _mm_setzero_ps();
 824         fiz0             = _mm_setzero_ps();
 825         fix1             = _mm_setzero_ps();
 826         fiy1             = _mm_setzero_ps();
 827         fiz1             = _mm_setzero_ps();
 828         fix2             = _mm_setzero_ps();
 829         fiy2             = _mm_setzero_ps();
 830         fiz2             = _mm_setzero_ps();
 831         fix3             = _mm_setzero_ps();
 832         fiy3             = _mm_setzero_ps();
 833         fiz3             = _mm_setzero_ps();
 834
 835         /* Start inner kernel loop */
 836         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 837         {
 838
 839             /* Get j neighbor index, and coordinate index */
 840             jnrA             = jjnr[jidx];
 841             jnrB             = jjnr[jidx+1];
 842             jnrC             = jjnr[jidx+2];
 843             jnrD             = jjnr[jidx+3];
 844             j_coord_offsetA  = DIM*jnrA;
 845             j_coord_offsetB  = DIM*jnrB;
 846             j_coord_offsetC  = DIM*jnrC;
 847             j_coord_offsetD  = DIM*jnrD;
 848
 849             /* load j atom coordinates */
 850             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 851                                               x+j_coord_offsetC,x+j_coord_offsetD,
 852                                               &jx0,&jy0,&jz0);
 853
 854             /* Calculate displacement vector */
 855             dx00             = _mm_sub_ps(ix0,jx0);
 856             dy00             = _mm_sub_ps(iy0,jy0);
 857             dz00             = _mm_sub_ps(iz0,jz0);
 858             dx10             = _mm_sub_ps(ix1,jx0);
 859             dy10             = _mm_sub_ps(iy1,jy0);
 860             dz10             = _mm_sub_ps(iz1,jz0);
 861             dx20             = _mm_sub_ps(ix2,jx0);
 862             dy20             = _mm_sub_ps(iy2,jy0);
 863             dz20             = _mm_sub_ps(iz2,jz0);
 864             dx30             = _mm_sub_ps(ix3,jx0);
 865             dy30             = _mm_sub_ps(iy3,jy0);
 866             dz30             = _mm_sub_ps(iz3,jz0);
 867
 868             /* Calculate squared distance and things based on it */
 869             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 870             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 871             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 872             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 873
 874             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 875             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 876             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 877
 878             rinvsq00         = gmx_mm_inv_ps(rsq00);
 879             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 880             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 881             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 882
 883             /* Load parameters for j particles */
 884             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 885                                                               charge+jnrC+0,charge+jnrD+0);
 886             vdwjidx0A        = 2*vdwtype[jnrA+0];
 887             vdwjidx0B        = 2*vdwtype[jnrB+0];
 888             vdwjidx0C        = 2*vdwtype[jnrC+0];
 889             vdwjidx0D        = 2*vdwtype[jnrD+0];
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             if (gmx_mm_any_lt(rsq00,rcutoff2))
 896             {
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 900                                          vdwparam+vdwioffset0+vdwjidx0B,
 901                                          vdwparam+vdwioffset0+vdwjidx0C,
 902                                          vdwparam+vdwioffset0+vdwjidx0D,
 903                                          &c6_00,&c12_00);
 904
 905             /* LENNARD-JONES DISPERSION/REPULSION */
 906
 907             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 908             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 909
 910             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 911
 912             fscal            = fvdw;
 913
 914             fscal            = _mm_and_ps(fscal,cutoff_mask);
 915
 916             /* Calculate temporary vectorial force */
 917             tx               = _mm_mul_ps(fscal,dx00);
 918             ty               = _mm_mul_ps(fscal,dy00);
 919             tz               = _mm_mul_ps(fscal,dz00);
 920
 921             /* Update vectorial force */
 922             fix0             = _mm_add_ps(fix0,tx);
 923             fiy0             = _mm_add_ps(fiy0,ty);
 924             fiz0             = _mm_add_ps(fiz0,tz);
 925
 926             fjptrA             = f+j_coord_offsetA;
 927             fjptrB             = f+j_coord_offsetB;
 928             fjptrC             = f+j_coord_offsetC;
 929             fjptrD             = f+j_coord_offsetD;
 930             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 931
 932             }
 933
 934             /**************************
 935              * CALCULATE INTERACTIONS *
 936              **************************/
 937
 938             if (gmx_mm_any_lt(rsq10,rcutoff2))
 939             {
 940
 941             /* Compute parameters for interactions between i and j atoms */
 942             qq10             = _mm_mul_ps(iq1,jq0);
 943
 944             /* REACTION-FIELD ELECTROSTATICS */
 945             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 946
 947             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 948
 949             fscal            = felec;
 950
 951             fscal            = _mm_and_ps(fscal,cutoff_mask);
 952
 953             /* Calculate temporary vectorial force */
 954             tx               = _mm_mul_ps(fscal,dx10);
 955             ty               = _mm_mul_ps(fscal,dy10);
 956             tz               = _mm_mul_ps(fscal,dz10);
 957
 958             /* Update vectorial force */
 959             fix1             = _mm_add_ps(fix1,tx);
 960             fiy1             = _mm_add_ps(fiy1,ty);
 961             fiz1             = _mm_add_ps(fiz1,tz);
 962
 963             fjptrA             = f+j_coord_offsetA;
 964             fjptrB             = f+j_coord_offsetB;
 965             fjptrC             = f+j_coord_offsetC;
 966             fjptrD             = f+j_coord_offsetD;
 967             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 968
 969             }
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             if (gmx_mm_any_lt(rsq20,rcutoff2))
 976             {
 977
 978             /* Compute parameters for interactions between i and j atoms */
 979             qq20             = _mm_mul_ps(iq2,jq0);
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 983
 984             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 985
 986             fscal            = felec;
 987
 988             fscal            = _mm_and_ps(fscal,cutoff_mask);
 989
 990             /* Calculate temporary vectorial force */
 991             tx               = _mm_mul_ps(fscal,dx20);
 992             ty               = _mm_mul_ps(fscal,dy20);
 993             tz               = _mm_mul_ps(fscal,dz20);
 994
 995             /* Update vectorial force */
 996             fix2             = _mm_add_ps(fix2,tx);
 997             fiy2             = _mm_add_ps(fiy2,ty);
 998             fiz2             = _mm_add_ps(fiz2,tz);
 999
1000             fjptrA             = f+j_coord_offsetA;
1001             fjptrB             = f+j_coord_offsetB;
1002             fjptrC             = f+j_coord_offsetC;
1003             fjptrD             = f+j_coord_offsetD;
1004             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1005
1006             }
1007
1008             /**************************
1009              * CALCULATE INTERACTIONS *
1010              **************************/
1011
1012             if (gmx_mm_any_lt(rsq30,rcutoff2))
1013             {
1014
1015             /* Compute parameters for interactions between i and j atoms */
1016             qq30             = _mm_mul_ps(iq3,jq0);
1017
1018             /* REACTION-FIELD ELECTROSTATICS */
1019             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1020
1021             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1022
1023             fscal            = felec;
1024
1025             fscal            = _mm_and_ps(fscal,cutoff_mask);
1026
1027             /* Calculate temporary vectorial force */
1028             tx               = _mm_mul_ps(fscal,dx30);
1029             ty               = _mm_mul_ps(fscal,dy30);
1030             tz               = _mm_mul_ps(fscal,dz30);
1031
1032             /* Update vectorial force */
1033             fix3             = _mm_add_ps(fix3,tx);
1034             fiy3             = _mm_add_ps(fiy3,ty);
1035             fiz3             = _mm_add_ps(fiz3,tz);
1036
1037             fjptrA             = f+j_coord_offsetA;
1038             fjptrB             = f+j_coord_offsetB;
1039             fjptrC             = f+j_coord_offsetC;
1040             fjptrD             = f+j_coord_offsetD;
1041             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1042
1043             }
1044
1045             /* Inner loop uses 120 flops */
1046         }
1047
1048         if(jidx<j_index_end)
1049         {
1050
1051             /* Get j neighbor index, and coordinate index */
1052             jnrlistA         = jjnr[jidx];
1053             jnrlistB         = jjnr[jidx+1];
1054             jnrlistC         = jjnr[jidx+2];
1055             jnrlistD         = jjnr[jidx+3];
1056             /* Sign of each element will be negative for non-real atoms.
1057              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1058              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1059              */
1060             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1061             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1062             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1063             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1064             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1065             j_coord_offsetA  = DIM*jnrA;
1066             j_coord_offsetB  = DIM*jnrB;
1067             j_coord_offsetC  = DIM*jnrC;
1068             j_coord_offsetD  = DIM*jnrD;
1069
1070             /* load j atom coordinates */
1071             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1072                                               x+j_coord_offsetC,x+j_coord_offsetD,
1073                                               &jx0,&jy0,&jz0);
1074
1075             /* Calculate displacement vector */
1076             dx00             = _mm_sub_ps(ix0,jx0);
1077             dy00             = _mm_sub_ps(iy0,jy0);
1078             dz00             = _mm_sub_ps(iz0,jz0);
1079             dx10             = _mm_sub_ps(ix1,jx0);
1080             dy10             = _mm_sub_ps(iy1,jy0);
1081             dz10             = _mm_sub_ps(iz1,jz0);
1082             dx20             = _mm_sub_ps(ix2,jx0);
1083             dy20             = _mm_sub_ps(iy2,jy0);
1084             dz20             = _mm_sub_ps(iz2,jz0);
1085             dx30             = _mm_sub_ps(ix3,jx0);
1086             dy30             = _mm_sub_ps(iy3,jy0);
1087             dz30             = _mm_sub_ps(iz3,jz0);
1088
1089             /* Calculate squared distance and things based on it */
1090             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1091             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1092             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1093             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1094
1095             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1096             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1097             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1098
1099             rinvsq00         = gmx_mm_inv_ps(rsq00);
1100             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1101             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1102             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1103
1104             /* Load parameters for j particles */
1105             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1106                                                               charge+jnrC+0,charge+jnrD+0);
1107             vdwjidx0A        = 2*vdwtype[jnrA+0];
1108             vdwjidx0B        = 2*vdwtype[jnrB+0];
1109             vdwjidx0C        = 2*vdwtype[jnrC+0];
1110             vdwjidx0D        = 2*vdwtype[jnrD+0];
1111
1112             /**************************
1113              * CALCULATE INTERACTIONS *
1114              **************************/
1115
1116             if (gmx_mm_any_lt(rsq00,rcutoff2))
1117             {
1118
1119             /* Compute parameters for interactions between i and j atoms */
1120             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1121                                          vdwparam+vdwioffset0+vdwjidx0B,
1122                                          vdwparam+vdwioffset0+vdwjidx0C,
1123                                          vdwparam+vdwioffset0+vdwjidx0D,
1124                                          &c6_00,&c12_00);
1125
1126             /* LENNARD-JONES DISPERSION/REPULSION */
1127
1128             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1129             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1130
1131             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1132
1133             fscal            = fvdw;
1134
1135             fscal            = _mm_and_ps(fscal,cutoff_mask);
1136
1137             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_ps(fscal,dx00);
1141             ty               = _mm_mul_ps(fscal,dy00);
1142             tz               = _mm_mul_ps(fscal,dz00);
1143
1144             /* Update vectorial force */
1145             fix0             = _mm_add_ps(fix0,tx);
1146             fiy0             = _mm_add_ps(fiy0,ty);
1147             fiz0             = _mm_add_ps(fiz0,tz);
1148
1149             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1150             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1151             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1152             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1153             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1154
1155             }
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             if (gmx_mm_any_lt(rsq10,rcutoff2))
1162             {
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq10             = _mm_mul_ps(iq1,jq0);
1166
1167             /* REACTION-FIELD ELECTROSTATICS */
1168             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1169
1170             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1171
1172             fscal            = felec;
1173
1174             fscal            = _mm_and_ps(fscal,cutoff_mask);
1175
1176             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1177
1178             /* Calculate temporary vectorial force */
1179             tx               = _mm_mul_ps(fscal,dx10);
1180             ty               = _mm_mul_ps(fscal,dy10);
1181             tz               = _mm_mul_ps(fscal,dz10);
1182
1183             /* Update vectorial force */
1184             fix1             = _mm_add_ps(fix1,tx);
1185             fiy1             = _mm_add_ps(fiy1,ty);
1186             fiz1             = _mm_add_ps(fiz1,tz);
1187
1188             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1189             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1190             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1191             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1192             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1193
1194             }
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             if (gmx_mm_any_lt(rsq20,rcutoff2))
1201             {
1202
1203             /* Compute parameters for interactions between i and j atoms */
1204             qq20             = _mm_mul_ps(iq2,jq0);
1205
1206             /* REACTION-FIELD ELECTROSTATICS */
1207             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1208
1209             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1210
1211             fscal            = felec;
1212
1213             fscal            = _mm_and_ps(fscal,cutoff_mask);
1214
1215             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1216
1217             /* Calculate temporary vectorial force */
1218             tx               = _mm_mul_ps(fscal,dx20);
1219             ty               = _mm_mul_ps(fscal,dy20);
1220             tz               = _mm_mul_ps(fscal,dz20);
1221
1222             /* Update vectorial force */
1223             fix2             = _mm_add_ps(fix2,tx);
1224             fiy2             = _mm_add_ps(fiy2,ty);
1225             fiz2             = _mm_add_ps(fiz2,tz);
1226
1227             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1228             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1229             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1230             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1231             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1232
1233             }
1234
1235             /**************************
1236              * CALCULATE INTERACTIONS *
1237              **************************/
1238
1239             if (gmx_mm_any_lt(rsq30,rcutoff2))
1240             {
1241
1242             /* Compute parameters for interactions between i and j atoms */
1243             qq30             = _mm_mul_ps(iq3,jq0);
1244
1245             /* REACTION-FIELD ELECTROSTATICS */
1246             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1247
1248             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1249
1250             fscal            = felec;
1251
1252             fscal            = _mm_and_ps(fscal,cutoff_mask);
1253
1254             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1255
1256             /* Calculate temporary vectorial force */
1257             tx               = _mm_mul_ps(fscal,dx30);
1258             ty               = _mm_mul_ps(fscal,dy30);
1259             tz               = _mm_mul_ps(fscal,dz30);
1260
1261             /* Update vectorial force */
1262             fix3             = _mm_add_ps(fix3,tx);
1263             fiy3             = _mm_add_ps(fiy3,ty);
1264             fiz3             = _mm_add_ps(fiz3,tz);
1265
1266             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1267             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1268             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1269             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1270             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1271
1272             }
1273
1274             /* Inner loop uses 120 flops */
1275         }
1276
1277         /* End of innermost loop */
1278
1279         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1280                                               f+i_coord_offset,fshift+i_shift_offset);
1281
1282         /* Increment number of inner iterations */
1283         inneriter                  += j_index_end - j_index_start;
1284
1285         /* Outer loop uses 24 flops */
1286     }
1287
1288     /* Increment number of outer iterations */
1289     outeriter        += nri;
1290
1291     /* Update outer/inner flops */
1292
1293     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*120);
1294 }