src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwLJSw_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_VdwLJSw_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_VdwLJSw_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           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  92     real             rswitch_scalar,d_scalar;
  93     __m128           dummy_mask,cutoff_mask;
  94     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  95     __m128           one     = _mm_set1_ps(1.0);
  96     __m128           two     = _mm_set1_ps(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_ps(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     krf              = _mm_set1_ps(fr->ic->k_rf);
 111     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 112     crf              = _mm_set1_ps(fr->ic->c_rf);
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 122     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->rcoulomb;
 126     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 128
 129     rswitch_scalar   = fr->rvdw_switch;
 130     rswitch          = _mm_set1_ps(rswitch_scalar);
 131     /* Setup switch parameters */
 132     d_scalar         = rcutoff_scalar-rswitch_scalar;
 133     d                = _mm_set1_ps(d_scalar);
 134     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 135     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 136     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 137     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 138     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 140
 141     /* Avoid stupid compiler warnings */
 142     jnrA = jnrB = jnrC = jnrD = 0;
 143     j_coord_offsetA = 0;
 144     j_coord_offsetB = 0;
 145     j_coord_offsetC = 0;
 146     j_coord_offsetD = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     for(iidx=0;iidx<4*DIM;iidx++)
 152     {
 153         scratch[iidx] = 0.0;
 154     }
 155
 156     /* Start outer loop over neighborlists */
 157     for(iidx=0; iidx<nri; iidx++)
 158     {
 159         /* Load shift vector for this list */
 160         i_shift_offset   = DIM*shiftidx[iidx];
 161
 162         /* Load limits for loop over neighbors */
 163         j_index_start    = jindex[iidx];
 164         j_index_end      = jindex[iidx+1];
 165
 166         /* Get outer coordinate index */
 167         inr              = iinr[iidx];
 168         i_coord_offset   = DIM*inr;
 169
 170         /* Load i particle coords and add shift vector */
 171         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 172                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 173
 174         fix0             = _mm_setzero_ps();
 175         fiy0             = _mm_setzero_ps();
 176         fiz0             = _mm_setzero_ps();
 177         fix1             = _mm_setzero_ps();
 178         fiy1             = _mm_setzero_ps();
 179         fiz1             = _mm_setzero_ps();
 180         fix2             = _mm_setzero_ps();
 181         fiy2             = _mm_setzero_ps();
 182         fiz2             = _mm_setzero_ps();
 183         fix3             = _mm_setzero_ps();
 184         fiy3             = _mm_setzero_ps();
 185         fiz3             = _mm_setzero_ps();
 186
 187         /* Reset potential sums */
 188         velecsum         = _mm_setzero_ps();
 189         vvdwsum          = _mm_setzero_ps();
 190
 191         /* Start inner kernel loop */
 192         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 193         {
 194
 195             /* Get j neighbor index, and coordinate index */
 196             jnrA             = jjnr[jidx];
 197             jnrB             = jjnr[jidx+1];
 198             jnrC             = jjnr[jidx+2];
 199             jnrD             = jjnr[jidx+3];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202             j_coord_offsetC  = DIM*jnrC;
 203             j_coord_offsetD  = DIM*jnrD;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               x+j_coord_offsetC,x+j_coord_offsetD,
 208                                               &jx0,&jy0,&jz0);
 209
 210             /* Calculate displacement vector */
 211             dx00             = _mm_sub_ps(ix0,jx0);
 212             dy00             = _mm_sub_ps(iy0,jy0);
 213             dz00             = _mm_sub_ps(iz0,jz0);
 214             dx10             = _mm_sub_ps(ix1,jx0);
 215             dy10             = _mm_sub_ps(iy1,jy0);
 216             dz10             = _mm_sub_ps(iz1,jz0);
 217             dx20             = _mm_sub_ps(ix2,jx0);
 218             dy20             = _mm_sub_ps(iy2,jy0);
 219             dz20             = _mm_sub_ps(iz2,jz0);
 220             dx30             = _mm_sub_ps(ix3,jx0);
 221             dy30             = _mm_sub_ps(iy3,jy0);
 222             dz30             = _mm_sub_ps(iz3,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 229
 230             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 231             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 232             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 233             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 234
 235             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 236             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 237             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 238             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 239
 240             /* Load parameters for j particles */
 241             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 242                                                               charge+jnrC+0,charge+jnrD+0);
 243             vdwjidx0A        = 2*vdwtype[jnrA+0];
 244             vdwjidx0B        = 2*vdwtype[jnrB+0];
 245             vdwjidx0C        = 2*vdwtype[jnrC+0];
 246             vdwjidx0D        = 2*vdwtype[jnrD+0];
 247
 248             fjx0             = _mm_setzero_ps();
 249             fjy0             = _mm_setzero_ps();
 250             fjz0             = _mm_setzero_ps();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             if (gmx_mm_any_lt(rsq00,rcutoff2))
 257             {
 258
 259             r00              = _mm_mul_ps(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 263                                          vdwparam+vdwioffset0+vdwjidx0B,
 264                                          vdwparam+vdwioffset0+vdwjidx0C,
 265                                          vdwparam+vdwioffset0+vdwjidx0D,
 266                                          &c6_00,&c12_00);
 267
 268             /* LENNARD-JONES DISPERSION/REPULSION */
 269
 270             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 271             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 272             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 273             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 274             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 275
 276             d                = _mm_sub_ps(r00,rswitch);
 277             d                = _mm_max_ps(d,_mm_setzero_ps());
 278             d2               = _mm_mul_ps(d,d);
 279             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 280
 281             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 282
 283             /* Evaluate switch function */
 284             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 285             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 286             vvdw             = _mm_mul_ps(vvdw,sw);
 287             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 291             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 292
 293             fscal            = fvdw;
 294
 295             fscal            = _mm_and_ps(fscal,cutoff_mask);
 296
 297             /* Calculate temporary vectorial force */
 298             tx               = _mm_mul_ps(fscal,dx00);
 299             ty               = _mm_mul_ps(fscal,dy00);
 300             tz               = _mm_mul_ps(fscal,dz00);
 301
 302             /* Update vectorial force */
 303             fix0             = _mm_add_ps(fix0,tx);
 304             fiy0             = _mm_add_ps(fiy0,ty);
 305             fiz0             = _mm_add_ps(fiz0,tz);
 306
 307             fjx0             = _mm_add_ps(fjx0,tx);
 308             fjy0             = _mm_add_ps(fjy0,ty);
 309             fjz0             = _mm_add_ps(fjz0,tz);
 310
 311             }
 312
 313             /**************************
 314              * CALCULATE INTERACTIONS *
 315              **************************/
 316
 317             if (gmx_mm_any_lt(rsq10,rcutoff2))
 318             {
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq10             = _mm_mul_ps(iq1,jq0);
 322
 323             /* REACTION-FIELD ELECTROSTATICS */
 324             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 325             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 326
 327             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velec            = _mm_and_ps(velec,cutoff_mask);
 331             velecsum         = _mm_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             fscal            = _mm_and_ps(fscal,cutoff_mask);
 336
 337             /* Calculate temporary vectorial force */
 338             tx               = _mm_mul_ps(fscal,dx10);
 339             ty               = _mm_mul_ps(fscal,dy10);
 340             tz               = _mm_mul_ps(fscal,dz10);
 341
 342             /* Update vectorial force */
 343             fix1             = _mm_add_ps(fix1,tx);
 344             fiy1             = _mm_add_ps(fiy1,ty);
 345             fiz1             = _mm_add_ps(fiz1,tz);
 346
 347             fjx0             = _mm_add_ps(fjx0,tx);
 348             fjy0             = _mm_add_ps(fjy0,ty);
 349             fjz0             = _mm_add_ps(fjz0,tz);
 350
 351             }
 352
 353             /**************************
 354              * CALCULATE INTERACTIONS *
 355              **************************/
 356
 357             if (gmx_mm_any_lt(rsq20,rcutoff2))
 358             {
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq20             = _mm_mul_ps(iq2,jq0);
 362
 363             /* REACTION-FIELD ELECTROSTATICS */
 364             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 365             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 366
 367             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velec            = _mm_and_ps(velec,cutoff_mask);
 371             velecsum         = _mm_add_ps(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375             fscal            = _mm_and_ps(fscal,cutoff_mask);
 376
 377             /* Calculate temporary vectorial force */
 378             tx               = _mm_mul_ps(fscal,dx20);
 379             ty               = _mm_mul_ps(fscal,dy20);
 380             tz               = _mm_mul_ps(fscal,dz20);
 381
 382             /* Update vectorial force */
 383             fix2             = _mm_add_ps(fix2,tx);
 384             fiy2             = _mm_add_ps(fiy2,ty);
 385             fiz2             = _mm_add_ps(fiz2,tz);
 386
 387             fjx0             = _mm_add_ps(fjx0,tx);
 388             fjy0             = _mm_add_ps(fjy0,ty);
 389             fjz0             = _mm_add_ps(fjz0,tz);
 390
 391             }
 392
 393             /**************************
 394              * CALCULATE INTERACTIONS *
 395              **************************/
 396
 397             if (gmx_mm_any_lt(rsq30,rcutoff2))
 398             {
 399
 400             /* Compute parameters for interactions between i and j atoms */
 401             qq30             = _mm_mul_ps(iq3,jq0);
 402
 403             /* REACTION-FIELD ELECTROSTATICS */
 404             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 405             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 406
 407             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_and_ps(velec,cutoff_mask);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_and_ps(fscal,cutoff_mask);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_ps(fscal,dx30);
 419             ty               = _mm_mul_ps(fscal,dy30);
 420             tz               = _mm_mul_ps(fscal,dz30);
 421
 422             /* Update vectorial force */
 423             fix3             = _mm_add_ps(fix3,tx);
 424             fiy3             = _mm_add_ps(fiy3,ty);
 425             fiz3             = _mm_add_ps(fiz3,tz);
 426
 427             fjx0             = _mm_add_ps(fjx0,tx);
 428             fjy0             = _mm_add_ps(fjy0,ty);
 429             fjz0             = _mm_add_ps(fjz0,tz);
 430
 431             }
 432
 433             fjptrA             = f+j_coord_offsetA;
 434             fjptrB             = f+j_coord_offsetB;
 435             fjptrC             = f+j_coord_offsetC;
 436             fjptrD             = f+j_coord_offsetD;
 437
 438             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 439
 440             /* Inner loop uses 167 flops */
 441         }
 442
 443         if(jidx<j_index_end)
 444         {
 445
 446             /* Get j neighbor index, and coordinate index */
 447             jnrlistA         = jjnr[jidx];
 448             jnrlistB         = jjnr[jidx+1];
 449             jnrlistC         = jjnr[jidx+2];
 450             jnrlistD         = jjnr[jidx+3];
 451             /* Sign of each element will be negative for non-real atoms.
 452              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 453              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 454              */
 455             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 456             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 457             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 458             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 459             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 460             j_coord_offsetA  = DIM*jnrA;
 461             j_coord_offsetB  = DIM*jnrB;
 462             j_coord_offsetC  = DIM*jnrC;
 463             j_coord_offsetD  = DIM*jnrD;
 464
 465             /* load j atom coordinates */
 466             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                               x+j_coord_offsetC,x+j_coord_offsetD,
 468                                               &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm_sub_ps(ix0,jx0);
 472             dy00             = _mm_sub_ps(iy0,jy0);
 473             dz00             = _mm_sub_ps(iz0,jz0);
 474             dx10             = _mm_sub_ps(ix1,jx0);
 475             dy10             = _mm_sub_ps(iy1,jy0);
 476             dz10             = _mm_sub_ps(iz1,jz0);
 477             dx20             = _mm_sub_ps(ix2,jx0);
 478             dy20             = _mm_sub_ps(iy2,jy0);
 479             dz20             = _mm_sub_ps(iz2,jz0);
 480             dx30             = _mm_sub_ps(ix3,jx0);
 481             dy30             = _mm_sub_ps(iy3,jy0);
 482             dz30             = _mm_sub_ps(iz3,jz0);
 483
 484             /* Calculate squared distance and things based on it */
 485             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 486             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 487             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 488             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 489
 490             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 491             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 492             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 493             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 494
 495             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 496             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 497             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 498             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 499
 500             /* Load parameters for j particles */
 501             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 502                                                               charge+jnrC+0,charge+jnrD+0);
 503             vdwjidx0A        = 2*vdwtype[jnrA+0];
 504             vdwjidx0B        = 2*vdwtype[jnrB+0];
 505             vdwjidx0C        = 2*vdwtype[jnrC+0];
 506             vdwjidx0D        = 2*vdwtype[jnrD+0];
 507
 508             fjx0             = _mm_setzero_ps();
 509             fjy0             = _mm_setzero_ps();
 510             fjz0             = _mm_setzero_ps();
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             if (gmx_mm_any_lt(rsq00,rcutoff2))
 517             {
 518
 519             r00              = _mm_mul_ps(rsq00,rinv00);
 520             r00              = _mm_andnot_ps(dummy_mask,r00);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 524                                          vdwparam+vdwioffset0+vdwjidx0B,
 525                                          vdwparam+vdwioffset0+vdwjidx0C,
 526                                          vdwparam+vdwioffset0+vdwjidx0D,
 527                                          &c6_00,&c12_00);
 528
 529             /* LENNARD-JONES DISPERSION/REPULSION */
 530
 531             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 532             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 533             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 534             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 535             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 536
 537             d                = _mm_sub_ps(r00,rswitch);
 538             d                = _mm_max_ps(d,_mm_setzero_ps());
 539             d2               = _mm_mul_ps(d,d);
 540             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 541
 542             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 543
 544             /* Evaluate switch function */
 545             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 546             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 547             vvdw             = _mm_mul_ps(vvdw,sw);
 548             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 549
 550             /* Update potential sum for this i atom from the interaction with this j atom. */
 551             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 552             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 553             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 554
 555             fscal            = fvdw;
 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,dx00);
 563             ty               = _mm_mul_ps(fscal,dy00);
 564             tz               = _mm_mul_ps(fscal,dz00);
 565
 566             /* Update vectorial force */
 567             fix0             = _mm_add_ps(fix0,tx);
 568             fiy0             = _mm_add_ps(fiy0,ty);
 569             fiz0             = _mm_add_ps(fiz0,tz);
 570
 571             fjx0             = _mm_add_ps(fjx0,tx);
 572             fjy0             = _mm_add_ps(fjy0,ty);
 573             fjz0             = _mm_add_ps(fjz0,tz);
 574
 575             }
 576
 577             /**************************
 578              * CALCULATE INTERACTIONS *
 579              **************************/
 580
 581             if (gmx_mm_any_lt(rsq10,rcutoff2))
 582             {
 583
 584             /* Compute parameters for interactions between i and j atoms */
 585             qq10             = _mm_mul_ps(iq1,jq0);
 586
 587             /* REACTION-FIELD ELECTROSTATICS */
 588             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 589             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 590
 591             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 592
 593             /* Update potential sum for this i atom from the interaction with this j atom. */
 594             velec            = _mm_and_ps(velec,cutoff_mask);
 595             velec            = _mm_andnot_ps(dummy_mask,velec);
 596             velecsum         = _mm_add_ps(velecsum,velec);
 597
 598             fscal            = felec;
 599
 600             fscal            = _mm_and_ps(fscal,cutoff_mask);
 601
 602             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 603
 604             /* Calculate temporary vectorial force */
 605             tx               = _mm_mul_ps(fscal,dx10);
 606             ty               = _mm_mul_ps(fscal,dy10);
 607             tz               = _mm_mul_ps(fscal,dz10);
 608
 609             /* Update vectorial force */
 610             fix1             = _mm_add_ps(fix1,tx);
 611             fiy1             = _mm_add_ps(fiy1,ty);
 612             fiz1             = _mm_add_ps(fiz1,tz);
 613
 614             fjx0             = _mm_add_ps(fjx0,tx);
 615             fjy0             = _mm_add_ps(fjy0,ty);
 616             fjz0             = _mm_add_ps(fjz0,tz);
 617
 618             }
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             if (gmx_mm_any_lt(rsq20,rcutoff2))
 625             {
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq20             = _mm_mul_ps(iq2,jq0);
 629
 630             /* REACTION-FIELD ELECTROSTATICS */
 631             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 632             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 633
 634             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 635
 636             /* Update potential sum for this i atom from the interaction with this j atom. */
 637             velec            = _mm_and_ps(velec,cutoff_mask);
 638             velec            = _mm_andnot_ps(dummy_mask,velec);
 639             velecsum         = _mm_add_ps(velecsum,velec);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm_and_ps(fscal,cutoff_mask);
 644
 645             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm_mul_ps(fscal,dx20);
 649             ty               = _mm_mul_ps(fscal,dy20);
 650             tz               = _mm_mul_ps(fscal,dz20);
 651
 652             /* Update vectorial force */
 653             fix2             = _mm_add_ps(fix2,tx);
 654             fiy2             = _mm_add_ps(fiy2,ty);
 655             fiz2             = _mm_add_ps(fiz2,tz);
 656
 657             fjx0             = _mm_add_ps(fjx0,tx);
 658             fjy0             = _mm_add_ps(fjy0,ty);
 659             fjz0             = _mm_add_ps(fjz0,tz);
 660
 661             }
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             if (gmx_mm_any_lt(rsq30,rcutoff2))
 668             {
 669
 670             /* Compute parameters for interactions between i and j atoms */
 671             qq30             = _mm_mul_ps(iq3,jq0);
 672
 673             /* REACTION-FIELD ELECTROSTATICS */
 674             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 675             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 676
 677             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 678
 679             /* Update potential sum for this i atom from the interaction with this j atom. */
 680             velec            = _mm_and_ps(velec,cutoff_mask);
 681             velec            = _mm_andnot_ps(dummy_mask,velec);
 682             velecsum         = _mm_add_ps(velecsum,velec);
 683
 684             fscal            = felec;
 685
 686             fscal            = _mm_and_ps(fscal,cutoff_mask);
 687
 688             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 689
 690             /* Calculate temporary vectorial force */
 691             tx               = _mm_mul_ps(fscal,dx30);
 692             ty               = _mm_mul_ps(fscal,dy30);
 693             tz               = _mm_mul_ps(fscal,dz30);
 694
 695             /* Update vectorial force */
 696             fix3             = _mm_add_ps(fix3,tx);
 697             fiy3             = _mm_add_ps(fiy3,ty);
 698             fiz3             = _mm_add_ps(fiz3,tz);
 699
 700             fjx0             = _mm_add_ps(fjx0,tx);
 701             fjy0             = _mm_add_ps(fjy0,ty);
 702             fjz0             = _mm_add_ps(fjz0,tz);
 703
 704             }
 705
 706             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 707             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 708             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 709             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 710
 711             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 712
 713             /* Inner loop uses 168 flops */
 714         }
 715
 716         /* End of innermost loop */
 717
 718         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 719                                               f+i_coord_offset,fshift+i_shift_offset);
 720
 721         ggid                        = gid[iidx];
 722         /* Update potential energies */
 723         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 724         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 725
 726         /* Increment number of inner iterations */
 727         inneriter                  += j_index_end - j_index_start;
 728
 729         /* Outer loop uses 26 flops */
 730     }
 731
 732     /* Increment number of outer iterations */
 733     outeriter        += nri;
 734
 735     /* Update outer/inner flops */
 736
 737     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*168);
 738 }
 739 /*
 740  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse2_single
 741  * Electrostatics interaction: ReactionField
 742  * VdW interaction:            LennardJones
 743  * Geometry:                   Water4-Particle
 744  * Calculate force/pot:        Force
 745  */
 746 void
 747 nb_kernel_ElecRFCut_VdwLJSw_GeomW4P1_F_sse2_single
 748                     (t_nblist * gmx_restrict                nlist,
 749                      rvec * gmx_restrict                    xx,
 750                      rvec * gmx_restrict                    ff,
 751                      t_forcerec * gmx_restrict              fr,
 752                      t_mdatoms * gmx_restrict               mdatoms,
 753                      nb_kernel_data_t * gmx_restrict        kernel_data,
 754                      t_nrnb * gmx_restrict                  nrnb)
 755 {
 756     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 757      * just 0 for non-waters.
 758      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 759      * jnr indices corresponding to data put in the four positions in the SIMD register.
 760      */
 761     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 762     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 763     int              jnrA,jnrB,jnrC,jnrD;
 764     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 765     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 766     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 767     real             rcutoff_scalar;
 768     real             *shiftvec,*fshift,*x,*f;
 769     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 770     real             scratch[4*DIM];
 771     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 772     int              vdwioffset0;
 773     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 774     int              vdwioffset1;
 775     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 776     int              vdwioffset2;
 777     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 778     int              vdwioffset3;
 779     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 780     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 781     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 782     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 783     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 784     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 785     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 786     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 787     real             *charge;
 788     int              nvdwtype;
 789     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 790     int              *vdwtype;
 791     real             *vdwparam;
 792     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 793     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 794     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 795     real             rswitch_scalar,d_scalar;
 796     __m128           dummy_mask,cutoff_mask;
 797     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 798     __m128           one     = _mm_set1_ps(1.0);
 799     __m128           two     = _mm_set1_ps(2.0);
 800     x                = xx[0];
 801     f                = ff[0];
 802
 803     nri              = nlist->nri;
 804     iinr             = nlist->iinr;
 805     jindex           = nlist->jindex;
 806     jjnr             = nlist->jjnr;
 807     shiftidx         = nlist->shift;
 808     gid              = nlist->gid;
 809     shiftvec         = fr->shift_vec[0];
 810     fshift           = fr->fshift[0];
 811     facel            = _mm_set1_ps(fr->epsfac);
 812     charge           = mdatoms->chargeA;
 813     krf              = _mm_set1_ps(fr->ic->k_rf);
 814     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 815     crf              = _mm_set1_ps(fr->ic->c_rf);
 816     nvdwtype         = fr->ntype;
 817     vdwparam         = fr->nbfp;
 818     vdwtype          = mdatoms->typeA;
 819
 820     /* Setup water-specific parameters */
 821     inr              = nlist->iinr[0];
 822     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 823     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 824     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 825     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 826
 827     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 828     rcutoff_scalar   = fr->rcoulomb;
 829     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 830     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 831
 832     rswitch_scalar   = fr->rvdw_switch;
 833     rswitch          = _mm_set1_ps(rswitch_scalar);
 834     /* Setup switch parameters */
 835     d_scalar         = rcutoff_scalar-rswitch_scalar;
 836     d                = _mm_set1_ps(d_scalar);
 837     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 838     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 839     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 840     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 841     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 842     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 843
 844     /* Avoid stupid compiler warnings */
 845     jnrA = jnrB = jnrC = jnrD = 0;
 846     j_coord_offsetA = 0;
 847     j_coord_offsetB = 0;
 848     j_coord_offsetC = 0;
 849     j_coord_offsetD = 0;
 850
 851     outeriter        = 0;
 852     inneriter        = 0;
 853
 854     for(iidx=0;iidx<4*DIM;iidx++)
 855     {
 856         scratch[iidx] = 0.0;
 857     }
 858
 859     /* Start outer loop over neighborlists */
 860     for(iidx=0; iidx<nri; iidx++)
 861     {
 862         /* Load shift vector for this list */
 863         i_shift_offset   = DIM*shiftidx[iidx];
 864
 865         /* Load limits for loop over neighbors */
 866         j_index_start    = jindex[iidx];
 867         j_index_end      = jindex[iidx+1];
 868
 869         /* Get outer coordinate index */
 870         inr              = iinr[iidx];
 871         i_coord_offset   = DIM*inr;
 872
 873         /* Load i particle coords and add shift vector */
 874         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 875                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 876
 877         fix0             = _mm_setzero_ps();
 878         fiy0             = _mm_setzero_ps();
 879         fiz0             = _mm_setzero_ps();
 880         fix1             = _mm_setzero_ps();
 881         fiy1             = _mm_setzero_ps();
 882         fiz1             = _mm_setzero_ps();
 883         fix2             = _mm_setzero_ps();
 884         fiy2             = _mm_setzero_ps();
 885         fiz2             = _mm_setzero_ps();
 886         fix3             = _mm_setzero_ps();
 887         fiy3             = _mm_setzero_ps();
 888         fiz3             = _mm_setzero_ps();
 889
 890         /* Start inner kernel loop */
 891         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 892         {
 893
 894             /* Get j neighbor index, and coordinate index */
 895             jnrA             = jjnr[jidx];
 896             jnrB             = jjnr[jidx+1];
 897             jnrC             = jjnr[jidx+2];
 898             jnrD             = jjnr[jidx+3];
 899             j_coord_offsetA  = DIM*jnrA;
 900             j_coord_offsetB  = DIM*jnrB;
 901             j_coord_offsetC  = DIM*jnrC;
 902             j_coord_offsetD  = DIM*jnrD;
 903
 904             /* load j atom coordinates */
 905             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 906                                               x+j_coord_offsetC,x+j_coord_offsetD,
 907                                               &jx0,&jy0,&jz0);
 908
 909             /* Calculate displacement vector */
 910             dx00             = _mm_sub_ps(ix0,jx0);
 911             dy00             = _mm_sub_ps(iy0,jy0);
 912             dz00             = _mm_sub_ps(iz0,jz0);
 913             dx10             = _mm_sub_ps(ix1,jx0);
 914             dy10             = _mm_sub_ps(iy1,jy0);
 915             dz10             = _mm_sub_ps(iz1,jz0);
 916             dx20             = _mm_sub_ps(ix2,jx0);
 917             dy20             = _mm_sub_ps(iy2,jy0);
 918             dz20             = _mm_sub_ps(iz2,jz0);
 919             dx30             = _mm_sub_ps(ix3,jx0);
 920             dy30             = _mm_sub_ps(iy3,jy0);
 921             dz30             = _mm_sub_ps(iz3,jz0);
 922
 923             /* Calculate squared distance and things based on it */
 924             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 925             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 926             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 927             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 928
 929             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 930             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 931             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 932             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 933
 934             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 935             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 936             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 937             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 938
 939             /* Load parameters for j particles */
 940             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 941                                                               charge+jnrC+0,charge+jnrD+0);
 942             vdwjidx0A        = 2*vdwtype[jnrA+0];
 943             vdwjidx0B        = 2*vdwtype[jnrB+0];
 944             vdwjidx0C        = 2*vdwtype[jnrC+0];
 945             vdwjidx0D        = 2*vdwtype[jnrD+0];
 946
 947             fjx0             = _mm_setzero_ps();
 948             fjy0             = _mm_setzero_ps();
 949             fjz0             = _mm_setzero_ps();
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             if (gmx_mm_any_lt(rsq00,rcutoff2))
 956             {
 957
 958             r00              = _mm_mul_ps(rsq00,rinv00);
 959
 960             /* Compute parameters for interactions between i and j atoms */
 961             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 962                                          vdwparam+vdwioffset0+vdwjidx0B,
 963                                          vdwparam+vdwioffset0+vdwjidx0C,
 964                                          vdwparam+vdwioffset0+vdwjidx0D,
 965                                          &c6_00,&c12_00);
 966
 967             /* LENNARD-JONES DISPERSION/REPULSION */
 968
 969             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 970             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 971             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 972             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 973             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 974
 975             d                = _mm_sub_ps(r00,rswitch);
 976             d                = _mm_max_ps(d,_mm_setzero_ps());
 977             d2               = _mm_mul_ps(d,d);
 978             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 979
 980             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 981
 982             /* Evaluate switch function */
 983             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 984             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
 985             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 986
 987             fscal            = fvdw;
 988
 989             fscal            = _mm_and_ps(fscal,cutoff_mask);
 990
 991             /* Calculate temporary vectorial force */
 992             tx               = _mm_mul_ps(fscal,dx00);
 993             ty               = _mm_mul_ps(fscal,dy00);
 994             tz               = _mm_mul_ps(fscal,dz00);
 995
 996             /* Update vectorial force */
 997             fix0             = _mm_add_ps(fix0,tx);
 998             fiy0             = _mm_add_ps(fiy0,ty);
 999             fiz0             = _mm_add_ps(fiz0,tz);
1000
1001             fjx0             = _mm_add_ps(fjx0,tx);
1002             fjy0             = _mm_add_ps(fjy0,ty);
1003             fjz0             = _mm_add_ps(fjz0,tz);
1004
1005             }
1006
1007             /**************************
1008              * CALCULATE INTERACTIONS *
1009              **************************/
1010
1011             if (gmx_mm_any_lt(rsq10,rcutoff2))
1012             {
1013
1014             /* Compute parameters for interactions between i and j atoms */
1015             qq10             = _mm_mul_ps(iq1,jq0);
1016
1017             /* REACTION-FIELD ELECTROSTATICS */
1018             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1019
1020             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm_and_ps(fscal,cutoff_mask);
1025
1026             /* Calculate temporary vectorial force */
1027             tx               = _mm_mul_ps(fscal,dx10);
1028             ty               = _mm_mul_ps(fscal,dy10);
1029             tz               = _mm_mul_ps(fscal,dz10);
1030
1031             /* Update vectorial force */
1032             fix1             = _mm_add_ps(fix1,tx);
1033             fiy1             = _mm_add_ps(fiy1,ty);
1034             fiz1             = _mm_add_ps(fiz1,tz);
1035
1036             fjx0             = _mm_add_ps(fjx0,tx);
1037             fjy0             = _mm_add_ps(fjy0,ty);
1038             fjz0             = _mm_add_ps(fjz0,tz);
1039
1040             }
1041
1042             /**************************
1043              * CALCULATE INTERACTIONS *
1044              **************************/
1045
1046             if (gmx_mm_any_lt(rsq20,rcutoff2))
1047             {
1048
1049             /* Compute parameters for interactions between i and j atoms */
1050             qq20             = _mm_mul_ps(iq2,jq0);
1051
1052             /* REACTION-FIELD ELECTROSTATICS */
1053             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1054
1055             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1056
1057             fscal            = felec;
1058
1059             fscal            = _mm_and_ps(fscal,cutoff_mask);
1060
1061             /* Calculate temporary vectorial force */
1062             tx               = _mm_mul_ps(fscal,dx20);
1063             ty               = _mm_mul_ps(fscal,dy20);
1064             tz               = _mm_mul_ps(fscal,dz20);
1065
1066             /* Update vectorial force */
1067             fix2             = _mm_add_ps(fix2,tx);
1068             fiy2             = _mm_add_ps(fiy2,ty);
1069             fiz2             = _mm_add_ps(fiz2,tz);
1070
1071             fjx0             = _mm_add_ps(fjx0,tx);
1072             fjy0             = _mm_add_ps(fjy0,ty);
1073             fjz0             = _mm_add_ps(fjz0,tz);
1074
1075             }
1076
1077             /**************************
1078              * CALCULATE INTERACTIONS *
1079              **************************/
1080
1081             if (gmx_mm_any_lt(rsq30,rcutoff2))
1082             {
1083
1084             /* Compute parameters for interactions between i and j atoms */
1085             qq30             = _mm_mul_ps(iq3,jq0);
1086
1087             /* REACTION-FIELD ELECTROSTATICS */
1088             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1089
1090             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm_and_ps(fscal,cutoff_mask);
1095
1096             /* Calculate temporary vectorial force */
1097             tx               = _mm_mul_ps(fscal,dx30);
1098             ty               = _mm_mul_ps(fscal,dy30);
1099             tz               = _mm_mul_ps(fscal,dz30);
1100
1101             /* Update vectorial force */
1102             fix3             = _mm_add_ps(fix3,tx);
1103             fiy3             = _mm_add_ps(fiy3,ty);
1104             fiz3             = _mm_add_ps(fiz3,tz);
1105
1106             fjx0             = _mm_add_ps(fjx0,tx);
1107             fjy0             = _mm_add_ps(fjy0,ty);
1108             fjz0             = _mm_add_ps(fjz0,tz);
1109
1110             }
1111
1112             fjptrA             = f+j_coord_offsetA;
1113             fjptrB             = f+j_coord_offsetB;
1114             fjptrC             = f+j_coord_offsetC;
1115             fjptrD             = f+j_coord_offsetD;
1116
1117             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1118
1119             /* Inner loop uses 146 flops */
1120         }
1121
1122         if(jidx<j_index_end)
1123         {
1124
1125             /* Get j neighbor index, and coordinate index */
1126             jnrlistA         = jjnr[jidx];
1127             jnrlistB         = jjnr[jidx+1];
1128             jnrlistC         = jjnr[jidx+2];
1129             jnrlistD         = jjnr[jidx+3];
1130             /* Sign of each element will be negative for non-real atoms.
1131              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1132              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1133              */
1134             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1135             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1136             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1137             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1138             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1139             j_coord_offsetA  = DIM*jnrA;
1140             j_coord_offsetB  = DIM*jnrB;
1141             j_coord_offsetC  = DIM*jnrC;
1142             j_coord_offsetD  = DIM*jnrD;
1143
1144             /* load j atom coordinates */
1145             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1146                                               x+j_coord_offsetC,x+j_coord_offsetD,
1147                                               &jx0,&jy0,&jz0);
1148
1149             /* Calculate displacement vector */
1150             dx00             = _mm_sub_ps(ix0,jx0);
1151             dy00             = _mm_sub_ps(iy0,jy0);
1152             dz00             = _mm_sub_ps(iz0,jz0);
1153             dx10             = _mm_sub_ps(ix1,jx0);
1154             dy10             = _mm_sub_ps(iy1,jy0);
1155             dz10             = _mm_sub_ps(iz1,jz0);
1156             dx20             = _mm_sub_ps(ix2,jx0);
1157             dy20             = _mm_sub_ps(iy2,jy0);
1158             dz20             = _mm_sub_ps(iz2,jz0);
1159             dx30             = _mm_sub_ps(ix3,jx0);
1160             dy30             = _mm_sub_ps(iy3,jy0);
1161             dz30             = _mm_sub_ps(iz3,jz0);
1162
1163             /* Calculate squared distance and things based on it */
1164             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1165             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1166             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1167             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1168
1169             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1170             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1171             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1172             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1173
1174             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1175             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1176             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1177             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1178
1179             /* Load parameters for j particles */
1180             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1181                                                               charge+jnrC+0,charge+jnrD+0);
1182             vdwjidx0A        = 2*vdwtype[jnrA+0];
1183             vdwjidx0B        = 2*vdwtype[jnrB+0];
1184             vdwjidx0C        = 2*vdwtype[jnrC+0];
1185             vdwjidx0D        = 2*vdwtype[jnrD+0];
1186
1187             fjx0             = _mm_setzero_ps();
1188             fjy0             = _mm_setzero_ps();
1189             fjz0             = _mm_setzero_ps();
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             if (gmx_mm_any_lt(rsq00,rcutoff2))
1196             {
1197
1198             r00              = _mm_mul_ps(rsq00,rinv00);
1199             r00              = _mm_andnot_ps(dummy_mask,r00);
1200
1201             /* Compute parameters for interactions between i and j atoms */
1202             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1203                                          vdwparam+vdwioffset0+vdwjidx0B,
1204                                          vdwparam+vdwioffset0+vdwjidx0C,
1205                                          vdwparam+vdwioffset0+vdwjidx0D,
1206                                          &c6_00,&c12_00);
1207
1208             /* LENNARD-JONES DISPERSION/REPULSION */
1209
1210             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1211             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
1212             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
1213             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
1214             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
1215
1216             d                = _mm_sub_ps(r00,rswitch);
1217             d                = _mm_max_ps(d,_mm_setzero_ps());
1218             d2               = _mm_mul_ps(d,d);
1219             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1220
1221             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1222
1223             /* Evaluate switch function */
1224             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1225             fvdw             = _mm_sub_ps( _mm_mul_ps(fvdw,sw) , _mm_mul_ps(rinv00,_mm_mul_ps(vvdw,dsw)) );
1226             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1227
1228             fscal            = fvdw;
1229
1230             fscal            = _mm_and_ps(fscal,cutoff_mask);
1231
1232             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1233
1234             /* Calculate temporary vectorial force */
1235             tx               = _mm_mul_ps(fscal,dx00);
1236             ty               = _mm_mul_ps(fscal,dy00);
1237             tz               = _mm_mul_ps(fscal,dz00);
1238
1239             /* Update vectorial force */
1240             fix0             = _mm_add_ps(fix0,tx);
1241             fiy0             = _mm_add_ps(fiy0,ty);
1242             fiz0             = _mm_add_ps(fiz0,tz);
1243
1244             fjx0             = _mm_add_ps(fjx0,tx);
1245             fjy0             = _mm_add_ps(fjy0,ty);
1246             fjz0             = _mm_add_ps(fjz0,tz);
1247
1248             }
1249
1250             /**************************
1251              * CALCULATE INTERACTIONS *
1252              **************************/
1253
1254             if (gmx_mm_any_lt(rsq10,rcutoff2))
1255             {
1256
1257             /* Compute parameters for interactions between i and j atoms */
1258             qq10             = _mm_mul_ps(iq1,jq0);
1259
1260             /* REACTION-FIELD ELECTROSTATICS */
1261             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1262
1263             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1264
1265             fscal            = felec;
1266
1267             fscal            = _mm_and_ps(fscal,cutoff_mask);
1268
1269             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1270
1271             /* Calculate temporary vectorial force */
1272             tx               = _mm_mul_ps(fscal,dx10);
1273             ty               = _mm_mul_ps(fscal,dy10);
1274             tz               = _mm_mul_ps(fscal,dz10);
1275
1276             /* Update vectorial force */
1277             fix1             = _mm_add_ps(fix1,tx);
1278             fiy1             = _mm_add_ps(fiy1,ty);
1279             fiz1             = _mm_add_ps(fiz1,tz);
1280
1281             fjx0             = _mm_add_ps(fjx0,tx);
1282             fjy0             = _mm_add_ps(fjy0,ty);
1283             fjz0             = _mm_add_ps(fjz0,tz);
1284
1285             }
1286
1287             /**************************
1288              * CALCULATE INTERACTIONS *
1289              **************************/
1290
1291             if (gmx_mm_any_lt(rsq20,rcutoff2))
1292             {
1293
1294             /* Compute parameters for interactions between i and j atoms */
1295             qq20             = _mm_mul_ps(iq2,jq0);
1296
1297             /* REACTION-FIELD ELECTROSTATICS */
1298             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1299
1300             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm_and_ps(fscal,cutoff_mask);
1305
1306             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1307
1308             /* Calculate temporary vectorial force */
1309             tx               = _mm_mul_ps(fscal,dx20);
1310             ty               = _mm_mul_ps(fscal,dy20);
1311             tz               = _mm_mul_ps(fscal,dz20);
1312
1313             /* Update vectorial force */
1314             fix2             = _mm_add_ps(fix2,tx);
1315             fiy2             = _mm_add_ps(fiy2,ty);
1316             fiz2             = _mm_add_ps(fiz2,tz);
1317
1318             fjx0             = _mm_add_ps(fjx0,tx);
1319             fjy0             = _mm_add_ps(fjy0,ty);
1320             fjz0             = _mm_add_ps(fjz0,tz);
1321
1322             }
1323
1324             /**************************
1325              * CALCULATE INTERACTIONS *
1326              **************************/
1327
1328             if (gmx_mm_any_lt(rsq30,rcutoff2))
1329             {
1330
1331             /* Compute parameters for interactions between i and j atoms */
1332             qq30             = _mm_mul_ps(iq3,jq0);
1333
1334             /* REACTION-FIELD ELECTROSTATICS */
1335             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1336
1337             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1338
1339             fscal            = felec;
1340
1341             fscal            = _mm_and_ps(fscal,cutoff_mask);
1342
1343             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1344
1345             /* Calculate temporary vectorial force */
1346             tx               = _mm_mul_ps(fscal,dx30);
1347             ty               = _mm_mul_ps(fscal,dy30);
1348             tz               = _mm_mul_ps(fscal,dz30);
1349
1350             /* Update vectorial force */
1351             fix3             = _mm_add_ps(fix3,tx);
1352             fiy3             = _mm_add_ps(fiy3,ty);
1353             fiz3             = _mm_add_ps(fiz3,tz);
1354
1355             fjx0             = _mm_add_ps(fjx0,tx);
1356             fjy0             = _mm_add_ps(fjy0,ty);
1357             fjz0             = _mm_add_ps(fjz0,tz);
1358
1359             }
1360
1361             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1362             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1363             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1364             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1365
1366             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1367
1368             /* Inner loop uses 147 flops */
1369         }
1370
1371         /* End of innermost loop */
1372
1373         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1374                                               f+i_coord_offset,fshift+i_shift_offset);
1375
1376         /* Increment number of inner iterations */
1377         inneriter                  += j_index_end - j_index_start;
1378
1379         /* Outer loop uses 24 flops */
1380     }
1381
1382     /* Increment number of outer iterations */
1383     outeriter        += nri;
1384
1385     /* Update outer/inner flops */
1386
1387     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*147);
1388 }