src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwCSTab_GeomW4P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_VF_sse4_1_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     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     krf              = _mm_set1_ps(fr->ic->k_rf);
 113     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 114     crf              = _mm_set1_ps(fr->ic->c_rf);
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 125     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 126     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 127     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = jnrC = jnrD = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133     j_coord_offsetC = 0;
 134     j_coord_offsetD = 0;
 135
 136     outeriter        = 0;
 137     inneriter        = 0;
 138
 139     for(iidx=0;iidx<4*DIM;iidx++)
 140     {
 141         scratch[iidx] = 0.0;
 142     }
 143
 144     /* Start outer loop over neighborlists */
 145     for(iidx=0; iidx<nri; iidx++)
 146     {
 147         /* Load shift vector for this list */
 148         i_shift_offset   = DIM*shiftidx[iidx];
 149
 150         /* Load limits for loop over neighbors */
 151         j_index_start    = jindex[iidx];
 152         j_index_end      = jindex[iidx+1];
 153
 154         /* Get outer coordinate index */
 155         inr              = iinr[iidx];
 156         i_coord_offset   = DIM*inr;
 157
 158         /* Load i particle coords and add shift vector */
 159         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 160                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 161
 162         fix0             = _mm_setzero_ps();
 163         fiy0             = _mm_setzero_ps();
 164         fiz0             = _mm_setzero_ps();
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171         fix3             = _mm_setzero_ps();
 172         fiy3             = _mm_setzero_ps();
 173         fiz3             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177         vvdwsum          = _mm_setzero_ps();
 178
 179         /* Start inner kernel loop */
 180         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 181         {
 182
 183             /* Get j neighbor index, and coordinate index */
 184             jnrA             = jjnr[jidx];
 185             jnrB             = jjnr[jidx+1];
 186             jnrC             = jjnr[jidx+2];
 187             jnrD             = jjnr[jidx+3];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190             j_coord_offsetC  = DIM*jnrC;
 191             j_coord_offsetD  = DIM*jnrD;
 192
 193             /* load j atom coordinates */
 194             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 195                                               x+j_coord_offsetC,x+j_coord_offsetD,
 196                                               &jx0,&jy0,&jz0);
 197
 198             /* Calculate displacement vector */
 199             dx00             = _mm_sub_ps(ix0,jx0);
 200             dy00             = _mm_sub_ps(iy0,jy0);
 201             dz00             = _mm_sub_ps(iz0,jz0);
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208             dx30             = _mm_sub_ps(ix3,jx0);
 209             dy30             = _mm_sub_ps(iy3,jy0);
 210             dz30             = _mm_sub_ps(iz3,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 216             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 217
 218             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 219             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 220             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 221             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 222
 223             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 224             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 225             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 226
 227             /* Load parameters for j particles */
 228             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 229                                                               charge+jnrC+0,charge+jnrD+0);
 230             vdwjidx0A        = 2*vdwtype[jnrA+0];
 231             vdwjidx0B        = 2*vdwtype[jnrB+0];
 232             vdwjidx0C        = 2*vdwtype[jnrC+0];
 233             vdwjidx0D        = 2*vdwtype[jnrD+0];
 234
 235             /**************************
 236              * CALCULATE INTERACTIONS *
 237              **************************/
 238
 239             r00              = _mm_mul_ps(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 243                                          vdwparam+vdwioffset0+vdwjidx0B,
 244                                          vdwparam+vdwioffset0+vdwjidx0C,
 245                                          vdwparam+vdwioffset0+vdwjidx0D,
 246                                          &c6_00,&c12_00);
 247
 248             /* Calculate table index by multiplying r with table scale and truncate to integer */
 249             rt               = _mm_mul_ps(r00,vftabscale);
 250             vfitab           = _mm_cvttps_epi32(rt);
 251             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 252             vfitab           = _mm_slli_epi32(vfitab,3);
 253
 254             /* CUBIC SPLINE TABLE DISPERSION */
 255             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 256             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 257             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 258             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 259             _MM_TRANSPOSE4_PS(Y,F,G,H);
 260             Heps             = _mm_mul_ps(vfeps,H);
 261             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 262             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 263             vvdw6            = _mm_mul_ps(c6_00,VV);
 264             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 265             fvdw6            = _mm_mul_ps(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw12           = _mm_mul_ps(c12_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw12           = _mm_mul_ps(c12_00,FF);
 280             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 281             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 285
 286             fscal            = fvdw;
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_ps(fscal,dx00);
 290             ty               = _mm_mul_ps(fscal,dy00);
 291             tz               = _mm_mul_ps(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm_add_ps(fix0,tx);
 295             fiy0             = _mm_add_ps(fiy0,ty);
 296             fiz0             = _mm_add_ps(fiz0,tz);
 297
 298             fjptrA             = f+j_coord_offsetA;
 299             fjptrB             = f+j_coord_offsetB;
 300             fjptrC             = f+j_coord_offsetC;
 301             fjptrD             = f+j_coord_offsetD;
 302             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_ps(iq1,jq0);
 310
 311             /* REACTION-FIELD ELECTROSTATICS */
 312             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 313             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317
 318             fscal            = felec;
 319
 320             /* Calculate temporary vectorial force */
 321             tx               = _mm_mul_ps(fscal,dx10);
 322             ty               = _mm_mul_ps(fscal,dy10);
 323             tz               = _mm_mul_ps(fscal,dz10);
 324
 325             /* Update vectorial force */
 326             fix1             = _mm_add_ps(fix1,tx);
 327             fiy1             = _mm_add_ps(fiy1,ty);
 328             fiz1             = _mm_add_ps(fiz1,tz);
 329
 330             fjptrA             = f+j_coord_offsetA;
 331             fjptrB             = f+j_coord_offsetB;
 332             fjptrC             = f+j_coord_offsetC;
 333             fjptrD             = f+j_coord_offsetD;
 334             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm_mul_ps(iq2,jq0);
 342
 343             /* REACTION-FIELD ELECTROSTATICS */
 344             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 345             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velecsum         = _mm_add_ps(velecsum,velec);
 349
 350             fscal            = felec;
 351
 352             /* Calculate temporary vectorial force */
 353             tx               = _mm_mul_ps(fscal,dx20);
 354             ty               = _mm_mul_ps(fscal,dy20);
 355             tz               = _mm_mul_ps(fscal,dz20);
 356
 357             /* Update vectorial force */
 358             fix2             = _mm_add_ps(fix2,tx);
 359             fiy2             = _mm_add_ps(fiy2,ty);
 360             fiz2             = _mm_add_ps(fiz2,tz);
 361
 362             fjptrA             = f+j_coord_offsetA;
 363             fjptrB             = f+j_coord_offsetB;
 364             fjptrC             = f+j_coord_offsetC;
 365             fjptrD             = f+j_coord_offsetD;
 366             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq30             = _mm_mul_ps(iq3,jq0);
 374
 375             /* REACTION-FIELD ELECTROSTATICS */
 376             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 377             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm_mul_ps(fscal,dx30);
 386             ty               = _mm_mul_ps(fscal,dy30);
 387             tz               = _mm_mul_ps(fscal,dz30);
 388
 389             /* Update vectorial force */
 390             fix3             = _mm_add_ps(fix3,tx);
 391             fiy3             = _mm_add_ps(fiy3,ty);
 392             fiz3             = _mm_add_ps(fiz3,tz);
 393
 394             fjptrA             = f+j_coord_offsetA;
 395             fjptrB             = f+j_coord_offsetB;
 396             fjptrC             = f+j_coord_offsetC;
 397             fjptrD             = f+j_coord_offsetD;
 398             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 399
 400             /* Inner loop uses 152 flops */
 401         }
 402
 403         if(jidx<j_index_end)
 404         {
 405
 406             /* Get j neighbor index, and coordinate index */
 407             jnrlistA         = jjnr[jidx];
 408             jnrlistB         = jjnr[jidx+1];
 409             jnrlistC         = jjnr[jidx+2];
 410             jnrlistD         = jjnr[jidx+3];
 411             /* Sign of each element will be negative for non-real atoms.
 412              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 413              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 414              */
 415             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 416             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 417             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 418             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 419             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               x+j_coord_offsetC,x+j_coord_offsetD,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_ps(ix0,jx0);
 432             dy00             = _mm_sub_ps(iy0,jy0);
 433             dz00             = _mm_sub_ps(iz0,jz0);
 434             dx10             = _mm_sub_ps(ix1,jx0);
 435             dy10             = _mm_sub_ps(iy1,jy0);
 436             dz10             = _mm_sub_ps(iz1,jz0);
 437             dx20             = _mm_sub_ps(ix2,jx0);
 438             dy20             = _mm_sub_ps(iy2,jy0);
 439             dz20             = _mm_sub_ps(iz2,jz0);
 440             dx30             = _mm_sub_ps(ix3,jx0);
 441             dy30             = _mm_sub_ps(iy3,jy0);
 442             dz30             = _mm_sub_ps(iz3,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 446             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 447             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 448             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 449
 450             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 451             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 452             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 453             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 454
 455             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 456             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 457             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 458
 459             /* Load parameters for j particles */
 460             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 461                                                               charge+jnrC+0,charge+jnrD+0);
 462             vdwjidx0A        = 2*vdwtype[jnrA+0];
 463             vdwjidx0B        = 2*vdwtype[jnrB+0];
 464             vdwjidx0C        = 2*vdwtype[jnrC+0];
 465             vdwjidx0D        = 2*vdwtype[jnrD+0];
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             r00              = _mm_mul_ps(rsq00,rinv00);
 472             r00              = _mm_andnot_ps(dummy_mask,r00);
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 476                                          vdwparam+vdwioffset0+vdwjidx0B,
 477                                          vdwparam+vdwioffset0+vdwjidx0C,
 478                                          vdwparam+vdwioffset0+vdwjidx0D,
 479                                          &c6_00,&c12_00);
 480
 481             /* Calculate table index by multiplying r with table scale and truncate to integer */
 482             rt               = _mm_mul_ps(r00,vftabscale);
 483             vfitab           = _mm_cvttps_epi32(rt);
 484             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 485             vfitab           = _mm_slli_epi32(vfitab,3);
 486
 487             /* CUBIC SPLINE TABLE DISPERSION */
 488             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 489             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 490             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 491             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 492             _MM_TRANSPOSE4_PS(Y,F,G,H);
 493             Heps             = _mm_mul_ps(vfeps,H);
 494             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 495             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 496             vvdw6            = _mm_mul_ps(c6_00,VV);
 497             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 498             fvdw6            = _mm_mul_ps(c6_00,FF);
 499
 500             /* CUBIC SPLINE TABLE REPULSION */
 501             vfitab           = _mm_add_epi32(vfitab,ifour);
 502             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 503             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 504             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 505             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 506             _MM_TRANSPOSE4_PS(Y,F,G,H);
 507             Heps             = _mm_mul_ps(vfeps,H);
 508             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 509             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 510             vvdw12           = _mm_mul_ps(c12_00,VV);
 511             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 512             fvdw12           = _mm_mul_ps(c12_00,FF);
 513             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 514             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 518             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 519
 520             fscal            = fvdw;
 521
 522             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm_mul_ps(fscal,dx00);
 526             ty               = _mm_mul_ps(fscal,dy00);
 527             tz               = _mm_mul_ps(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm_add_ps(fix0,tx);
 531             fiy0             = _mm_add_ps(fiy0,ty);
 532             fiz0             = _mm_add_ps(fiz0,tz);
 533
 534             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 535             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 536             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 537             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 538             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq10             = _mm_mul_ps(iq1,jq0);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 549             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx10);
 561             ty               = _mm_mul_ps(fscal,dy10);
 562             tz               = _mm_mul_ps(fscal,dz10);
 563
 564             /* Update vectorial force */
 565             fix1             = _mm_add_ps(fix1,tx);
 566             fiy1             = _mm_add_ps(fiy1,ty);
 567             fiz1             = _mm_add_ps(fiz1,tz);
 568
 569             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 570             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 571             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 572             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 573             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 574
 575             /**************************
 576              * CALCULATE INTERACTIONS *
 577              **************************/
 578
 579             /* Compute parameters for interactions between i and j atoms */
 580             qq20             = _mm_mul_ps(iq2,jq0);
 581
 582             /* REACTION-FIELD ELECTROSTATICS */
 583             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 584             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 585
 586             /* Update potential sum for this i atom from the interaction with this j atom. */
 587             velec            = _mm_andnot_ps(dummy_mask,velec);
 588             velecsum         = _mm_add_ps(velecsum,velec);
 589
 590             fscal            = felec;
 591
 592             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 593
 594             /* Calculate temporary vectorial force */
 595             tx               = _mm_mul_ps(fscal,dx20);
 596             ty               = _mm_mul_ps(fscal,dy20);
 597             tz               = _mm_mul_ps(fscal,dz20);
 598
 599             /* Update vectorial force */
 600             fix2             = _mm_add_ps(fix2,tx);
 601             fiy2             = _mm_add_ps(fiy2,ty);
 602             fiz2             = _mm_add_ps(fiz2,tz);
 603
 604             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 605             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 606             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 607             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 608             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 609
 610             /**************************
 611              * CALCULATE INTERACTIONS *
 612              **************************/
 613
 614             /* Compute parameters for interactions between i and j atoms */
 615             qq30             = _mm_mul_ps(iq3,jq0);
 616
 617             /* REACTION-FIELD ELECTROSTATICS */
 618             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 619             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_andnot_ps(dummy_mask,velec);
 623             velecsum         = _mm_add_ps(velecsum,velec);
 624
 625             fscal            = felec;
 626
 627             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 628
 629             /* Calculate temporary vectorial force */
 630             tx               = _mm_mul_ps(fscal,dx30);
 631             ty               = _mm_mul_ps(fscal,dy30);
 632             tz               = _mm_mul_ps(fscal,dz30);
 633
 634             /* Update vectorial force */
 635             fix3             = _mm_add_ps(fix3,tx);
 636             fiy3             = _mm_add_ps(fiy3,ty);
 637             fiz3             = _mm_add_ps(fiz3,tz);
 638
 639             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 640             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 641             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 642             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 643             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 644
 645             /* Inner loop uses 153 flops */
 646         }
 647
 648         /* End of innermost loop */
 649
 650         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 651                                               f+i_coord_offset,fshift+i_shift_offset);
 652
 653         ggid                        = gid[iidx];
 654         /* Update potential energies */
 655         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 656         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 657
 658         /* Increment number of inner iterations */
 659         inneriter                  += j_index_end - j_index_start;
 660
 661         /* Outer loop uses 26 flops */
 662     }
 663
 664     /* Increment number of outer iterations */
 665     outeriter        += nri;
 666
 667     /* Update outer/inner flops */
 668
 669     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*153);
 670 }
 671 /*
 672  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_single
 673  * Electrostatics interaction: ReactionField
 674  * VdW interaction:            CubicSplineTable
 675  * Geometry:                   Water4-Particle
 676  * Calculate force/pot:        Force
 677  */
 678 void
 679 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse4_1_single
 680                     (t_nblist * gmx_restrict                nlist,
 681                      rvec * gmx_restrict                    xx,
 682                      rvec * gmx_restrict                    ff,
 683                      t_forcerec * gmx_restrict              fr,
 684                      t_mdatoms * gmx_restrict               mdatoms,
 685                      nb_kernel_data_t * gmx_restrict        kernel_data,
 686                      t_nrnb * gmx_restrict                  nrnb)
 687 {
 688     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 689      * just 0 for non-waters.
 690      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 691      * jnr indices corresponding to data put in the four positions in the SIMD register.
 692      */
 693     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 694     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 695     int              jnrA,jnrB,jnrC,jnrD;
 696     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 697     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 698     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 699     real             rcutoff_scalar;
 700     real             *shiftvec,*fshift,*x,*f;
 701     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 702     real             scratch[4*DIM];
 703     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 704     int              vdwioffset0;
 705     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 706     int              vdwioffset1;
 707     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 708     int              vdwioffset2;
 709     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 710     int              vdwioffset3;
 711     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 712     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 713     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 714     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 715     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 716     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 717     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 718     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 719     real             *charge;
 720     int              nvdwtype;
 721     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 722     int              *vdwtype;
 723     real             *vdwparam;
 724     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 725     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 726     __m128i          vfitab;
 727     __m128i          ifour       = _mm_set1_epi32(4);
 728     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 729     real             *vftab;
 730     __m128           dummy_mask,cutoff_mask;
 731     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 732     __m128           one     = _mm_set1_ps(1.0);
 733     __m128           two     = _mm_set1_ps(2.0);
 734     x                = xx[0];
 735     f                = ff[0];
 736
 737     nri              = nlist->nri;
 738     iinr             = nlist->iinr;
 739     jindex           = nlist->jindex;
 740     jjnr             = nlist->jjnr;
 741     shiftidx         = nlist->shift;
 742     gid              = nlist->gid;
 743     shiftvec         = fr->shift_vec[0];
 744     fshift           = fr->fshift[0];
 745     facel            = _mm_set1_ps(fr->epsfac);
 746     charge           = mdatoms->chargeA;
 747     krf              = _mm_set1_ps(fr->ic->k_rf);
 748     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 749     crf              = _mm_set1_ps(fr->ic->c_rf);
 750     nvdwtype         = fr->ntype;
 751     vdwparam         = fr->nbfp;
 752     vdwtype          = mdatoms->typeA;
 753
 754     vftab            = kernel_data->table_vdw->data;
 755     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 756
 757     /* Setup water-specific parameters */
 758     inr              = nlist->iinr[0];
 759     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 760     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 761     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 762     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 763
 764     /* Avoid stupid compiler warnings */
 765     jnrA = jnrB = jnrC = jnrD = 0;
 766     j_coord_offsetA = 0;
 767     j_coord_offsetB = 0;
 768     j_coord_offsetC = 0;
 769     j_coord_offsetD = 0;
 770
 771     outeriter        = 0;
 772     inneriter        = 0;
 773
 774     for(iidx=0;iidx<4*DIM;iidx++)
 775     {
 776         scratch[iidx] = 0.0;
 777     }
 778
 779     /* Start outer loop over neighborlists */
 780     for(iidx=0; iidx<nri; iidx++)
 781     {
 782         /* Load shift vector for this list */
 783         i_shift_offset   = DIM*shiftidx[iidx];
 784
 785         /* Load limits for loop over neighbors */
 786         j_index_start    = jindex[iidx];
 787         j_index_end      = jindex[iidx+1];
 788
 789         /* Get outer coordinate index */
 790         inr              = iinr[iidx];
 791         i_coord_offset   = DIM*inr;
 792
 793         /* Load i particle coords and add shift vector */
 794         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 795                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 796
 797         fix0             = _mm_setzero_ps();
 798         fiy0             = _mm_setzero_ps();
 799         fiz0             = _mm_setzero_ps();
 800         fix1             = _mm_setzero_ps();
 801         fiy1             = _mm_setzero_ps();
 802         fiz1             = _mm_setzero_ps();
 803         fix2             = _mm_setzero_ps();
 804         fiy2             = _mm_setzero_ps();
 805         fiz2             = _mm_setzero_ps();
 806         fix3             = _mm_setzero_ps();
 807         fiy3             = _mm_setzero_ps();
 808         fiz3             = _mm_setzero_ps();
 809
 810         /* Start inner kernel loop */
 811         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 812         {
 813
 814             /* Get j neighbor index, and coordinate index */
 815             jnrA             = jjnr[jidx];
 816             jnrB             = jjnr[jidx+1];
 817             jnrC             = jjnr[jidx+2];
 818             jnrD             = jjnr[jidx+3];
 819             j_coord_offsetA  = DIM*jnrA;
 820             j_coord_offsetB  = DIM*jnrB;
 821             j_coord_offsetC  = DIM*jnrC;
 822             j_coord_offsetD  = DIM*jnrD;
 823
 824             /* load j atom coordinates */
 825             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 826                                               x+j_coord_offsetC,x+j_coord_offsetD,
 827                                               &jx0,&jy0,&jz0);
 828
 829             /* Calculate displacement vector */
 830             dx00             = _mm_sub_ps(ix0,jx0);
 831             dy00             = _mm_sub_ps(iy0,jy0);
 832             dz00             = _mm_sub_ps(iz0,jz0);
 833             dx10             = _mm_sub_ps(ix1,jx0);
 834             dy10             = _mm_sub_ps(iy1,jy0);
 835             dz10             = _mm_sub_ps(iz1,jz0);
 836             dx20             = _mm_sub_ps(ix2,jx0);
 837             dy20             = _mm_sub_ps(iy2,jy0);
 838             dz20             = _mm_sub_ps(iz2,jz0);
 839             dx30             = _mm_sub_ps(ix3,jx0);
 840             dy30             = _mm_sub_ps(iy3,jy0);
 841             dz30             = _mm_sub_ps(iz3,jz0);
 842
 843             /* Calculate squared distance and things based on it */
 844             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 845             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 846             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 847             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 848
 849             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 850             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 851             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 852             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 853
 854             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 855             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 856             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 857
 858             /* Load parameters for j particles */
 859             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 860                                                               charge+jnrC+0,charge+jnrD+0);
 861             vdwjidx0A        = 2*vdwtype[jnrA+0];
 862             vdwjidx0B        = 2*vdwtype[jnrB+0];
 863             vdwjidx0C        = 2*vdwtype[jnrC+0];
 864             vdwjidx0D        = 2*vdwtype[jnrD+0];
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             r00              = _mm_mul_ps(rsq00,rinv00);
 871
 872             /* Compute parameters for interactions between i and j atoms */
 873             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 874                                          vdwparam+vdwioffset0+vdwjidx0B,
 875                                          vdwparam+vdwioffset0+vdwjidx0C,
 876                                          vdwparam+vdwioffset0+vdwjidx0D,
 877                                          &c6_00,&c12_00);
 878
 879             /* Calculate table index by multiplying r with table scale and truncate to integer */
 880             rt               = _mm_mul_ps(r00,vftabscale);
 881             vfitab           = _mm_cvttps_epi32(rt);
 882             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 883             vfitab           = _mm_slli_epi32(vfitab,3);
 884
 885             /* CUBIC SPLINE TABLE DISPERSION */
 886             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 887             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 888             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 889             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 890             _MM_TRANSPOSE4_PS(Y,F,G,H);
 891             Heps             = _mm_mul_ps(vfeps,H);
 892             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 893             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 894             fvdw6            = _mm_mul_ps(c6_00,FF);
 895
 896             /* CUBIC SPLINE TABLE REPULSION */
 897             vfitab           = _mm_add_epi32(vfitab,ifour);
 898             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 899             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 900             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 901             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 902             _MM_TRANSPOSE4_PS(Y,F,G,H);
 903             Heps             = _mm_mul_ps(vfeps,H);
 904             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 905             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 906             fvdw12           = _mm_mul_ps(c12_00,FF);
 907             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 908
 909             fscal            = fvdw;
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_ps(fscal,dx00);
 913             ty               = _mm_mul_ps(fscal,dy00);
 914             tz               = _mm_mul_ps(fscal,dz00);
 915
 916             /* Update vectorial force */
 917             fix0             = _mm_add_ps(fix0,tx);
 918             fiy0             = _mm_add_ps(fiy0,ty);
 919             fiz0             = _mm_add_ps(fiz0,tz);
 920
 921             fjptrA             = f+j_coord_offsetA;
 922             fjptrB             = f+j_coord_offsetB;
 923             fjptrC             = f+j_coord_offsetC;
 924             fjptrD             = f+j_coord_offsetD;
 925             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 926
 927             /**************************
 928              * CALCULATE INTERACTIONS *
 929              **************************/
 930
 931             /* Compute parameters for interactions between i and j atoms */
 932             qq10             = _mm_mul_ps(iq1,jq0);
 933
 934             /* REACTION-FIELD ELECTROSTATICS */
 935             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 936
 937             fscal            = felec;
 938
 939             /* Calculate temporary vectorial force */
 940             tx               = _mm_mul_ps(fscal,dx10);
 941             ty               = _mm_mul_ps(fscal,dy10);
 942             tz               = _mm_mul_ps(fscal,dz10);
 943
 944             /* Update vectorial force */
 945             fix1             = _mm_add_ps(fix1,tx);
 946             fiy1             = _mm_add_ps(fiy1,ty);
 947             fiz1             = _mm_add_ps(fiz1,tz);
 948
 949             fjptrA             = f+j_coord_offsetA;
 950             fjptrB             = f+j_coord_offsetB;
 951             fjptrC             = f+j_coord_offsetC;
 952             fjptrD             = f+j_coord_offsetD;
 953             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             qq20             = _mm_mul_ps(iq2,jq0);
 961
 962             /* REACTION-FIELD ELECTROSTATICS */
 963             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 964
 965             fscal            = felec;
 966
 967             /* Calculate temporary vectorial force */
 968             tx               = _mm_mul_ps(fscal,dx20);
 969             ty               = _mm_mul_ps(fscal,dy20);
 970             tz               = _mm_mul_ps(fscal,dz20);
 971
 972             /* Update vectorial force */
 973             fix2             = _mm_add_ps(fix2,tx);
 974             fiy2             = _mm_add_ps(fiy2,ty);
 975             fiz2             = _mm_add_ps(fiz2,tz);
 976
 977             fjptrA             = f+j_coord_offsetA;
 978             fjptrB             = f+j_coord_offsetB;
 979             fjptrC             = f+j_coord_offsetC;
 980             fjptrD             = f+j_coord_offsetD;
 981             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 982
 983             /**************************
 984              * CALCULATE INTERACTIONS *
 985              **************************/
 986
 987             /* Compute parameters for interactions between i and j atoms */
 988             qq30             = _mm_mul_ps(iq3,jq0);
 989
 990             /* REACTION-FIELD ELECTROSTATICS */
 991             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 992
 993             fscal            = felec;
 994
 995             /* Calculate temporary vectorial force */
 996             tx               = _mm_mul_ps(fscal,dx30);
 997             ty               = _mm_mul_ps(fscal,dy30);
 998             tz               = _mm_mul_ps(fscal,dz30);
 999
1000             /* Update vectorial force */
1001             fix3             = _mm_add_ps(fix3,tx);
1002             fiy3             = _mm_add_ps(fiy3,ty);
1003             fiz3             = _mm_add_ps(fiz3,tz);
1004
1005             fjptrA             = f+j_coord_offsetA;
1006             fjptrB             = f+j_coord_offsetB;
1007             fjptrC             = f+j_coord_offsetC;
1008             fjptrD             = f+j_coord_offsetD;
1009             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1010
1011             /* Inner loop uses 129 flops */
1012         }
1013
1014         if(jidx<j_index_end)
1015         {
1016
1017             /* Get j neighbor index, and coordinate index */
1018             jnrlistA         = jjnr[jidx];
1019             jnrlistB         = jjnr[jidx+1];
1020             jnrlistC         = jjnr[jidx+2];
1021             jnrlistD         = jjnr[jidx+3];
1022             /* Sign of each element will be negative for non-real atoms.
1023              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1024              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1025              */
1026             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1027             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1028             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1029             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1030             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1031             j_coord_offsetA  = DIM*jnrA;
1032             j_coord_offsetB  = DIM*jnrB;
1033             j_coord_offsetC  = DIM*jnrC;
1034             j_coord_offsetD  = DIM*jnrD;
1035
1036             /* load j atom coordinates */
1037             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1038                                               x+j_coord_offsetC,x+j_coord_offsetD,
1039                                               &jx0,&jy0,&jz0);
1040
1041             /* Calculate displacement vector */
1042             dx00             = _mm_sub_ps(ix0,jx0);
1043             dy00             = _mm_sub_ps(iy0,jy0);
1044             dz00             = _mm_sub_ps(iz0,jz0);
1045             dx10             = _mm_sub_ps(ix1,jx0);
1046             dy10             = _mm_sub_ps(iy1,jy0);
1047             dz10             = _mm_sub_ps(iz1,jz0);
1048             dx20             = _mm_sub_ps(ix2,jx0);
1049             dy20             = _mm_sub_ps(iy2,jy0);
1050             dz20             = _mm_sub_ps(iz2,jz0);
1051             dx30             = _mm_sub_ps(ix3,jx0);
1052             dy30             = _mm_sub_ps(iy3,jy0);
1053             dz30             = _mm_sub_ps(iz3,jz0);
1054
1055             /* Calculate squared distance and things based on it */
1056             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1057             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1058             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1059             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1060
1061             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1062             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1063             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1064             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1065
1066             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1067             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1068             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1069
1070             /* Load parameters for j particles */
1071             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1072                                                               charge+jnrC+0,charge+jnrD+0);
1073             vdwjidx0A        = 2*vdwtype[jnrA+0];
1074             vdwjidx0B        = 2*vdwtype[jnrB+0];
1075             vdwjidx0C        = 2*vdwtype[jnrC+0];
1076             vdwjidx0D        = 2*vdwtype[jnrD+0];
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             r00              = _mm_mul_ps(rsq00,rinv00);
1083             r00              = _mm_andnot_ps(dummy_mask,r00);
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1087                                          vdwparam+vdwioffset0+vdwjidx0B,
1088                                          vdwparam+vdwioffset0+vdwjidx0C,
1089                                          vdwparam+vdwioffset0+vdwjidx0D,
1090                                          &c6_00,&c12_00);
1091
1092             /* Calculate table index by multiplying r with table scale and truncate to integer */
1093             rt               = _mm_mul_ps(r00,vftabscale);
1094             vfitab           = _mm_cvttps_epi32(rt);
1095             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1096             vfitab           = _mm_slli_epi32(vfitab,3);
1097
1098             /* CUBIC SPLINE TABLE DISPERSION */
1099             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1100             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1101             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1102             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1103             _MM_TRANSPOSE4_PS(Y,F,G,H);
1104             Heps             = _mm_mul_ps(vfeps,H);
1105             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1106             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1107             fvdw6            = _mm_mul_ps(c6_00,FF);
1108
1109             /* CUBIC SPLINE TABLE REPULSION */
1110             vfitab           = _mm_add_epi32(vfitab,ifour);
1111             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1112             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1113             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1114             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1115             _MM_TRANSPOSE4_PS(Y,F,G,H);
1116             Heps             = _mm_mul_ps(vfeps,H);
1117             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1118             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1119             fvdw12           = _mm_mul_ps(c12_00,FF);
1120             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1121
1122             fscal            = fvdw;
1123
1124             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm_mul_ps(fscal,dx00);
1128             ty               = _mm_mul_ps(fscal,dy00);
1129             tz               = _mm_mul_ps(fscal,dz00);
1130
1131             /* Update vectorial force */
1132             fix0             = _mm_add_ps(fix0,tx);
1133             fiy0             = _mm_add_ps(fiy0,ty);
1134             fiz0             = _mm_add_ps(fiz0,tz);
1135
1136             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1137             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1138             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1139             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1140             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             /* Compute parameters for interactions between i and j atoms */
1147             qq10             = _mm_mul_ps(iq1,jq0);
1148
1149             /* REACTION-FIELD ELECTROSTATICS */
1150             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm_mul_ps(fscal,dx10);
1158             ty               = _mm_mul_ps(fscal,dy10);
1159             tz               = _mm_mul_ps(fscal,dz10);
1160
1161             /* Update vectorial force */
1162             fix1             = _mm_add_ps(fix1,tx);
1163             fiy1             = _mm_add_ps(fiy1,ty);
1164             fiz1             = _mm_add_ps(fiz1,tz);
1165
1166             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1167             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1168             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1169             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1170             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             /* Compute parameters for interactions between i and j atoms */
1177             qq20             = _mm_mul_ps(iq2,jq0);
1178
1179             /* REACTION-FIELD ELECTROSTATICS */
1180             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1181
1182             fscal            = felec;
1183
1184             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1185
1186             /* Calculate temporary vectorial force */
1187             tx               = _mm_mul_ps(fscal,dx20);
1188             ty               = _mm_mul_ps(fscal,dy20);
1189             tz               = _mm_mul_ps(fscal,dz20);
1190
1191             /* Update vectorial force */
1192             fix2             = _mm_add_ps(fix2,tx);
1193             fiy2             = _mm_add_ps(fiy2,ty);
1194             fiz2             = _mm_add_ps(fiz2,tz);
1195
1196             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1197             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1198             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1199             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1200             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1201
1202             /**************************
1203              * CALCULATE INTERACTIONS *
1204              **************************/
1205
1206             /* Compute parameters for interactions between i and j atoms */
1207             qq30             = _mm_mul_ps(iq3,jq0);
1208
1209             /* REACTION-FIELD ELECTROSTATICS */
1210             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1211
1212             fscal            = felec;
1213
1214             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1215
1216             /* Calculate temporary vectorial force */
1217             tx               = _mm_mul_ps(fscal,dx30);
1218             ty               = _mm_mul_ps(fscal,dy30);
1219             tz               = _mm_mul_ps(fscal,dz30);
1220
1221             /* Update vectorial force */
1222             fix3             = _mm_add_ps(fix3,tx);
1223             fiy3             = _mm_add_ps(fiy3,ty);
1224             fiz3             = _mm_add_ps(fiz3,tz);
1225
1226             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1227             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1228             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1229             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1230             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1231
1232             /* Inner loop uses 130 flops */
1233         }
1234
1235         /* End of innermost loop */
1236
1237         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1238                                               f+i_coord_offset,fshift+i_shift_offset);
1239
1240         /* Increment number of inner iterations */
1241         inneriter                  += j_index_end - j_index_start;
1242
1243         /* Outer loop uses 24 flops */
1244     }
1245
1246     /* Increment number of outer iterations */
1247     outeriter        += nri;
1248
1249     /* Update outer/inner flops */
1250
1251     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*130);
1252 }