src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_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_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_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              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 130
 131     /* Avoid stupid compiler warnings */
 132     jnrA = jnrB = jnrC = jnrD = 0;
 133     j_coord_offsetA = 0;
 134     j_coord_offsetB = 0;
 135     j_coord_offsetC = 0;
 136     j_coord_offsetD = 0;
 137
 138     outeriter        = 0;
 139     inneriter        = 0;
 140
 141     for(iidx=0;iidx<4*DIM;iidx++)
 142     {
 143         scratch[iidx] = 0.0;
 144     }
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167         fix1             = _mm_setzero_ps();
 168         fiy1             = _mm_setzero_ps();
 169         fiz1             = _mm_setzero_ps();
 170         fix2             = _mm_setzero_ps();
 171         fiy2             = _mm_setzero_ps();
 172         fiz2             = _mm_setzero_ps();
 173
 174         /* Reset potential sums */
 175         velecsum         = _mm_setzero_ps();
 176         vvdwsum          = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx00             = _mm_sub_ps(ix0,jx0);
 199             dy00             = _mm_sub_ps(iy0,jy0);
 200             dz00             = _mm_sub_ps(iz0,jz0);
 201             dx10             = _mm_sub_ps(ix1,jx0);
 202             dy10             = _mm_sub_ps(iy1,jy0);
 203             dz10             = _mm_sub_ps(iz1,jz0);
 204             dx20             = _mm_sub_ps(ix2,jx0);
 205             dy20             = _mm_sub_ps(iy2,jy0);
 206             dz20             = _mm_sub_ps(iz2,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 210             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 211             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 212
 213             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 214             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 215             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 216
 217             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 218             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 219             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq00,rcutoff2))
 234             {
 235
 236             r00              = _mm_mul_ps(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq00             = _mm_mul_ps(iq0,jq0);
 240             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 241                                          vdwparam+vdwioffset0+vdwjidx0B,
 242                                          vdwparam+vdwioffset0+vdwjidx0C,
 243                                          vdwparam+vdwioffset0+vdwjidx0D,
 244                                          &c6_00,&c12_00);
 245
 246             /* Calculate table index by multiplying r with table scale and truncate to integer */
 247             rt               = _mm_mul_ps(r00,vftabscale);
 248             vfitab           = _mm_cvttps_epi32(rt);
 249             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 250             vfitab           = _mm_slli_epi32(vfitab,3);
 251
 252             /* REACTION-FIELD ELECTROSTATICS */
 253             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 254             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 255
 256             /* CUBIC SPLINE TABLE DISPERSION */
 257             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 259             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 260             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 261             _MM_TRANSPOSE4_PS(Y,F,G,H);
 262             Heps             = _mm_mul_ps(vfeps,H);
 263             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 264             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 265             vvdw6            = _mm_mul_ps(c6_00,VV);
 266             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 267             fvdw6            = _mm_mul_ps(c6_00,FF);
 268
 269             /* CUBIC SPLINE TABLE REPULSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 271             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 273             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 274             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 275             _MM_TRANSPOSE4_PS(Y,F,G,H);
 276             Heps             = _mm_mul_ps(vfeps,H);
 277             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 278             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 279             vvdw12           = _mm_mul_ps(c12_00,VV);
 280             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 281             fvdw12           = _mm_mul_ps(c12_00,FF);
 282             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 283             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 284
 285             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velec            = _mm_and_ps(velec,cutoff_mask);
 289             velecsum         = _mm_add_ps(velecsum,velec);
 290             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 291             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 292
 293             fscal            = _mm_add_ps(felec,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             fjptrA             = f+j_coord_offsetA;
 308             fjptrB             = f+j_coord_offsetB;
 309             fjptrC             = f+j_coord_offsetC;
 310             fjptrD             = f+j_coord_offsetD;
 311             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 312
 313             }
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             if (gmx_mm_any_lt(rsq10,rcutoff2))
 320             {
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* REACTION-FIELD ELECTROSTATICS */
 326             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 327             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 328
 329             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 330
 331             /* Update potential sum for this i atom from the interaction with this j atom. */
 332             velec            = _mm_and_ps(velec,cutoff_mask);
 333             velecsum         = _mm_add_ps(velecsum,velec);
 334
 335             fscal            = felec;
 336
 337             fscal            = _mm_and_ps(fscal,cutoff_mask);
 338
 339             /* Calculate temporary vectorial force */
 340             tx               = _mm_mul_ps(fscal,dx10);
 341             ty               = _mm_mul_ps(fscal,dy10);
 342             tz               = _mm_mul_ps(fscal,dz10);
 343
 344             /* Update vectorial force */
 345             fix1             = _mm_add_ps(fix1,tx);
 346             fiy1             = _mm_add_ps(fiy1,ty);
 347             fiz1             = _mm_add_ps(fiz1,tz);
 348
 349             fjptrA             = f+j_coord_offsetA;
 350             fjptrB             = f+j_coord_offsetB;
 351             fjptrC             = f+j_coord_offsetC;
 352             fjptrD             = f+j_coord_offsetD;
 353             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 354
 355             }
 356
 357             /**************************
 358              * CALCULATE INTERACTIONS *
 359              **************************/
 360
 361             if (gmx_mm_any_lt(rsq20,rcutoff2))
 362             {
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq20             = _mm_mul_ps(iq2,jq0);
 366
 367             /* REACTION-FIELD ELECTROSTATICS */
 368             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 369             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 370
 371             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_ps(velec,cutoff_mask);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx20);
 383             ty               = _mm_mul_ps(fscal,dy20);
 384             tz               = _mm_mul_ps(fscal,dz20);
 385
 386             /* Update vectorial force */
 387             fix2             = _mm_add_ps(fix2,tx);
 388             fiy2             = _mm_add_ps(fiy2,ty);
 389             fiz2             = _mm_add_ps(fiz2,tz);
 390
 391             fjptrA             = f+j_coord_offsetA;
 392             fjptrB             = f+j_coord_offsetB;
 393             fjptrC             = f+j_coord_offsetC;
 394             fjptrD             = f+j_coord_offsetD;
 395             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 396
 397             }
 398
 399             /* Inner loop uses 144 flops */
 400         }
 401
 402         if(jidx<j_index_end)
 403         {
 404
 405             /* Get j neighbor index, and coordinate index */
 406             jnrlistA         = jjnr[jidx];
 407             jnrlistB         = jjnr[jidx+1];
 408             jnrlistC         = jjnr[jidx+2];
 409             jnrlistD         = jjnr[jidx+3];
 410             /* Sign of each element will be negative for non-real atoms.
 411              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 412              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 413              */
 414             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 415             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 416             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 417             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 418             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 419             j_coord_offsetA  = DIM*jnrA;
 420             j_coord_offsetB  = DIM*jnrB;
 421             j_coord_offsetC  = DIM*jnrC;
 422             j_coord_offsetD  = DIM*jnrD;
 423
 424             /* load j atom coordinates */
 425             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 426                                               x+j_coord_offsetC,x+j_coord_offsetD,
 427                                               &jx0,&jy0,&jz0);
 428
 429             /* Calculate displacement vector */
 430             dx00             = _mm_sub_ps(ix0,jx0);
 431             dy00             = _mm_sub_ps(iy0,jy0);
 432             dz00             = _mm_sub_ps(iz0,jz0);
 433             dx10             = _mm_sub_ps(ix1,jx0);
 434             dy10             = _mm_sub_ps(iy1,jy0);
 435             dz10             = _mm_sub_ps(iz1,jz0);
 436             dx20             = _mm_sub_ps(ix2,jx0);
 437             dy20             = _mm_sub_ps(iy2,jy0);
 438             dz20             = _mm_sub_ps(iz2,jz0);
 439
 440             /* Calculate squared distance and things based on it */
 441             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 442             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 443             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 444
 445             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 446             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 447             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 448
 449             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 450             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 451             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 452
 453             /* Load parameters for j particles */
 454             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 455                                                               charge+jnrC+0,charge+jnrD+0);
 456             vdwjidx0A        = 2*vdwtype[jnrA+0];
 457             vdwjidx0B        = 2*vdwtype[jnrB+0];
 458             vdwjidx0C        = 2*vdwtype[jnrC+0];
 459             vdwjidx0D        = 2*vdwtype[jnrD+0];
 460
 461             /**************************
 462              * CALCULATE INTERACTIONS *
 463              **************************/
 464
 465             if (gmx_mm_any_lt(rsq00,rcutoff2))
 466             {
 467
 468             r00              = _mm_mul_ps(rsq00,rinv00);
 469             r00              = _mm_andnot_ps(dummy_mask,r00);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 472             qq00             = _mm_mul_ps(iq0,jq0);
 473             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 474                                          vdwparam+vdwioffset0+vdwjidx0B,
 475                                          vdwparam+vdwioffset0+vdwjidx0C,
 476                                          vdwparam+vdwioffset0+vdwjidx0D,
 477                                          &c6_00,&c12_00);
 478
 479             /* Calculate table index by multiplying r with table scale and truncate to integer */
 480             rt               = _mm_mul_ps(r00,vftabscale);
 481             vfitab           = _mm_cvttps_epi32(rt);
 482             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 483             vfitab           = _mm_slli_epi32(vfitab,3);
 484
 485             /* REACTION-FIELD ELECTROSTATICS */
 486             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 487             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 488
 489             /* CUBIC SPLINE TABLE DISPERSION */
 490             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 491             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 492             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 493             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 494             _MM_TRANSPOSE4_PS(Y,F,G,H);
 495             Heps             = _mm_mul_ps(vfeps,H);
 496             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 497             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 498             vvdw6            = _mm_mul_ps(c6_00,VV);
 499             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 500             fvdw6            = _mm_mul_ps(c6_00,FF);
 501
 502             /* CUBIC SPLINE TABLE REPULSION */
 503             vfitab           = _mm_add_epi32(vfitab,ifour);
 504             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 505             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 506             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 507             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 508             _MM_TRANSPOSE4_PS(Y,F,G,H);
 509             Heps             = _mm_mul_ps(vfeps,H);
 510             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 511             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 512             vvdw12           = _mm_mul_ps(c12_00,VV);
 513             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 514             fvdw12           = _mm_mul_ps(c12_00,FF);
 515             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 516             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 517
 518             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 519
 520             /* Update potential sum for this i atom from the interaction with this j atom. */
 521             velec            = _mm_and_ps(velec,cutoff_mask);
 522             velec            = _mm_andnot_ps(dummy_mask,velec);
 523             velecsum         = _mm_add_ps(velecsum,velec);
 524             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 525             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 526             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 527
 528             fscal            = _mm_add_ps(felec,fvdw);
 529
 530             fscal            = _mm_and_ps(fscal,cutoff_mask);
 531
 532             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 533
 534             /* Calculate temporary vectorial force */
 535             tx               = _mm_mul_ps(fscal,dx00);
 536             ty               = _mm_mul_ps(fscal,dy00);
 537             tz               = _mm_mul_ps(fscal,dz00);
 538
 539             /* Update vectorial force */
 540             fix0             = _mm_add_ps(fix0,tx);
 541             fiy0             = _mm_add_ps(fiy0,ty);
 542             fiz0             = _mm_add_ps(fiz0,tz);
 543
 544             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 545             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 546             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 547             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 548             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 549
 550             }
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             if (gmx_mm_any_lt(rsq10,rcutoff2))
 557             {
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq10             = _mm_mul_ps(iq1,jq0);
 561
 562             /* REACTION-FIELD ELECTROSTATICS */
 563             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 564             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 565
 566             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 567
 568             /* Update potential sum for this i atom from the interaction with this j atom. */
 569             velec            = _mm_and_ps(velec,cutoff_mask);
 570             velec            = _mm_andnot_ps(dummy_mask,velec);
 571             velecsum         = _mm_add_ps(velecsum,velec);
 572
 573             fscal            = felec;
 574
 575             fscal            = _mm_and_ps(fscal,cutoff_mask);
 576
 577             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_ps(fscal,dx10);
 581             ty               = _mm_mul_ps(fscal,dy10);
 582             tz               = _mm_mul_ps(fscal,dz10);
 583
 584             /* Update vectorial force */
 585             fix1             = _mm_add_ps(fix1,tx);
 586             fiy1             = _mm_add_ps(fiy1,ty);
 587             fiz1             = _mm_add_ps(fiz1,tz);
 588
 589             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 590             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 591             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 592             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 593             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 594
 595             }
 596
 597             /**************************
 598              * CALCULATE INTERACTIONS *
 599              **************************/
 600
 601             if (gmx_mm_any_lt(rsq20,rcutoff2))
 602             {
 603
 604             /* Compute parameters for interactions between i and j atoms */
 605             qq20             = _mm_mul_ps(iq2,jq0);
 606
 607             /* REACTION-FIELD ELECTROSTATICS */
 608             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 609             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 610
 611             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm_and_ps(velec,cutoff_mask);
 615             velec            = _mm_andnot_ps(dummy_mask,velec);
 616             velecsum         = _mm_add_ps(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm_and_ps(fscal,cutoff_mask);
 621
 622             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm_mul_ps(fscal,dx20);
 626             ty               = _mm_mul_ps(fscal,dy20);
 627             tz               = _mm_mul_ps(fscal,dz20);
 628
 629             /* Update vectorial force */
 630             fix2             = _mm_add_ps(fix2,tx);
 631             fiy2             = _mm_add_ps(fiy2,ty);
 632             fiz2             = _mm_add_ps(fiz2,tz);
 633
 634             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 635             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 636             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 637             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 638             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 639
 640             }
 641
 642             /* Inner loop uses 145 flops */
 643         }
 644
 645         /* End of innermost loop */
 646
 647         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 648                                               f+i_coord_offset,fshift+i_shift_offset);
 649
 650         ggid                        = gid[iidx];
 651         /* Update potential energies */
 652         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 653         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 654
 655         /* Increment number of inner iterations */
 656         inneriter                  += j_index_end - j_index_start;
 657
 658         /* Outer loop uses 20 flops */
 659     }
 660
 661     /* Increment number of outer iterations */
 662     outeriter        += nri;
 663
 664     /* Update outer/inner flops */
 665
 666     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 667 }
 668 /*
 669  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_single
 670  * Electrostatics interaction: ReactionField
 671  * VdW interaction:            CubicSplineTable
 672  * Geometry:                   Water3-Particle
 673  * Calculate force/pot:        Force
 674  */
 675 void
 676 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse4_1_single
 677                     (t_nblist * gmx_restrict                nlist,
 678                      rvec * gmx_restrict                    xx,
 679                      rvec * gmx_restrict                    ff,
 680                      t_forcerec * gmx_restrict              fr,
 681                      t_mdatoms * gmx_restrict               mdatoms,
 682                      nb_kernel_data_t * gmx_restrict        kernel_data,
 683                      t_nrnb * gmx_restrict                  nrnb)
 684 {
 685     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 686      * just 0 for non-waters.
 687      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 688      * jnr indices corresponding to data put in the four positions in the SIMD register.
 689      */
 690     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 691     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 692     int              jnrA,jnrB,jnrC,jnrD;
 693     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 694     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 695     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 696     real             rcutoff_scalar;
 697     real             *shiftvec,*fshift,*x,*f;
 698     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 699     real             scratch[4*DIM];
 700     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 701     int              vdwioffset0;
 702     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 703     int              vdwioffset1;
 704     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 705     int              vdwioffset2;
 706     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 707     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 708     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 709     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 710     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 711     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 712     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 713     real             *charge;
 714     int              nvdwtype;
 715     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 716     int              *vdwtype;
 717     real             *vdwparam;
 718     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 719     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 720     __m128i          vfitab;
 721     __m128i          ifour       = _mm_set1_epi32(4);
 722     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 723     real             *vftab;
 724     __m128           dummy_mask,cutoff_mask;
 725     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 726     __m128           one     = _mm_set1_ps(1.0);
 727     __m128           two     = _mm_set1_ps(2.0);
 728     x                = xx[0];
 729     f                = ff[0];
 730
 731     nri              = nlist->nri;
 732     iinr             = nlist->iinr;
 733     jindex           = nlist->jindex;
 734     jjnr             = nlist->jjnr;
 735     shiftidx         = nlist->shift;
 736     gid              = nlist->gid;
 737     shiftvec         = fr->shift_vec[0];
 738     fshift           = fr->fshift[0];
 739     facel            = _mm_set1_ps(fr->epsfac);
 740     charge           = mdatoms->chargeA;
 741     krf              = _mm_set1_ps(fr->ic->k_rf);
 742     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 743     crf              = _mm_set1_ps(fr->ic->c_rf);
 744     nvdwtype         = fr->ntype;
 745     vdwparam         = fr->nbfp;
 746     vdwtype          = mdatoms->typeA;
 747
 748     vftab            = kernel_data->table_vdw->data;
 749     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 750
 751     /* Setup water-specific parameters */
 752     inr              = nlist->iinr[0];
 753     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 754     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 755     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 756     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 757
 758     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 759     rcutoff_scalar   = fr->rcoulomb;
 760     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 761     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 762
 763     /* Avoid stupid compiler warnings */
 764     jnrA = jnrB = jnrC = jnrD = 0;
 765     j_coord_offsetA = 0;
 766     j_coord_offsetB = 0;
 767     j_coord_offsetC = 0;
 768     j_coord_offsetD = 0;
 769
 770     outeriter        = 0;
 771     inneriter        = 0;
 772
 773     for(iidx=0;iidx<4*DIM;iidx++)
 774     {
 775         scratch[iidx] = 0.0;
 776     }
 777
 778     /* Start outer loop over neighborlists */
 779     for(iidx=0; iidx<nri; iidx++)
 780     {
 781         /* Load shift vector for this list */
 782         i_shift_offset   = DIM*shiftidx[iidx];
 783
 784         /* Load limits for loop over neighbors */
 785         j_index_start    = jindex[iidx];
 786         j_index_end      = jindex[iidx+1];
 787
 788         /* Get outer coordinate index */
 789         inr              = iinr[iidx];
 790         i_coord_offset   = DIM*inr;
 791
 792         /* Load i particle coords and add shift vector */
 793         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 794                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 795
 796         fix0             = _mm_setzero_ps();
 797         fiy0             = _mm_setzero_ps();
 798         fiz0             = _mm_setzero_ps();
 799         fix1             = _mm_setzero_ps();
 800         fiy1             = _mm_setzero_ps();
 801         fiz1             = _mm_setzero_ps();
 802         fix2             = _mm_setzero_ps();
 803         fiy2             = _mm_setzero_ps();
 804         fiz2             = _mm_setzero_ps();
 805
 806         /* Start inner kernel loop */
 807         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 808         {
 809
 810             /* Get j neighbor index, and coordinate index */
 811             jnrA             = jjnr[jidx];
 812             jnrB             = jjnr[jidx+1];
 813             jnrC             = jjnr[jidx+2];
 814             jnrD             = jjnr[jidx+3];
 815             j_coord_offsetA  = DIM*jnrA;
 816             j_coord_offsetB  = DIM*jnrB;
 817             j_coord_offsetC  = DIM*jnrC;
 818             j_coord_offsetD  = DIM*jnrD;
 819
 820             /* load j atom coordinates */
 821             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 822                                               x+j_coord_offsetC,x+j_coord_offsetD,
 823                                               &jx0,&jy0,&jz0);
 824
 825             /* Calculate displacement vector */
 826             dx00             = _mm_sub_ps(ix0,jx0);
 827             dy00             = _mm_sub_ps(iy0,jy0);
 828             dz00             = _mm_sub_ps(iz0,jz0);
 829             dx10             = _mm_sub_ps(ix1,jx0);
 830             dy10             = _mm_sub_ps(iy1,jy0);
 831             dz10             = _mm_sub_ps(iz1,jz0);
 832             dx20             = _mm_sub_ps(ix2,jx0);
 833             dy20             = _mm_sub_ps(iy2,jy0);
 834             dz20             = _mm_sub_ps(iz2,jz0);
 835
 836             /* Calculate squared distance and things based on it */
 837             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 838             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 839             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 840
 841             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 842             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 843             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 844
 845             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 846             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 847             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 848
 849             /* Load parameters for j particles */
 850             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 851                                                               charge+jnrC+0,charge+jnrD+0);
 852             vdwjidx0A        = 2*vdwtype[jnrA+0];
 853             vdwjidx0B        = 2*vdwtype[jnrB+0];
 854             vdwjidx0C        = 2*vdwtype[jnrC+0];
 855             vdwjidx0D        = 2*vdwtype[jnrD+0];
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             if (gmx_mm_any_lt(rsq00,rcutoff2))
 862             {
 863
 864             r00              = _mm_mul_ps(rsq00,rinv00);
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             qq00             = _mm_mul_ps(iq0,jq0);
 868             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 869                                          vdwparam+vdwioffset0+vdwjidx0B,
 870                                          vdwparam+vdwioffset0+vdwjidx0C,
 871                                          vdwparam+vdwioffset0+vdwjidx0D,
 872                                          &c6_00,&c12_00);
 873
 874             /* Calculate table index by multiplying r with table scale and truncate to integer */
 875             rt               = _mm_mul_ps(r00,vftabscale);
 876             vfitab           = _mm_cvttps_epi32(rt);
 877             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 878             vfitab           = _mm_slli_epi32(vfitab,3);
 879
 880             /* REACTION-FIELD ELECTROSTATICS */
 881             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 882
 883             /* CUBIC SPLINE TABLE DISPERSION */
 884             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 885             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 886             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 887             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 888             _MM_TRANSPOSE4_PS(Y,F,G,H);
 889             Heps             = _mm_mul_ps(vfeps,H);
 890             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 891             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 892             fvdw6            = _mm_mul_ps(c6_00,FF);
 893
 894             /* CUBIC SPLINE TABLE REPULSION */
 895             vfitab           = _mm_add_epi32(vfitab,ifour);
 896             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 897             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 898             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 899             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 900             _MM_TRANSPOSE4_PS(Y,F,G,H);
 901             Heps             = _mm_mul_ps(vfeps,H);
 902             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 903             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 904             fvdw12           = _mm_mul_ps(c12_00,FF);
 905             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 906
 907             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 908
 909             fscal            = _mm_add_ps(felec,fvdw);
 910
 911             fscal            = _mm_and_ps(fscal,cutoff_mask);
 912
 913             /* Calculate temporary vectorial force */
 914             tx               = _mm_mul_ps(fscal,dx00);
 915             ty               = _mm_mul_ps(fscal,dy00);
 916             tz               = _mm_mul_ps(fscal,dz00);
 917
 918             /* Update vectorial force */
 919             fix0             = _mm_add_ps(fix0,tx);
 920             fiy0             = _mm_add_ps(fiy0,ty);
 921             fiz0             = _mm_add_ps(fiz0,tz);
 922
 923             fjptrA             = f+j_coord_offsetA;
 924             fjptrB             = f+j_coord_offsetB;
 925             fjptrC             = f+j_coord_offsetC;
 926             fjptrD             = f+j_coord_offsetD;
 927             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 928
 929             }
 930
 931             /**************************
 932              * CALCULATE INTERACTIONS *
 933              **************************/
 934
 935             if (gmx_mm_any_lt(rsq10,rcutoff2))
 936             {
 937
 938             /* Compute parameters for interactions between i and j atoms */
 939             qq10             = _mm_mul_ps(iq1,jq0);
 940
 941             /* REACTION-FIELD ELECTROSTATICS */
 942             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 943
 944             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 945
 946             fscal            = felec;
 947
 948             fscal            = _mm_and_ps(fscal,cutoff_mask);
 949
 950             /* Calculate temporary vectorial force */
 951             tx               = _mm_mul_ps(fscal,dx10);
 952             ty               = _mm_mul_ps(fscal,dy10);
 953             tz               = _mm_mul_ps(fscal,dz10);
 954
 955             /* Update vectorial force */
 956             fix1             = _mm_add_ps(fix1,tx);
 957             fiy1             = _mm_add_ps(fiy1,ty);
 958             fiz1             = _mm_add_ps(fiz1,tz);
 959
 960             fjptrA             = f+j_coord_offsetA;
 961             fjptrB             = f+j_coord_offsetB;
 962             fjptrC             = f+j_coord_offsetC;
 963             fjptrD             = f+j_coord_offsetD;
 964             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 965
 966             }
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             if (gmx_mm_any_lt(rsq20,rcutoff2))
 973             {
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq20             = _mm_mul_ps(iq2,jq0);
 977
 978             /* REACTION-FIELD ELECTROSTATICS */
 979             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 980
 981             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 982
 983             fscal            = felec;
 984
 985             fscal            = _mm_and_ps(fscal,cutoff_mask);
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm_mul_ps(fscal,dx20);
 989             ty               = _mm_mul_ps(fscal,dy20);
 990             tz               = _mm_mul_ps(fscal,dz20);
 991
 992             /* Update vectorial force */
 993             fix2             = _mm_add_ps(fix2,tx);
 994             fiy2             = _mm_add_ps(fiy2,ty);
 995             fiz2             = _mm_add_ps(fiz2,tz);
 996
 997             fjptrA             = f+j_coord_offsetA;
 998             fjptrB             = f+j_coord_offsetB;
 999             fjptrC             = f+j_coord_offsetC;
1000             fjptrD             = f+j_coord_offsetD;
1001             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1002
1003             }
1004
1005             /* Inner loop uses 117 flops */
1006         }
1007
1008         if(jidx<j_index_end)
1009         {
1010
1011             /* Get j neighbor index, and coordinate index */
1012             jnrlistA         = jjnr[jidx];
1013             jnrlistB         = jjnr[jidx+1];
1014             jnrlistC         = jjnr[jidx+2];
1015             jnrlistD         = jjnr[jidx+3];
1016             /* Sign of each element will be negative for non-real atoms.
1017              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1018              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1019              */
1020             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1021             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1022             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1023             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1024             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1025             j_coord_offsetA  = DIM*jnrA;
1026             j_coord_offsetB  = DIM*jnrB;
1027             j_coord_offsetC  = DIM*jnrC;
1028             j_coord_offsetD  = DIM*jnrD;
1029
1030             /* load j atom coordinates */
1031             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1032                                               x+j_coord_offsetC,x+j_coord_offsetD,
1033                                               &jx0,&jy0,&jz0);
1034
1035             /* Calculate displacement vector */
1036             dx00             = _mm_sub_ps(ix0,jx0);
1037             dy00             = _mm_sub_ps(iy0,jy0);
1038             dz00             = _mm_sub_ps(iz0,jz0);
1039             dx10             = _mm_sub_ps(ix1,jx0);
1040             dy10             = _mm_sub_ps(iy1,jy0);
1041             dz10             = _mm_sub_ps(iz1,jz0);
1042             dx20             = _mm_sub_ps(ix2,jx0);
1043             dy20             = _mm_sub_ps(iy2,jy0);
1044             dz20             = _mm_sub_ps(iz2,jz0);
1045
1046             /* Calculate squared distance and things based on it */
1047             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1048             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1049             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1050
1051             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1052             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1053             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1054
1055             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1056             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1057             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1058
1059             /* Load parameters for j particles */
1060             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1061                                                               charge+jnrC+0,charge+jnrD+0);
1062             vdwjidx0A        = 2*vdwtype[jnrA+0];
1063             vdwjidx0B        = 2*vdwtype[jnrB+0];
1064             vdwjidx0C        = 2*vdwtype[jnrC+0];
1065             vdwjidx0D        = 2*vdwtype[jnrD+0];
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             if (gmx_mm_any_lt(rsq00,rcutoff2))
1072             {
1073
1074             r00              = _mm_mul_ps(rsq00,rinv00);
1075             r00              = _mm_andnot_ps(dummy_mask,r00);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq00             = _mm_mul_ps(iq0,jq0);
1079             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1080                                          vdwparam+vdwioffset0+vdwjidx0B,
1081                                          vdwparam+vdwioffset0+vdwjidx0C,
1082                                          vdwparam+vdwioffset0+vdwjidx0D,
1083                                          &c6_00,&c12_00);
1084
1085             /* Calculate table index by multiplying r with table scale and truncate to integer */
1086             rt               = _mm_mul_ps(r00,vftabscale);
1087             vfitab           = _mm_cvttps_epi32(rt);
1088             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1089             vfitab           = _mm_slli_epi32(vfitab,3);
1090
1091             /* REACTION-FIELD ELECTROSTATICS */
1092             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1093
1094             /* CUBIC SPLINE TABLE DISPERSION */
1095             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1096             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1097             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1098             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1099             _MM_TRANSPOSE4_PS(Y,F,G,H);
1100             Heps             = _mm_mul_ps(vfeps,H);
1101             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1102             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1103             fvdw6            = _mm_mul_ps(c6_00,FF);
1104
1105             /* CUBIC SPLINE TABLE REPULSION */
1106             vfitab           = _mm_add_epi32(vfitab,ifour);
1107             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1108             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1109             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1110             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1111             _MM_TRANSPOSE4_PS(Y,F,G,H);
1112             Heps             = _mm_mul_ps(vfeps,H);
1113             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1114             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1115             fvdw12           = _mm_mul_ps(c12_00,FF);
1116             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1117
1118             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1119
1120             fscal            = _mm_add_ps(felec,fvdw);
1121
1122             fscal            = _mm_and_ps(fscal,cutoff_mask);
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
1144             /**************************
1145              * CALCULATE INTERACTIONS *
1146              **************************/
1147
1148             if (gmx_mm_any_lt(rsq10,rcutoff2))
1149             {
1150
1151             /* Compute parameters for interactions between i and j atoms */
1152             qq10             = _mm_mul_ps(iq1,jq0);
1153
1154             /* REACTION-FIELD ELECTROSTATICS */
1155             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1156
1157             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1158
1159             fscal            = felec;
1160
1161             fscal            = _mm_and_ps(fscal,cutoff_mask);
1162
1163             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_ps(fscal,dx10);
1167             ty               = _mm_mul_ps(fscal,dy10);
1168             tz               = _mm_mul_ps(fscal,dz10);
1169
1170             /* Update vectorial force */
1171             fix1             = _mm_add_ps(fix1,tx);
1172             fiy1             = _mm_add_ps(fiy1,ty);
1173             fiz1             = _mm_add_ps(fiz1,tz);
1174
1175             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1176             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1177             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1178             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1179             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1180
1181             }
1182
1183             /**************************
1184              * CALCULATE INTERACTIONS *
1185              **************************/
1186
1187             if (gmx_mm_any_lt(rsq20,rcutoff2))
1188             {
1189
1190             /* Compute parameters for interactions between i and j atoms */
1191             qq20             = _mm_mul_ps(iq2,jq0);
1192
1193             /* REACTION-FIELD ELECTROSTATICS */
1194             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1195
1196             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm_and_ps(fscal,cutoff_mask);
1201
1202             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1203
1204             /* Calculate temporary vectorial force */
1205             tx               = _mm_mul_ps(fscal,dx20);
1206             ty               = _mm_mul_ps(fscal,dy20);
1207             tz               = _mm_mul_ps(fscal,dz20);
1208
1209             /* Update vectorial force */
1210             fix2             = _mm_add_ps(fix2,tx);
1211             fiy2             = _mm_add_ps(fiy2,ty);
1212             fiz2             = _mm_add_ps(fiz2,tz);
1213
1214             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1215             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1216             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1217             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1218             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1219
1220             }
1221
1222             /* Inner loop uses 118 flops */
1223         }
1224
1225         /* End of innermost loop */
1226
1227         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1228                                               f+i_coord_offset,fshift+i_shift_offset);
1229
1230         /* Increment number of inner iterations */
1231         inneriter                  += j_index_end - j_index_start;
1232
1233         /* Outer loop uses 18 flops */
1234     }
1235
1236     /* Increment number of outer iterations */
1237     outeriter        += nri;
1238
1239     /* Update outer/inner flops */
1240
1241     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1242 }