src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128           dummy_mask,cutoff_mask;
  90     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  91     __m128           one     = _mm_set1_ps(1.0);
  92     __m128           two     = _mm_set1_ps(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_ps(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_ps(fr->ic->k_rf);
 107     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_ps(fr->ic->c_rf);
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138         shX              = shiftvec[i_shift_offset+XX];
 139         shY              = shiftvec[i_shift_offset+YY];
 140         shZ              = shiftvec[i_shift_offset+ZZ];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 152         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 153         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 154         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 155         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 156         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 157         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 158         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 159         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164         fix1             = _mm_setzero_ps();
 165         fiy1             = _mm_setzero_ps();
 166         fiz1             = _mm_setzero_ps();
 167         fix2             = _mm_setzero_ps();
 168         fiy2             = _mm_setzero_ps();
 169         fiz2             = _mm_setzero_ps();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173         vvdwsum          = _mm_setzero_ps();
 174
 175         /* Start inner kernel loop */
 176         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 177         {
 178
 179             /* Get j neighbor index, and coordinate index */
 180             jnrA             = jjnr[jidx];
 181             jnrB             = jjnr[jidx+1];
 182             jnrC             = jjnr[jidx+2];
 183             jnrD             = jjnr[jidx+3];
 184
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199             dx10             = _mm_sub_ps(ix1,jx0);
 200             dy10             = _mm_sub_ps(iy1,jy0);
 201             dz10             = _mm_sub_ps(iz1,jz0);
 202             dx20             = _mm_sub_ps(ix2,jx0);
 203             dy20             = _mm_sub_ps(iy2,jy0);
 204             dz20             = _mm_sub_ps(iz2,jz0);
 205
 206             /* Calculate squared distance and things based on it */
 207             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 208             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 209             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 210
 211             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 212             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 213             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 214
 215             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 216             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 217             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 218
 219             /* Load parameters for j particles */
 220             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 221                                                               charge+jnrC+0,charge+jnrD+0);
 222             vdwjidx0A        = 2*vdwtype[jnrA+0];
 223             vdwjidx0B        = 2*vdwtype[jnrB+0];
 224             vdwjidx0C        = 2*vdwtype[jnrC+0];
 225             vdwjidx0D        = 2*vdwtype[jnrD+0];
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             r00              = _mm_mul_ps(rsq00,rinv00);
 232
 233             /* Compute parameters for interactions between i and j atoms */
 234             qq00             = _mm_mul_ps(iq0,jq0);
 235             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 236                                          vdwparam+vdwioffset0+vdwjidx0B,
 237                                          vdwparam+vdwioffset0+vdwjidx0C,
 238                                          vdwparam+vdwioffset0+vdwjidx0D,
 239                                          &c6_00,&c12_00);
 240
 241             /* Calculate table index by multiplying r with table scale and truncate to integer */
 242             rt               = _mm_mul_ps(r00,vftabscale);
 243             vfitab           = _mm_cvttps_epi32(rt);
 244             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 245             vfitab           = _mm_slli_epi32(vfitab,3);
 246
 247             /* REACTION-FIELD ELECTROSTATICS */
 248             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 249             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 250
 251             /* CUBIC SPLINE TABLE DISPERSION */
 252             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 253             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 254             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 255             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 256             _MM_TRANSPOSE4_PS(Y,F,G,H);
 257             Heps             = _mm_mul_ps(vfeps,H);
 258             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 259             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 260             vvdw6            = _mm_mul_ps(c6_00,VV);
 261             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 262             fvdw6            = _mm_mul_ps(c6_00,FF);
 263
 264             /* CUBIC SPLINE TABLE REPULSION */
 265             vfitab           = _mm_add_epi32(vfitab,ifour);
 266             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 267             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 268             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 269             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 270             _MM_TRANSPOSE4_PS(Y,F,G,H);
 271             Heps             = _mm_mul_ps(vfeps,H);
 272             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 273             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 274             vvdw12           = _mm_mul_ps(c12_00,VV);
 275             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 276             fvdw12           = _mm_mul_ps(c12_00,FF);
 277             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 278             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 279
 280             /* Update potential sum for this i atom from the interaction with this j atom. */
 281             velecsum         = _mm_add_ps(velecsum,velec);
 282             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 283
 284             fscal            = _mm_add_ps(felec,fvdw);
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx00);
 288             ty               = _mm_mul_ps(fscal,dy00);
 289             tz               = _mm_mul_ps(fscal,dz00);
 290
 291             /* Update vectorial force */
 292             fix0             = _mm_add_ps(fix0,tx);
 293             fiy0             = _mm_add_ps(fiy0,ty);
 294             fiz0             = _mm_add_ps(fiz0,tz);
 295
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 297                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 298                                                    tx,ty,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_ps(iq1,jq0);
 306
 307             /* REACTION-FIELD ELECTROSTATICS */
 308             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 309             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx10);
 318             ty               = _mm_mul_ps(fscal,dy10);
 319             tz               = _mm_mul_ps(fscal,dz10);
 320
 321             /* Update vectorial force */
 322             fix1             = _mm_add_ps(fix1,tx);
 323             fiy1             = _mm_add_ps(fiy1,ty);
 324             fiz1             = _mm_add_ps(fiz1,tz);
 325
 326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 327                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 328                                                    tx,ty,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq20             = _mm_mul_ps(iq2,jq0);
 336
 337             /* REACTION-FIELD ELECTROSTATICS */
 338             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 339             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx20);
 348             ty               = _mm_mul_ps(fscal,dy20);
 349             tz               = _mm_mul_ps(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_ps(fix2,tx);
 353             fiy2             = _mm_add_ps(fiy2,ty);
 354             fiz2             = _mm_add_ps(fiz2,tz);
 355
 356             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 357                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 358                                                    tx,ty,tz);
 359
 360             /* Inner loop uses 131 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             /* Get j neighbor index, and coordinate index */
 367             jnrA             = jjnr[jidx];
 368             jnrB             = jjnr[jidx+1];
 369             jnrC             = jjnr[jidx+2];
 370             jnrD             = jjnr[jidx+3];
 371
 372             /* Sign of each element will be negative for non-real atoms.
 373              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 374              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 375              */
 376             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 377             jnrA       = (jnrA>=0) ? jnrA : 0;
 378             jnrB       = (jnrB>=0) ? jnrB : 0;
 379             jnrC       = (jnrC>=0) ? jnrC : 0;
 380             jnrD       = (jnrD>=0) ? jnrD : 0;
 381
 382             j_coord_offsetA  = DIM*jnrA;
 383             j_coord_offsetB  = DIM*jnrB;
 384             j_coord_offsetC  = DIM*jnrC;
 385             j_coord_offsetD  = DIM*jnrD;
 386
 387             /* load j atom coordinates */
 388             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 389                                               x+j_coord_offsetC,x+j_coord_offsetD,
 390                                               &jx0,&jy0,&jz0);
 391
 392             /* Calculate displacement vector */
 393             dx00             = _mm_sub_ps(ix0,jx0);
 394             dy00             = _mm_sub_ps(iy0,jy0);
 395             dz00             = _mm_sub_ps(iz0,jz0);
 396             dx10             = _mm_sub_ps(ix1,jx0);
 397             dy10             = _mm_sub_ps(iy1,jy0);
 398             dz10             = _mm_sub_ps(iz1,jz0);
 399             dx20             = _mm_sub_ps(ix2,jx0);
 400             dy20             = _mm_sub_ps(iy2,jy0);
 401             dz20             = _mm_sub_ps(iz2,jz0);
 402
 403             /* Calculate squared distance and things based on it */
 404             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 405             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 406             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 407
 408             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 409             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 410             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 411
 412             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 413             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 414             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 415
 416             /* Load parameters for j particles */
 417             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 418                                                               charge+jnrC+0,charge+jnrD+0);
 419             vdwjidx0A        = 2*vdwtype[jnrA+0];
 420             vdwjidx0B        = 2*vdwtype[jnrB+0];
 421             vdwjidx0C        = 2*vdwtype[jnrC+0];
 422             vdwjidx0D        = 2*vdwtype[jnrD+0];
 423
 424             /**************************
 425              * CALCULATE INTERACTIONS *
 426              **************************/
 427
 428             r00              = _mm_mul_ps(rsq00,rinv00);
 429             r00              = _mm_andnot_ps(dummy_mask,r00);
 430
 431             /* Compute parameters for interactions between i and j atoms */
 432             qq00             = _mm_mul_ps(iq0,jq0);
 433             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 434                                          vdwparam+vdwioffset0+vdwjidx0B,
 435                                          vdwparam+vdwioffset0+vdwjidx0C,
 436                                          vdwparam+vdwioffset0+vdwjidx0D,
 437                                          &c6_00,&c12_00);
 438
 439             /* Calculate table index by multiplying r with table scale and truncate to integer */
 440             rt               = _mm_mul_ps(r00,vftabscale);
 441             vfitab           = _mm_cvttps_epi32(rt);
 442             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 443             vfitab           = _mm_slli_epi32(vfitab,3);
 444
 445             /* REACTION-FIELD ELECTROSTATICS */
 446             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 447             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 448
 449             /* CUBIC SPLINE TABLE DISPERSION */
 450             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 451             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 452             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 453             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 454             _MM_TRANSPOSE4_PS(Y,F,G,H);
 455             Heps             = _mm_mul_ps(vfeps,H);
 456             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 457             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 458             vvdw6            = _mm_mul_ps(c6_00,VV);
 459             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 460             fvdw6            = _mm_mul_ps(c6_00,FF);
 461
 462             /* CUBIC SPLINE TABLE REPULSION */
 463             vfitab           = _mm_add_epi32(vfitab,ifour);
 464             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 465             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 466             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 467             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 468             _MM_TRANSPOSE4_PS(Y,F,G,H);
 469             Heps             = _mm_mul_ps(vfeps,H);
 470             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 471             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 472             vvdw12           = _mm_mul_ps(c12_00,VV);
 473             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 474             fvdw12           = _mm_mul_ps(c12_00,FF);
 475             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 476             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm_andnot_ps(dummy_mask,velec);
 480             velecsum         = _mm_add_ps(velecsum,velec);
 481             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 482             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 483
 484             fscal            = _mm_add_ps(felec,fvdw);
 485
 486             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm_mul_ps(fscal,dx00);
 490             ty               = _mm_mul_ps(fscal,dy00);
 491             tz               = _mm_mul_ps(fscal,dz00);
 492
 493             /* Update vectorial force */
 494             fix0             = _mm_add_ps(fix0,tx);
 495             fiy0             = _mm_add_ps(fiy0,ty);
 496             fiz0             = _mm_add_ps(fiz0,tz);
 497
 498             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 499                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 500                                                    tx,ty,tz);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             qq10             = _mm_mul_ps(iq1,jq0);
 508
 509             /* REACTION-FIELD ELECTROSTATICS */
 510             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 511             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm_andnot_ps(dummy_mask,velec);
 515             velecsum         = _mm_add_ps(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm_mul_ps(fscal,dx10);
 523             ty               = _mm_mul_ps(fscal,dy10);
 524             tz               = _mm_mul_ps(fscal,dz10);
 525
 526             /* Update vectorial force */
 527             fix1             = _mm_add_ps(fix1,tx);
 528             fiy1             = _mm_add_ps(fiy1,ty);
 529             fiz1             = _mm_add_ps(fiz1,tz);
 530
 531             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 532                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 533                                                    tx,ty,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq20             = _mm_mul_ps(iq2,jq0);
 541
 542             /* REACTION-FIELD ELECTROSTATICS */
 543             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 544             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm_andnot_ps(dummy_mask,velec);
 548             velecsum         = _mm_add_ps(velecsum,velec);
 549
 550             fscal            = felec;
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm_mul_ps(fscal,dx20);
 556             ty               = _mm_mul_ps(fscal,dy20);
 557             tz               = _mm_mul_ps(fscal,dz20);
 558
 559             /* Update vectorial force */
 560             fix2             = _mm_add_ps(fix2,tx);
 561             fiy2             = _mm_add_ps(fiy2,ty);
 562             fiz2             = _mm_add_ps(fiz2,tz);
 563
 564             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 565                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 566                                                    tx,ty,tz);
 567
 568             /* Inner loop uses 132 flops */
 569         }
 570
 571         /* End of innermost loop */
 572
 573         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 574                                               f+i_coord_offset,fshift+i_shift_offset);
 575
 576         ggid                        = gid[iidx];
 577         /* Update potential energies */
 578         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 579         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 580
 581         /* Increment number of inner iterations */
 582         inneriter                  += j_index_end - j_index_start;
 583
 584         /* Outer loop uses 29 flops */
 585     }
 586
 587     /* Increment number of outer iterations */
 588     outeriter        += nri;
 589
 590     /* Update outer/inner flops */
 591
 592     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*132);
 593 }
 594 /*
 595  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 596  * Electrostatics interaction: ReactionField
 597  * VdW interaction:            CubicSplineTable
 598  * Geometry:                   Water3-Particle
 599  * Calculate force/pot:        Force
 600  */
 601 void
 602 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 603                     (t_nblist * gmx_restrict                nlist,
 604                      rvec * gmx_restrict                    xx,
 605                      rvec * gmx_restrict                    ff,
 606                      t_forcerec * gmx_restrict              fr,
 607                      t_mdatoms * gmx_restrict               mdatoms,
 608                      nb_kernel_data_t * gmx_restrict        kernel_data,
 609                      t_nrnb * gmx_restrict                  nrnb)
 610 {
 611     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 612      * just 0 for non-waters.
 613      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 614      * jnr indices corresponding to data put in the four positions in the SIMD register.
 615      */
 616     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 617     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 618     int              jnrA,jnrB,jnrC,jnrD;
 619     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 620     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 621     real             shX,shY,shZ,rcutoff_scalar;
 622     real             *shiftvec,*fshift,*x,*f;
 623     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 624     int              vdwioffset0;
 625     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 626     int              vdwioffset1;
 627     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 628     int              vdwioffset2;
 629     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 630     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 631     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 632     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 633     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 634     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 635     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 636     real             *charge;
 637     int              nvdwtype;
 638     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 639     int              *vdwtype;
 640     real             *vdwparam;
 641     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 642     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 643     __m128i          vfitab;
 644     __m128i          ifour       = _mm_set1_epi32(4);
 645     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 646     real             *vftab;
 647     __m128           dummy_mask,cutoff_mask;
 648     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 649     __m128           one     = _mm_set1_ps(1.0);
 650     __m128           two     = _mm_set1_ps(2.0);
 651     x                = xx[0];
 652     f                = ff[0];
 653
 654     nri              = nlist->nri;
 655     iinr             = nlist->iinr;
 656     jindex           = nlist->jindex;
 657     jjnr             = nlist->jjnr;
 658     shiftidx         = nlist->shift;
 659     gid              = nlist->gid;
 660     shiftvec         = fr->shift_vec[0];
 661     fshift           = fr->fshift[0];
 662     facel            = _mm_set1_ps(fr->epsfac);
 663     charge           = mdatoms->chargeA;
 664     krf              = _mm_set1_ps(fr->ic->k_rf);
 665     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 666     crf              = _mm_set1_ps(fr->ic->c_rf);
 667     nvdwtype         = fr->ntype;
 668     vdwparam         = fr->nbfp;
 669     vdwtype          = mdatoms->typeA;
 670
 671     vftab            = kernel_data->table_vdw->data;
 672     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 673
 674     /* Setup water-specific parameters */
 675     inr              = nlist->iinr[0];
 676     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 677     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 678     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 679     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 680
 681     /* Avoid stupid compiler warnings */
 682     jnrA = jnrB = jnrC = jnrD = 0;
 683     j_coord_offsetA = 0;
 684     j_coord_offsetB = 0;
 685     j_coord_offsetC = 0;
 686     j_coord_offsetD = 0;
 687
 688     outeriter        = 0;
 689     inneriter        = 0;
 690
 691     /* Start outer loop over neighborlists */
 692     for(iidx=0; iidx<nri; iidx++)
 693     {
 694         /* Load shift vector for this list */
 695         i_shift_offset   = DIM*shiftidx[iidx];
 696         shX              = shiftvec[i_shift_offset+XX];
 697         shY              = shiftvec[i_shift_offset+YY];
 698         shZ              = shiftvec[i_shift_offset+ZZ];
 699
 700         /* Load limits for loop over neighbors */
 701         j_index_start    = jindex[iidx];
 702         j_index_end      = jindex[iidx+1];
 703
 704         /* Get outer coordinate index */
 705         inr              = iinr[iidx];
 706         i_coord_offset   = DIM*inr;
 707
 708         /* Load i particle coords and add shift vector */
 709         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 710         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 711         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 712         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 713         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 714         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 715         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 716         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 717         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 718
 719         fix0             = _mm_setzero_ps();
 720         fiy0             = _mm_setzero_ps();
 721         fiz0             = _mm_setzero_ps();
 722         fix1             = _mm_setzero_ps();
 723         fiy1             = _mm_setzero_ps();
 724         fiz1             = _mm_setzero_ps();
 725         fix2             = _mm_setzero_ps();
 726         fiy2             = _mm_setzero_ps();
 727         fiz2             = _mm_setzero_ps();
 728
 729         /* Start inner kernel loop */
 730         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 731         {
 732
 733             /* Get j neighbor index, and coordinate index */
 734             jnrA             = jjnr[jidx];
 735             jnrB             = jjnr[jidx+1];
 736             jnrC             = jjnr[jidx+2];
 737             jnrD             = jjnr[jidx+3];
 738
 739             j_coord_offsetA  = DIM*jnrA;
 740             j_coord_offsetB  = DIM*jnrB;
 741             j_coord_offsetC  = DIM*jnrC;
 742             j_coord_offsetD  = DIM*jnrD;
 743
 744             /* load j atom coordinates */
 745             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 746                                               x+j_coord_offsetC,x+j_coord_offsetD,
 747                                               &jx0,&jy0,&jz0);
 748
 749             /* Calculate displacement vector */
 750             dx00             = _mm_sub_ps(ix0,jx0);
 751             dy00             = _mm_sub_ps(iy0,jy0);
 752             dz00             = _mm_sub_ps(iz0,jz0);
 753             dx10             = _mm_sub_ps(ix1,jx0);
 754             dy10             = _mm_sub_ps(iy1,jy0);
 755             dz10             = _mm_sub_ps(iz1,jz0);
 756             dx20             = _mm_sub_ps(ix2,jx0);
 757             dy20             = _mm_sub_ps(iy2,jy0);
 758             dz20             = _mm_sub_ps(iz2,jz0);
 759
 760             /* Calculate squared distance and things based on it */
 761             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 762             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 763             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 764
 765             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 766             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 767             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 768
 769             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 770             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 771             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 772
 773             /* Load parameters for j particles */
 774             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 775                                                               charge+jnrC+0,charge+jnrD+0);
 776             vdwjidx0A        = 2*vdwtype[jnrA+0];
 777             vdwjidx0B        = 2*vdwtype[jnrB+0];
 778             vdwjidx0C        = 2*vdwtype[jnrC+0];
 779             vdwjidx0D        = 2*vdwtype[jnrD+0];
 780
 781             /**************************
 782              * CALCULATE INTERACTIONS *
 783              **************************/
 784
 785             r00              = _mm_mul_ps(rsq00,rinv00);
 786
 787             /* Compute parameters for interactions between i and j atoms */
 788             qq00             = _mm_mul_ps(iq0,jq0);
 789             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 790                                          vdwparam+vdwioffset0+vdwjidx0B,
 791                                          vdwparam+vdwioffset0+vdwjidx0C,
 792                                          vdwparam+vdwioffset0+vdwjidx0D,
 793                                          &c6_00,&c12_00);
 794
 795             /* Calculate table index by multiplying r with table scale and truncate to integer */
 796             rt               = _mm_mul_ps(r00,vftabscale);
 797             vfitab           = _mm_cvttps_epi32(rt);
 798             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 799             vfitab           = _mm_slli_epi32(vfitab,3);
 800
 801             /* REACTION-FIELD ELECTROSTATICS */
 802             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 803
 804             /* CUBIC SPLINE TABLE DISPERSION */
 805             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 806             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 807             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 808             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 809             _MM_TRANSPOSE4_PS(Y,F,G,H);
 810             Heps             = _mm_mul_ps(vfeps,H);
 811             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 812             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 813             fvdw6            = _mm_mul_ps(c6_00,FF);
 814
 815             /* CUBIC SPLINE TABLE REPULSION */
 816             vfitab           = _mm_add_epi32(vfitab,ifour);
 817             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 818             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 819             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 820             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 821             _MM_TRANSPOSE4_PS(Y,F,G,H);
 822             Heps             = _mm_mul_ps(vfeps,H);
 823             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 824             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 825             fvdw12           = _mm_mul_ps(c12_00,FF);
 826             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 827
 828             fscal            = _mm_add_ps(felec,fvdw);
 829
 830             /* Calculate temporary vectorial force */
 831             tx               = _mm_mul_ps(fscal,dx00);
 832             ty               = _mm_mul_ps(fscal,dy00);
 833             tz               = _mm_mul_ps(fscal,dz00);
 834
 835             /* Update vectorial force */
 836             fix0             = _mm_add_ps(fix0,tx);
 837             fiy0             = _mm_add_ps(fiy0,ty);
 838             fiz0             = _mm_add_ps(fiz0,tz);
 839
 840             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 841                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 842                                                    tx,ty,tz);
 843
 844             /**************************
 845              * CALCULATE INTERACTIONS *
 846              **************************/
 847
 848             /* Compute parameters for interactions between i and j atoms */
 849             qq10             = _mm_mul_ps(iq1,jq0);
 850
 851             /* REACTION-FIELD ELECTROSTATICS */
 852             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 853
 854             fscal            = felec;
 855
 856             /* Calculate temporary vectorial force */
 857             tx               = _mm_mul_ps(fscal,dx10);
 858             ty               = _mm_mul_ps(fscal,dy10);
 859             tz               = _mm_mul_ps(fscal,dz10);
 860
 861             /* Update vectorial force */
 862             fix1             = _mm_add_ps(fix1,tx);
 863             fiy1             = _mm_add_ps(fiy1,ty);
 864             fiz1             = _mm_add_ps(fiz1,tz);
 865
 866             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 867                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 868                                                    tx,ty,tz);
 869
 870             /**************************
 871              * CALCULATE INTERACTIONS *
 872              **************************/
 873
 874             /* Compute parameters for interactions between i and j atoms */
 875             qq20             = _mm_mul_ps(iq2,jq0);
 876
 877             /* REACTION-FIELD ELECTROSTATICS */
 878             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 879
 880             fscal            = felec;
 881
 882             /* Calculate temporary vectorial force */
 883             tx               = _mm_mul_ps(fscal,dx20);
 884             ty               = _mm_mul_ps(fscal,dy20);
 885             tz               = _mm_mul_ps(fscal,dz20);
 886
 887             /* Update vectorial force */
 888             fix2             = _mm_add_ps(fix2,tx);
 889             fiy2             = _mm_add_ps(fiy2,ty);
 890             fiz2             = _mm_add_ps(fiz2,tz);
 891
 892             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 893                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 894                                                    tx,ty,tz);
 895
 896             /* Inner loop uses 108 flops */
 897         }
 898
 899         if(jidx<j_index_end)
 900         {
 901
 902             /* Get j neighbor index, and coordinate index */
 903             jnrA             = jjnr[jidx];
 904             jnrB             = jjnr[jidx+1];
 905             jnrC             = jjnr[jidx+2];
 906             jnrD             = jjnr[jidx+3];
 907
 908             /* Sign of each element will be negative for non-real atoms.
 909              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 910              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 911              */
 912             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 913             jnrA       = (jnrA>=0) ? jnrA : 0;
 914             jnrB       = (jnrB>=0) ? jnrB : 0;
 915             jnrC       = (jnrC>=0) ? jnrC : 0;
 916             jnrD       = (jnrD>=0) ? jnrD : 0;
 917
 918             j_coord_offsetA  = DIM*jnrA;
 919             j_coord_offsetB  = DIM*jnrB;
 920             j_coord_offsetC  = DIM*jnrC;
 921             j_coord_offsetD  = DIM*jnrD;
 922
 923             /* load j atom coordinates */
 924             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 925                                               x+j_coord_offsetC,x+j_coord_offsetD,
 926                                               &jx0,&jy0,&jz0);
 927
 928             /* Calculate displacement vector */
 929             dx00             = _mm_sub_ps(ix0,jx0);
 930             dy00             = _mm_sub_ps(iy0,jy0);
 931             dz00             = _mm_sub_ps(iz0,jz0);
 932             dx10             = _mm_sub_ps(ix1,jx0);
 933             dy10             = _mm_sub_ps(iy1,jy0);
 934             dz10             = _mm_sub_ps(iz1,jz0);
 935             dx20             = _mm_sub_ps(ix2,jx0);
 936             dy20             = _mm_sub_ps(iy2,jy0);
 937             dz20             = _mm_sub_ps(iz2,jz0);
 938
 939             /* Calculate squared distance and things based on it */
 940             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 941             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 942             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 943
 944             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 945             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 946             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 947
 948             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 949             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 950             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 951
 952             /* Load parameters for j particles */
 953             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 954                                                               charge+jnrC+0,charge+jnrD+0);
 955             vdwjidx0A        = 2*vdwtype[jnrA+0];
 956             vdwjidx0B        = 2*vdwtype[jnrB+0];
 957             vdwjidx0C        = 2*vdwtype[jnrC+0];
 958             vdwjidx0D        = 2*vdwtype[jnrD+0];
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r00              = _mm_mul_ps(rsq00,rinv00);
 965             r00              = _mm_andnot_ps(dummy_mask,r00);
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq00             = _mm_mul_ps(iq0,jq0);
 969             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 970                                          vdwparam+vdwioffset0+vdwjidx0B,
 971                                          vdwparam+vdwioffset0+vdwjidx0C,
 972                                          vdwparam+vdwioffset0+vdwjidx0D,
 973                                          &c6_00,&c12_00);
 974
 975             /* Calculate table index by multiplying r with table scale and truncate to integer */
 976             rt               = _mm_mul_ps(r00,vftabscale);
 977             vfitab           = _mm_cvttps_epi32(rt);
 978             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 979             vfitab           = _mm_slli_epi32(vfitab,3);
 980
 981             /* REACTION-FIELD ELECTROSTATICS */
 982             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 983
 984             /* CUBIC SPLINE TABLE DISPERSION */
 985             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 986             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 987             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 988             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 989             _MM_TRANSPOSE4_PS(Y,F,G,H);
 990             Heps             = _mm_mul_ps(vfeps,H);
 991             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 992             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 993             fvdw6            = _mm_mul_ps(c6_00,FF);
 994
 995             /* CUBIC SPLINE TABLE REPULSION */
 996             vfitab           = _mm_add_epi32(vfitab,ifour);
 997             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 998             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 999             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1000             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1001             _MM_TRANSPOSE4_PS(Y,F,G,H);
1002             Heps             = _mm_mul_ps(vfeps,H);
1003             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1004             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1005             fvdw12           = _mm_mul_ps(c12_00,FF);
1006             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1007
1008             fscal            = _mm_add_ps(felec,fvdw);
1009
1010             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1011
1012             /* Calculate temporary vectorial force */
1013             tx               = _mm_mul_ps(fscal,dx00);
1014             ty               = _mm_mul_ps(fscal,dy00);
1015             tz               = _mm_mul_ps(fscal,dz00);
1016
1017             /* Update vectorial force */
1018             fix0             = _mm_add_ps(fix0,tx);
1019             fiy0             = _mm_add_ps(fiy0,ty);
1020             fiz0             = _mm_add_ps(fiz0,tz);
1021
1022             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1023                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1024                                                    tx,ty,tz);
1025
1026             /**************************
1027              * CALCULATE INTERACTIONS *
1028              **************************/
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq10             = _mm_mul_ps(iq1,jq0);
1032
1033             /* REACTION-FIELD ELECTROSTATICS */
1034             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1035
1036             fscal            = felec;
1037
1038             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm_mul_ps(fscal,dx10);
1042             ty               = _mm_mul_ps(fscal,dy10);
1043             tz               = _mm_mul_ps(fscal,dz10);
1044
1045             /* Update vectorial force */
1046             fix1             = _mm_add_ps(fix1,tx);
1047             fiy1             = _mm_add_ps(fiy1,ty);
1048             fiz1             = _mm_add_ps(fiz1,tz);
1049
1050             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1051                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1052                                                    tx,ty,tz);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             /* Compute parameters for interactions between i and j atoms */
1059             qq20             = _mm_mul_ps(iq2,jq0);
1060
1061             /* REACTION-FIELD ELECTROSTATICS */
1062             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm_mul_ps(fscal,dx20);
1070             ty               = _mm_mul_ps(fscal,dy20);
1071             tz               = _mm_mul_ps(fscal,dz20);
1072
1073             /* Update vectorial force */
1074             fix2             = _mm_add_ps(fix2,tx);
1075             fiy2             = _mm_add_ps(fiy2,ty);
1076             fiz2             = _mm_add_ps(fiz2,tz);
1077
1078             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1079                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1080                                                    tx,ty,tz);
1081
1082             /* Inner loop uses 109 flops */
1083         }
1084
1085         /* End of innermost loop */
1086
1087         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1088                                               f+i_coord_offset,fshift+i_shift_offset);
1089
1090         /* Increment number of inner iterations */
1091         inneriter                  += j_index_end - j_index_start;
1092
1093         /* Outer loop uses 27 flops */
1094     }
1095
1096     /* Increment number of outer iterations */
1097     outeriter        += nri;
1098
1099     /* Update outer/inner flops */
1100
1101     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*109);
1102 }