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              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     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_ps();
 171         vvdwsum          = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196             dx10             = _mm_sub_ps(ix1,jx0);
 197             dy10             = _mm_sub_ps(iy1,jy0);
 198             dz10             = _mm_sub_ps(iz1,jz0);
 199             dx20             = _mm_sub_ps(ix2,jx0);
 200             dy20             = _mm_sub_ps(iy2,jy0);
 201             dz20             = _mm_sub_ps(iz2,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 207
 208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 209             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 210             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 211
 212             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 213             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 218                                                               charge+jnrC+0,charge+jnrD+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221             vdwjidx0C        = 2*vdwtype[jnrC+0];
 222             vdwjidx0D        = 2*vdwtype[jnrD+0];
 223
 224             fjx0             = _mm_setzero_ps();
 225             fjy0             = _mm_setzero_ps();
 226             fjz0             = _mm_setzero_ps();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             r00              = _mm_mul_ps(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_ps(iq0,jq0);
 236             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,
 238                                          vdwparam+vdwioffset0+vdwjidx0C,
 239                                          vdwparam+vdwioffset0+vdwjidx0D,
 240                                          &c6_00,&c12_00);
 241
 242             /* Calculate table index by multiplying r with table scale and truncate to integer */
 243             rt               = _mm_mul_ps(r00,vftabscale);
 244             vfitab           = _mm_cvttps_epi32(rt);
 245             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 246             vfitab           = _mm_slli_epi32(vfitab,3);
 247
 248             /* REACTION-FIELD ELECTROSTATICS */
 249             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 250             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 251
 252             /* CUBIC SPLINE TABLE DISPERSION */
 253             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 254             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 255             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 256             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 257             _MM_TRANSPOSE4_PS(Y,F,G,H);
 258             Heps             = _mm_mul_ps(vfeps,H);
 259             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 260             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 261             vvdw6            = _mm_mul_ps(c6_00,VV);
 262             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 263             fvdw6            = _mm_mul_ps(c6_00,FF);
 264
 265             /* CUBIC SPLINE TABLE REPULSION */
 266             vfitab           = _mm_add_epi32(vfitab,ifour);
 267             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 268             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 269             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 270             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 271             _MM_TRANSPOSE4_PS(Y,F,G,H);
 272             Heps             = _mm_mul_ps(vfeps,H);
 273             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 274             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 275             vvdw12           = _mm_mul_ps(c12_00,VV);
 276             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 277             fvdw12           = _mm_mul_ps(c12_00,FF);
 278             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 279             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 284
 285             fscal            = _mm_add_ps(felec,fvdw);
 286
 287             /* Calculate temporary vectorial force */
 288             tx               = _mm_mul_ps(fscal,dx00);
 289             ty               = _mm_mul_ps(fscal,dy00);
 290             tz               = _mm_mul_ps(fscal,dz00);
 291
 292             /* Update vectorial force */
 293             fix0             = _mm_add_ps(fix0,tx);
 294             fiy0             = _mm_add_ps(fiy0,ty);
 295             fiz0             = _mm_add_ps(fiz0,tz);
 296
 297             fjx0             = _mm_add_ps(fjx0,tx);
 298             fjy0             = _mm_add_ps(fjy0,ty);
 299             fjz0             = _mm_add_ps(fjz0,tz);
 300
 301             /**************************
 302              * CALCULATE INTERACTIONS *
 303              **************************/
 304
 305             /* Compute parameters for interactions between i and j atoms */
 306             qq10             = _mm_mul_ps(iq1,jq0);
 307
 308             /* REACTION-FIELD ELECTROSTATICS */
 309             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 310             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velecsum         = _mm_add_ps(velecsum,velec);
 314
 315             fscal            = felec;
 316
 317             /* Calculate temporary vectorial force */
 318             tx               = _mm_mul_ps(fscal,dx10);
 319             ty               = _mm_mul_ps(fscal,dy10);
 320             tz               = _mm_mul_ps(fscal,dz10);
 321
 322             /* Update vectorial force */
 323             fix1             = _mm_add_ps(fix1,tx);
 324             fiy1             = _mm_add_ps(fiy1,ty);
 325             fiz1             = _mm_add_ps(fiz1,tz);
 326
 327             fjx0             = _mm_add_ps(fjx0,tx);
 328             fjy0             = _mm_add_ps(fjy0,ty);
 329             fjz0             = _mm_add_ps(fjz0,tz);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq20             = _mm_mul_ps(iq2,jq0);
 337
 338             /* REACTION-FIELD ELECTROSTATICS */
 339             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 340             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 341
 342             /* Update potential sum for this i atom from the interaction with this j atom. */
 343             velecsum         = _mm_add_ps(velecsum,velec);
 344
 345             fscal            = felec;
 346
 347             /* Calculate temporary vectorial force */
 348             tx               = _mm_mul_ps(fscal,dx20);
 349             ty               = _mm_mul_ps(fscal,dy20);
 350             tz               = _mm_mul_ps(fscal,dz20);
 351
 352             /* Update vectorial force */
 353             fix2             = _mm_add_ps(fix2,tx);
 354             fiy2             = _mm_add_ps(fiy2,ty);
 355             fiz2             = _mm_add_ps(fiz2,tz);
 356
 357             fjx0             = _mm_add_ps(fjx0,tx);
 358             fjy0             = _mm_add_ps(fjy0,ty);
 359             fjz0             = _mm_add_ps(fjz0,tz);
 360
 361             fjptrA             = f+j_coord_offsetA;
 362             fjptrB             = f+j_coord_offsetB;
 363             fjptrC             = f+j_coord_offsetC;
 364             fjptrD             = f+j_coord_offsetD;
 365
 366             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 367
 368             /* Inner loop uses 131 flops */
 369         }
 370
 371         if(jidx<j_index_end)
 372         {
 373
 374             /* Get j neighbor index, and coordinate index */
 375             jnrlistA         = jjnr[jidx];
 376             jnrlistB         = jjnr[jidx+1];
 377             jnrlistC         = jjnr[jidx+2];
 378             jnrlistD         = jjnr[jidx+3];
 379             /* Sign of each element will be negative for non-real atoms.
 380              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 381              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 382              */
 383             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 384             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 385             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 386             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 387             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 388             j_coord_offsetA  = DIM*jnrA;
 389             j_coord_offsetB  = DIM*jnrB;
 390             j_coord_offsetC  = DIM*jnrC;
 391             j_coord_offsetD  = DIM*jnrD;
 392
 393             /* load j atom coordinates */
 394             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 395                                               x+j_coord_offsetC,x+j_coord_offsetD,
 396                                               &jx0,&jy0,&jz0);
 397
 398             /* Calculate displacement vector */
 399             dx00             = _mm_sub_ps(ix0,jx0);
 400             dy00             = _mm_sub_ps(iy0,jy0);
 401             dz00             = _mm_sub_ps(iz0,jz0);
 402             dx10             = _mm_sub_ps(ix1,jx0);
 403             dy10             = _mm_sub_ps(iy1,jy0);
 404             dz10             = _mm_sub_ps(iz1,jz0);
 405             dx20             = _mm_sub_ps(ix2,jx0);
 406             dy20             = _mm_sub_ps(iy2,jy0);
 407             dz20             = _mm_sub_ps(iz2,jz0);
 408
 409             /* Calculate squared distance and things based on it */
 410             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 411             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 412             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 413
 414             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 415             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 416             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 417
 418             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 419             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 420             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 421
 422             /* Load parameters for j particles */
 423             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 424                                                               charge+jnrC+0,charge+jnrD+0);
 425             vdwjidx0A        = 2*vdwtype[jnrA+0];
 426             vdwjidx0B        = 2*vdwtype[jnrB+0];
 427             vdwjidx0C        = 2*vdwtype[jnrC+0];
 428             vdwjidx0D        = 2*vdwtype[jnrD+0];
 429
 430             fjx0             = _mm_setzero_ps();
 431             fjy0             = _mm_setzero_ps();
 432             fjz0             = _mm_setzero_ps();
 433
 434             /**************************
 435              * CALCULATE INTERACTIONS *
 436              **************************/
 437
 438             r00              = _mm_mul_ps(rsq00,rinv00);
 439             r00              = _mm_andnot_ps(dummy_mask,r00);
 440
 441             /* Compute parameters for interactions between i and j atoms */
 442             qq00             = _mm_mul_ps(iq0,jq0);
 443             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 444                                          vdwparam+vdwioffset0+vdwjidx0B,
 445                                          vdwparam+vdwioffset0+vdwjidx0C,
 446                                          vdwparam+vdwioffset0+vdwjidx0D,
 447                                          &c6_00,&c12_00);
 448
 449             /* Calculate table index by multiplying r with table scale and truncate to integer */
 450             rt               = _mm_mul_ps(r00,vftabscale);
 451             vfitab           = _mm_cvttps_epi32(rt);
 452             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 453             vfitab           = _mm_slli_epi32(vfitab,3);
 454
 455             /* REACTION-FIELD ELECTROSTATICS */
 456             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 457             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 458
 459             /* CUBIC SPLINE TABLE DISPERSION */
 460             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 461             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 462             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 463             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 464             _MM_TRANSPOSE4_PS(Y,F,G,H);
 465             Heps             = _mm_mul_ps(vfeps,H);
 466             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 467             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 468             vvdw6            = _mm_mul_ps(c6_00,VV);
 469             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 470             fvdw6            = _mm_mul_ps(c6_00,FF);
 471
 472             /* CUBIC SPLINE TABLE REPULSION */
 473             vfitab           = _mm_add_epi32(vfitab,ifour);
 474             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 475             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 476             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 477             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 478             _MM_TRANSPOSE4_PS(Y,F,G,H);
 479             Heps             = _mm_mul_ps(vfeps,H);
 480             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 481             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 482             vvdw12           = _mm_mul_ps(c12_00,VV);
 483             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 484             fvdw12           = _mm_mul_ps(c12_00,FF);
 485             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 486             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm_andnot_ps(dummy_mask,velec);
 490             velecsum         = _mm_add_ps(velecsum,velec);
 491             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 492             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 493
 494             fscal            = _mm_add_ps(felec,fvdw);
 495
 496             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_ps(fscal,dx00);
 500             ty               = _mm_mul_ps(fscal,dy00);
 501             tz               = _mm_mul_ps(fscal,dz00);
 502
 503             /* Update vectorial force */
 504             fix0             = _mm_add_ps(fix0,tx);
 505             fiy0             = _mm_add_ps(fiy0,ty);
 506             fiz0             = _mm_add_ps(fiz0,tz);
 507
 508             fjx0             = _mm_add_ps(fjx0,tx);
 509             fjy0             = _mm_add_ps(fjy0,ty);
 510             fjz0             = _mm_add_ps(fjz0,tz);
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             qq10             = _mm_mul_ps(iq1,jq0);
 518
 519             /* REACTION-FIELD ELECTROSTATICS */
 520             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 521             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             velec            = _mm_andnot_ps(dummy_mask,velec);
 525             velecsum         = _mm_add_ps(velecsum,velec);
 526
 527             fscal            = felec;
 528
 529             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 530
 531             /* Calculate temporary vectorial force */
 532             tx               = _mm_mul_ps(fscal,dx10);
 533             ty               = _mm_mul_ps(fscal,dy10);
 534             tz               = _mm_mul_ps(fscal,dz10);
 535
 536             /* Update vectorial force */
 537             fix1             = _mm_add_ps(fix1,tx);
 538             fiy1             = _mm_add_ps(fiy1,ty);
 539             fiz1             = _mm_add_ps(fiz1,tz);
 540
 541             fjx0             = _mm_add_ps(fjx0,tx);
 542             fjy0             = _mm_add_ps(fjy0,ty);
 543             fjz0             = _mm_add_ps(fjz0,tz);
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             /* Compute parameters for interactions between i and j atoms */
 550             qq20             = _mm_mul_ps(iq2,jq0);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 554             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 555
 556             /* Update potential sum for this i atom from the interaction with this j atom. */
 557             velec            = _mm_andnot_ps(dummy_mask,velec);
 558             velecsum         = _mm_add_ps(velecsum,velec);
 559
 560             fscal            = felec;
 561
 562             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 563
 564             /* Calculate temporary vectorial force */
 565             tx               = _mm_mul_ps(fscal,dx20);
 566             ty               = _mm_mul_ps(fscal,dy20);
 567             tz               = _mm_mul_ps(fscal,dz20);
 568
 569             /* Update vectorial force */
 570             fix2             = _mm_add_ps(fix2,tx);
 571             fiy2             = _mm_add_ps(fiy2,ty);
 572             fiz2             = _mm_add_ps(fiz2,tz);
 573
 574             fjx0             = _mm_add_ps(fjx0,tx);
 575             fjy0             = _mm_add_ps(fjy0,ty);
 576             fjz0             = _mm_add_ps(fjz0,tz);
 577
 578             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 579             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 580             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 581             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 582
 583             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 584
 585             /* Inner loop uses 132 flops */
 586         }
 587
 588         /* End of innermost loop */
 589
 590         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 591                                               f+i_coord_offset,fshift+i_shift_offset);
 592
 593         ggid                        = gid[iidx];
 594         /* Update potential energies */
 595         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 596         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 597
 598         /* Increment number of inner iterations */
 599         inneriter                  += j_index_end - j_index_start;
 600
 601         /* Outer loop uses 20 flops */
 602     }
 603
 604     /* Increment number of outer iterations */
 605     outeriter        += nri;
 606
 607     /* Update outer/inner flops */
 608
 609     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*132);
 610 }
 611 /*
 612  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 613  * Electrostatics interaction: ReactionField
 614  * VdW interaction:            CubicSplineTable
 615  * Geometry:                   Water3-Particle
 616  * Calculate force/pot:        Force
 617  */
 618 void
 619 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse2_single
 620                     (t_nblist * gmx_restrict                nlist,
 621                      rvec * gmx_restrict                    xx,
 622                      rvec * gmx_restrict                    ff,
 623                      t_forcerec * gmx_restrict              fr,
 624                      t_mdatoms * gmx_restrict               mdatoms,
 625                      nb_kernel_data_t * gmx_restrict        kernel_data,
 626                      t_nrnb * gmx_restrict                  nrnb)
 627 {
 628     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 629      * just 0 for non-waters.
 630      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 631      * jnr indices corresponding to data put in the four positions in the SIMD register.
 632      */
 633     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 634     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 635     int              jnrA,jnrB,jnrC,jnrD;
 636     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 637     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 638     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 639     real             rcutoff_scalar;
 640     real             *shiftvec,*fshift,*x,*f;
 641     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 642     real             scratch[4*DIM];
 643     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 644     int              vdwioffset0;
 645     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 646     int              vdwioffset1;
 647     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 648     int              vdwioffset2;
 649     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 650     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 651     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 652     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 653     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 654     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 655     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 656     real             *charge;
 657     int              nvdwtype;
 658     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 659     int              *vdwtype;
 660     real             *vdwparam;
 661     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 662     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 663     __m128i          vfitab;
 664     __m128i          ifour       = _mm_set1_epi32(4);
 665     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 666     real             *vftab;
 667     __m128           dummy_mask,cutoff_mask;
 668     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 669     __m128           one     = _mm_set1_ps(1.0);
 670     __m128           two     = _mm_set1_ps(2.0);
 671     x                = xx[0];
 672     f                = ff[0];
 673
 674     nri              = nlist->nri;
 675     iinr             = nlist->iinr;
 676     jindex           = nlist->jindex;
 677     jjnr             = nlist->jjnr;
 678     shiftidx         = nlist->shift;
 679     gid              = nlist->gid;
 680     shiftvec         = fr->shift_vec[0];
 681     fshift           = fr->fshift[0];
 682     facel            = _mm_set1_ps(fr->epsfac);
 683     charge           = mdatoms->chargeA;
 684     krf              = _mm_set1_ps(fr->ic->k_rf);
 685     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 686     crf              = _mm_set1_ps(fr->ic->c_rf);
 687     nvdwtype         = fr->ntype;
 688     vdwparam         = fr->nbfp;
 689     vdwtype          = mdatoms->typeA;
 690
 691     vftab            = kernel_data->table_vdw->data;
 692     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 693
 694     /* Setup water-specific parameters */
 695     inr              = nlist->iinr[0];
 696     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 697     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 698     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 699     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 700
 701     /* Avoid stupid compiler warnings */
 702     jnrA = jnrB = jnrC = jnrD = 0;
 703     j_coord_offsetA = 0;
 704     j_coord_offsetB = 0;
 705     j_coord_offsetC = 0;
 706     j_coord_offsetD = 0;
 707
 708     outeriter        = 0;
 709     inneriter        = 0;
 710
 711     for(iidx=0;iidx<4*DIM;iidx++)
 712     {
 713         scratch[iidx] = 0.0;
 714     }
 715
 716     /* Start outer loop over neighborlists */
 717     for(iidx=0; iidx<nri; iidx++)
 718     {
 719         /* Load shift vector for this list */
 720         i_shift_offset   = DIM*shiftidx[iidx];
 721
 722         /* Load limits for loop over neighbors */
 723         j_index_start    = jindex[iidx];
 724         j_index_end      = jindex[iidx+1];
 725
 726         /* Get outer coordinate index */
 727         inr              = iinr[iidx];
 728         i_coord_offset   = DIM*inr;
 729
 730         /* Load i particle coords and add shift vector */
 731         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 732                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 733
 734         fix0             = _mm_setzero_ps();
 735         fiy0             = _mm_setzero_ps();
 736         fiz0             = _mm_setzero_ps();
 737         fix1             = _mm_setzero_ps();
 738         fiy1             = _mm_setzero_ps();
 739         fiz1             = _mm_setzero_ps();
 740         fix2             = _mm_setzero_ps();
 741         fiy2             = _mm_setzero_ps();
 742         fiz2             = _mm_setzero_ps();
 743
 744         /* Start inner kernel loop */
 745         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 746         {
 747
 748             /* Get j neighbor index, and coordinate index */
 749             jnrA             = jjnr[jidx];
 750             jnrB             = jjnr[jidx+1];
 751             jnrC             = jjnr[jidx+2];
 752             jnrD             = jjnr[jidx+3];
 753             j_coord_offsetA  = DIM*jnrA;
 754             j_coord_offsetB  = DIM*jnrB;
 755             j_coord_offsetC  = DIM*jnrC;
 756             j_coord_offsetD  = DIM*jnrD;
 757
 758             /* load j atom coordinates */
 759             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 760                                               x+j_coord_offsetC,x+j_coord_offsetD,
 761                                               &jx0,&jy0,&jz0);
 762
 763             /* Calculate displacement vector */
 764             dx00             = _mm_sub_ps(ix0,jx0);
 765             dy00             = _mm_sub_ps(iy0,jy0);
 766             dz00             = _mm_sub_ps(iz0,jz0);
 767             dx10             = _mm_sub_ps(ix1,jx0);
 768             dy10             = _mm_sub_ps(iy1,jy0);
 769             dz10             = _mm_sub_ps(iz1,jz0);
 770             dx20             = _mm_sub_ps(ix2,jx0);
 771             dy20             = _mm_sub_ps(iy2,jy0);
 772             dz20             = _mm_sub_ps(iz2,jz0);
 773
 774             /* Calculate squared distance and things based on it */
 775             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 776             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 777             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 778
 779             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 780             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 781             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 782
 783             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 784             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 785             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 786
 787             /* Load parameters for j particles */
 788             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 789                                                               charge+jnrC+0,charge+jnrD+0);
 790             vdwjidx0A        = 2*vdwtype[jnrA+0];
 791             vdwjidx0B        = 2*vdwtype[jnrB+0];
 792             vdwjidx0C        = 2*vdwtype[jnrC+0];
 793             vdwjidx0D        = 2*vdwtype[jnrD+0];
 794
 795             fjx0             = _mm_setzero_ps();
 796             fjy0             = _mm_setzero_ps();
 797             fjz0             = _mm_setzero_ps();
 798
 799             /**************************
 800              * CALCULATE INTERACTIONS *
 801              **************************/
 802
 803             r00              = _mm_mul_ps(rsq00,rinv00);
 804
 805             /* Compute parameters for interactions between i and j atoms */
 806             qq00             = _mm_mul_ps(iq0,jq0);
 807             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 808                                          vdwparam+vdwioffset0+vdwjidx0B,
 809                                          vdwparam+vdwioffset0+vdwjidx0C,
 810                                          vdwparam+vdwioffset0+vdwjidx0D,
 811                                          &c6_00,&c12_00);
 812
 813             /* Calculate table index by multiplying r with table scale and truncate to integer */
 814             rt               = _mm_mul_ps(r00,vftabscale);
 815             vfitab           = _mm_cvttps_epi32(rt);
 816             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 817             vfitab           = _mm_slli_epi32(vfitab,3);
 818
 819             /* REACTION-FIELD ELECTROSTATICS */
 820             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 821
 822             /* CUBIC SPLINE TABLE DISPERSION */
 823             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 824             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 825             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 826             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 827             _MM_TRANSPOSE4_PS(Y,F,G,H);
 828             Heps             = _mm_mul_ps(vfeps,H);
 829             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 830             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 831             fvdw6            = _mm_mul_ps(c6_00,FF);
 832
 833             /* CUBIC SPLINE TABLE REPULSION */
 834             vfitab           = _mm_add_epi32(vfitab,ifour);
 835             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 836             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 837             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 838             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 839             _MM_TRANSPOSE4_PS(Y,F,G,H);
 840             Heps             = _mm_mul_ps(vfeps,H);
 841             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 842             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 843             fvdw12           = _mm_mul_ps(c12_00,FF);
 844             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 845
 846             fscal            = _mm_add_ps(felec,fvdw);
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm_mul_ps(fscal,dx00);
 850             ty               = _mm_mul_ps(fscal,dy00);
 851             tz               = _mm_mul_ps(fscal,dz00);
 852
 853             /* Update vectorial force */
 854             fix0             = _mm_add_ps(fix0,tx);
 855             fiy0             = _mm_add_ps(fiy0,ty);
 856             fiz0             = _mm_add_ps(fiz0,tz);
 857
 858             fjx0             = _mm_add_ps(fjx0,tx);
 859             fjy0             = _mm_add_ps(fjy0,ty);
 860             fjz0             = _mm_add_ps(fjz0,tz);
 861
 862             /**************************
 863              * CALCULATE INTERACTIONS *
 864              **************************/
 865
 866             /* Compute parameters for interactions between i and j atoms */
 867             qq10             = _mm_mul_ps(iq1,jq0);
 868
 869             /* REACTION-FIELD ELECTROSTATICS */
 870             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 871
 872             fscal            = felec;
 873
 874             /* Calculate temporary vectorial force */
 875             tx               = _mm_mul_ps(fscal,dx10);
 876             ty               = _mm_mul_ps(fscal,dy10);
 877             tz               = _mm_mul_ps(fscal,dz10);
 878
 879             /* Update vectorial force */
 880             fix1             = _mm_add_ps(fix1,tx);
 881             fiy1             = _mm_add_ps(fiy1,ty);
 882             fiz1             = _mm_add_ps(fiz1,tz);
 883
 884             fjx0             = _mm_add_ps(fjx0,tx);
 885             fjy0             = _mm_add_ps(fjy0,ty);
 886             fjz0             = _mm_add_ps(fjz0,tz);
 887
 888             /**************************
 889              * CALCULATE INTERACTIONS *
 890              **************************/
 891
 892             /* Compute parameters for interactions between i and j atoms */
 893             qq20             = _mm_mul_ps(iq2,jq0);
 894
 895             /* REACTION-FIELD ELECTROSTATICS */
 896             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 897
 898             fscal            = felec;
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm_mul_ps(fscal,dx20);
 902             ty               = _mm_mul_ps(fscal,dy20);
 903             tz               = _mm_mul_ps(fscal,dz20);
 904
 905             /* Update vectorial force */
 906             fix2             = _mm_add_ps(fix2,tx);
 907             fiy2             = _mm_add_ps(fiy2,ty);
 908             fiz2             = _mm_add_ps(fiz2,tz);
 909
 910             fjx0             = _mm_add_ps(fjx0,tx);
 911             fjy0             = _mm_add_ps(fjy0,ty);
 912             fjz0             = _mm_add_ps(fjz0,tz);
 913
 914             fjptrA             = f+j_coord_offsetA;
 915             fjptrB             = f+j_coord_offsetB;
 916             fjptrC             = f+j_coord_offsetC;
 917             fjptrD             = f+j_coord_offsetD;
 918
 919             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 920
 921             /* Inner loop uses 108 flops */
 922         }
 923
 924         if(jidx<j_index_end)
 925         {
 926
 927             /* Get j neighbor index, and coordinate index */
 928             jnrlistA         = jjnr[jidx];
 929             jnrlistB         = jjnr[jidx+1];
 930             jnrlistC         = jjnr[jidx+2];
 931             jnrlistD         = jjnr[jidx+3];
 932             /* Sign of each element will be negative for non-real atoms.
 933              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 934              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 935              */
 936             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 937             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 938             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 939             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 940             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 941             j_coord_offsetA  = DIM*jnrA;
 942             j_coord_offsetB  = DIM*jnrB;
 943             j_coord_offsetC  = DIM*jnrC;
 944             j_coord_offsetD  = DIM*jnrD;
 945
 946             /* load j atom coordinates */
 947             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 948                                               x+j_coord_offsetC,x+j_coord_offsetD,
 949                                               &jx0,&jy0,&jz0);
 950
 951             /* Calculate displacement vector */
 952             dx00             = _mm_sub_ps(ix0,jx0);
 953             dy00             = _mm_sub_ps(iy0,jy0);
 954             dz00             = _mm_sub_ps(iz0,jz0);
 955             dx10             = _mm_sub_ps(ix1,jx0);
 956             dy10             = _mm_sub_ps(iy1,jy0);
 957             dz10             = _mm_sub_ps(iz1,jz0);
 958             dx20             = _mm_sub_ps(ix2,jx0);
 959             dy20             = _mm_sub_ps(iy2,jy0);
 960             dz20             = _mm_sub_ps(iz2,jz0);
 961
 962             /* Calculate squared distance and things based on it */
 963             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 964             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 965             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 966
 967             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 968             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 969             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 970
 971             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 972             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 973             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 974
 975             /* Load parameters for j particles */
 976             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 977                                                               charge+jnrC+0,charge+jnrD+0);
 978             vdwjidx0A        = 2*vdwtype[jnrA+0];
 979             vdwjidx0B        = 2*vdwtype[jnrB+0];
 980             vdwjidx0C        = 2*vdwtype[jnrC+0];
 981             vdwjidx0D        = 2*vdwtype[jnrD+0];
 982
 983             fjx0             = _mm_setzero_ps();
 984             fjy0             = _mm_setzero_ps();
 985             fjz0             = _mm_setzero_ps();
 986
 987             /**************************
 988              * CALCULATE INTERACTIONS *
 989              **************************/
 990
 991             r00              = _mm_mul_ps(rsq00,rinv00);
 992             r00              = _mm_andnot_ps(dummy_mask,r00);
 993
 994             /* Compute parameters for interactions between i and j atoms */
 995             qq00             = _mm_mul_ps(iq0,jq0);
 996             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 997                                          vdwparam+vdwioffset0+vdwjidx0B,
 998                                          vdwparam+vdwioffset0+vdwjidx0C,
 999                                          vdwparam+vdwioffset0+vdwjidx0D,
1000                                          &c6_00,&c12_00);
1001
1002             /* Calculate table index by multiplying r with table scale and truncate to integer */
1003             rt               = _mm_mul_ps(r00,vftabscale);
1004             vfitab           = _mm_cvttps_epi32(rt);
1005             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1006             vfitab           = _mm_slli_epi32(vfitab,3);
1007
1008             /* REACTION-FIELD ELECTROSTATICS */
1009             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1010
1011             /* CUBIC SPLINE TABLE DISPERSION */
1012             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1013             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1014             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1015             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1016             _MM_TRANSPOSE4_PS(Y,F,G,H);
1017             Heps             = _mm_mul_ps(vfeps,H);
1018             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1019             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1020             fvdw6            = _mm_mul_ps(c6_00,FF);
1021
1022             /* CUBIC SPLINE TABLE REPULSION */
1023             vfitab           = _mm_add_epi32(vfitab,ifour);
1024             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1025             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1026             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1027             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1028             _MM_TRANSPOSE4_PS(Y,F,G,H);
1029             Heps             = _mm_mul_ps(vfeps,H);
1030             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1031             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1032             fvdw12           = _mm_mul_ps(c12_00,FF);
1033             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1034
1035             fscal            = _mm_add_ps(felec,fvdw);
1036
1037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm_mul_ps(fscal,dx00);
1041             ty               = _mm_mul_ps(fscal,dy00);
1042             tz               = _mm_mul_ps(fscal,dz00);
1043
1044             /* Update vectorial force */
1045             fix0             = _mm_add_ps(fix0,tx);
1046             fiy0             = _mm_add_ps(fiy0,ty);
1047             fiz0             = _mm_add_ps(fiz0,tz);
1048
1049             fjx0             = _mm_add_ps(fjx0,tx);
1050             fjy0             = _mm_add_ps(fjy0,ty);
1051             fjz0             = _mm_add_ps(fjz0,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq10             = _mm_mul_ps(iq1,jq0);
1059
1060             /* REACTION-FIELD ELECTROSTATICS */
1061             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1062
1063             fscal            = felec;
1064
1065             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1066
1067             /* Calculate temporary vectorial force */
1068             tx               = _mm_mul_ps(fscal,dx10);
1069             ty               = _mm_mul_ps(fscal,dy10);
1070             tz               = _mm_mul_ps(fscal,dz10);
1071
1072             /* Update vectorial force */
1073             fix1             = _mm_add_ps(fix1,tx);
1074             fiy1             = _mm_add_ps(fiy1,ty);
1075             fiz1             = _mm_add_ps(fiz1,tz);
1076
1077             fjx0             = _mm_add_ps(fjx0,tx);
1078             fjy0             = _mm_add_ps(fjy0,ty);
1079             fjz0             = _mm_add_ps(fjz0,tz);
1080
1081             /**************************
1082              * CALCULATE INTERACTIONS *
1083              **************************/
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq20             = _mm_mul_ps(iq2,jq0);
1087
1088             /* REACTION-FIELD ELECTROSTATICS */
1089             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1090
1091             fscal            = felec;
1092
1093             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_ps(fscal,dx20);
1097             ty               = _mm_mul_ps(fscal,dy20);
1098             tz               = _mm_mul_ps(fscal,dz20);
1099
1100             /* Update vectorial force */
1101             fix2             = _mm_add_ps(fix2,tx);
1102             fiy2             = _mm_add_ps(fiy2,ty);
1103             fiz2             = _mm_add_ps(fiz2,tz);
1104
1105             fjx0             = _mm_add_ps(fjx0,tx);
1106             fjy0             = _mm_add_ps(fjy0,ty);
1107             fjz0             = _mm_add_ps(fjz0,tz);
1108
1109             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1110             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1111             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1112             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1113
1114             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1115
1116             /* Inner loop uses 109 flops */
1117         }
1118
1119         /* End of innermost loop */
1120
1121         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1122                                               f+i_coord_offset,fshift+i_shift_offset);
1123
1124         /* Increment number of inner iterations */
1125         inneriter                  += j_index_end - j_index_start;
1126
1127         /* Outer loop uses 18 flops */
1128     }
1129
1130     /* Increment number of outer iterations */
1131     outeriter        += nri;
1132
1133     /* Update outer/inner flops */
1134
1135     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1136 }