src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_VF_sse4_1_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* 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             /**************************
 225              * CALCULATE INTERACTIONS *
 226              **************************/
 227
 228             r00              = _mm_mul_ps(rsq00,rinv00);
 229
 230             /* Compute parameters for interactions between i and j atoms */
 231             qq00             = _mm_mul_ps(iq0,jq0);
 232             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 233                                          vdwparam+vdwioffset0+vdwjidx0B,
 234                                          vdwparam+vdwioffset0+vdwjidx0C,
 235                                          vdwparam+vdwioffset0+vdwjidx0D,
 236                                          &c6_00,&c12_00);
 237
 238             /* Calculate table index by multiplying r with table scale and truncate to integer */
 239             rt               = _mm_mul_ps(r00,vftabscale);
 240             vfitab           = _mm_cvttps_epi32(rt);
 241             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 242             vfitab           = _mm_slli_epi32(vfitab,3);
 243
 244             /* REACTION-FIELD ELECTROSTATICS */
 245             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 246             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 247
 248             /* CUBIC SPLINE TABLE DISPERSION */
 249             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 250             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 251             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 252             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 253             _MM_TRANSPOSE4_PS(Y,F,G,H);
 254             Heps             = _mm_mul_ps(vfeps,H);
 255             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 256             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 257             vvdw6            = _mm_mul_ps(c6_00,VV);
 258             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 259             fvdw6            = _mm_mul_ps(c6_00,FF);
 260
 261             /* CUBIC SPLINE TABLE REPULSION */
 262             vfitab           = _mm_add_epi32(vfitab,ifour);
 263             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 264             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 265             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 266             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 267             _MM_TRANSPOSE4_PS(Y,F,G,H);
 268             Heps             = _mm_mul_ps(vfeps,H);
 269             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 270             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 271             vvdw12           = _mm_mul_ps(c12_00,VV);
 272             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 273             fvdw12           = _mm_mul_ps(c12_00,FF);
 274             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 275             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_ps(felec,fvdw);
 282
 283             /* Calculate temporary vectorial force */
 284             tx               = _mm_mul_ps(fscal,dx00);
 285             ty               = _mm_mul_ps(fscal,dy00);
 286             tz               = _mm_mul_ps(fscal,dz00);
 287
 288             /* Update vectorial force */
 289             fix0             = _mm_add_ps(fix0,tx);
 290             fiy0             = _mm_add_ps(fiy0,ty);
 291             fiz0             = _mm_add_ps(fiz0,tz);
 292
 293             fjptrA             = f+j_coord_offsetA;
 294             fjptrB             = f+j_coord_offsetB;
 295             fjptrC             = f+j_coord_offsetC;
 296             fjptrD             = f+j_coord_offsetD;
 297             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 308             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm_add_ps(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_ps(fscal,dx10);
 317             ty               = _mm_mul_ps(fscal,dy10);
 318             tz               = _mm_mul_ps(fscal,dz10);
 319
 320             /* Update vectorial force */
 321             fix1             = _mm_add_ps(fix1,tx);
 322             fiy1             = _mm_add_ps(fiy1,ty);
 323             fiz1             = _mm_add_ps(fiz1,tz);
 324
 325             fjptrA             = f+j_coord_offsetA;
 326             fjptrB             = f+j_coord_offsetB;
 327             fjptrC             = f+j_coord_offsetC;
 328             fjptrD             = f+j_coord_offsetD;
 329             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,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             fjptrA             = f+j_coord_offsetA;
 358             fjptrB             = f+j_coord_offsetB;
 359             fjptrC             = f+j_coord_offsetC;
 360             fjptrD             = f+j_coord_offsetD;
 361             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 362
 363             /* Inner loop uses 131 flops */
 364         }
 365
 366         if(jidx<j_index_end)
 367         {
 368
 369             /* Get j neighbor index, and coordinate index */
 370             jnrlistA         = jjnr[jidx];
 371             jnrlistB         = jjnr[jidx+1];
 372             jnrlistC         = jjnr[jidx+2];
 373             jnrlistD         = jjnr[jidx+3];
 374             /* Sign of each element will be negative for non-real atoms.
 375              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 376              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 377              */
 378             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 379             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 380             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 381             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 382             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 383             j_coord_offsetA  = DIM*jnrA;
 384             j_coord_offsetB  = DIM*jnrB;
 385             j_coord_offsetC  = DIM*jnrC;
 386             j_coord_offsetD  = DIM*jnrD;
 387
 388             /* load j atom coordinates */
 389             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 390                                               x+j_coord_offsetC,x+j_coord_offsetD,
 391                                               &jx0,&jy0,&jz0);
 392
 393             /* Calculate displacement vector */
 394             dx00             = _mm_sub_ps(ix0,jx0);
 395             dy00             = _mm_sub_ps(iy0,jy0);
 396             dz00             = _mm_sub_ps(iz0,jz0);
 397             dx10             = _mm_sub_ps(ix1,jx0);
 398             dy10             = _mm_sub_ps(iy1,jy0);
 399             dz10             = _mm_sub_ps(iz1,jz0);
 400             dx20             = _mm_sub_ps(ix2,jx0);
 401             dy20             = _mm_sub_ps(iy2,jy0);
 402             dz20             = _mm_sub_ps(iz2,jz0);
 403
 404             /* Calculate squared distance and things based on it */
 405             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 406             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 407             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 408
 409             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 410             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 411             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 412
 413             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 414             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 415             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 416
 417             /* Load parameters for j particles */
 418             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 419                                                               charge+jnrC+0,charge+jnrD+0);
 420             vdwjidx0A        = 2*vdwtype[jnrA+0];
 421             vdwjidx0B        = 2*vdwtype[jnrB+0];
 422             vdwjidx0C        = 2*vdwtype[jnrC+0];
 423             vdwjidx0D        = 2*vdwtype[jnrD+0];
 424
 425             /**************************
 426              * CALCULATE INTERACTIONS *
 427              **************************/
 428
 429             r00              = _mm_mul_ps(rsq00,rinv00);
 430             r00              = _mm_andnot_ps(dummy_mask,r00);
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq00             = _mm_mul_ps(iq0,jq0);
 434             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 435                                          vdwparam+vdwioffset0+vdwjidx0B,
 436                                          vdwparam+vdwioffset0+vdwjidx0C,
 437                                          vdwparam+vdwioffset0+vdwjidx0D,
 438                                          &c6_00,&c12_00);
 439
 440             /* Calculate table index by multiplying r with table scale and truncate to integer */
 441             rt               = _mm_mul_ps(r00,vftabscale);
 442             vfitab           = _mm_cvttps_epi32(rt);
 443             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 444             vfitab           = _mm_slli_epi32(vfitab,3);
 445
 446             /* REACTION-FIELD ELECTROSTATICS */
 447             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 448             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 449
 450             /* CUBIC SPLINE TABLE DISPERSION */
 451             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 452             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 453             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 454             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 455             _MM_TRANSPOSE4_PS(Y,F,G,H);
 456             Heps             = _mm_mul_ps(vfeps,H);
 457             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 458             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 459             vvdw6            = _mm_mul_ps(c6_00,VV);
 460             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 461             fvdw6            = _mm_mul_ps(c6_00,FF);
 462
 463             /* CUBIC SPLINE TABLE REPULSION */
 464             vfitab           = _mm_add_epi32(vfitab,ifour);
 465             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 466             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 467             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 468             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 469             _MM_TRANSPOSE4_PS(Y,F,G,H);
 470             Heps             = _mm_mul_ps(vfeps,H);
 471             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 472             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 473             vvdw12           = _mm_mul_ps(c12_00,VV);
 474             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 475             fvdw12           = _mm_mul_ps(c12_00,FF);
 476             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 477             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm_andnot_ps(dummy_mask,velec);
 481             velecsum         = _mm_add_ps(velecsum,velec);
 482             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 483             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 484
 485             fscal            = _mm_add_ps(felec,fvdw);
 486
 487             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm_mul_ps(fscal,dx00);
 491             ty               = _mm_mul_ps(fscal,dy00);
 492             tz               = _mm_mul_ps(fscal,dz00);
 493
 494             /* Update vectorial force */
 495             fix0             = _mm_add_ps(fix0,tx);
 496             fiy0             = _mm_add_ps(fiy0,ty);
 497             fiz0             = _mm_add_ps(fiz0,tz);
 498
 499             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 500             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 501             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 502             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 503             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq10             = _mm_mul_ps(iq1,jq0);
 511
 512             /* REACTION-FIELD ELECTROSTATICS */
 513             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 514             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_andnot_ps(dummy_mask,velec);
 518             velecsum         = _mm_add_ps(velecsum,velec);
 519
 520             fscal            = felec;
 521
 522             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm_mul_ps(fscal,dx10);
 526             ty               = _mm_mul_ps(fscal,dy10);
 527             tz               = _mm_mul_ps(fscal,dz10);
 528
 529             /* Update vectorial force */
 530             fix1             = _mm_add_ps(fix1,tx);
 531             fiy1             = _mm_add_ps(fiy1,ty);
 532             fiz1             = _mm_add_ps(fiz1,tz);
 533
 534             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 535             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 536             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 537             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 538             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 539
 540             /**************************
 541              * CALCULATE INTERACTIONS *
 542              **************************/
 543
 544             /* Compute parameters for interactions between i and j atoms */
 545             qq20             = _mm_mul_ps(iq2,jq0);
 546
 547             /* REACTION-FIELD ELECTROSTATICS */
 548             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 549             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_andnot_ps(dummy_mask,velec);
 553             velecsum         = _mm_add_ps(velecsum,velec);
 554
 555             fscal            = felec;
 556
 557             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 558
 559             /* Calculate temporary vectorial force */
 560             tx               = _mm_mul_ps(fscal,dx20);
 561             ty               = _mm_mul_ps(fscal,dy20);
 562             tz               = _mm_mul_ps(fscal,dz20);
 563
 564             /* Update vectorial force */
 565             fix2             = _mm_add_ps(fix2,tx);
 566             fiy2             = _mm_add_ps(fiy2,ty);
 567             fiz2             = _mm_add_ps(fiz2,tz);
 568
 569             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 570             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 571             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 572             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 573             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 574
 575             /* Inner loop uses 132 flops */
 576         }
 577
 578         /* End of innermost loop */
 579
 580         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 581                                               f+i_coord_offset,fshift+i_shift_offset);
 582
 583         ggid                        = gid[iidx];
 584         /* Update potential energies */
 585         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 586         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 587
 588         /* Increment number of inner iterations */
 589         inneriter                  += j_index_end - j_index_start;
 590
 591         /* Outer loop uses 20 flops */
 592     }
 593
 594     /* Increment number of outer iterations */
 595     outeriter        += nri;
 596
 597     /* Update outer/inner flops */
 598
 599     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*132);
 600 }
 601 /*
 602  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_single
 603  * Electrostatics interaction: ReactionField
 604  * VdW interaction:            CubicSplineTable
 605  * Geometry:                   Water3-Particle
 606  * Calculate force/pot:        Force
 607  */
 608 void
 609 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_sse4_1_single
 610                     (t_nblist * gmx_restrict                nlist,
 611                      rvec * gmx_restrict                    xx,
 612                      rvec * gmx_restrict                    ff,
 613                      t_forcerec * gmx_restrict              fr,
 614                      t_mdatoms * gmx_restrict               mdatoms,
 615                      nb_kernel_data_t * gmx_restrict        kernel_data,
 616                      t_nrnb * gmx_restrict                  nrnb)
 617 {
 618     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 619      * just 0 for non-waters.
 620      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 621      * jnr indices corresponding to data put in the four positions in the SIMD register.
 622      */
 623     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 624     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 625     int              jnrA,jnrB,jnrC,jnrD;
 626     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 627     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 628     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 629     real             rcutoff_scalar;
 630     real             *shiftvec,*fshift,*x,*f;
 631     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 632     real             scratch[4*DIM];
 633     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 634     int              vdwioffset0;
 635     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 636     int              vdwioffset1;
 637     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 638     int              vdwioffset2;
 639     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 640     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 641     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 642     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 643     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 644     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 645     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 646     real             *charge;
 647     int              nvdwtype;
 648     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 649     int              *vdwtype;
 650     real             *vdwparam;
 651     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 652     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 653     __m128i          vfitab;
 654     __m128i          ifour       = _mm_set1_epi32(4);
 655     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 656     real             *vftab;
 657     __m128           dummy_mask,cutoff_mask;
 658     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 659     __m128           one     = _mm_set1_ps(1.0);
 660     __m128           two     = _mm_set1_ps(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_ps(fr->epsfac);
 673     charge           = mdatoms->chargeA;
 674     krf              = _mm_set1_ps(fr->ic->k_rf);
 675     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 676     crf              = _mm_set1_ps(fr->ic->c_rf);
 677     nvdwtype         = fr->ntype;
 678     vdwparam         = fr->nbfp;
 679     vdwtype          = mdatoms->typeA;
 680
 681     vftab            = kernel_data->table_vdw->data;
 682     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 683
 684     /* Setup water-specific parameters */
 685     inr              = nlist->iinr[0];
 686     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 687     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 688     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 689     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 690
 691     /* Avoid stupid compiler warnings */
 692     jnrA = jnrB = jnrC = jnrD = 0;
 693     j_coord_offsetA = 0;
 694     j_coord_offsetB = 0;
 695     j_coord_offsetC = 0;
 696     j_coord_offsetD = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     for(iidx=0;iidx<4*DIM;iidx++)
 702     {
 703         scratch[iidx] = 0.0;
 704     }
 705
 706     /* Start outer loop over neighborlists */
 707     for(iidx=0; iidx<nri; iidx++)
 708     {
 709         /* Load shift vector for this list */
 710         i_shift_offset   = DIM*shiftidx[iidx];
 711
 712         /* Load limits for loop over neighbors */
 713         j_index_start    = jindex[iidx];
 714         j_index_end      = jindex[iidx+1];
 715
 716         /* Get outer coordinate index */
 717         inr              = iinr[iidx];
 718         i_coord_offset   = DIM*inr;
 719
 720         /* Load i particle coords and add shift vector */
 721         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 722                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 723
 724         fix0             = _mm_setzero_ps();
 725         fiy0             = _mm_setzero_ps();
 726         fiz0             = _mm_setzero_ps();
 727         fix1             = _mm_setzero_ps();
 728         fiy1             = _mm_setzero_ps();
 729         fiz1             = _mm_setzero_ps();
 730         fix2             = _mm_setzero_ps();
 731         fiy2             = _mm_setzero_ps();
 732         fiz2             = _mm_setzero_ps();
 733
 734         /* Start inner kernel loop */
 735         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 736         {
 737
 738             /* Get j neighbor index, and coordinate index */
 739             jnrA             = jjnr[jidx];
 740             jnrB             = jjnr[jidx+1];
 741             jnrC             = jjnr[jidx+2];
 742             jnrD             = jjnr[jidx+3];
 743             j_coord_offsetA  = DIM*jnrA;
 744             j_coord_offsetB  = DIM*jnrB;
 745             j_coord_offsetC  = DIM*jnrC;
 746             j_coord_offsetD  = DIM*jnrD;
 747
 748             /* load j atom coordinates */
 749             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 750                                               x+j_coord_offsetC,x+j_coord_offsetD,
 751                                               &jx0,&jy0,&jz0);
 752
 753             /* Calculate displacement vector */
 754             dx00             = _mm_sub_ps(ix0,jx0);
 755             dy00             = _mm_sub_ps(iy0,jy0);
 756             dz00             = _mm_sub_ps(iz0,jz0);
 757             dx10             = _mm_sub_ps(ix1,jx0);
 758             dy10             = _mm_sub_ps(iy1,jy0);
 759             dz10             = _mm_sub_ps(iz1,jz0);
 760             dx20             = _mm_sub_ps(ix2,jx0);
 761             dy20             = _mm_sub_ps(iy2,jy0);
 762             dz20             = _mm_sub_ps(iz2,jz0);
 763
 764             /* Calculate squared distance and things based on it */
 765             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 766             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 767             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 768
 769             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 770             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 771             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 772
 773             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 774             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 775             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 776
 777             /* Load parameters for j particles */
 778             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 779                                                               charge+jnrC+0,charge+jnrD+0);
 780             vdwjidx0A        = 2*vdwtype[jnrA+0];
 781             vdwjidx0B        = 2*vdwtype[jnrB+0];
 782             vdwjidx0C        = 2*vdwtype[jnrC+0];
 783             vdwjidx0D        = 2*vdwtype[jnrD+0];
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             r00              = _mm_mul_ps(rsq00,rinv00);
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             qq00             = _mm_mul_ps(iq0,jq0);
 793             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 794                                          vdwparam+vdwioffset0+vdwjidx0B,
 795                                          vdwparam+vdwioffset0+vdwjidx0C,
 796                                          vdwparam+vdwioffset0+vdwjidx0D,
 797                                          &c6_00,&c12_00);
 798
 799             /* Calculate table index by multiplying r with table scale and truncate to integer */
 800             rt               = _mm_mul_ps(r00,vftabscale);
 801             vfitab           = _mm_cvttps_epi32(rt);
 802             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 803             vfitab           = _mm_slli_epi32(vfitab,3);
 804
 805             /* REACTION-FIELD ELECTROSTATICS */
 806             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 807
 808             /* CUBIC SPLINE TABLE DISPERSION */
 809             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 810             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 811             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 812             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 813             _MM_TRANSPOSE4_PS(Y,F,G,H);
 814             Heps             = _mm_mul_ps(vfeps,H);
 815             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 816             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 817             fvdw6            = _mm_mul_ps(c6_00,FF);
 818
 819             /* CUBIC SPLINE TABLE REPULSION */
 820             vfitab           = _mm_add_epi32(vfitab,ifour);
 821             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 822             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 823             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 824             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 825             _MM_TRANSPOSE4_PS(Y,F,G,H);
 826             Heps             = _mm_mul_ps(vfeps,H);
 827             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 828             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 829             fvdw12           = _mm_mul_ps(c12_00,FF);
 830             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 831
 832             fscal            = _mm_add_ps(felec,fvdw);
 833
 834             /* Calculate temporary vectorial force */
 835             tx               = _mm_mul_ps(fscal,dx00);
 836             ty               = _mm_mul_ps(fscal,dy00);
 837             tz               = _mm_mul_ps(fscal,dz00);
 838
 839             /* Update vectorial force */
 840             fix0             = _mm_add_ps(fix0,tx);
 841             fiy0             = _mm_add_ps(fiy0,ty);
 842             fiz0             = _mm_add_ps(fiz0,tz);
 843
 844             fjptrA             = f+j_coord_offsetA;
 845             fjptrB             = f+j_coord_offsetB;
 846             fjptrC             = f+j_coord_offsetC;
 847             fjptrD             = f+j_coord_offsetD;
 848             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 849
 850             /**************************
 851              * CALCULATE INTERACTIONS *
 852              **************************/
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq10             = _mm_mul_ps(iq1,jq0);
 856
 857             /* REACTION-FIELD ELECTROSTATICS */
 858             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 859
 860             fscal            = felec;
 861
 862             /* Calculate temporary vectorial force */
 863             tx               = _mm_mul_ps(fscal,dx10);
 864             ty               = _mm_mul_ps(fscal,dy10);
 865             tz               = _mm_mul_ps(fscal,dz10);
 866
 867             /* Update vectorial force */
 868             fix1             = _mm_add_ps(fix1,tx);
 869             fiy1             = _mm_add_ps(fiy1,ty);
 870             fiz1             = _mm_add_ps(fiz1,tz);
 871
 872             fjptrA             = f+j_coord_offsetA;
 873             fjptrB             = f+j_coord_offsetB;
 874             fjptrC             = f+j_coord_offsetC;
 875             fjptrD             = f+j_coord_offsetD;
 876             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq20             = _mm_mul_ps(iq2,jq0);
 884
 885             /* REACTION-FIELD ELECTROSTATICS */
 886             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 887
 888             fscal            = felec;
 889
 890             /* Calculate temporary vectorial force */
 891             tx               = _mm_mul_ps(fscal,dx20);
 892             ty               = _mm_mul_ps(fscal,dy20);
 893             tz               = _mm_mul_ps(fscal,dz20);
 894
 895             /* Update vectorial force */
 896             fix2             = _mm_add_ps(fix2,tx);
 897             fiy2             = _mm_add_ps(fiy2,ty);
 898             fiz2             = _mm_add_ps(fiz2,tz);
 899
 900             fjptrA             = f+j_coord_offsetA;
 901             fjptrB             = f+j_coord_offsetB;
 902             fjptrC             = f+j_coord_offsetC;
 903             fjptrD             = f+j_coord_offsetD;
 904             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 905
 906             /* Inner loop uses 108 flops */
 907         }
 908
 909         if(jidx<j_index_end)
 910         {
 911
 912             /* Get j neighbor index, and coordinate index */
 913             jnrlistA         = jjnr[jidx];
 914             jnrlistB         = jjnr[jidx+1];
 915             jnrlistC         = jjnr[jidx+2];
 916             jnrlistD         = jjnr[jidx+3];
 917             /* Sign of each element will be negative for non-real atoms.
 918              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 919              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 920              */
 921             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 922             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 923             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 924             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 925             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 926             j_coord_offsetA  = DIM*jnrA;
 927             j_coord_offsetB  = DIM*jnrB;
 928             j_coord_offsetC  = DIM*jnrC;
 929             j_coord_offsetD  = DIM*jnrD;
 930
 931             /* load j atom coordinates */
 932             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 933                                               x+j_coord_offsetC,x+j_coord_offsetD,
 934                                               &jx0,&jy0,&jz0);
 935
 936             /* Calculate displacement vector */
 937             dx00             = _mm_sub_ps(ix0,jx0);
 938             dy00             = _mm_sub_ps(iy0,jy0);
 939             dz00             = _mm_sub_ps(iz0,jz0);
 940             dx10             = _mm_sub_ps(ix1,jx0);
 941             dy10             = _mm_sub_ps(iy1,jy0);
 942             dz10             = _mm_sub_ps(iz1,jz0);
 943             dx20             = _mm_sub_ps(ix2,jx0);
 944             dy20             = _mm_sub_ps(iy2,jy0);
 945             dz20             = _mm_sub_ps(iz2,jz0);
 946
 947             /* Calculate squared distance and things based on it */
 948             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 949             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 950             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 951
 952             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 953             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 954             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 955
 956             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 957             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 958             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 959
 960             /* Load parameters for j particles */
 961             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 962                                                               charge+jnrC+0,charge+jnrD+0);
 963             vdwjidx0A        = 2*vdwtype[jnrA+0];
 964             vdwjidx0B        = 2*vdwtype[jnrB+0];
 965             vdwjidx0C        = 2*vdwtype[jnrC+0];
 966             vdwjidx0D        = 2*vdwtype[jnrD+0];
 967
 968             /**************************
 969              * CALCULATE INTERACTIONS *
 970              **************************/
 971
 972             r00              = _mm_mul_ps(rsq00,rinv00);
 973             r00              = _mm_andnot_ps(dummy_mask,r00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq00             = _mm_mul_ps(iq0,jq0);
 977             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 978                                          vdwparam+vdwioffset0+vdwjidx0B,
 979                                          vdwparam+vdwioffset0+vdwjidx0C,
 980                                          vdwparam+vdwioffset0+vdwjidx0D,
 981                                          &c6_00,&c12_00);
 982
 983             /* Calculate table index by multiplying r with table scale and truncate to integer */
 984             rt               = _mm_mul_ps(r00,vftabscale);
 985             vfitab           = _mm_cvttps_epi32(rt);
 986             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 987             vfitab           = _mm_slli_epi32(vfitab,3);
 988
 989             /* REACTION-FIELD ELECTROSTATICS */
 990             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 991
 992             /* CUBIC SPLINE TABLE DISPERSION */
 993             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 994             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 995             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 996             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 997             _MM_TRANSPOSE4_PS(Y,F,G,H);
 998             Heps             = _mm_mul_ps(vfeps,H);
 999             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1000             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1001             fvdw6            = _mm_mul_ps(c6_00,FF);
1002
1003             /* CUBIC SPLINE TABLE REPULSION */
1004             vfitab           = _mm_add_epi32(vfitab,ifour);
1005             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1006             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1007             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1008             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1009             _MM_TRANSPOSE4_PS(Y,F,G,H);
1010             Heps             = _mm_mul_ps(vfeps,H);
1011             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1012             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1013             fvdw12           = _mm_mul_ps(c12_00,FF);
1014             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1015
1016             fscal            = _mm_add_ps(felec,fvdw);
1017
1018             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1019
1020             /* Calculate temporary vectorial force */
1021             tx               = _mm_mul_ps(fscal,dx00);
1022             ty               = _mm_mul_ps(fscal,dy00);
1023             tz               = _mm_mul_ps(fscal,dz00);
1024
1025             /* Update vectorial force */
1026             fix0             = _mm_add_ps(fix0,tx);
1027             fiy0             = _mm_add_ps(fiy0,ty);
1028             fiz0             = _mm_add_ps(fiz0,tz);
1029
1030             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1031             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1032             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1033             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1034             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1035
1036             /**************************
1037              * CALCULATE INTERACTIONS *
1038              **************************/
1039
1040             /* Compute parameters for interactions between i and j atoms */
1041             qq10             = _mm_mul_ps(iq1,jq0);
1042
1043             /* REACTION-FIELD ELECTROSTATICS */
1044             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1049
1050             /* Calculate temporary vectorial force */
1051             tx               = _mm_mul_ps(fscal,dx10);
1052             ty               = _mm_mul_ps(fscal,dy10);
1053             tz               = _mm_mul_ps(fscal,dz10);
1054
1055             /* Update vectorial force */
1056             fix1             = _mm_add_ps(fix1,tx);
1057             fiy1             = _mm_add_ps(fiy1,ty);
1058             fiz1             = _mm_add_ps(fiz1,tz);
1059
1060             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1061             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1062             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1063             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1064             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1065
1066             /**************************
1067              * CALCULATE INTERACTIONS *
1068              **************************/
1069
1070             /* Compute parameters for interactions between i and j atoms */
1071             qq20             = _mm_mul_ps(iq2,jq0);
1072
1073             /* REACTION-FIELD ELECTROSTATICS */
1074             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_ps(fscal,dx20);
1082             ty               = _mm_mul_ps(fscal,dy20);
1083             tz               = _mm_mul_ps(fscal,dz20);
1084
1085             /* Update vectorial force */
1086             fix2             = _mm_add_ps(fix2,tx);
1087             fiy2             = _mm_add_ps(fiy2,ty);
1088             fiz2             = _mm_add_ps(fiz2,tz);
1089
1090             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1091             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1092             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1093             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1094             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1095
1096             /* Inner loop uses 109 flops */
1097         }
1098
1099         /* End of innermost loop */
1100
1101         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1102                                               f+i_coord_offset,fshift+i_shift_offset);
1103
1104         /* Increment number of inner iterations */
1105         inneriter                  += j_index_end - j_index_start;
1106
1107         /* Outer loop uses 18 flops */
1108     }
1109
1110     /* Increment number of outer iterations */
1111     outeriter        += nri;
1112
1113     /* Update outer/inner flops */
1114
1115     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1116 }