src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_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_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128           dummy_mask,cutoff_mask;
  90     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  91     __m128           one     = _mm_set1_ps(1.0);
  92     __m128           two     = _mm_set1_ps(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_ps(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     nvdwtype         = fr->ntype;
 107     vdwparam         = fr->nbfp;
 108     vdwtype          = mdatoms->typeA;
 109
 110     vftab            = kernel_data->table_vdw->data;
 111     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 116     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 117     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 118     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135         shX              = shiftvec[i_shift_offset+XX];
 136         shY              = shiftvec[i_shift_offset+YY];
 137         shZ              = shiftvec[i_shift_offset+ZZ];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 149         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 150         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 151         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 152         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 153         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 154         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 155         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 156         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 157
 158         fix0             = _mm_setzero_ps();
 159         fiy0             = _mm_setzero_ps();
 160         fiz0             = _mm_setzero_ps();
 161         fix1             = _mm_setzero_ps();
 162         fiy1             = _mm_setzero_ps();
 163         fiz1             = _mm_setzero_ps();
 164         fix2             = _mm_setzero_ps();
 165         fiy2             = _mm_setzero_ps();
 166         fiz2             = _mm_setzero_ps();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_ps();
 170         vvdwsum          = _mm_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181
 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_cvtepi32_ps(vfitab));
 242             vfitab           = _mm_slli_epi32(vfitab,3);
 243
 244             /* COULOMB ELECTROSTATICS */
 245             velec            = _mm_mul_ps(qq00,rinv00);
 246             felec            = _mm_mul_ps(velec,rinvsq00);
 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             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 294                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 295                                                    tx,ty,tz);
 296
 297             /**************************
 298              * CALCULATE INTERACTIONS *
 299              **************************/
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_ps(iq1,jq0);
 303
 304             /* COULOMB ELECTROSTATICS */
 305             velec            = _mm_mul_ps(qq10,rinv10);
 306             felec            = _mm_mul_ps(velec,rinvsq10);
 307
 308             /* Update potential sum for this i atom from the interaction with this j atom. */
 309             velecsum         = _mm_add_ps(velecsum,velec);
 310
 311             fscal            = felec;
 312
 313             /* Calculate temporary vectorial force */
 314             tx               = _mm_mul_ps(fscal,dx10);
 315             ty               = _mm_mul_ps(fscal,dy10);
 316             tz               = _mm_mul_ps(fscal,dz10);
 317
 318             /* Update vectorial force */
 319             fix1             = _mm_add_ps(fix1,tx);
 320             fiy1             = _mm_add_ps(fiy1,ty);
 321             fiz1             = _mm_add_ps(fiz1,tz);
 322
 323             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 324                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 325                                                    tx,ty,tz);
 326
 327             /**************************
 328              * CALCULATE INTERACTIONS *
 329              **************************/
 330
 331             /* Compute parameters for interactions between i and j atoms */
 332             qq20             = _mm_mul_ps(iq2,jq0);
 333
 334             /* COULOMB ELECTROSTATICS */
 335             velec            = _mm_mul_ps(qq20,rinv20);
 336             felec            = _mm_mul_ps(velec,rinvsq20);
 337
 338             /* Update potential sum for this i atom from the interaction with this j atom. */
 339             velecsum         = _mm_add_ps(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             /* Calculate temporary vectorial force */
 344             tx               = _mm_mul_ps(fscal,dx20);
 345             ty               = _mm_mul_ps(fscal,dy20);
 346             tz               = _mm_mul_ps(fscal,dz20);
 347
 348             /* Update vectorial force */
 349             fix2             = _mm_add_ps(fix2,tx);
 350             fiy2             = _mm_add_ps(fiy2,ty);
 351             fiz2             = _mm_add_ps(fiz2,tz);
 352
 353             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 354                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 355                                                    tx,ty,tz);
 356
 357             /* Inner loop uses 119 flops */
 358         }
 359
 360         if(jidx<j_index_end)
 361         {
 362
 363             /* Get j neighbor index, and coordinate index */
 364             jnrA             = jjnr[jidx];
 365             jnrB             = jjnr[jidx+1];
 366             jnrC             = jjnr[jidx+2];
 367             jnrD             = jjnr[jidx+3];
 368
 369             /* Sign of each element will be negative for non-real atoms.
 370              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 371              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 372              */
 373             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 374             jnrA       = (jnrA>=0) ? jnrA : 0;
 375             jnrB       = (jnrB>=0) ? jnrB : 0;
 376             jnrC       = (jnrC>=0) ? jnrC : 0;
 377             jnrD       = (jnrD>=0) ? jnrD : 0;
 378
 379             j_coord_offsetA  = DIM*jnrA;
 380             j_coord_offsetB  = DIM*jnrB;
 381             j_coord_offsetC  = DIM*jnrC;
 382             j_coord_offsetD  = DIM*jnrD;
 383
 384             /* load j atom coordinates */
 385             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 386                                               x+j_coord_offsetC,x+j_coord_offsetD,
 387                                               &jx0,&jy0,&jz0);
 388
 389             /* Calculate displacement vector */
 390             dx00             = _mm_sub_ps(ix0,jx0);
 391             dy00             = _mm_sub_ps(iy0,jy0);
 392             dz00             = _mm_sub_ps(iz0,jz0);
 393             dx10             = _mm_sub_ps(ix1,jx0);
 394             dy10             = _mm_sub_ps(iy1,jy0);
 395             dz10             = _mm_sub_ps(iz1,jz0);
 396             dx20             = _mm_sub_ps(ix2,jx0);
 397             dy20             = _mm_sub_ps(iy2,jy0);
 398             dz20             = _mm_sub_ps(iz2,jz0);
 399
 400             /* Calculate squared distance and things based on it */
 401             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 402             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 403             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 404
 405             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 406             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 407             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 408
 409             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 410             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 411             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 412
 413             /* Load parameters for j particles */
 414             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 415                                                               charge+jnrC+0,charge+jnrD+0);
 416             vdwjidx0A        = 2*vdwtype[jnrA+0];
 417             vdwjidx0B        = 2*vdwtype[jnrB+0];
 418             vdwjidx0C        = 2*vdwtype[jnrC+0];
 419             vdwjidx0D        = 2*vdwtype[jnrD+0];
 420
 421             /**************************
 422              * CALCULATE INTERACTIONS *
 423              **************************/
 424
 425             r00              = _mm_mul_ps(rsq00,rinv00);
 426             r00              = _mm_andnot_ps(dummy_mask,r00);
 427
 428             /* Compute parameters for interactions between i and j atoms */
 429             qq00             = _mm_mul_ps(iq0,jq0);
 430             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 431                                          vdwparam+vdwioffset0+vdwjidx0B,
 432                                          vdwparam+vdwioffset0+vdwjidx0C,
 433                                          vdwparam+vdwioffset0+vdwjidx0D,
 434                                          &c6_00,&c12_00);
 435
 436             /* Calculate table index by multiplying r with table scale and truncate to integer */
 437             rt               = _mm_mul_ps(r00,vftabscale);
 438             vfitab           = _mm_cvttps_epi32(rt);
 439             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 440             vfitab           = _mm_slli_epi32(vfitab,3);
 441
 442             /* COULOMB ELECTROSTATICS */
 443             velec            = _mm_mul_ps(qq00,rinv00);
 444             felec            = _mm_mul_ps(velec,rinvsq00);
 445
 446             /* CUBIC SPLINE TABLE DISPERSION */
 447             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 448             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 449             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 450             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 451             _MM_TRANSPOSE4_PS(Y,F,G,H);
 452             Heps             = _mm_mul_ps(vfeps,H);
 453             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 454             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 455             vvdw6            = _mm_mul_ps(c6_00,VV);
 456             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 457             fvdw6            = _mm_mul_ps(c6_00,FF);
 458
 459             /* CUBIC SPLINE TABLE REPULSION */
 460             vfitab           = _mm_add_epi32(vfitab,ifour);
 461             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 462             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 463             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 464             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 465             _MM_TRANSPOSE4_PS(Y,F,G,H);
 466             Heps             = _mm_mul_ps(vfeps,H);
 467             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 468             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 469             vvdw12           = _mm_mul_ps(c12_00,VV);
 470             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 471             fvdw12           = _mm_mul_ps(c12_00,FF);
 472             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 473             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velec            = _mm_andnot_ps(dummy_mask,velec);
 477             velecsum         = _mm_add_ps(velecsum,velec);
 478             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 479             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 480
 481             fscal            = _mm_add_ps(felec,fvdw);
 482
 483             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 484
 485             /* Calculate temporary vectorial force */
 486             tx               = _mm_mul_ps(fscal,dx00);
 487             ty               = _mm_mul_ps(fscal,dy00);
 488             tz               = _mm_mul_ps(fscal,dz00);
 489
 490             /* Update vectorial force */
 491             fix0             = _mm_add_ps(fix0,tx);
 492             fiy0             = _mm_add_ps(fiy0,ty);
 493             fiz0             = _mm_add_ps(fiz0,tz);
 494
 495             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 496                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 497                                                    tx,ty,tz);
 498
 499             /**************************
 500              * CALCULATE INTERACTIONS *
 501              **************************/
 502
 503             /* Compute parameters for interactions between i and j atoms */
 504             qq10             = _mm_mul_ps(iq1,jq0);
 505
 506             /* COULOMB ELECTROSTATICS */
 507             velec            = _mm_mul_ps(qq10,rinv10);
 508             felec            = _mm_mul_ps(velec,rinvsq10);
 509
 510             /* Update potential sum for this i atom from the interaction with this j atom. */
 511             velec            = _mm_andnot_ps(dummy_mask,velec);
 512             velecsum         = _mm_add_ps(velecsum,velec);
 513
 514             fscal            = felec;
 515
 516             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 517
 518             /* Calculate temporary vectorial force */
 519             tx               = _mm_mul_ps(fscal,dx10);
 520             ty               = _mm_mul_ps(fscal,dy10);
 521             tz               = _mm_mul_ps(fscal,dz10);
 522
 523             /* Update vectorial force */
 524             fix1             = _mm_add_ps(fix1,tx);
 525             fiy1             = _mm_add_ps(fiy1,ty);
 526             fiz1             = _mm_add_ps(fiz1,tz);
 527
 528             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 529                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 530                                                    tx,ty,tz);
 531
 532             /**************************
 533              * CALCULATE INTERACTIONS *
 534              **************************/
 535
 536             /* Compute parameters for interactions between i and j atoms */
 537             qq20             = _mm_mul_ps(iq2,jq0);
 538
 539             /* COULOMB ELECTROSTATICS */
 540             velec            = _mm_mul_ps(qq20,rinv20);
 541             felec            = _mm_mul_ps(velec,rinvsq20);
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_andnot_ps(dummy_mask,velec);
 545             velecsum         = _mm_add_ps(velecsum,velec);
 546
 547             fscal            = felec;
 548
 549             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm_mul_ps(fscal,dx20);
 553             ty               = _mm_mul_ps(fscal,dy20);
 554             tz               = _mm_mul_ps(fscal,dz20);
 555
 556             /* Update vectorial force */
 557             fix2             = _mm_add_ps(fix2,tx);
 558             fiy2             = _mm_add_ps(fiy2,ty);
 559             fiz2             = _mm_add_ps(fiz2,tz);
 560
 561             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 562                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 563                                                    tx,ty,tz);
 564
 565             /* Inner loop uses 120 flops */
 566         }
 567
 568         /* End of innermost loop */
 569
 570         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 571                                               f+i_coord_offset,fshift+i_shift_offset);
 572
 573         ggid                        = gid[iidx];
 574         /* Update potential energies */
 575         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 576         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 577
 578         /* Increment number of inner iterations */
 579         inneriter                  += j_index_end - j_index_start;
 580
 581         /* Outer loop uses 29 flops */
 582     }
 583
 584     /* Increment number of outer iterations */
 585     outeriter        += nri;
 586
 587     /* Update outer/inner flops */
 588
 589     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*120);
 590 }
 591 /*
 592  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_single
 593  * Electrostatics interaction: Coulomb
 594  * VdW interaction:            CubicSplineTable
 595  * Geometry:                   Water3-Particle
 596  * Calculate force/pot:        Force
 597  */
 598 void
 599 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse2_single
 600                     (t_nblist * gmx_restrict                nlist,
 601                      rvec * gmx_restrict                    xx,
 602                      rvec * gmx_restrict                    ff,
 603                      t_forcerec * gmx_restrict              fr,
 604                      t_mdatoms * gmx_restrict               mdatoms,
 605                      nb_kernel_data_t * gmx_restrict        kernel_data,
 606                      t_nrnb * gmx_restrict                  nrnb)
 607 {
 608     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 609      * just 0 for non-waters.
 610      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 611      * jnr indices corresponding to data put in the four positions in the SIMD register.
 612      */
 613     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 614     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 615     int              jnrA,jnrB,jnrC,jnrD;
 616     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 617     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 618     real             shX,shY,shZ,rcutoff_scalar;
 619     real             *shiftvec,*fshift,*x,*f;
 620     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 621     int              vdwioffset0;
 622     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 623     int              vdwioffset1;
 624     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 625     int              vdwioffset2;
 626     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 627     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 628     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 629     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 630     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 631     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 632     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 633     real             *charge;
 634     int              nvdwtype;
 635     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 636     int              *vdwtype;
 637     real             *vdwparam;
 638     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 639     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 640     __m128i          vfitab;
 641     __m128i          ifour       = _mm_set1_epi32(4);
 642     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 643     real             *vftab;
 644     __m128           dummy_mask,cutoff_mask;
 645     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 646     __m128           one     = _mm_set1_ps(1.0);
 647     __m128           two     = _mm_set1_ps(2.0);
 648     x                = xx[0];
 649     f                = ff[0];
 650
 651     nri              = nlist->nri;
 652     iinr             = nlist->iinr;
 653     jindex           = nlist->jindex;
 654     jjnr             = nlist->jjnr;
 655     shiftidx         = nlist->shift;
 656     gid              = nlist->gid;
 657     shiftvec         = fr->shift_vec[0];
 658     fshift           = fr->fshift[0];
 659     facel            = _mm_set1_ps(fr->epsfac);
 660     charge           = mdatoms->chargeA;
 661     nvdwtype         = fr->ntype;
 662     vdwparam         = fr->nbfp;
 663     vdwtype          = mdatoms->typeA;
 664
 665     vftab            = kernel_data->table_vdw->data;
 666     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 667
 668     /* Setup water-specific parameters */
 669     inr              = nlist->iinr[0];
 670     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 671     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 672     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 673     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 674
 675     /* Avoid stupid compiler warnings */
 676     jnrA = jnrB = jnrC = jnrD = 0;
 677     j_coord_offsetA = 0;
 678     j_coord_offsetB = 0;
 679     j_coord_offsetC = 0;
 680     j_coord_offsetD = 0;
 681
 682     outeriter        = 0;
 683     inneriter        = 0;
 684
 685     /* Start outer loop over neighborlists */
 686     for(iidx=0; iidx<nri; iidx++)
 687     {
 688         /* Load shift vector for this list */
 689         i_shift_offset   = DIM*shiftidx[iidx];
 690         shX              = shiftvec[i_shift_offset+XX];
 691         shY              = shiftvec[i_shift_offset+YY];
 692         shZ              = shiftvec[i_shift_offset+ZZ];
 693
 694         /* Load limits for loop over neighbors */
 695         j_index_start    = jindex[iidx];
 696         j_index_end      = jindex[iidx+1];
 697
 698         /* Get outer coordinate index */
 699         inr              = iinr[iidx];
 700         i_coord_offset   = DIM*inr;
 701
 702         /* Load i particle coords and add shift vector */
 703         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 704         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 705         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 706         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 707         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 708         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 709         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 710         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 711         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 712
 713         fix0             = _mm_setzero_ps();
 714         fiy0             = _mm_setzero_ps();
 715         fiz0             = _mm_setzero_ps();
 716         fix1             = _mm_setzero_ps();
 717         fiy1             = _mm_setzero_ps();
 718         fiz1             = _mm_setzero_ps();
 719         fix2             = _mm_setzero_ps();
 720         fiy2             = _mm_setzero_ps();
 721         fiz2             = _mm_setzero_ps();
 722
 723         /* Start inner kernel loop */
 724         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 725         {
 726
 727             /* Get j neighbor index, and coordinate index */
 728             jnrA             = jjnr[jidx];
 729             jnrB             = jjnr[jidx+1];
 730             jnrC             = jjnr[jidx+2];
 731             jnrD             = jjnr[jidx+3];
 732
 733             j_coord_offsetA  = DIM*jnrA;
 734             j_coord_offsetB  = DIM*jnrB;
 735             j_coord_offsetC  = DIM*jnrC;
 736             j_coord_offsetD  = DIM*jnrD;
 737
 738             /* load j atom coordinates */
 739             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 740                                               x+j_coord_offsetC,x+j_coord_offsetD,
 741                                               &jx0,&jy0,&jz0);
 742
 743             /* Calculate displacement vector */
 744             dx00             = _mm_sub_ps(ix0,jx0);
 745             dy00             = _mm_sub_ps(iy0,jy0);
 746             dz00             = _mm_sub_ps(iz0,jz0);
 747             dx10             = _mm_sub_ps(ix1,jx0);
 748             dy10             = _mm_sub_ps(iy1,jy0);
 749             dz10             = _mm_sub_ps(iz1,jz0);
 750             dx20             = _mm_sub_ps(ix2,jx0);
 751             dy20             = _mm_sub_ps(iy2,jy0);
 752             dz20             = _mm_sub_ps(iz2,jz0);
 753
 754             /* Calculate squared distance and things based on it */
 755             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 756             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 757             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 758
 759             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 760             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 761             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 762
 763             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 764             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 765             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 766
 767             /* Load parameters for j particles */
 768             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 769                                                               charge+jnrC+0,charge+jnrD+0);
 770             vdwjidx0A        = 2*vdwtype[jnrA+0];
 771             vdwjidx0B        = 2*vdwtype[jnrB+0];
 772             vdwjidx0C        = 2*vdwtype[jnrC+0];
 773             vdwjidx0D        = 2*vdwtype[jnrD+0];
 774
 775             /**************************
 776              * CALCULATE INTERACTIONS *
 777              **************************/
 778
 779             r00              = _mm_mul_ps(rsq00,rinv00);
 780
 781             /* Compute parameters for interactions between i and j atoms */
 782             qq00             = _mm_mul_ps(iq0,jq0);
 783             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 784                                          vdwparam+vdwioffset0+vdwjidx0B,
 785                                          vdwparam+vdwioffset0+vdwjidx0C,
 786                                          vdwparam+vdwioffset0+vdwjidx0D,
 787                                          &c6_00,&c12_00);
 788
 789             /* Calculate table index by multiplying r with table scale and truncate to integer */
 790             rt               = _mm_mul_ps(r00,vftabscale);
 791             vfitab           = _mm_cvttps_epi32(rt);
 792             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 793             vfitab           = _mm_slli_epi32(vfitab,3);
 794
 795             /* COULOMB ELECTROSTATICS */
 796             velec            = _mm_mul_ps(qq00,rinv00);
 797             felec            = _mm_mul_ps(velec,rinvsq00);
 798
 799             /* CUBIC SPLINE TABLE DISPERSION */
 800             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 801             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 802             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 803             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 804             _MM_TRANSPOSE4_PS(Y,F,G,H);
 805             Heps             = _mm_mul_ps(vfeps,H);
 806             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 807             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 808             fvdw6            = _mm_mul_ps(c6_00,FF);
 809
 810             /* CUBIC SPLINE TABLE REPULSION */
 811             vfitab           = _mm_add_epi32(vfitab,ifour);
 812             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 813             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 814             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 815             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 816             _MM_TRANSPOSE4_PS(Y,F,G,H);
 817             Heps             = _mm_mul_ps(vfeps,H);
 818             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 819             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 820             fvdw12           = _mm_mul_ps(c12_00,FF);
 821             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 822
 823             fscal            = _mm_add_ps(felec,fvdw);
 824
 825             /* Calculate temporary vectorial force */
 826             tx               = _mm_mul_ps(fscal,dx00);
 827             ty               = _mm_mul_ps(fscal,dy00);
 828             tz               = _mm_mul_ps(fscal,dz00);
 829
 830             /* Update vectorial force */
 831             fix0             = _mm_add_ps(fix0,tx);
 832             fiy0             = _mm_add_ps(fiy0,ty);
 833             fiz0             = _mm_add_ps(fiz0,tz);
 834
 835             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 836                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 837                                                    tx,ty,tz);
 838
 839             /**************************
 840              * CALCULATE INTERACTIONS *
 841              **************************/
 842
 843             /* Compute parameters for interactions between i and j atoms */
 844             qq10             = _mm_mul_ps(iq1,jq0);
 845
 846             /* COULOMB ELECTROSTATICS */
 847             velec            = _mm_mul_ps(qq10,rinv10);
 848             felec            = _mm_mul_ps(velec,rinvsq10);
 849
 850             fscal            = felec;
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm_mul_ps(fscal,dx10);
 854             ty               = _mm_mul_ps(fscal,dy10);
 855             tz               = _mm_mul_ps(fscal,dz10);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm_add_ps(fix1,tx);
 859             fiy1             = _mm_add_ps(fiy1,ty);
 860             fiz1             = _mm_add_ps(fiz1,tz);
 861
 862             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 863                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 864                                                    tx,ty,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq20             = _mm_mul_ps(iq2,jq0);
 872
 873             /* COULOMB ELECTROSTATICS */
 874             velec            = _mm_mul_ps(qq20,rinv20);
 875             felec            = _mm_mul_ps(velec,rinvsq20);
 876
 877             fscal            = felec;
 878
 879             /* Calculate temporary vectorial force */
 880             tx               = _mm_mul_ps(fscal,dx20);
 881             ty               = _mm_mul_ps(fscal,dy20);
 882             tz               = _mm_mul_ps(fscal,dz20);
 883
 884             /* Update vectorial force */
 885             fix2             = _mm_add_ps(fix2,tx);
 886             fiy2             = _mm_add_ps(fiy2,ty);
 887             fiz2             = _mm_add_ps(fiz2,tz);
 888
 889             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 890                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 891                                                    tx,ty,tz);
 892
 893             /* Inner loop uses 108 flops */
 894         }
 895
 896         if(jidx<j_index_end)
 897         {
 898
 899             /* Get j neighbor index, and coordinate index */
 900             jnrA             = jjnr[jidx];
 901             jnrB             = jjnr[jidx+1];
 902             jnrC             = jjnr[jidx+2];
 903             jnrD             = jjnr[jidx+3];
 904
 905             /* Sign of each element will be negative for non-real atoms.
 906              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 907              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 908              */
 909             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 910             jnrA       = (jnrA>=0) ? jnrA : 0;
 911             jnrB       = (jnrB>=0) ? jnrB : 0;
 912             jnrC       = (jnrC>=0) ? jnrC : 0;
 913             jnrD       = (jnrD>=0) ? jnrD : 0;
 914
 915             j_coord_offsetA  = DIM*jnrA;
 916             j_coord_offsetB  = DIM*jnrB;
 917             j_coord_offsetC  = DIM*jnrC;
 918             j_coord_offsetD  = DIM*jnrD;
 919
 920             /* load j atom coordinates */
 921             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 922                                               x+j_coord_offsetC,x+j_coord_offsetD,
 923                                               &jx0,&jy0,&jz0);
 924
 925             /* Calculate displacement vector */
 926             dx00             = _mm_sub_ps(ix0,jx0);
 927             dy00             = _mm_sub_ps(iy0,jy0);
 928             dz00             = _mm_sub_ps(iz0,jz0);
 929             dx10             = _mm_sub_ps(ix1,jx0);
 930             dy10             = _mm_sub_ps(iy1,jy0);
 931             dz10             = _mm_sub_ps(iz1,jz0);
 932             dx20             = _mm_sub_ps(ix2,jx0);
 933             dy20             = _mm_sub_ps(iy2,jy0);
 934             dz20             = _mm_sub_ps(iz2,jz0);
 935
 936             /* Calculate squared distance and things based on it */
 937             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 938             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 939             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 940
 941             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 942             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 943             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 944
 945             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 946             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 947             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 948
 949             /* Load parameters for j particles */
 950             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 951                                                               charge+jnrC+0,charge+jnrD+0);
 952             vdwjidx0A        = 2*vdwtype[jnrA+0];
 953             vdwjidx0B        = 2*vdwtype[jnrB+0];
 954             vdwjidx0C        = 2*vdwtype[jnrC+0];
 955             vdwjidx0D        = 2*vdwtype[jnrD+0];
 956
 957             /**************************
 958              * CALCULATE INTERACTIONS *
 959              **************************/
 960
 961             r00              = _mm_mul_ps(rsq00,rinv00);
 962             r00              = _mm_andnot_ps(dummy_mask,r00);
 963
 964             /* Compute parameters for interactions between i and j atoms */
 965             qq00             = _mm_mul_ps(iq0,jq0);
 966             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 967                                          vdwparam+vdwioffset0+vdwjidx0B,
 968                                          vdwparam+vdwioffset0+vdwjidx0C,
 969                                          vdwparam+vdwioffset0+vdwjidx0D,
 970                                          &c6_00,&c12_00);
 971
 972             /* Calculate table index by multiplying r with table scale and truncate to integer */
 973             rt               = _mm_mul_ps(r00,vftabscale);
 974             vfitab           = _mm_cvttps_epi32(rt);
 975             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 976             vfitab           = _mm_slli_epi32(vfitab,3);
 977
 978             /* COULOMB ELECTROSTATICS */
 979             velec            = _mm_mul_ps(qq00,rinv00);
 980             felec            = _mm_mul_ps(velec,rinvsq00);
 981
 982             /* CUBIC SPLINE TABLE DISPERSION */
 983             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 984             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 985             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 986             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 987             _MM_TRANSPOSE4_PS(Y,F,G,H);
 988             Heps             = _mm_mul_ps(vfeps,H);
 989             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 990             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 991             fvdw6            = _mm_mul_ps(c6_00,FF);
 992
 993             /* CUBIC SPLINE TABLE REPULSION */
 994             vfitab           = _mm_add_epi32(vfitab,ifour);
 995             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 996             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 997             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 998             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 999             _MM_TRANSPOSE4_PS(Y,F,G,H);
1000             Heps             = _mm_mul_ps(vfeps,H);
1001             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1002             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1003             fvdw12           = _mm_mul_ps(c12_00,FF);
1004             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1005
1006             fscal            = _mm_add_ps(felec,fvdw);
1007
1008             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1009
1010             /* Calculate temporary vectorial force */
1011             tx               = _mm_mul_ps(fscal,dx00);
1012             ty               = _mm_mul_ps(fscal,dy00);
1013             tz               = _mm_mul_ps(fscal,dz00);
1014
1015             /* Update vectorial force */
1016             fix0             = _mm_add_ps(fix0,tx);
1017             fiy0             = _mm_add_ps(fiy0,ty);
1018             fiz0             = _mm_add_ps(fiz0,tz);
1019
1020             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1021                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1022                                                    tx,ty,tz);
1023
1024             /**************************
1025              * CALCULATE INTERACTIONS *
1026              **************************/
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq10             = _mm_mul_ps(iq1,jq0);
1030
1031             /* COULOMB ELECTROSTATICS */
1032             velec            = _mm_mul_ps(qq10,rinv10);
1033             felec            = _mm_mul_ps(velec,rinvsq10);
1034
1035             fscal            = felec;
1036
1037             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1038
1039             /* Calculate temporary vectorial force */
1040             tx               = _mm_mul_ps(fscal,dx10);
1041             ty               = _mm_mul_ps(fscal,dy10);
1042             tz               = _mm_mul_ps(fscal,dz10);
1043
1044             /* Update vectorial force */
1045             fix1             = _mm_add_ps(fix1,tx);
1046             fiy1             = _mm_add_ps(fiy1,ty);
1047             fiz1             = _mm_add_ps(fiz1,tz);
1048
1049             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1050                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1051                                                    tx,ty,tz);
1052
1053             /**************************
1054              * CALCULATE INTERACTIONS *
1055              **************************/
1056
1057             /* Compute parameters for interactions between i and j atoms */
1058             qq20             = _mm_mul_ps(iq2,jq0);
1059
1060             /* COULOMB ELECTROSTATICS */
1061             velec            = _mm_mul_ps(qq20,rinv20);
1062             felec            = _mm_mul_ps(velec,rinvsq20);
1063
1064             fscal            = felec;
1065
1066             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1067
1068             /* Calculate temporary vectorial force */
1069             tx               = _mm_mul_ps(fscal,dx20);
1070             ty               = _mm_mul_ps(fscal,dy20);
1071             tz               = _mm_mul_ps(fscal,dz20);
1072
1073             /* Update vectorial force */
1074             fix2             = _mm_add_ps(fix2,tx);
1075             fiy2             = _mm_add_ps(fiy2,ty);
1076             fiz2             = _mm_add_ps(fiz2,tz);
1077
1078             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1079                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1080                                                    tx,ty,tz);
1081
1082             /* Inner loop uses 109 flops */
1083         }
1084
1085         /* End of innermost loop */
1086
1087         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1088                                               f+i_coord_offset,fshift+i_shift_offset);
1089
1090         /* Increment number of inner iterations */
1091         inneriter                  += j_index_end - j_index_start;
1092
1093         /* Outer loop uses 27 flops */
1094     }
1095
1096     /* Increment number of outer iterations */
1097     outeriter        += nri;
1098
1099     /* Update outer/inner flops */
1100
1101     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*109);
1102 }