src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwCSTab_GeomW4P1_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_ElecEw_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  94     real             *vftab;
  95     __m128i          ewitab;
  96     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115     nvdwtype         = fr->ntype;
 116     vdwparam         = fr->nbfp;
 117     vdwtype          = mdatoms->typeA;
 118
 119     vftab            = kernel_data->table_vdw->data;
 120     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 121
 122     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 123     ewtab            = fr->ic->tabq_coul_FDV0;
 124     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 125     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 126
 127     /* Setup water-specific parameters */
 128     inr              = nlist->iinr[0];
 129     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 130     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 131     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 132     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 133
 134     /* Avoid stupid compiler warnings */
 135     jnrA = jnrB = jnrC = jnrD = 0;
 136     j_coord_offsetA = 0;
 137     j_coord_offsetB = 0;
 138     j_coord_offsetC = 0;
 139     j_coord_offsetD = 0;
 140
 141     outeriter        = 0;
 142     inneriter        = 0;
 143
 144     for(iidx=0;iidx<4*DIM;iidx++)
 145     {
 146         scratch[iidx] = 0.0;
 147     }
 148
 149     /* Start outer loop over neighborlists */
 150     for(iidx=0; iidx<nri; iidx++)
 151     {
 152         /* Load shift vector for this list */
 153         i_shift_offset   = DIM*shiftidx[iidx];
 154
 155         /* Load limits for loop over neighbors */
 156         j_index_start    = jindex[iidx];
 157         j_index_end      = jindex[iidx+1];
 158
 159         /* Get outer coordinate index */
 160         inr              = iinr[iidx];
 161         i_coord_offset   = DIM*inr;
 162
 163         /* Load i particle coords and add shift vector */
 164         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 165                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176         fix3             = _mm_setzero_ps();
 177         fiy3             = _mm_setzero_ps();
 178         fiz3             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195             j_coord_offsetC  = DIM*jnrC;
 196             j_coord_offsetD  = DIM*jnrD;
 197
 198             /* load j atom coordinates */
 199             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 200                                               x+j_coord_offsetC,x+j_coord_offsetD,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_ps(ix0,jx0);
 205             dy00             = _mm_sub_ps(iy0,jy0);
 206             dz00             = _mm_sub_ps(iz0,jz0);
 207             dx10             = _mm_sub_ps(ix1,jx0);
 208             dy10             = _mm_sub_ps(iy1,jy0);
 209             dz10             = _mm_sub_ps(iz1,jz0);
 210             dx20             = _mm_sub_ps(ix2,jx0);
 211             dy20             = _mm_sub_ps(iy2,jy0);
 212             dz20             = _mm_sub_ps(iz2,jz0);
 213             dx30             = _mm_sub_ps(ix3,jx0);
 214             dy30             = _mm_sub_ps(iy3,jy0);
 215             dz30             = _mm_sub_ps(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 222
 223             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 224             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 225             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 226             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 227
 228             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 234                                                               charge+jnrC+0,charge+jnrD+0);
 235             vdwjidx0A        = 2*vdwtype[jnrA+0];
 236             vdwjidx0B        = 2*vdwtype[jnrB+0];
 237             vdwjidx0C        = 2*vdwtype[jnrC+0];
 238             vdwjidx0D        = 2*vdwtype[jnrD+0];
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             r00              = _mm_mul_ps(rsq00,rinv00);
 245
 246             /* Compute parameters for interactions between i and j atoms */
 247             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 248                                          vdwparam+vdwioffset0+vdwjidx0B,
 249                                          vdwparam+vdwioffset0+vdwjidx0C,
 250                                          vdwparam+vdwioffset0+vdwjidx0D,
 251                                          &c6_00,&c12_00);
 252
 253             /* Calculate table index by multiplying r with table scale and truncate to integer */
 254             rt               = _mm_mul_ps(r00,vftabscale);
 255             vfitab           = _mm_cvttps_epi32(rt);
 256             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 257             vfitab           = _mm_slli_epi32(vfitab,3);
 258
 259             /* CUBIC SPLINE TABLE DISPERSION */
 260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 264             _MM_TRANSPOSE4_PS(Y,F,G,H);
 265             Heps             = _mm_mul_ps(vfeps,H);
 266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 267             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 268             vvdw6            = _mm_mul_ps(c6_00,VV);
 269             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 270             fvdw6            = _mm_mul_ps(c6_00,FF);
 271
 272             /* CUBIC SPLINE TABLE REPULSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Heps             = _mm_mul_ps(vfeps,H);
 280             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 281             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 282             vvdw12           = _mm_mul_ps(c12_00,VV);
 283             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 284             fvdw12           = _mm_mul_ps(c12_00,FF);
 285             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 286             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 287
 288             /* Update potential sum for this i atom from the interaction with this j atom. */
 289             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 290
 291             fscal            = fvdw;
 292
 293             /* Calculate temporary vectorial force */
 294             tx               = _mm_mul_ps(fscal,dx00);
 295             ty               = _mm_mul_ps(fscal,dy00);
 296             tz               = _mm_mul_ps(fscal,dz00);
 297
 298             /* Update vectorial force */
 299             fix0             = _mm_add_ps(fix0,tx);
 300             fiy0             = _mm_add_ps(fiy0,ty);
 301             fiz0             = _mm_add_ps(fiz0,tz);
 302
 303             fjptrA             = f+j_coord_offsetA;
 304             fjptrB             = f+j_coord_offsetB;
 305             fjptrC             = f+j_coord_offsetC;
 306             fjptrD             = f+j_coord_offsetD;
 307             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 308
 309             /**************************
 310              * CALCULATE INTERACTIONS *
 311              **************************/
 312
 313             r10              = _mm_mul_ps(rsq10,rinv10);
 314
 315             /* Compute parameters for interactions between i and j atoms */
 316             qq10             = _mm_mul_ps(iq1,jq0);
 317
 318             /* EWALD ELECTROSTATICS */
 319
 320             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 321             ewrt             = _mm_mul_ps(r10,ewtabscale);
 322             ewitab           = _mm_cvttps_epi32(ewrt);
 323             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 324             ewitab           = _mm_slli_epi32(ewitab,2);
 325             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 326             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 327             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 328             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 329             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 330             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 331             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 332             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 333             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm_mul_ps(fscal,dx10);
 342             ty               = _mm_mul_ps(fscal,dy10);
 343             tz               = _mm_mul_ps(fscal,dz10);
 344
 345             /* Update vectorial force */
 346             fix1             = _mm_add_ps(fix1,tx);
 347             fiy1             = _mm_add_ps(fiy1,ty);
 348             fiz1             = _mm_add_ps(fiz1,tz);
 349
 350             fjptrA             = f+j_coord_offsetA;
 351             fjptrB             = f+j_coord_offsetB;
 352             fjptrC             = f+j_coord_offsetC;
 353             fjptrD             = f+j_coord_offsetD;
 354             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             r20              = _mm_mul_ps(rsq20,rinv20);
 361
 362             /* Compute parameters for interactions between i and j atoms */
 363             qq20             = _mm_mul_ps(iq2,jq0);
 364
 365             /* EWALD ELECTROSTATICS */
 366
 367             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 368             ewrt             = _mm_mul_ps(r20,ewtabscale);
 369             ewitab           = _mm_cvttps_epi32(ewrt);
 370             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 371             ewitab           = _mm_slli_epi32(ewitab,2);
 372             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 373             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 374             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 375             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 376             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 377             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 378             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 379             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 380             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velecsum         = _mm_add_ps(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             /* Calculate temporary vectorial force */
 388             tx               = _mm_mul_ps(fscal,dx20);
 389             ty               = _mm_mul_ps(fscal,dy20);
 390             tz               = _mm_mul_ps(fscal,dz20);
 391
 392             /* Update vectorial force */
 393             fix2             = _mm_add_ps(fix2,tx);
 394             fiy2             = _mm_add_ps(fiy2,ty);
 395             fiz2             = _mm_add_ps(fiz2,tz);
 396
 397             fjptrA             = f+j_coord_offsetA;
 398             fjptrB             = f+j_coord_offsetB;
 399             fjptrC             = f+j_coord_offsetC;
 400             fjptrD             = f+j_coord_offsetD;
 401             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 402
 403             /**************************
 404              * CALCULATE INTERACTIONS *
 405              **************************/
 406
 407             r30              = _mm_mul_ps(rsq30,rinv30);
 408
 409             /* Compute parameters for interactions between i and j atoms */
 410             qq30             = _mm_mul_ps(iq3,jq0);
 411
 412             /* EWALD ELECTROSTATICS */
 413
 414             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 415             ewrt             = _mm_mul_ps(r30,ewtabscale);
 416             ewitab           = _mm_cvttps_epi32(ewrt);
 417             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 418             ewitab           = _mm_slli_epi32(ewitab,2);
 419             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 420             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 421             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 422             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 423             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 424             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 425             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 426             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 427             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 428
 429             /* Update potential sum for this i atom from the interaction with this j atom. */
 430             velecsum         = _mm_add_ps(velecsum,velec);
 431
 432             fscal            = felec;
 433
 434             /* Calculate temporary vectorial force */
 435             tx               = _mm_mul_ps(fscal,dx30);
 436             ty               = _mm_mul_ps(fscal,dy30);
 437             tz               = _mm_mul_ps(fscal,dz30);
 438
 439             /* Update vectorial force */
 440             fix3             = _mm_add_ps(fix3,tx);
 441             fiy3             = _mm_add_ps(fiy3,ty);
 442             fiz3             = _mm_add_ps(fiz3,tz);
 443
 444             fjptrA             = f+j_coord_offsetA;
 445             fjptrB             = f+j_coord_offsetB;
 446             fjptrC             = f+j_coord_offsetC;
 447             fjptrD             = f+j_coord_offsetD;
 448             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 449
 450             /* Inner loop uses 179 flops */
 451         }
 452
 453         if(jidx<j_index_end)
 454         {
 455
 456             /* Get j neighbor index, and coordinate index */
 457             jnrlistA         = jjnr[jidx];
 458             jnrlistB         = jjnr[jidx+1];
 459             jnrlistC         = jjnr[jidx+2];
 460             jnrlistD         = jjnr[jidx+3];
 461             /* Sign of each element will be negative for non-real atoms.
 462              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 463              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 464              */
 465             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 466             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 467             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 468             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 469             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 470             j_coord_offsetA  = DIM*jnrA;
 471             j_coord_offsetB  = DIM*jnrB;
 472             j_coord_offsetC  = DIM*jnrC;
 473             j_coord_offsetD  = DIM*jnrD;
 474
 475             /* load j atom coordinates */
 476             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 477                                               x+j_coord_offsetC,x+j_coord_offsetD,
 478                                               &jx0,&jy0,&jz0);
 479
 480             /* Calculate displacement vector */
 481             dx00             = _mm_sub_ps(ix0,jx0);
 482             dy00             = _mm_sub_ps(iy0,jy0);
 483             dz00             = _mm_sub_ps(iz0,jz0);
 484             dx10             = _mm_sub_ps(ix1,jx0);
 485             dy10             = _mm_sub_ps(iy1,jy0);
 486             dz10             = _mm_sub_ps(iz1,jz0);
 487             dx20             = _mm_sub_ps(ix2,jx0);
 488             dy20             = _mm_sub_ps(iy2,jy0);
 489             dz20             = _mm_sub_ps(iz2,jz0);
 490             dx30             = _mm_sub_ps(ix3,jx0);
 491             dy30             = _mm_sub_ps(iy3,jy0);
 492             dz30             = _mm_sub_ps(iz3,jz0);
 493
 494             /* Calculate squared distance and things based on it */
 495             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 496             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 497             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 498             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 499
 500             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 501             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 502             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 503             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 504
 505             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 506             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 507             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 508
 509             /* Load parameters for j particles */
 510             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 511                                                               charge+jnrC+0,charge+jnrD+0);
 512             vdwjidx0A        = 2*vdwtype[jnrA+0];
 513             vdwjidx0B        = 2*vdwtype[jnrB+0];
 514             vdwjidx0C        = 2*vdwtype[jnrC+0];
 515             vdwjidx0D        = 2*vdwtype[jnrD+0];
 516
 517             /**************************
 518              * CALCULATE INTERACTIONS *
 519              **************************/
 520
 521             r00              = _mm_mul_ps(rsq00,rinv00);
 522             r00              = _mm_andnot_ps(dummy_mask,r00);
 523
 524             /* Compute parameters for interactions between i and j atoms */
 525             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 526                                          vdwparam+vdwioffset0+vdwjidx0B,
 527                                          vdwparam+vdwioffset0+vdwjidx0C,
 528                                          vdwparam+vdwioffset0+vdwjidx0D,
 529                                          &c6_00,&c12_00);
 530
 531             /* Calculate table index by multiplying r with table scale and truncate to integer */
 532             rt               = _mm_mul_ps(r00,vftabscale);
 533             vfitab           = _mm_cvttps_epi32(rt);
 534             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 535             vfitab           = _mm_slli_epi32(vfitab,3);
 536
 537             /* CUBIC SPLINE TABLE DISPERSION */
 538             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 539             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 540             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 541             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 542             _MM_TRANSPOSE4_PS(Y,F,G,H);
 543             Heps             = _mm_mul_ps(vfeps,H);
 544             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 545             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 546             vvdw6            = _mm_mul_ps(c6_00,VV);
 547             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 548             fvdw6            = _mm_mul_ps(c6_00,FF);
 549
 550             /* CUBIC SPLINE TABLE REPULSION */
 551             vfitab           = _mm_add_epi32(vfitab,ifour);
 552             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 553             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 554             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 555             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 556             _MM_TRANSPOSE4_PS(Y,F,G,H);
 557             Heps             = _mm_mul_ps(vfeps,H);
 558             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 559             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 560             vvdw12           = _mm_mul_ps(c12_00,VV);
 561             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 562             fvdw12           = _mm_mul_ps(c12_00,FF);
 563             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 564             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 565
 566             /* Update potential sum for this i atom from the interaction with this j atom. */
 567             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 568             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 569
 570             fscal            = fvdw;
 571
 572             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 573
 574             /* Calculate temporary vectorial force */
 575             tx               = _mm_mul_ps(fscal,dx00);
 576             ty               = _mm_mul_ps(fscal,dy00);
 577             tz               = _mm_mul_ps(fscal,dz00);
 578
 579             /* Update vectorial force */
 580             fix0             = _mm_add_ps(fix0,tx);
 581             fiy0             = _mm_add_ps(fiy0,ty);
 582             fiz0             = _mm_add_ps(fiz0,tz);
 583
 584             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 585             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 586             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 587             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 588             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 589
 590             /**************************
 591              * CALCULATE INTERACTIONS *
 592              **************************/
 593
 594             r10              = _mm_mul_ps(rsq10,rinv10);
 595             r10              = _mm_andnot_ps(dummy_mask,r10);
 596
 597             /* Compute parameters for interactions between i and j atoms */
 598             qq10             = _mm_mul_ps(iq1,jq0);
 599
 600             /* EWALD ELECTROSTATICS */
 601
 602             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 603             ewrt             = _mm_mul_ps(r10,ewtabscale);
 604             ewitab           = _mm_cvttps_epi32(ewrt);
 605             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 606             ewitab           = _mm_slli_epi32(ewitab,2);
 607             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 608             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 609             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 610             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 611             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 612             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 613             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 614             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 615             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 616
 617             /* Update potential sum for this i atom from the interaction with this j atom. */
 618             velec            = _mm_andnot_ps(dummy_mask,velec);
 619             velecsum         = _mm_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 624
 625             /* Calculate temporary vectorial force */
 626             tx               = _mm_mul_ps(fscal,dx10);
 627             ty               = _mm_mul_ps(fscal,dy10);
 628             tz               = _mm_mul_ps(fscal,dz10);
 629
 630             /* Update vectorial force */
 631             fix1             = _mm_add_ps(fix1,tx);
 632             fiy1             = _mm_add_ps(fiy1,ty);
 633             fiz1             = _mm_add_ps(fiz1,tz);
 634
 635             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 636             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 637             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 638             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 639             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 640
 641             /**************************
 642              * CALCULATE INTERACTIONS *
 643              **************************/
 644
 645             r20              = _mm_mul_ps(rsq20,rinv20);
 646             r20              = _mm_andnot_ps(dummy_mask,r20);
 647
 648             /* Compute parameters for interactions between i and j atoms */
 649             qq20             = _mm_mul_ps(iq2,jq0);
 650
 651             /* EWALD ELECTROSTATICS */
 652
 653             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 654             ewrt             = _mm_mul_ps(r20,ewtabscale);
 655             ewitab           = _mm_cvttps_epi32(ewrt);
 656             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 657             ewitab           = _mm_slli_epi32(ewitab,2);
 658             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 659             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 660             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 661             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 662             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 663             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 664             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 665             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 666             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 667
 668             /* Update potential sum for this i atom from the interaction with this j atom. */
 669             velec            = _mm_andnot_ps(dummy_mask,velec);
 670             velecsum         = _mm_add_ps(velecsum,velec);
 671
 672             fscal            = felec;
 673
 674             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 675
 676             /* Calculate temporary vectorial force */
 677             tx               = _mm_mul_ps(fscal,dx20);
 678             ty               = _mm_mul_ps(fscal,dy20);
 679             tz               = _mm_mul_ps(fscal,dz20);
 680
 681             /* Update vectorial force */
 682             fix2             = _mm_add_ps(fix2,tx);
 683             fiy2             = _mm_add_ps(fiy2,ty);
 684             fiz2             = _mm_add_ps(fiz2,tz);
 685
 686             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 687             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 688             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 689             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 690             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 691
 692             /**************************
 693              * CALCULATE INTERACTIONS *
 694              **************************/
 695
 696             r30              = _mm_mul_ps(rsq30,rinv30);
 697             r30              = _mm_andnot_ps(dummy_mask,r30);
 698
 699             /* Compute parameters for interactions between i and j atoms */
 700             qq30             = _mm_mul_ps(iq3,jq0);
 701
 702             /* EWALD ELECTROSTATICS */
 703
 704             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 705             ewrt             = _mm_mul_ps(r30,ewtabscale);
 706             ewitab           = _mm_cvttps_epi32(ewrt);
 707             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 708             ewitab           = _mm_slli_epi32(ewitab,2);
 709             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 710             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 711             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 712             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 713             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 714             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 715             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 716             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 717             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 718
 719             /* Update potential sum for this i atom from the interaction with this j atom. */
 720             velec            = _mm_andnot_ps(dummy_mask,velec);
 721             velecsum         = _mm_add_ps(velecsum,velec);
 722
 723             fscal            = felec;
 724
 725             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 726
 727             /* Calculate temporary vectorial force */
 728             tx               = _mm_mul_ps(fscal,dx30);
 729             ty               = _mm_mul_ps(fscal,dy30);
 730             tz               = _mm_mul_ps(fscal,dz30);
 731
 732             /* Update vectorial force */
 733             fix3             = _mm_add_ps(fix3,tx);
 734             fiy3             = _mm_add_ps(fiy3,ty);
 735             fiz3             = _mm_add_ps(fiz3,tz);
 736
 737             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 738             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 739             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 740             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 741             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 742
 743             /* Inner loop uses 183 flops */
 744         }
 745
 746         /* End of innermost loop */
 747
 748         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 749                                               f+i_coord_offset,fshift+i_shift_offset);
 750
 751         ggid                        = gid[iidx];
 752         /* Update potential energies */
 753         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 754         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 755
 756         /* Increment number of inner iterations */
 757         inneriter                  += j_index_end - j_index_start;
 758
 759         /* Outer loop uses 26 flops */
 760     }
 761
 762     /* Increment number of outer iterations */
 763     outeriter        += nri;
 764
 765     /* Update outer/inner flops */
 766
 767     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*183);
 768 }
 769 /*
 770  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse2_single
 771  * Electrostatics interaction: Ewald
 772  * VdW interaction:            CubicSplineTable
 773  * Geometry:                   Water4-Particle
 774  * Calculate force/pot:        Force
 775  */
 776 void
 777 nb_kernel_ElecEw_VdwCSTab_GeomW4P1_F_sse2_single
 778                     (t_nblist * gmx_restrict                nlist,
 779                      rvec * gmx_restrict                    xx,
 780                      rvec * gmx_restrict                    ff,
 781                      t_forcerec * gmx_restrict              fr,
 782                      t_mdatoms * gmx_restrict               mdatoms,
 783                      nb_kernel_data_t * gmx_restrict        kernel_data,
 784                      t_nrnb * gmx_restrict                  nrnb)
 785 {
 786     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 787      * just 0 for non-waters.
 788      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 789      * jnr indices corresponding to data put in the four positions in the SIMD register.
 790      */
 791     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 792     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 793     int              jnrA,jnrB,jnrC,jnrD;
 794     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 795     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 796     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 797     real             rcutoff_scalar;
 798     real             *shiftvec,*fshift,*x,*f;
 799     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 800     real             scratch[4*DIM];
 801     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 802     int              vdwioffset0;
 803     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 804     int              vdwioffset1;
 805     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 806     int              vdwioffset2;
 807     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 808     int              vdwioffset3;
 809     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 810     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 811     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 812     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 813     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 814     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 815     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 816     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 817     real             *charge;
 818     int              nvdwtype;
 819     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 820     int              *vdwtype;
 821     real             *vdwparam;
 822     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 823     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 824     __m128i          vfitab;
 825     __m128i          ifour       = _mm_set1_epi32(4);
 826     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 827     real             *vftab;
 828     __m128i          ewitab;
 829     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 830     real             *ewtab;
 831     __m128           dummy_mask,cutoff_mask;
 832     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 833     __m128           one     = _mm_set1_ps(1.0);
 834     __m128           two     = _mm_set1_ps(2.0);
 835     x                = xx[0];
 836     f                = ff[0];
 837
 838     nri              = nlist->nri;
 839     iinr             = nlist->iinr;
 840     jindex           = nlist->jindex;
 841     jjnr             = nlist->jjnr;
 842     shiftidx         = nlist->shift;
 843     gid              = nlist->gid;
 844     shiftvec         = fr->shift_vec[0];
 845     fshift           = fr->fshift[0];
 846     facel            = _mm_set1_ps(fr->epsfac);
 847     charge           = mdatoms->chargeA;
 848     nvdwtype         = fr->ntype;
 849     vdwparam         = fr->nbfp;
 850     vdwtype          = mdatoms->typeA;
 851
 852     vftab            = kernel_data->table_vdw->data;
 853     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 854
 855     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 856     ewtab            = fr->ic->tabq_coul_F;
 857     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 858     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 859
 860     /* Setup water-specific parameters */
 861     inr              = nlist->iinr[0];
 862     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 863     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 864     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 865     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 866
 867     /* Avoid stupid compiler warnings */
 868     jnrA = jnrB = jnrC = jnrD = 0;
 869     j_coord_offsetA = 0;
 870     j_coord_offsetB = 0;
 871     j_coord_offsetC = 0;
 872     j_coord_offsetD = 0;
 873
 874     outeriter        = 0;
 875     inneriter        = 0;
 876
 877     for(iidx=0;iidx<4*DIM;iidx++)
 878     {
 879         scratch[iidx] = 0.0;
 880     }
 881
 882     /* Start outer loop over neighborlists */
 883     for(iidx=0; iidx<nri; iidx++)
 884     {
 885         /* Load shift vector for this list */
 886         i_shift_offset   = DIM*shiftidx[iidx];
 887
 888         /* Load limits for loop over neighbors */
 889         j_index_start    = jindex[iidx];
 890         j_index_end      = jindex[iidx+1];
 891
 892         /* Get outer coordinate index */
 893         inr              = iinr[iidx];
 894         i_coord_offset   = DIM*inr;
 895
 896         /* Load i particle coords and add shift vector */
 897         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 898                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 899
 900         fix0             = _mm_setzero_ps();
 901         fiy0             = _mm_setzero_ps();
 902         fiz0             = _mm_setzero_ps();
 903         fix1             = _mm_setzero_ps();
 904         fiy1             = _mm_setzero_ps();
 905         fiz1             = _mm_setzero_ps();
 906         fix2             = _mm_setzero_ps();
 907         fiy2             = _mm_setzero_ps();
 908         fiz2             = _mm_setzero_ps();
 909         fix3             = _mm_setzero_ps();
 910         fiy3             = _mm_setzero_ps();
 911         fiz3             = _mm_setzero_ps();
 912
 913         /* Start inner kernel loop */
 914         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 915         {
 916
 917             /* Get j neighbor index, and coordinate index */
 918             jnrA             = jjnr[jidx];
 919             jnrB             = jjnr[jidx+1];
 920             jnrC             = jjnr[jidx+2];
 921             jnrD             = jjnr[jidx+3];
 922             j_coord_offsetA  = DIM*jnrA;
 923             j_coord_offsetB  = DIM*jnrB;
 924             j_coord_offsetC  = DIM*jnrC;
 925             j_coord_offsetD  = DIM*jnrD;
 926
 927             /* load j atom coordinates */
 928             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 929                                               x+j_coord_offsetC,x+j_coord_offsetD,
 930                                               &jx0,&jy0,&jz0);
 931
 932             /* Calculate displacement vector */
 933             dx00             = _mm_sub_ps(ix0,jx0);
 934             dy00             = _mm_sub_ps(iy0,jy0);
 935             dz00             = _mm_sub_ps(iz0,jz0);
 936             dx10             = _mm_sub_ps(ix1,jx0);
 937             dy10             = _mm_sub_ps(iy1,jy0);
 938             dz10             = _mm_sub_ps(iz1,jz0);
 939             dx20             = _mm_sub_ps(ix2,jx0);
 940             dy20             = _mm_sub_ps(iy2,jy0);
 941             dz20             = _mm_sub_ps(iz2,jz0);
 942             dx30             = _mm_sub_ps(ix3,jx0);
 943             dy30             = _mm_sub_ps(iy3,jy0);
 944             dz30             = _mm_sub_ps(iz3,jz0);
 945
 946             /* Calculate squared distance and things based on it */
 947             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 948             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 949             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 950             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 951
 952             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 953             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 954             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 955             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 956
 957             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 958             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 959             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 960
 961             /* Load parameters for j particles */
 962             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 963                                                               charge+jnrC+0,charge+jnrD+0);
 964             vdwjidx0A        = 2*vdwtype[jnrA+0];
 965             vdwjidx0B        = 2*vdwtype[jnrB+0];
 966             vdwjidx0C        = 2*vdwtype[jnrC+0];
 967             vdwjidx0D        = 2*vdwtype[jnrD+0];
 968
 969             /**************************
 970              * CALCULATE INTERACTIONS *
 971              **************************/
 972
 973             r00              = _mm_mul_ps(rsq00,rinv00);
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 977                                          vdwparam+vdwioffset0+vdwjidx0B,
 978                                          vdwparam+vdwioffset0+vdwjidx0C,
 979                                          vdwparam+vdwioffset0+vdwjidx0D,
 980                                          &c6_00,&c12_00);
 981
 982             /* Calculate table index by multiplying r with table scale and truncate to integer */
 983             rt               = _mm_mul_ps(r00,vftabscale);
 984             vfitab           = _mm_cvttps_epi32(rt);
 985             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 986             vfitab           = _mm_slli_epi32(vfitab,3);
 987
 988             /* CUBIC SPLINE TABLE DISPERSION */
 989             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 990             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 991             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 992             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 993             _MM_TRANSPOSE4_PS(Y,F,G,H);
 994             Heps             = _mm_mul_ps(vfeps,H);
 995             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 996             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 997             fvdw6            = _mm_mul_ps(c6_00,FF);
 998
 999             /* CUBIC SPLINE TABLE REPULSION */
1000             vfitab           = _mm_add_epi32(vfitab,ifour);
1001             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1002             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1003             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1004             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1005             _MM_TRANSPOSE4_PS(Y,F,G,H);
1006             Heps             = _mm_mul_ps(vfeps,H);
1007             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1008             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1009             fvdw12           = _mm_mul_ps(c12_00,FF);
1010             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1011
1012             fscal            = fvdw;
1013
1014             /* Calculate temporary vectorial force */
1015             tx               = _mm_mul_ps(fscal,dx00);
1016             ty               = _mm_mul_ps(fscal,dy00);
1017             tz               = _mm_mul_ps(fscal,dz00);
1018
1019             /* Update vectorial force */
1020             fix0             = _mm_add_ps(fix0,tx);
1021             fiy0             = _mm_add_ps(fiy0,ty);
1022             fiz0             = _mm_add_ps(fiz0,tz);
1023
1024             fjptrA             = f+j_coord_offsetA;
1025             fjptrB             = f+j_coord_offsetB;
1026             fjptrC             = f+j_coord_offsetC;
1027             fjptrD             = f+j_coord_offsetD;
1028             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1029
1030             /**************************
1031              * CALCULATE INTERACTIONS *
1032              **************************/
1033
1034             r10              = _mm_mul_ps(rsq10,rinv10);
1035
1036             /* Compute parameters for interactions between i and j atoms */
1037             qq10             = _mm_mul_ps(iq1,jq0);
1038
1039             /* EWALD ELECTROSTATICS */
1040
1041             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1042             ewrt             = _mm_mul_ps(r10,ewtabscale);
1043             ewitab           = _mm_cvttps_epi32(ewrt);
1044             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1045             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1046                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1047                                          &ewtabF,&ewtabFn);
1048             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1049             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1050
1051             fscal            = felec;
1052
1053             /* Calculate temporary vectorial force */
1054             tx               = _mm_mul_ps(fscal,dx10);
1055             ty               = _mm_mul_ps(fscal,dy10);
1056             tz               = _mm_mul_ps(fscal,dz10);
1057
1058             /* Update vectorial force */
1059             fix1             = _mm_add_ps(fix1,tx);
1060             fiy1             = _mm_add_ps(fiy1,ty);
1061             fiz1             = _mm_add_ps(fiz1,tz);
1062
1063             fjptrA             = f+j_coord_offsetA;
1064             fjptrB             = f+j_coord_offsetB;
1065             fjptrC             = f+j_coord_offsetC;
1066             fjptrD             = f+j_coord_offsetD;
1067             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1068
1069             /**************************
1070              * CALCULATE INTERACTIONS *
1071              **************************/
1072
1073             r20              = _mm_mul_ps(rsq20,rinv20);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq20             = _mm_mul_ps(iq2,jq0);
1077
1078             /* EWALD ELECTROSTATICS */
1079
1080             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1081             ewrt             = _mm_mul_ps(r20,ewtabscale);
1082             ewitab           = _mm_cvttps_epi32(ewrt);
1083             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1084             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1085                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1086                                          &ewtabF,&ewtabFn);
1087             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1088             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1089
1090             fscal            = felec;
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm_mul_ps(fscal,dx20);
1094             ty               = _mm_mul_ps(fscal,dy20);
1095             tz               = _mm_mul_ps(fscal,dz20);
1096
1097             /* Update vectorial force */
1098             fix2             = _mm_add_ps(fix2,tx);
1099             fiy2             = _mm_add_ps(fiy2,ty);
1100             fiz2             = _mm_add_ps(fiz2,tz);
1101
1102             fjptrA             = f+j_coord_offsetA;
1103             fjptrB             = f+j_coord_offsetB;
1104             fjptrC             = f+j_coord_offsetC;
1105             fjptrD             = f+j_coord_offsetD;
1106             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             r30              = _mm_mul_ps(rsq30,rinv30);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq30             = _mm_mul_ps(iq3,jq0);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm_mul_ps(r30,ewtabscale);
1121             ewitab           = _mm_cvttps_epi32(ewrt);
1122             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1123             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1124                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1125                                          &ewtabF,&ewtabFn);
1126             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1127             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1128
1129             fscal            = felec;
1130
1131             /* Calculate temporary vectorial force */
1132             tx               = _mm_mul_ps(fscal,dx30);
1133             ty               = _mm_mul_ps(fscal,dy30);
1134             tz               = _mm_mul_ps(fscal,dz30);
1135
1136             /* Update vectorial force */
1137             fix3             = _mm_add_ps(fix3,tx);
1138             fiy3             = _mm_add_ps(fiy3,ty);
1139             fiz3             = _mm_add_ps(fiz3,tz);
1140
1141             fjptrA             = f+j_coord_offsetA;
1142             fjptrB             = f+j_coord_offsetB;
1143             fjptrC             = f+j_coord_offsetC;
1144             fjptrD             = f+j_coord_offsetD;
1145             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1146
1147             /* Inner loop uses 156 flops */
1148         }
1149
1150         if(jidx<j_index_end)
1151         {
1152
1153             /* Get j neighbor index, and coordinate index */
1154             jnrlistA         = jjnr[jidx];
1155             jnrlistB         = jjnr[jidx+1];
1156             jnrlistC         = jjnr[jidx+2];
1157             jnrlistD         = jjnr[jidx+3];
1158             /* Sign of each element will be negative for non-real atoms.
1159              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1160              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1161              */
1162             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1163             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1164             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1165             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1166             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1167             j_coord_offsetA  = DIM*jnrA;
1168             j_coord_offsetB  = DIM*jnrB;
1169             j_coord_offsetC  = DIM*jnrC;
1170             j_coord_offsetD  = DIM*jnrD;
1171
1172             /* load j atom coordinates */
1173             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1174                                               x+j_coord_offsetC,x+j_coord_offsetD,
1175                                               &jx0,&jy0,&jz0);
1176
1177             /* Calculate displacement vector */
1178             dx00             = _mm_sub_ps(ix0,jx0);
1179             dy00             = _mm_sub_ps(iy0,jy0);
1180             dz00             = _mm_sub_ps(iz0,jz0);
1181             dx10             = _mm_sub_ps(ix1,jx0);
1182             dy10             = _mm_sub_ps(iy1,jy0);
1183             dz10             = _mm_sub_ps(iz1,jz0);
1184             dx20             = _mm_sub_ps(ix2,jx0);
1185             dy20             = _mm_sub_ps(iy2,jy0);
1186             dz20             = _mm_sub_ps(iz2,jz0);
1187             dx30             = _mm_sub_ps(ix3,jx0);
1188             dy30             = _mm_sub_ps(iy3,jy0);
1189             dz30             = _mm_sub_ps(iz3,jz0);
1190
1191             /* Calculate squared distance and things based on it */
1192             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1193             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1194             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1195             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1196
1197             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1198             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1199             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1200             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1201
1202             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1203             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1204             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1205
1206             /* Load parameters for j particles */
1207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1208                                                               charge+jnrC+0,charge+jnrD+0);
1209             vdwjidx0A        = 2*vdwtype[jnrA+0];
1210             vdwjidx0B        = 2*vdwtype[jnrB+0];
1211             vdwjidx0C        = 2*vdwtype[jnrC+0];
1212             vdwjidx0D        = 2*vdwtype[jnrD+0];
1213
1214             /**************************
1215              * CALCULATE INTERACTIONS *
1216              **************************/
1217
1218             r00              = _mm_mul_ps(rsq00,rinv00);
1219             r00              = _mm_andnot_ps(dummy_mask,r00);
1220
1221             /* Compute parameters for interactions between i and j atoms */
1222             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1223                                          vdwparam+vdwioffset0+vdwjidx0B,
1224                                          vdwparam+vdwioffset0+vdwjidx0C,
1225                                          vdwparam+vdwioffset0+vdwjidx0D,
1226                                          &c6_00,&c12_00);
1227
1228             /* Calculate table index by multiplying r with table scale and truncate to integer */
1229             rt               = _mm_mul_ps(r00,vftabscale);
1230             vfitab           = _mm_cvttps_epi32(rt);
1231             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1232             vfitab           = _mm_slli_epi32(vfitab,3);
1233
1234             /* CUBIC SPLINE TABLE DISPERSION */
1235             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1236             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1237             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1238             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1239             _MM_TRANSPOSE4_PS(Y,F,G,H);
1240             Heps             = _mm_mul_ps(vfeps,H);
1241             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1242             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1243             fvdw6            = _mm_mul_ps(c6_00,FF);
1244
1245             /* CUBIC SPLINE TABLE REPULSION */
1246             vfitab           = _mm_add_epi32(vfitab,ifour);
1247             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1248             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1249             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1250             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1251             _MM_TRANSPOSE4_PS(Y,F,G,H);
1252             Heps             = _mm_mul_ps(vfeps,H);
1253             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1254             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1255             fvdw12           = _mm_mul_ps(c12_00,FF);
1256             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1257
1258             fscal            = fvdw;
1259
1260             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1261
1262             /* Calculate temporary vectorial force */
1263             tx               = _mm_mul_ps(fscal,dx00);
1264             ty               = _mm_mul_ps(fscal,dy00);
1265             tz               = _mm_mul_ps(fscal,dz00);
1266
1267             /* Update vectorial force */
1268             fix0             = _mm_add_ps(fix0,tx);
1269             fiy0             = _mm_add_ps(fiy0,ty);
1270             fiz0             = _mm_add_ps(fiz0,tz);
1271
1272             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1273             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1274             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1275             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1276             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1277
1278             /**************************
1279              * CALCULATE INTERACTIONS *
1280              **************************/
1281
1282             r10              = _mm_mul_ps(rsq10,rinv10);
1283             r10              = _mm_andnot_ps(dummy_mask,r10);
1284
1285             /* Compute parameters for interactions between i and j atoms */
1286             qq10             = _mm_mul_ps(iq1,jq0);
1287
1288             /* EWALD ELECTROSTATICS */
1289
1290             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1291             ewrt             = _mm_mul_ps(r10,ewtabscale);
1292             ewitab           = _mm_cvttps_epi32(ewrt);
1293             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1294             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1295                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1296                                          &ewtabF,&ewtabFn);
1297             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1298             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1299
1300             fscal            = felec;
1301
1302             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1303
1304             /* Calculate temporary vectorial force */
1305             tx               = _mm_mul_ps(fscal,dx10);
1306             ty               = _mm_mul_ps(fscal,dy10);
1307             tz               = _mm_mul_ps(fscal,dz10);
1308
1309             /* Update vectorial force */
1310             fix1             = _mm_add_ps(fix1,tx);
1311             fiy1             = _mm_add_ps(fiy1,ty);
1312             fiz1             = _mm_add_ps(fiz1,tz);
1313
1314             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1315             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1316             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1317             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1318             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1319
1320             /**************************
1321              * CALCULATE INTERACTIONS *
1322              **************************/
1323
1324             r20              = _mm_mul_ps(rsq20,rinv20);
1325             r20              = _mm_andnot_ps(dummy_mask,r20);
1326
1327             /* Compute parameters for interactions between i and j atoms */
1328             qq20             = _mm_mul_ps(iq2,jq0);
1329
1330             /* EWALD ELECTROSTATICS */
1331
1332             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1333             ewrt             = _mm_mul_ps(r20,ewtabscale);
1334             ewitab           = _mm_cvttps_epi32(ewrt);
1335             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1336             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1337                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1338                                          &ewtabF,&ewtabFn);
1339             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1340             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1341
1342             fscal            = felec;
1343
1344             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1345
1346             /* Calculate temporary vectorial force */
1347             tx               = _mm_mul_ps(fscal,dx20);
1348             ty               = _mm_mul_ps(fscal,dy20);
1349             tz               = _mm_mul_ps(fscal,dz20);
1350
1351             /* Update vectorial force */
1352             fix2             = _mm_add_ps(fix2,tx);
1353             fiy2             = _mm_add_ps(fiy2,ty);
1354             fiz2             = _mm_add_ps(fiz2,tz);
1355
1356             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1357             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1358             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1359             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1360             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1361
1362             /**************************
1363              * CALCULATE INTERACTIONS *
1364              **************************/
1365
1366             r30              = _mm_mul_ps(rsq30,rinv30);
1367             r30              = _mm_andnot_ps(dummy_mask,r30);
1368
1369             /* Compute parameters for interactions between i and j atoms */
1370             qq30             = _mm_mul_ps(iq3,jq0);
1371
1372             /* EWALD ELECTROSTATICS */
1373
1374             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1375             ewrt             = _mm_mul_ps(r30,ewtabscale);
1376             ewitab           = _mm_cvttps_epi32(ewrt);
1377             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1378             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1379                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1380                                          &ewtabF,&ewtabFn);
1381             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1382             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1383
1384             fscal            = felec;
1385
1386             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1387
1388             /* Calculate temporary vectorial force */
1389             tx               = _mm_mul_ps(fscal,dx30);
1390             ty               = _mm_mul_ps(fscal,dy30);
1391             tz               = _mm_mul_ps(fscal,dz30);
1392
1393             /* Update vectorial force */
1394             fix3             = _mm_add_ps(fix3,tx);
1395             fiy3             = _mm_add_ps(fiy3,ty);
1396             fiz3             = _mm_add_ps(fiz3,tz);
1397
1398             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1399             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1400             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1401             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1402             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1403
1404             /* Inner loop uses 160 flops */
1405         }
1406
1407         /* End of innermost loop */
1408
1409         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1410                                               f+i_coord_offset,fshift+i_shift_offset);
1411
1412         /* Increment number of inner iterations */
1413         inneriter                  += j_index_end - j_index_start;
1414
1415         /* Outer loop uses 24 flops */
1416     }
1417
1418     /* Increment number of outer iterations */
1419     outeriter        += nri;
1420
1421     /* Update outer/inner flops */
1422
1423     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*160);
1424 }