src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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              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              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_elec_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 119     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 120     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138         shX              = shiftvec[i_shift_offset+XX];
 139         shY              = shiftvec[i_shift_offset+YY];
 140         shZ              = shiftvec[i_shift_offset+ZZ];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 152         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 153         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 154         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 155         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 156         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 157         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 158         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 159         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 160         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 161         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 162         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167         fix1             = _mm_setzero_ps();
 168         fiy1             = _mm_setzero_ps();
 169         fiz1             = _mm_setzero_ps();
 170         fix2             = _mm_setzero_ps();
 171         fiy2             = _mm_setzero_ps();
 172         fiz2             = _mm_setzero_ps();
 173         fix3             = _mm_setzero_ps();
 174         fiy3             = _mm_setzero_ps();
 175         fiz3             = _mm_setzero_ps();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_ps();
 179         vvdwsum          = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_ps(ix0,jx0);
 203             dy00             = _mm_sub_ps(iy0,jy0);
 204             dz00             = _mm_sub_ps(iz0,jz0);
 205             dx10             = _mm_sub_ps(ix1,jx0);
 206             dy10             = _mm_sub_ps(iy1,jy0);
 207             dz10             = _mm_sub_ps(iz1,jz0);
 208             dx20             = _mm_sub_ps(ix2,jx0);
 209             dy20             = _mm_sub_ps(iy2,jy0);
 210             dz20             = _mm_sub_ps(iz2,jz0);
 211             dx30             = _mm_sub_ps(ix3,jx0);
 212             dy30             = _mm_sub_ps(iy3,jy0);
 213             dz30             = _mm_sub_ps(iz3,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 225
 226             /* Load parameters for j particles */
 227             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 228                                                               charge+jnrC+0,charge+jnrD+0);
 229             vdwjidx0A        = 2*vdwtype[jnrA+0];
 230             vdwjidx0B        = 2*vdwtype[jnrB+0];
 231             vdwjidx0C        = 2*vdwtype[jnrC+0];
 232             vdwjidx0D        = 2*vdwtype[jnrD+0];
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             r00              = _mm_mul_ps(rsq00,rinv00);
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,
 243                                          vdwparam+vdwioffset0+vdwjidx0C,
 244                                          vdwparam+vdwioffset0+vdwjidx0D,
 245                                          &c6_00,&c12_00);
 246
 247             /* Calculate table index by multiplying r with table scale and truncate to integer */
 248             rt               = _mm_mul_ps(r00,vftabscale);
 249             vfitab           = _mm_cvttps_epi32(rt);
 250             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 251             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 252
 253             /* CUBIC SPLINE TABLE DISPERSION */
 254             vfitab           = _mm_add_epi32(vfitab,ifour);
 255             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 256             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 257             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 258             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 259             _MM_TRANSPOSE4_PS(Y,F,G,H);
 260             Heps             = _mm_mul_ps(vfeps,H);
 261             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 262             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 263             vvdw6            = _mm_mul_ps(c6_00,VV);
 264             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 265             fvdw6            = _mm_mul_ps(c6_00,FF);
 266
 267             /* CUBIC SPLINE TABLE REPULSION */
 268             vfitab           = _mm_add_epi32(vfitab,ifour);
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             vvdw12           = _mm_mul_ps(c12_00,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             fvdw12           = _mm_mul_ps(c12_00,FF);
 280             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 281             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 285
 286             fscal            = fvdw;
 287
 288             /* Calculate temporary vectorial force */
 289             tx               = _mm_mul_ps(fscal,dx00);
 290             ty               = _mm_mul_ps(fscal,dy00);
 291             tz               = _mm_mul_ps(fscal,dz00);
 292
 293             /* Update vectorial force */
 294             fix0             = _mm_add_ps(fix0,tx);
 295             fiy0             = _mm_add_ps(fiy0,ty);
 296             fiz0             = _mm_add_ps(fiz0,tz);
 297
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 299                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 300                                                    tx,ty,tz);
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r10              = _mm_mul_ps(rsq10,rinv10);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq10             = _mm_mul_ps(iq1,jq0);
 310
 311             /* Calculate table index by multiplying r with table scale and truncate to integer */
 312             rt               = _mm_mul_ps(r10,vftabscale);
 313             vfitab           = _mm_cvttps_epi32(rt);
 314             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 315             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 316
 317             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 318             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 319             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 320             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 321             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 322             _MM_TRANSPOSE4_PS(Y,F,G,H);
 323             Heps             = _mm_mul_ps(vfeps,H);
 324             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 325             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 326             velec            = _mm_mul_ps(qq10,VV);
 327             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 328             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Calculate temporary vectorial force */
 336             tx               = _mm_mul_ps(fscal,dx10);
 337             ty               = _mm_mul_ps(fscal,dy10);
 338             tz               = _mm_mul_ps(fscal,dz10);
 339
 340             /* Update vectorial force */
 341             fix1             = _mm_add_ps(fix1,tx);
 342             fiy1             = _mm_add_ps(fiy1,ty);
 343             fiz1             = _mm_add_ps(fiz1,tz);
 344
 345             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 346                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 347                                                    tx,ty,tz);
 348
 349             /**************************
 350              * CALCULATE INTERACTIONS *
 351              **************************/
 352
 353             r20              = _mm_mul_ps(rsq20,rinv20);
 354
 355             /* Compute parameters for interactions between i and j atoms */
 356             qq20             = _mm_mul_ps(iq2,jq0);
 357
 358             /* Calculate table index by multiplying r with table scale and truncate to integer */
 359             rt               = _mm_mul_ps(r20,vftabscale);
 360             vfitab           = _mm_cvttps_epi32(rt);
 361             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 362             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 363
 364             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 365             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 366             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 367             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 368             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 369             _MM_TRANSPOSE4_PS(Y,F,G,H);
 370             Heps             = _mm_mul_ps(vfeps,H);
 371             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 372             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 373             velec            = _mm_mul_ps(qq20,VV);
 374             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 375             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 376
 377             /* Update potential sum for this i atom from the interaction with this j atom. */
 378             velecsum         = _mm_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             /* Calculate temporary vectorial force */
 383             tx               = _mm_mul_ps(fscal,dx20);
 384             ty               = _mm_mul_ps(fscal,dy20);
 385             tz               = _mm_mul_ps(fscal,dz20);
 386
 387             /* Update vectorial force */
 388             fix2             = _mm_add_ps(fix2,tx);
 389             fiy2             = _mm_add_ps(fiy2,ty);
 390             fiz2             = _mm_add_ps(fiz2,tz);
 391
 392             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 393                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 394                                                    tx,ty,tz);
 395
 396             /**************************
 397              * CALCULATE INTERACTIONS *
 398              **************************/
 399
 400             r30              = _mm_mul_ps(rsq30,rinv30);
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq30             = _mm_mul_ps(iq3,jq0);
 404
 405             /* Calculate table index by multiplying r with table scale and truncate to integer */
 406             rt               = _mm_mul_ps(r30,vftabscale);
 407             vfitab           = _mm_cvttps_epi32(rt);
 408             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 409             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 410
 411             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 412             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 413             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 414             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 415             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 416             _MM_TRANSPOSE4_PS(Y,F,G,H);
 417             Heps             = _mm_mul_ps(vfeps,H);
 418             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 419             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 420             velec            = _mm_mul_ps(qq30,VV);
 421             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 422             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 423
 424             /* Update potential sum for this i atom from the interaction with this j atom. */
 425             velecsum         = _mm_add_ps(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             /* Calculate temporary vectorial force */
 430             tx               = _mm_mul_ps(fscal,dx30);
 431             ty               = _mm_mul_ps(fscal,dy30);
 432             tz               = _mm_mul_ps(fscal,dz30);
 433
 434             /* Update vectorial force */
 435             fix3             = _mm_add_ps(fix3,tx);
 436             fiy3             = _mm_add_ps(fiy3,ty);
 437             fiz3             = _mm_add_ps(fiz3,tz);
 438
 439             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 440                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 441                                                    tx,ty,tz);
 442
 443             /* Inner loop uses 185 flops */
 444         }
 445
 446         if(jidx<j_index_end)
 447         {
 448
 449             /* Get j neighbor index, and coordinate index */
 450             jnrA             = jjnr[jidx];
 451             jnrB             = jjnr[jidx+1];
 452             jnrC             = jjnr[jidx+2];
 453             jnrD             = jjnr[jidx+3];
 454
 455             /* Sign of each element will be negative for non-real atoms.
 456              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 457              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 458              */
 459             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 460             jnrA       = (jnrA>=0) ? jnrA : 0;
 461             jnrB       = (jnrB>=0) ? jnrB : 0;
 462             jnrC       = (jnrC>=0) ? jnrC : 0;
 463             jnrD       = (jnrD>=0) ? jnrD : 0;
 464
 465             j_coord_offsetA  = DIM*jnrA;
 466             j_coord_offsetB  = DIM*jnrB;
 467             j_coord_offsetC  = DIM*jnrC;
 468             j_coord_offsetD  = DIM*jnrD;
 469
 470             /* load j atom coordinates */
 471             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 472                                               x+j_coord_offsetC,x+j_coord_offsetD,
 473                                               &jx0,&jy0,&jz0);
 474
 475             /* Calculate displacement vector */
 476             dx00             = _mm_sub_ps(ix0,jx0);
 477             dy00             = _mm_sub_ps(iy0,jy0);
 478             dz00             = _mm_sub_ps(iz0,jz0);
 479             dx10             = _mm_sub_ps(ix1,jx0);
 480             dy10             = _mm_sub_ps(iy1,jy0);
 481             dz10             = _mm_sub_ps(iz1,jz0);
 482             dx20             = _mm_sub_ps(ix2,jx0);
 483             dy20             = _mm_sub_ps(iy2,jy0);
 484             dz20             = _mm_sub_ps(iz2,jz0);
 485             dx30             = _mm_sub_ps(ix3,jx0);
 486             dy30             = _mm_sub_ps(iy3,jy0);
 487             dz30             = _mm_sub_ps(iz3,jz0);
 488
 489             /* Calculate squared distance and things based on it */
 490             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 491             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 492             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 493             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 494
 495             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 496             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 497             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 498             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 499
 500             /* Load parameters for j particles */
 501             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 502                                                               charge+jnrC+0,charge+jnrD+0);
 503             vdwjidx0A        = 2*vdwtype[jnrA+0];
 504             vdwjidx0B        = 2*vdwtype[jnrB+0];
 505             vdwjidx0C        = 2*vdwtype[jnrC+0];
 506             vdwjidx0D        = 2*vdwtype[jnrD+0];
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r00              = _mm_mul_ps(rsq00,rinv00);
 513             r00              = _mm_andnot_ps(dummy_mask,r00);
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 517                                          vdwparam+vdwioffset0+vdwjidx0B,
 518                                          vdwparam+vdwioffset0+vdwjidx0C,
 519                                          vdwparam+vdwioffset0+vdwjidx0D,
 520                                          &c6_00,&c12_00);
 521
 522             /* Calculate table index by multiplying r with table scale and truncate to integer */
 523             rt               = _mm_mul_ps(r00,vftabscale);
 524             vfitab           = _mm_cvttps_epi32(rt);
 525             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 526             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 527
 528             /* CUBIC SPLINE TABLE DISPERSION */
 529             vfitab           = _mm_add_epi32(vfitab,ifour);
 530             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 531             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 532             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 533             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 534             _MM_TRANSPOSE4_PS(Y,F,G,H);
 535             Heps             = _mm_mul_ps(vfeps,H);
 536             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 537             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 538             vvdw6            = _mm_mul_ps(c6_00,VV);
 539             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 540             fvdw6            = _mm_mul_ps(c6_00,FF);
 541
 542             /* CUBIC SPLINE TABLE REPULSION */
 543             vfitab           = _mm_add_epi32(vfitab,ifour);
 544             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 545             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 546             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 547             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 548             _MM_TRANSPOSE4_PS(Y,F,G,H);
 549             Heps             = _mm_mul_ps(vfeps,H);
 550             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 551             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 552             vvdw12           = _mm_mul_ps(c12_00,VV);
 553             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 554             fvdw12           = _mm_mul_ps(c12_00,FF);
 555             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 556             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 557
 558             /* Update potential sum for this i atom from the interaction with this j atom. */
 559             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 560             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 561
 562             fscal            = fvdw;
 563
 564             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 565
 566             /* Calculate temporary vectorial force */
 567             tx               = _mm_mul_ps(fscal,dx00);
 568             ty               = _mm_mul_ps(fscal,dy00);
 569             tz               = _mm_mul_ps(fscal,dz00);
 570
 571             /* Update vectorial force */
 572             fix0             = _mm_add_ps(fix0,tx);
 573             fiy0             = _mm_add_ps(fiy0,ty);
 574             fiz0             = _mm_add_ps(fiz0,tz);
 575
 576             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 577                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 578                                                    tx,ty,tz);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             r10              = _mm_mul_ps(rsq10,rinv10);
 585             r10              = _mm_andnot_ps(dummy_mask,r10);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq10             = _mm_mul_ps(iq1,jq0);
 589
 590             /* Calculate table index by multiplying r with table scale and truncate to integer */
 591             rt               = _mm_mul_ps(r10,vftabscale);
 592             vfitab           = _mm_cvttps_epi32(rt);
 593             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 594             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 595
 596             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 597             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 598             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 599             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 600             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 601             _MM_TRANSPOSE4_PS(Y,F,G,H);
 602             Heps             = _mm_mul_ps(vfeps,H);
 603             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 604             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 605             velec            = _mm_mul_ps(qq10,VV);
 606             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 607             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 608
 609             /* Update potential sum for this i atom from the interaction with this j atom. */
 610             velec            = _mm_andnot_ps(dummy_mask,velec);
 611             velecsum         = _mm_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 616
 617             /* Calculate temporary vectorial force */
 618             tx               = _mm_mul_ps(fscal,dx10);
 619             ty               = _mm_mul_ps(fscal,dy10);
 620             tz               = _mm_mul_ps(fscal,dz10);
 621
 622             /* Update vectorial force */
 623             fix1             = _mm_add_ps(fix1,tx);
 624             fiy1             = _mm_add_ps(fiy1,ty);
 625             fiz1             = _mm_add_ps(fiz1,tz);
 626
 627             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 628                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 629                                                    tx,ty,tz);
 630
 631             /**************************
 632              * CALCULATE INTERACTIONS *
 633              **************************/
 634
 635             r20              = _mm_mul_ps(rsq20,rinv20);
 636             r20              = _mm_andnot_ps(dummy_mask,r20);
 637
 638             /* Compute parameters for interactions between i and j atoms */
 639             qq20             = _mm_mul_ps(iq2,jq0);
 640
 641             /* Calculate table index by multiplying r with table scale and truncate to integer */
 642             rt               = _mm_mul_ps(r20,vftabscale);
 643             vfitab           = _mm_cvttps_epi32(rt);
 644             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 645             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 646
 647             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 648             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 649             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 650             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 651             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 652             _MM_TRANSPOSE4_PS(Y,F,G,H);
 653             Heps             = _mm_mul_ps(vfeps,H);
 654             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 655             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 656             velec            = _mm_mul_ps(qq20,VV);
 657             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 658             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 659
 660             /* Update potential sum for this i atom from the interaction with this j atom. */
 661             velec            = _mm_andnot_ps(dummy_mask,velec);
 662             velecsum         = _mm_add_ps(velecsum,velec);
 663
 664             fscal            = felec;
 665
 666             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 667
 668             /* Calculate temporary vectorial force */
 669             tx               = _mm_mul_ps(fscal,dx20);
 670             ty               = _mm_mul_ps(fscal,dy20);
 671             tz               = _mm_mul_ps(fscal,dz20);
 672
 673             /* Update vectorial force */
 674             fix2             = _mm_add_ps(fix2,tx);
 675             fiy2             = _mm_add_ps(fiy2,ty);
 676             fiz2             = _mm_add_ps(fiz2,tz);
 677
 678             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 679                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 680                                                    tx,ty,tz);
 681
 682             /**************************
 683              * CALCULATE INTERACTIONS *
 684              **************************/
 685
 686             r30              = _mm_mul_ps(rsq30,rinv30);
 687             r30              = _mm_andnot_ps(dummy_mask,r30);
 688
 689             /* Compute parameters for interactions between i and j atoms */
 690             qq30             = _mm_mul_ps(iq3,jq0);
 691
 692             /* Calculate table index by multiplying r with table scale and truncate to integer */
 693             rt               = _mm_mul_ps(r30,vftabscale);
 694             vfitab           = _mm_cvttps_epi32(rt);
 695             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 696             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 697
 698             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 699             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 700             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 701             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 702             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 703             _MM_TRANSPOSE4_PS(Y,F,G,H);
 704             Heps             = _mm_mul_ps(vfeps,H);
 705             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 706             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 707             velec            = _mm_mul_ps(qq30,VV);
 708             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 709             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 710
 711             /* Update potential sum for this i atom from the interaction with this j atom. */
 712             velec            = _mm_andnot_ps(dummy_mask,velec);
 713             velecsum         = _mm_add_ps(velecsum,velec);
 714
 715             fscal            = felec;
 716
 717             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 718
 719             /* Calculate temporary vectorial force */
 720             tx               = _mm_mul_ps(fscal,dx30);
 721             ty               = _mm_mul_ps(fscal,dy30);
 722             tz               = _mm_mul_ps(fscal,dz30);
 723
 724             /* Update vectorial force */
 725             fix3             = _mm_add_ps(fix3,tx);
 726             fiy3             = _mm_add_ps(fiy3,ty);
 727             fiz3             = _mm_add_ps(fiz3,tz);
 728
 729             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 730                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 731                                                    tx,ty,tz);
 732
 733             /* Inner loop uses 189 flops */
 734         }
 735
 736         /* End of innermost loop */
 737
 738         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 739                                               f+i_coord_offset,fshift+i_shift_offset);
 740
 741         ggid                        = gid[iidx];
 742         /* Update potential energies */
 743         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 744         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 745
 746         /* Increment number of inner iterations */
 747         inneriter                  += j_index_end - j_index_start;
 748
 749         /* Outer loop uses 38 flops */
 750     }
 751
 752     /* Increment number of outer iterations */
 753     outeriter        += nri;
 754
 755     /* Update outer/inner flops */
 756
 757     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*189);
 758 }
 759 /*
 760  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_single
 761  * Electrostatics interaction: CubicSplineTable
 762  * VdW interaction:            CubicSplineTable
 763  * Geometry:                   Water4-Particle
 764  * Calculate force/pot:        Force
 765  */
 766 void
 767 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_sse2_single
 768                     (t_nblist * gmx_restrict                nlist,
 769                      rvec * gmx_restrict                    xx,
 770                      rvec * gmx_restrict                    ff,
 771                      t_forcerec * gmx_restrict              fr,
 772                      t_mdatoms * gmx_restrict               mdatoms,
 773                      nb_kernel_data_t * gmx_restrict        kernel_data,
 774                      t_nrnb * gmx_restrict                  nrnb)
 775 {
 776     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 777      * just 0 for non-waters.
 778      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 779      * jnr indices corresponding to data put in the four positions in the SIMD register.
 780      */
 781     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 782     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 783     int              jnrA,jnrB,jnrC,jnrD;
 784     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 785     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 786     real             shX,shY,shZ,rcutoff_scalar;
 787     real             *shiftvec,*fshift,*x,*f;
 788     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 789     int              vdwioffset0;
 790     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 791     int              vdwioffset1;
 792     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 793     int              vdwioffset2;
 794     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 795     int              vdwioffset3;
 796     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 797     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 798     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 799     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 800     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 801     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 802     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 803     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 804     real             *charge;
 805     int              nvdwtype;
 806     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 807     int              *vdwtype;
 808     real             *vdwparam;
 809     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 810     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 811     __m128i          vfitab;
 812     __m128i          ifour       = _mm_set1_epi32(4);
 813     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 814     real             *vftab;
 815     __m128           dummy_mask,cutoff_mask;
 816     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 817     __m128           one     = _mm_set1_ps(1.0);
 818     __m128           two     = _mm_set1_ps(2.0);
 819     x                = xx[0];
 820     f                = ff[0];
 821
 822     nri              = nlist->nri;
 823     iinr             = nlist->iinr;
 824     jindex           = nlist->jindex;
 825     jjnr             = nlist->jjnr;
 826     shiftidx         = nlist->shift;
 827     gid              = nlist->gid;
 828     shiftvec         = fr->shift_vec[0];
 829     fshift           = fr->fshift[0];
 830     facel            = _mm_set1_ps(fr->epsfac);
 831     charge           = mdatoms->chargeA;
 832     nvdwtype         = fr->ntype;
 833     vdwparam         = fr->nbfp;
 834     vdwtype          = mdatoms->typeA;
 835
 836     vftab            = kernel_data->table_elec_vdw->data;
 837     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 838
 839     /* Setup water-specific parameters */
 840     inr              = nlist->iinr[0];
 841     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 842     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 843     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 844     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 845
 846     /* Avoid stupid compiler warnings */
 847     jnrA = jnrB = jnrC = jnrD = 0;
 848     j_coord_offsetA = 0;
 849     j_coord_offsetB = 0;
 850     j_coord_offsetC = 0;
 851     j_coord_offsetD = 0;
 852
 853     outeriter        = 0;
 854     inneriter        = 0;
 855
 856     /* Start outer loop over neighborlists */
 857     for(iidx=0; iidx<nri; iidx++)
 858     {
 859         /* Load shift vector for this list */
 860         i_shift_offset   = DIM*shiftidx[iidx];
 861         shX              = shiftvec[i_shift_offset+XX];
 862         shY              = shiftvec[i_shift_offset+YY];
 863         shZ              = shiftvec[i_shift_offset+ZZ];
 864
 865         /* Load limits for loop over neighbors */
 866         j_index_start    = jindex[iidx];
 867         j_index_end      = jindex[iidx+1];
 868
 869         /* Get outer coordinate index */
 870         inr              = iinr[iidx];
 871         i_coord_offset   = DIM*inr;
 872
 873         /* Load i particle coords and add shift vector */
 874         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 875         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 876         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 877         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 878         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 879         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 880         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 881         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 882         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 883         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 884         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 885         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 886
 887         fix0             = _mm_setzero_ps();
 888         fiy0             = _mm_setzero_ps();
 889         fiz0             = _mm_setzero_ps();
 890         fix1             = _mm_setzero_ps();
 891         fiy1             = _mm_setzero_ps();
 892         fiz1             = _mm_setzero_ps();
 893         fix2             = _mm_setzero_ps();
 894         fiy2             = _mm_setzero_ps();
 895         fiz2             = _mm_setzero_ps();
 896         fix3             = _mm_setzero_ps();
 897         fiy3             = _mm_setzero_ps();
 898         fiz3             = _mm_setzero_ps();
 899
 900         /* Start inner kernel loop */
 901         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 902         {
 903
 904             /* Get j neighbor index, and coordinate index */
 905             jnrA             = jjnr[jidx];
 906             jnrB             = jjnr[jidx+1];
 907             jnrC             = jjnr[jidx+2];
 908             jnrD             = jjnr[jidx+3];
 909
 910             j_coord_offsetA  = DIM*jnrA;
 911             j_coord_offsetB  = DIM*jnrB;
 912             j_coord_offsetC  = DIM*jnrC;
 913             j_coord_offsetD  = DIM*jnrD;
 914
 915             /* load j atom coordinates */
 916             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 917                                               x+j_coord_offsetC,x+j_coord_offsetD,
 918                                               &jx0,&jy0,&jz0);
 919
 920             /* Calculate displacement vector */
 921             dx00             = _mm_sub_ps(ix0,jx0);
 922             dy00             = _mm_sub_ps(iy0,jy0);
 923             dz00             = _mm_sub_ps(iz0,jz0);
 924             dx10             = _mm_sub_ps(ix1,jx0);
 925             dy10             = _mm_sub_ps(iy1,jy0);
 926             dz10             = _mm_sub_ps(iz1,jz0);
 927             dx20             = _mm_sub_ps(ix2,jx0);
 928             dy20             = _mm_sub_ps(iy2,jy0);
 929             dz20             = _mm_sub_ps(iz2,jz0);
 930             dx30             = _mm_sub_ps(ix3,jx0);
 931             dy30             = _mm_sub_ps(iy3,jy0);
 932             dz30             = _mm_sub_ps(iz3,jz0);
 933
 934             /* Calculate squared distance and things based on it */
 935             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 936             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 937             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 938             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 939
 940             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 941             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 942             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 943             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 944
 945             /* Load parameters for j particles */
 946             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 947                                                               charge+jnrC+0,charge+jnrD+0);
 948             vdwjidx0A        = 2*vdwtype[jnrA+0];
 949             vdwjidx0B        = 2*vdwtype[jnrB+0];
 950             vdwjidx0C        = 2*vdwtype[jnrC+0];
 951             vdwjidx0D        = 2*vdwtype[jnrD+0];
 952
 953             /**************************
 954              * CALCULATE INTERACTIONS *
 955              **************************/
 956
 957             r00              = _mm_mul_ps(rsq00,rinv00);
 958
 959             /* Compute parameters for interactions between i and j atoms */
 960             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 961                                          vdwparam+vdwioffset0+vdwjidx0B,
 962                                          vdwparam+vdwioffset0+vdwjidx0C,
 963                                          vdwparam+vdwioffset0+vdwjidx0D,
 964                                          &c6_00,&c12_00);
 965
 966             /* Calculate table index by multiplying r with table scale and truncate to integer */
 967             rt               = _mm_mul_ps(r00,vftabscale);
 968             vfitab           = _mm_cvttps_epi32(rt);
 969             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 970             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 971
 972             /* CUBIC SPLINE TABLE DISPERSION */
 973             vfitab           = _mm_add_epi32(vfitab,ifour);
 974             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 975             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 976             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 977             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 978             _MM_TRANSPOSE4_PS(Y,F,G,H);
 979             Heps             = _mm_mul_ps(vfeps,H);
 980             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 981             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 982             fvdw6            = _mm_mul_ps(c6_00,FF);
 983
 984             /* CUBIC SPLINE TABLE REPULSION */
 985             vfitab           = _mm_add_epi32(vfitab,ifour);
 986             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 987             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 988             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 989             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 990             _MM_TRANSPOSE4_PS(Y,F,G,H);
 991             Heps             = _mm_mul_ps(vfeps,H);
 992             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 993             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 994             fvdw12           = _mm_mul_ps(c12_00,FF);
 995             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 996
 997             fscal            = fvdw;
 998
 999             /* Calculate temporary vectorial force */
1000             tx               = _mm_mul_ps(fscal,dx00);
1001             ty               = _mm_mul_ps(fscal,dy00);
1002             tz               = _mm_mul_ps(fscal,dz00);
1003
1004             /* Update vectorial force */
1005             fix0             = _mm_add_ps(fix0,tx);
1006             fiy0             = _mm_add_ps(fiy0,ty);
1007             fiz0             = _mm_add_ps(fiz0,tz);
1008
1009             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1010                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1011                                                    tx,ty,tz);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r10              = _mm_mul_ps(rsq10,rinv10);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq10             = _mm_mul_ps(iq1,jq0);
1021
1022             /* Calculate table index by multiplying r with table scale and truncate to integer */
1023             rt               = _mm_mul_ps(r10,vftabscale);
1024             vfitab           = _mm_cvttps_epi32(rt);
1025             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1026             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1027
1028             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1029             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1030             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1031             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1032             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1033             _MM_TRANSPOSE4_PS(Y,F,G,H);
1034             Heps             = _mm_mul_ps(vfeps,H);
1035             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1036             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1037             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1038
1039             fscal            = felec;
1040
1041             /* Calculate temporary vectorial force */
1042             tx               = _mm_mul_ps(fscal,dx10);
1043             ty               = _mm_mul_ps(fscal,dy10);
1044             tz               = _mm_mul_ps(fscal,dz10);
1045
1046             /* Update vectorial force */
1047             fix1             = _mm_add_ps(fix1,tx);
1048             fiy1             = _mm_add_ps(fiy1,ty);
1049             fiz1             = _mm_add_ps(fiz1,tz);
1050
1051             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1052                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1053                                                    tx,ty,tz);
1054
1055             /**************************
1056              * CALCULATE INTERACTIONS *
1057              **************************/
1058
1059             r20              = _mm_mul_ps(rsq20,rinv20);
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq20             = _mm_mul_ps(iq2,jq0);
1063
1064             /* Calculate table index by multiplying r with table scale and truncate to integer */
1065             rt               = _mm_mul_ps(r20,vftabscale);
1066             vfitab           = _mm_cvttps_epi32(rt);
1067             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1068             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1069
1070             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1071             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1072             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1073             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1074             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1075             _MM_TRANSPOSE4_PS(Y,F,G,H);
1076             Heps             = _mm_mul_ps(vfeps,H);
1077             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1078             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1079             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1080
1081             fscal            = felec;
1082
1083             /* Calculate temporary vectorial force */
1084             tx               = _mm_mul_ps(fscal,dx20);
1085             ty               = _mm_mul_ps(fscal,dy20);
1086             tz               = _mm_mul_ps(fscal,dz20);
1087
1088             /* Update vectorial force */
1089             fix2             = _mm_add_ps(fix2,tx);
1090             fiy2             = _mm_add_ps(fiy2,ty);
1091             fiz2             = _mm_add_ps(fiz2,tz);
1092
1093             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1094                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1095                                                    tx,ty,tz);
1096
1097             /**************************
1098              * CALCULATE INTERACTIONS *
1099              **************************/
1100
1101             r30              = _mm_mul_ps(rsq30,rinv30);
1102
1103             /* Compute parameters for interactions between i and j atoms */
1104             qq30             = _mm_mul_ps(iq3,jq0);
1105
1106             /* Calculate table index by multiplying r with table scale and truncate to integer */
1107             rt               = _mm_mul_ps(r30,vftabscale);
1108             vfitab           = _mm_cvttps_epi32(rt);
1109             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1110             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1111
1112             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1113             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1114             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1115             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1116             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1117             _MM_TRANSPOSE4_PS(Y,F,G,H);
1118             Heps             = _mm_mul_ps(vfeps,H);
1119             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1120             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1121             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1122
1123             fscal            = felec;
1124
1125             /* Calculate temporary vectorial force */
1126             tx               = _mm_mul_ps(fscal,dx30);
1127             ty               = _mm_mul_ps(fscal,dy30);
1128             tz               = _mm_mul_ps(fscal,dz30);
1129
1130             /* Update vectorial force */
1131             fix3             = _mm_add_ps(fix3,tx);
1132             fiy3             = _mm_add_ps(fiy3,ty);
1133             fiz3             = _mm_add_ps(fiz3,tz);
1134
1135             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1136                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1137                                                    tx,ty,tz);
1138
1139             /* Inner loop uses 165 flops */
1140         }
1141
1142         if(jidx<j_index_end)
1143         {
1144
1145             /* Get j neighbor index, and coordinate index */
1146             jnrA             = jjnr[jidx];
1147             jnrB             = jjnr[jidx+1];
1148             jnrC             = jjnr[jidx+2];
1149             jnrD             = jjnr[jidx+3];
1150
1151             /* Sign of each element will be negative for non-real atoms.
1152              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1153              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1154              */
1155             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1156             jnrA       = (jnrA>=0) ? jnrA : 0;
1157             jnrB       = (jnrB>=0) ? jnrB : 0;
1158             jnrC       = (jnrC>=0) ? jnrC : 0;
1159             jnrD       = (jnrD>=0) ? jnrD : 0;
1160
1161             j_coord_offsetA  = DIM*jnrA;
1162             j_coord_offsetB  = DIM*jnrB;
1163             j_coord_offsetC  = DIM*jnrC;
1164             j_coord_offsetD  = DIM*jnrD;
1165
1166             /* load j atom coordinates */
1167             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1168                                               x+j_coord_offsetC,x+j_coord_offsetD,
1169                                               &jx0,&jy0,&jz0);
1170
1171             /* Calculate displacement vector */
1172             dx00             = _mm_sub_ps(ix0,jx0);
1173             dy00             = _mm_sub_ps(iy0,jy0);
1174             dz00             = _mm_sub_ps(iz0,jz0);
1175             dx10             = _mm_sub_ps(ix1,jx0);
1176             dy10             = _mm_sub_ps(iy1,jy0);
1177             dz10             = _mm_sub_ps(iz1,jz0);
1178             dx20             = _mm_sub_ps(ix2,jx0);
1179             dy20             = _mm_sub_ps(iy2,jy0);
1180             dz20             = _mm_sub_ps(iz2,jz0);
1181             dx30             = _mm_sub_ps(ix3,jx0);
1182             dy30             = _mm_sub_ps(iy3,jy0);
1183             dz30             = _mm_sub_ps(iz3,jz0);
1184
1185             /* Calculate squared distance and things based on it */
1186             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1187             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1188             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1189             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1190
1191             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1192             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1193             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1194             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1195
1196             /* Load parameters for j particles */
1197             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1198                                                               charge+jnrC+0,charge+jnrD+0);
1199             vdwjidx0A        = 2*vdwtype[jnrA+0];
1200             vdwjidx0B        = 2*vdwtype[jnrB+0];
1201             vdwjidx0C        = 2*vdwtype[jnrC+0];
1202             vdwjidx0D        = 2*vdwtype[jnrD+0];
1203
1204             /**************************
1205              * CALCULATE INTERACTIONS *
1206              **************************/
1207
1208             r00              = _mm_mul_ps(rsq00,rinv00);
1209             r00              = _mm_andnot_ps(dummy_mask,r00);
1210
1211             /* Compute parameters for interactions between i and j atoms */
1212             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1213                                          vdwparam+vdwioffset0+vdwjidx0B,
1214                                          vdwparam+vdwioffset0+vdwjidx0C,
1215                                          vdwparam+vdwioffset0+vdwjidx0D,
1216                                          &c6_00,&c12_00);
1217
1218             /* Calculate table index by multiplying r with table scale and truncate to integer */
1219             rt               = _mm_mul_ps(r00,vftabscale);
1220             vfitab           = _mm_cvttps_epi32(rt);
1221             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1222             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1223
1224             /* CUBIC SPLINE TABLE DISPERSION */
1225             vfitab           = _mm_add_epi32(vfitab,ifour);
1226             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1227             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1228             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1229             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1230             _MM_TRANSPOSE4_PS(Y,F,G,H);
1231             Heps             = _mm_mul_ps(vfeps,H);
1232             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1233             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1234             fvdw6            = _mm_mul_ps(c6_00,FF);
1235
1236             /* CUBIC SPLINE TABLE REPULSION */
1237             vfitab           = _mm_add_epi32(vfitab,ifour);
1238             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1239             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1240             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1241             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1242             _MM_TRANSPOSE4_PS(Y,F,G,H);
1243             Heps             = _mm_mul_ps(vfeps,H);
1244             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1245             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1246             fvdw12           = _mm_mul_ps(c12_00,FF);
1247             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1248
1249             fscal            = fvdw;
1250
1251             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1252
1253             /* Calculate temporary vectorial force */
1254             tx               = _mm_mul_ps(fscal,dx00);
1255             ty               = _mm_mul_ps(fscal,dy00);
1256             tz               = _mm_mul_ps(fscal,dz00);
1257
1258             /* Update vectorial force */
1259             fix0             = _mm_add_ps(fix0,tx);
1260             fiy0             = _mm_add_ps(fiy0,ty);
1261             fiz0             = _mm_add_ps(fiz0,tz);
1262
1263             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1264                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1265                                                    tx,ty,tz);
1266
1267             /**************************
1268              * CALCULATE INTERACTIONS *
1269              **************************/
1270
1271             r10              = _mm_mul_ps(rsq10,rinv10);
1272             r10              = _mm_andnot_ps(dummy_mask,r10);
1273
1274             /* Compute parameters for interactions between i and j atoms */
1275             qq10             = _mm_mul_ps(iq1,jq0);
1276
1277             /* Calculate table index by multiplying r with table scale and truncate to integer */
1278             rt               = _mm_mul_ps(r10,vftabscale);
1279             vfitab           = _mm_cvttps_epi32(rt);
1280             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1281             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1282
1283             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1284             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1285             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1286             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1287             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1288             _MM_TRANSPOSE4_PS(Y,F,G,H);
1289             Heps             = _mm_mul_ps(vfeps,H);
1290             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1291             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1292             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1293
1294             fscal            = felec;
1295
1296             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1297
1298             /* Calculate temporary vectorial force */
1299             tx               = _mm_mul_ps(fscal,dx10);
1300             ty               = _mm_mul_ps(fscal,dy10);
1301             tz               = _mm_mul_ps(fscal,dz10);
1302
1303             /* Update vectorial force */
1304             fix1             = _mm_add_ps(fix1,tx);
1305             fiy1             = _mm_add_ps(fiy1,ty);
1306             fiz1             = _mm_add_ps(fiz1,tz);
1307
1308             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1309                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1310                                                    tx,ty,tz);
1311
1312             /**************************
1313              * CALCULATE INTERACTIONS *
1314              **************************/
1315
1316             r20              = _mm_mul_ps(rsq20,rinv20);
1317             r20              = _mm_andnot_ps(dummy_mask,r20);
1318
1319             /* Compute parameters for interactions between i and j atoms */
1320             qq20             = _mm_mul_ps(iq2,jq0);
1321
1322             /* Calculate table index by multiplying r with table scale and truncate to integer */
1323             rt               = _mm_mul_ps(r20,vftabscale);
1324             vfitab           = _mm_cvttps_epi32(rt);
1325             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1326             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1327
1328             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1329             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1330             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1331             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1332             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1333             _MM_TRANSPOSE4_PS(Y,F,G,H);
1334             Heps             = _mm_mul_ps(vfeps,H);
1335             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1336             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1337             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1338
1339             fscal            = felec;
1340
1341             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1342
1343             /* Calculate temporary vectorial force */
1344             tx               = _mm_mul_ps(fscal,dx20);
1345             ty               = _mm_mul_ps(fscal,dy20);
1346             tz               = _mm_mul_ps(fscal,dz20);
1347
1348             /* Update vectorial force */
1349             fix2             = _mm_add_ps(fix2,tx);
1350             fiy2             = _mm_add_ps(fiy2,ty);
1351             fiz2             = _mm_add_ps(fiz2,tz);
1352
1353             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1354                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1355                                                    tx,ty,tz);
1356
1357             /**************************
1358              * CALCULATE INTERACTIONS *
1359              **************************/
1360
1361             r30              = _mm_mul_ps(rsq30,rinv30);
1362             r30              = _mm_andnot_ps(dummy_mask,r30);
1363
1364             /* Compute parameters for interactions between i and j atoms */
1365             qq30             = _mm_mul_ps(iq3,jq0);
1366
1367             /* Calculate table index by multiplying r with table scale and truncate to integer */
1368             rt               = _mm_mul_ps(r30,vftabscale);
1369             vfitab           = _mm_cvttps_epi32(rt);
1370             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1371             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1372
1373             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1374             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1375             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1376             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1377             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1378             _MM_TRANSPOSE4_PS(Y,F,G,H);
1379             Heps             = _mm_mul_ps(vfeps,H);
1380             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1381             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1382             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
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             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1399                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1400                                                    tx,ty,tz);
1401
1402             /* Inner loop uses 169 flops */
1403         }
1404
1405         /* End of innermost loop */
1406
1407         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1408                                               f+i_coord_offset,fshift+i_shift_offset);
1409
1410         /* Increment number of inner iterations */
1411         inneriter                  += j_index_end - j_index_start;
1412
1413         /* Outer loop uses 36 flops */
1414     }
1415
1416     /* Increment number of outer iterations */
1417     outeriter        += nri;
1418
1419     /* Update outer/inner flops */
1420
1421     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*169);
1422 }