src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128           dummy_mask,cutoff_mask;
  90     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  91     __m128           one     = _mm_set1_ps(1.0);
  92     __m128           two     = _mm_set1_ps(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_ps(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     nvdwtype         = fr->ntype;
 107     vdwparam         = fr->nbfp;
 108     vdwtype          = mdatoms->typeA;
 109
 110     vftab            = kernel_data->table_elec->data;
 111     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 112
 113     /* Setup water-specific parameters */
 114     inr              = nlist->iinr[0];
 115     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 116     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 117     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 118     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 119
 120     /* Avoid stupid compiler warnings */
 121     jnrA = jnrB = jnrC = jnrD = 0;
 122     j_coord_offsetA = 0;
 123     j_coord_offsetB = 0;
 124     j_coord_offsetC = 0;
 125     j_coord_offsetD = 0;
 126
 127     outeriter        = 0;
 128     inneriter        = 0;
 129
 130     /* Start outer loop over neighborlists */
 131     for(iidx=0; iidx<nri; iidx++)
 132     {
 133         /* Load shift vector for this list */
 134         i_shift_offset   = DIM*shiftidx[iidx];
 135         shX              = shiftvec[i_shift_offset+XX];
 136         shY              = shiftvec[i_shift_offset+YY];
 137         shZ              = shiftvec[i_shift_offset+ZZ];
 138
 139         /* Load limits for loop over neighbors */
 140         j_index_start    = jindex[iidx];
 141         j_index_end      = jindex[iidx+1];
 142
 143         /* Get outer coordinate index */
 144         inr              = iinr[iidx];
 145         i_coord_offset   = DIM*inr;
 146
 147         /* Load i particle coords and add shift vector */
 148         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 149         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 150         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 151         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 152         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 153         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 154         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 155         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 156         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 157
 158         fix0             = _mm_setzero_ps();
 159         fiy0             = _mm_setzero_ps();
 160         fiz0             = _mm_setzero_ps();
 161         fix1             = _mm_setzero_ps();
 162         fiy1             = _mm_setzero_ps();
 163         fiz1             = _mm_setzero_ps();
 164         fix2             = _mm_setzero_ps();
 165         fiy2             = _mm_setzero_ps();
 166         fiz2             = _mm_setzero_ps();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_ps();
 170         vvdwsum          = _mm_setzero_ps();
 171
 172         /* Start inner kernel loop */
 173         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 174         {
 175
 176             /* Get j neighbor index, and coordinate index */
 177             jnrA             = jjnr[jidx];
 178             jnrB             = jjnr[jidx+1];
 179             jnrC             = jjnr[jidx+2];
 180             jnrD             = jjnr[jidx+3];
 181
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196             dx10             = _mm_sub_ps(ix1,jx0);
 197             dy10             = _mm_sub_ps(iy1,jy0);
 198             dz10             = _mm_sub_ps(iz1,jz0);
 199             dx20             = _mm_sub_ps(ix2,jx0);
 200             dy20             = _mm_sub_ps(iy2,jy0);
 201             dz20             = _mm_sub_ps(iz2,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 207
 208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 209             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 210             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 211
 212             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219             vdwjidx0C        = 2*vdwtype[jnrC+0];
 220             vdwjidx0D        = 2*vdwtype[jnrD+0];
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r00              = _mm_mul_ps(rsq00,rinv00);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq00             = _mm_mul_ps(iq0,jq0);
 230             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 231                                          vdwparam+vdwioffset0+vdwjidx0B,
 232                                          vdwparam+vdwioffset0+vdwjidx0C,
 233                                          vdwparam+vdwioffset0+vdwjidx0D,
 234                                          &c6_00,&c12_00);
 235
 236             /* Calculate table index by multiplying r with table scale and truncate to integer */
 237             rt               = _mm_mul_ps(r00,vftabscale);
 238             vfitab           = _mm_cvttps_epi32(rt);
 239             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 240             vfitab           = _mm_slli_epi32(vfitab,2);
 241
 242             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 243             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 245             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 246             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 247             _MM_TRANSPOSE4_PS(Y,F,G,H);
 248             Heps             = _mm_mul_ps(vfeps,H);
 249             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 250             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 251             velec            = _mm_mul_ps(qq00,VV);
 252             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 253             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 254
 255             /* LENNARD-JONES DISPERSION/REPULSION */
 256
 257             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 258             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 259             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 260             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 261             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velecsum         = _mm_add_ps(velecsum,velec);
 265             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 266
 267             fscal            = _mm_add_ps(felec,fvdw);
 268
 269             /* Calculate temporary vectorial force */
 270             tx               = _mm_mul_ps(fscal,dx00);
 271             ty               = _mm_mul_ps(fscal,dy00);
 272             tz               = _mm_mul_ps(fscal,dz00);
 273
 274             /* Update vectorial force */
 275             fix0             = _mm_add_ps(fix0,tx);
 276             fiy0             = _mm_add_ps(fiy0,ty);
 277             fiz0             = _mm_add_ps(fiz0,tz);
 278
 279             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 280                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 281                                                    tx,ty,tz);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r10              = _mm_mul_ps(rsq10,rinv10);
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_ps(iq1,jq0);
 291
 292             /* Calculate table index by multiplying r with table scale and truncate to integer */
 293             rt               = _mm_mul_ps(r10,vftabscale);
 294             vfitab           = _mm_cvttps_epi32(rt);
 295             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 296             vfitab           = _mm_slli_epi32(vfitab,2);
 297
 298             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 299             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 300             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 301             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 302             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 303             _MM_TRANSPOSE4_PS(Y,F,G,H);
 304             Heps             = _mm_mul_ps(vfeps,H);
 305             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 306             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 307             velec            = _mm_mul_ps(qq10,VV);
 308             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 309             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx10);
 318             ty               = _mm_mul_ps(fscal,dy10);
 319             tz               = _mm_mul_ps(fscal,dz10);
 320
 321             /* Update vectorial force */
 322             fix1             = _mm_add_ps(fix1,tx);
 323             fiy1             = _mm_add_ps(fiy1,ty);
 324             fiz1             = _mm_add_ps(fiz1,tz);
 325
 326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 327                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 328                                                    tx,ty,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             r20              = _mm_mul_ps(rsq20,rinv20);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* Calculate table index by multiplying r with table scale and truncate to integer */
 340             rt               = _mm_mul_ps(r20,vftabscale);
 341             vfitab           = _mm_cvttps_epi32(rt);
 342             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 343             vfitab           = _mm_slli_epi32(vfitab,2);
 344
 345             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 346             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 347             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 348             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 349             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 350             _MM_TRANSPOSE4_PS(Y,F,G,H);
 351             Heps             = _mm_mul_ps(vfeps,H);
 352             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 353             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 354             velec            = _mm_mul_ps(qq20,VV);
 355             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 356             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 357
 358             /* Update potential sum for this i atom from the interaction with this j atom. */
 359             velecsum         = _mm_add_ps(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             /* Calculate temporary vectorial force */
 364             tx               = _mm_mul_ps(fscal,dx20);
 365             ty               = _mm_mul_ps(fscal,dy20);
 366             tz               = _mm_mul_ps(fscal,dz20);
 367
 368             /* Update vectorial force */
 369             fix2             = _mm_add_ps(fix2,tx);
 370             fiy2             = _mm_add_ps(fiy2,ty);
 371             fiz2             = _mm_add_ps(fiz2,tz);
 372
 373             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 374                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 375                                                    tx,ty,tz);
 376
 377             /* Inner loop uses 142 flops */
 378         }
 379
 380         if(jidx<j_index_end)
 381         {
 382
 383             /* Get j neighbor index, and coordinate index */
 384             jnrA             = jjnr[jidx];
 385             jnrB             = jjnr[jidx+1];
 386             jnrC             = jjnr[jidx+2];
 387             jnrD             = jjnr[jidx+3];
 388
 389             /* Sign of each element will be negative for non-real atoms.
 390              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 391              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 392              */
 393             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 394             jnrA       = (jnrA>=0) ? jnrA : 0;
 395             jnrB       = (jnrB>=0) ? jnrB : 0;
 396             jnrC       = (jnrC>=0) ? jnrC : 0;
 397             jnrD       = (jnrD>=0) ? jnrD : 0;
 398
 399             j_coord_offsetA  = DIM*jnrA;
 400             j_coord_offsetB  = DIM*jnrB;
 401             j_coord_offsetC  = DIM*jnrC;
 402             j_coord_offsetD  = DIM*jnrD;
 403
 404             /* load j atom coordinates */
 405             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 406                                               x+j_coord_offsetC,x+j_coord_offsetD,
 407                                               &jx0,&jy0,&jz0);
 408
 409             /* Calculate displacement vector */
 410             dx00             = _mm_sub_ps(ix0,jx0);
 411             dy00             = _mm_sub_ps(iy0,jy0);
 412             dz00             = _mm_sub_ps(iz0,jz0);
 413             dx10             = _mm_sub_ps(ix1,jx0);
 414             dy10             = _mm_sub_ps(iy1,jy0);
 415             dz10             = _mm_sub_ps(iz1,jz0);
 416             dx20             = _mm_sub_ps(ix2,jx0);
 417             dy20             = _mm_sub_ps(iy2,jy0);
 418             dz20             = _mm_sub_ps(iz2,jz0);
 419
 420             /* Calculate squared distance and things based on it */
 421             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 422             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 423             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 424
 425             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 426             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 427             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 428
 429             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 430
 431             /* Load parameters for j particles */
 432             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 433                                                               charge+jnrC+0,charge+jnrD+0);
 434             vdwjidx0A        = 2*vdwtype[jnrA+0];
 435             vdwjidx0B        = 2*vdwtype[jnrB+0];
 436             vdwjidx0C        = 2*vdwtype[jnrC+0];
 437             vdwjidx0D        = 2*vdwtype[jnrD+0];
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r00              = _mm_mul_ps(rsq00,rinv00);
 444             r00              = _mm_andnot_ps(dummy_mask,r00);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq00             = _mm_mul_ps(iq0,jq0);
 448             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 449                                          vdwparam+vdwioffset0+vdwjidx0B,
 450                                          vdwparam+vdwioffset0+vdwjidx0C,
 451                                          vdwparam+vdwioffset0+vdwjidx0D,
 452                                          &c6_00,&c12_00);
 453
 454             /* Calculate table index by multiplying r with table scale and truncate to integer */
 455             rt               = _mm_mul_ps(r00,vftabscale);
 456             vfitab           = _mm_cvttps_epi32(rt);
 457             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 458             vfitab           = _mm_slli_epi32(vfitab,2);
 459
 460             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 461             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 462             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 463             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 464             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 465             _MM_TRANSPOSE4_PS(Y,F,G,H);
 466             Heps             = _mm_mul_ps(vfeps,H);
 467             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 468             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 469             velec            = _mm_mul_ps(qq00,VV);
 470             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 471             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 472
 473             /* LENNARD-JONES DISPERSION/REPULSION */
 474
 475             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 476             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 477             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 478             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 479             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm_andnot_ps(dummy_mask,velec);
 483             velecsum         = _mm_add_ps(velecsum,velec);
 484             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 485             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 486
 487             fscal            = _mm_add_ps(felec,fvdw);
 488
 489             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 490
 491             /* Calculate temporary vectorial force */
 492             tx               = _mm_mul_ps(fscal,dx00);
 493             ty               = _mm_mul_ps(fscal,dy00);
 494             tz               = _mm_mul_ps(fscal,dz00);
 495
 496             /* Update vectorial force */
 497             fix0             = _mm_add_ps(fix0,tx);
 498             fiy0             = _mm_add_ps(fiy0,ty);
 499             fiz0             = _mm_add_ps(fiz0,tz);
 500
 501             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 502                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 503                                                    tx,ty,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             r10              = _mm_mul_ps(rsq10,rinv10);
 510             r10              = _mm_andnot_ps(dummy_mask,r10);
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm_mul_ps(iq1,jq0);
 514
 515             /* Calculate table index by multiplying r with table scale and truncate to integer */
 516             rt               = _mm_mul_ps(r10,vftabscale);
 517             vfitab           = _mm_cvttps_epi32(rt);
 518             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 519             vfitab           = _mm_slli_epi32(vfitab,2);
 520
 521             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 522             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 523             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 524             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 525             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 526             _MM_TRANSPOSE4_PS(Y,F,G,H);
 527             Heps             = _mm_mul_ps(vfeps,H);
 528             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 529             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 530             velec            = _mm_mul_ps(qq10,VV);
 531             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 532             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 533
 534             /* Update potential sum for this i atom from the interaction with this j atom. */
 535             velec            = _mm_andnot_ps(dummy_mask,velec);
 536             velecsum         = _mm_add_ps(velecsum,velec);
 537
 538             fscal            = felec;
 539
 540             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 541
 542             /* Calculate temporary vectorial force */
 543             tx               = _mm_mul_ps(fscal,dx10);
 544             ty               = _mm_mul_ps(fscal,dy10);
 545             tz               = _mm_mul_ps(fscal,dz10);
 546
 547             /* Update vectorial force */
 548             fix1             = _mm_add_ps(fix1,tx);
 549             fiy1             = _mm_add_ps(fiy1,ty);
 550             fiz1             = _mm_add_ps(fiz1,tz);
 551
 552             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 553                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 554                                                    tx,ty,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r20              = _mm_mul_ps(rsq20,rinv20);
 561             r20              = _mm_andnot_ps(dummy_mask,r20);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq20             = _mm_mul_ps(iq2,jq0);
 565
 566             /* Calculate table index by multiplying r with table scale and truncate to integer */
 567             rt               = _mm_mul_ps(r20,vftabscale);
 568             vfitab           = _mm_cvttps_epi32(rt);
 569             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 570             vfitab           = _mm_slli_epi32(vfitab,2);
 571
 572             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 573             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 574             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 575             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 576             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 577             _MM_TRANSPOSE4_PS(Y,F,G,H);
 578             Heps             = _mm_mul_ps(vfeps,H);
 579             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 580             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 581             velec            = _mm_mul_ps(qq20,VV);
 582             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 583             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm_andnot_ps(dummy_mask,velec);
 587             velecsum         = _mm_add_ps(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 592
 593             /* Calculate temporary vectorial force */
 594             tx               = _mm_mul_ps(fscal,dx20);
 595             ty               = _mm_mul_ps(fscal,dy20);
 596             tz               = _mm_mul_ps(fscal,dz20);
 597
 598             /* Update vectorial force */
 599             fix2             = _mm_add_ps(fix2,tx);
 600             fiy2             = _mm_add_ps(fiy2,ty);
 601             fiz2             = _mm_add_ps(fiz2,tz);
 602
 603             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 604                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 605                                                    tx,ty,tz);
 606
 607             /* Inner loop uses 145 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         ggid                        = gid[iidx];
 616         /* Update potential energies */
 617         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 618         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 619
 620         /* Increment number of inner iterations */
 621         inneriter                  += j_index_end - j_index_start;
 622
 623         /* Outer loop uses 29 flops */
 624     }
 625
 626     /* Increment number of outer iterations */
 627     outeriter        += nri;
 628
 629     /* Update outer/inner flops */
 630
 631     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*145);
 632 }
 633 /*
 634  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 635  * Electrostatics interaction: CubicSplineTable
 636  * VdW interaction:            LennardJones
 637  * Geometry:                   Water3-Particle
 638  * Calculate force/pot:        Force
 639  */
 640 void
 641 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse2_single
 642                     (t_nblist * gmx_restrict                nlist,
 643                      rvec * gmx_restrict                    xx,
 644                      rvec * gmx_restrict                    ff,
 645                      t_forcerec * gmx_restrict              fr,
 646                      t_mdatoms * gmx_restrict               mdatoms,
 647                      nb_kernel_data_t * gmx_restrict        kernel_data,
 648                      t_nrnb * gmx_restrict                  nrnb)
 649 {
 650     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 651      * just 0 for non-waters.
 652      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 653      * jnr indices corresponding to data put in the four positions in the SIMD register.
 654      */
 655     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 656     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 657     int              jnrA,jnrB,jnrC,jnrD;
 658     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 659     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 660     real             shX,shY,shZ,rcutoff_scalar;
 661     real             *shiftvec,*fshift,*x,*f;
 662     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 663     int              vdwioffset0;
 664     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 665     int              vdwioffset1;
 666     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 667     int              vdwioffset2;
 668     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 669     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 670     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 671     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 672     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 673     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 674     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 675     real             *charge;
 676     int              nvdwtype;
 677     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 678     int              *vdwtype;
 679     real             *vdwparam;
 680     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 681     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 682     __m128i          vfitab;
 683     __m128i          ifour       = _mm_set1_epi32(4);
 684     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 685     real             *vftab;
 686     __m128           dummy_mask,cutoff_mask;
 687     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 688     __m128           one     = _mm_set1_ps(1.0);
 689     __m128           two     = _mm_set1_ps(2.0);
 690     x                = xx[0];
 691     f                = ff[0];
 692
 693     nri              = nlist->nri;
 694     iinr             = nlist->iinr;
 695     jindex           = nlist->jindex;
 696     jjnr             = nlist->jjnr;
 697     shiftidx         = nlist->shift;
 698     gid              = nlist->gid;
 699     shiftvec         = fr->shift_vec[0];
 700     fshift           = fr->fshift[0];
 701     facel            = _mm_set1_ps(fr->epsfac);
 702     charge           = mdatoms->chargeA;
 703     nvdwtype         = fr->ntype;
 704     vdwparam         = fr->nbfp;
 705     vdwtype          = mdatoms->typeA;
 706
 707     vftab            = kernel_data->table_elec->data;
 708     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 709
 710     /* Setup water-specific parameters */
 711     inr              = nlist->iinr[0];
 712     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 713     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 714     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 715     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 716
 717     /* Avoid stupid compiler warnings */
 718     jnrA = jnrB = jnrC = jnrD = 0;
 719     j_coord_offsetA = 0;
 720     j_coord_offsetB = 0;
 721     j_coord_offsetC = 0;
 722     j_coord_offsetD = 0;
 723
 724     outeriter        = 0;
 725     inneriter        = 0;
 726
 727     /* Start outer loop over neighborlists */
 728     for(iidx=0; iidx<nri; iidx++)
 729     {
 730         /* Load shift vector for this list */
 731         i_shift_offset   = DIM*shiftidx[iidx];
 732         shX              = shiftvec[i_shift_offset+XX];
 733         shY              = shiftvec[i_shift_offset+YY];
 734         shZ              = shiftvec[i_shift_offset+ZZ];
 735
 736         /* Load limits for loop over neighbors */
 737         j_index_start    = jindex[iidx];
 738         j_index_end      = jindex[iidx+1];
 739
 740         /* Get outer coordinate index */
 741         inr              = iinr[iidx];
 742         i_coord_offset   = DIM*inr;
 743
 744         /* Load i particle coords and add shift vector */
 745         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 746         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 747         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 748         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 749         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 750         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 751         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 752         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 753         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 754
 755         fix0             = _mm_setzero_ps();
 756         fiy0             = _mm_setzero_ps();
 757         fiz0             = _mm_setzero_ps();
 758         fix1             = _mm_setzero_ps();
 759         fiy1             = _mm_setzero_ps();
 760         fiz1             = _mm_setzero_ps();
 761         fix2             = _mm_setzero_ps();
 762         fiy2             = _mm_setzero_ps();
 763         fiz2             = _mm_setzero_ps();
 764
 765         /* Start inner kernel loop */
 766         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 767         {
 768
 769             /* Get j neighbor index, and coordinate index */
 770             jnrA             = jjnr[jidx];
 771             jnrB             = jjnr[jidx+1];
 772             jnrC             = jjnr[jidx+2];
 773             jnrD             = jjnr[jidx+3];
 774
 775             j_coord_offsetA  = DIM*jnrA;
 776             j_coord_offsetB  = DIM*jnrB;
 777             j_coord_offsetC  = DIM*jnrC;
 778             j_coord_offsetD  = DIM*jnrD;
 779
 780             /* load j atom coordinates */
 781             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 782                                               x+j_coord_offsetC,x+j_coord_offsetD,
 783                                               &jx0,&jy0,&jz0);
 784
 785             /* Calculate displacement vector */
 786             dx00             = _mm_sub_ps(ix0,jx0);
 787             dy00             = _mm_sub_ps(iy0,jy0);
 788             dz00             = _mm_sub_ps(iz0,jz0);
 789             dx10             = _mm_sub_ps(ix1,jx0);
 790             dy10             = _mm_sub_ps(iy1,jy0);
 791             dz10             = _mm_sub_ps(iz1,jz0);
 792             dx20             = _mm_sub_ps(ix2,jx0);
 793             dy20             = _mm_sub_ps(iy2,jy0);
 794             dz20             = _mm_sub_ps(iz2,jz0);
 795
 796             /* Calculate squared distance and things based on it */
 797             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 798             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 799             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 800
 801             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 802             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 803             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 804
 805             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 806
 807             /* Load parameters for j particles */
 808             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 809                                                               charge+jnrC+0,charge+jnrD+0);
 810             vdwjidx0A        = 2*vdwtype[jnrA+0];
 811             vdwjidx0B        = 2*vdwtype[jnrB+0];
 812             vdwjidx0C        = 2*vdwtype[jnrC+0];
 813             vdwjidx0D        = 2*vdwtype[jnrD+0];
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r00              = _mm_mul_ps(rsq00,rinv00);
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             qq00             = _mm_mul_ps(iq0,jq0);
 823             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 824                                          vdwparam+vdwioffset0+vdwjidx0B,
 825                                          vdwparam+vdwioffset0+vdwjidx0C,
 826                                          vdwparam+vdwioffset0+vdwjidx0D,
 827                                          &c6_00,&c12_00);
 828
 829             /* Calculate table index by multiplying r with table scale and truncate to integer */
 830             rt               = _mm_mul_ps(r00,vftabscale);
 831             vfitab           = _mm_cvttps_epi32(rt);
 832             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 833             vfitab           = _mm_slli_epi32(vfitab,2);
 834
 835             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 836             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 837             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 838             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 839             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 840             _MM_TRANSPOSE4_PS(Y,F,G,H);
 841             Heps             = _mm_mul_ps(vfeps,H);
 842             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 843             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 844             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 845
 846             /* LENNARD-JONES DISPERSION/REPULSION */
 847
 848             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 849             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 850
 851             fscal            = _mm_add_ps(felec,fvdw);
 852
 853             /* Calculate temporary vectorial force */
 854             tx               = _mm_mul_ps(fscal,dx00);
 855             ty               = _mm_mul_ps(fscal,dy00);
 856             tz               = _mm_mul_ps(fscal,dz00);
 857
 858             /* Update vectorial force */
 859             fix0             = _mm_add_ps(fix0,tx);
 860             fiy0             = _mm_add_ps(fiy0,ty);
 861             fiz0             = _mm_add_ps(fiz0,tz);
 862
 863             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 864                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 865                                                    tx,ty,tz);
 866
 867             /**************************
 868              * CALCULATE INTERACTIONS *
 869              **************************/
 870
 871             r10              = _mm_mul_ps(rsq10,rinv10);
 872
 873             /* Compute parameters for interactions between i and j atoms */
 874             qq10             = _mm_mul_ps(iq1,jq0);
 875
 876             /* Calculate table index by multiplying r with table scale and truncate to integer */
 877             rt               = _mm_mul_ps(r10,vftabscale);
 878             vfitab           = _mm_cvttps_epi32(rt);
 879             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 880             vfitab           = _mm_slli_epi32(vfitab,2);
 881
 882             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 883             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 884             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 885             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 886             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 887             _MM_TRANSPOSE4_PS(Y,F,G,H);
 888             Heps             = _mm_mul_ps(vfeps,H);
 889             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 890             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 891             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 892
 893             fscal            = felec;
 894
 895             /* Calculate temporary vectorial force */
 896             tx               = _mm_mul_ps(fscal,dx10);
 897             ty               = _mm_mul_ps(fscal,dy10);
 898             tz               = _mm_mul_ps(fscal,dz10);
 899
 900             /* Update vectorial force */
 901             fix1             = _mm_add_ps(fix1,tx);
 902             fiy1             = _mm_add_ps(fiy1,ty);
 903             fiz1             = _mm_add_ps(fiz1,tz);
 904
 905             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 906                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 907                                                    tx,ty,tz);
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             r20              = _mm_mul_ps(rsq20,rinv20);
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq20             = _mm_mul_ps(iq2,jq0);
 917
 918             /* Calculate table index by multiplying r with table scale and truncate to integer */
 919             rt               = _mm_mul_ps(r20,vftabscale);
 920             vfitab           = _mm_cvttps_epi32(rt);
 921             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 922             vfitab           = _mm_slli_epi32(vfitab,2);
 923
 924             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 925             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 926             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 927             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 928             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 929             _MM_TRANSPOSE4_PS(Y,F,G,H);
 930             Heps             = _mm_mul_ps(vfeps,H);
 931             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 932             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 933             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 934
 935             fscal            = felec;
 936
 937             /* Calculate temporary vectorial force */
 938             tx               = _mm_mul_ps(fscal,dx20);
 939             ty               = _mm_mul_ps(fscal,dy20);
 940             tz               = _mm_mul_ps(fscal,dz20);
 941
 942             /* Update vectorial force */
 943             fix2             = _mm_add_ps(fix2,tx);
 944             fiy2             = _mm_add_ps(fiy2,ty);
 945             fiz2             = _mm_add_ps(fiz2,tz);
 946
 947             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 948                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 949                                                    tx,ty,tz);
 950
 951             /* Inner loop uses 125 flops */
 952         }
 953
 954         if(jidx<j_index_end)
 955         {
 956
 957             /* Get j neighbor index, and coordinate index */
 958             jnrA             = jjnr[jidx];
 959             jnrB             = jjnr[jidx+1];
 960             jnrC             = jjnr[jidx+2];
 961             jnrD             = jjnr[jidx+3];
 962
 963             /* Sign of each element will be negative for non-real atoms.
 964              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 965              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 966              */
 967             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 968             jnrA       = (jnrA>=0) ? jnrA : 0;
 969             jnrB       = (jnrB>=0) ? jnrB : 0;
 970             jnrC       = (jnrC>=0) ? jnrC : 0;
 971             jnrD       = (jnrD>=0) ? jnrD : 0;
 972
 973             j_coord_offsetA  = DIM*jnrA;
 974             j_coord_offsetB  = DIM*jnrB;
 975             j_coord_offsetC  = DIM*jnrC;
 976             j_coord_offsetD  = DIM*jnrD;
 977
 978             /* load j atom coordinates */
 979             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 980                                               x+j_coord_offsetC,x+j_coord_offsetD,
 981                                               &jx0,&jy0,&jz0);
 982
 983             /* Calculate displacement vector */
 984             dx00             = _mm_sub_ps(ix0,jx0);
 985             dy00             = _mm_sub_ps(iy0,jy0);
 986             dz00             = _mm_sub_ps(iz0,jz0);
 987             dx10             = _mm_sub_ps(ix1,jx0);
 988             dy10             = _mm_sub_ps(iy1,jy0);
 989             dz10             = _mm_sub_ps(iz1,jz0);
 990             dx20             = _mm_sub_ps(ix2,jx0);
 991             dy20             = _mm_sub_ps(iy2,jy0);
 992             dz20             = _mm_sub_ps(iz2,jz0);
 993
 994             /* Calculate squared distance and things based on it */
 995             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 996             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 997             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 998
 999             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1000             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1001             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1002
1003             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1004
1005             /* Load parameters for j particles */
1006             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1007                                                               charge+jnrC+0,charge+jnrD+0);
1008             vdwjidx0A        = 2*vdwtype[jnrA+0];
1009             vdwjidx0B        = 2*vdwtype[jnrB+0];
1010             vdwjidx0C        = 2*vdwtype[jnrC+0];
1011             vdwjidx0D        = 2*vdwtype[jnrD+0];
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r00              = _mm_mul_ps(rsq00,rinv00);
1018             r00              = _mm_andnot_ps(dummy_mask,r00);
1019
1020             /* Compute parameters for interactions between i and j atoms */
1021             qq00             = _mm_mul_ps(iq0,jq0);
1022             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1023                                          vdwparam+vdwioffset0+vdwjidx0B,
1024                                          vdwparam+vdwioffset0+vdwjidx0C,
1025                                          vdwparam+vdwioffset0+vdwjidx0D,
1026                                          &c6_00,&c12_00);
1027
1028             /* Calculate table index by multiplying r with table scale and truncate to integer */
1029             rt               = _mm_mul_ps(r00,vftabscale);
1030             vfitab           = _mm_cvttps_epi32(rt);
1031             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1032             vfitab           = _mm_slli_epi32(vfitab,2);
1033
1034             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1035             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1036             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1037             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1038             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1039             _MM_TRANSPOSE4_PS(Y,F,G,H);
1040             Heps             = _mm_mul_ps(vfeps,H);
1041             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1042             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1043             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1044
1045             /* LENNARD-JONES DISPERSION/REPULSION */
1046
1047             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1048             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1049
1050             fscal            = _mm_add_ps(felec,fvdw);
1051
1052             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm_mul_ps(fscal,dx00);
1056             ty               = _mm_mul_ps(fscal,dy00);
1057             tz               = _mm_mul_ps(fscal,dz00);
1058
1059             /* Update vectorial force */
1060             fix0             = _mm_add_ps(fix0,tx);
1061             fiy0             = _mm_add_ps(fiy0,ty);
1062             fiz0             = _mm_add_ps(fiz0,tz);
1063
1064             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1065                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1066                                                    tx,ty,tz);
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             r10              = _mm_mul_ps(rsq10,rinv10);
1073             r10              = _mm_andnot_ps(dummy_mask,r10);
1074
1075             /* Compute parameters for interactions between i and j atoms */
1076             qq10             = _mm_mul_ps(iq1,jq0);
1077
1078             /* Calculate table index by multiplying r with table scale and truncate to integer */
1079             rt               = _mm_mul_ps(r10,vftabscale);
1080             vfitab           = _mm_cvttps_epi32(rt);
1081             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1082             vfitab           = _mm_slli_epi32(vfitab,2);
1083
1084             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1085             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1086             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1087             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1088             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1089             _MM_TRANSPOSE4_PS(Y,F,G,H);
1090             Heps             = _mm_mul_ps(vfeps,H);
1091             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1092             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1093             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm_mul_ps(fscal,dx10);
1101             ty               = _mm_mul_ps(fscal,dy10);
1102             tz               = _mm_mul_ps(fscal,dz10);
1103
1104             /* Update vectorial force */
1105             fix1             = _mm_add_ps(fix1,tx);
1106             fiy1             = _mm_add_ps(fiy1,ty);
1107             fiz1             = _mm_add_ps(fiz1,tz);
1108
1109             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1110                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1111                                                    tx,ty,tz);
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             r20              = _mm_mul_ps(rsq20,rinv20);
1118             r20              = _mm_andnot_ps(dummy_mask,r20);
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq20             = _mm_mul_ps(iq2,jq0);
1122
1123             /* Calculate table index by multiplying r with table scale and truncate to integer */
1124             rt               = _mm_mul_ps(r20,vftabscale);
1125             vfitab           = _mm_cvttps_epi32(rt);
1126             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1127             vfitab           = _mm_slli_epi32(vfitab,2);
1128
1129             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1130             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1131             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1132             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1133             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1134             _MM_TRANSPOSE4_PS(Y,F,G,H);
1135             Heps             = _mm_mul_ps(vfeps,H);
1136             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1137             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1138             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1139
1140             fscal            = felec;
1141
1142             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1143
1144             /* Calculate temporary vectorial force */
1145             tx               = _mm_mul_ps(fscal,dx20);
1146             ty               = _mm_mul_ps(fscal,dy20);
1147             tz               = _mm_mul_ps(fscal,dz20);
1148
1149             /* Update vectorial force */
1150             fix2             = _mm_add_ps(fix2,tx);
1151             fiy2             = _mm_add_ps(fiy2,ty);
1152             fiz2             = _mm_add_ps(fiz2,tz);
1153
1154             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1155                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1156                                                    tx,ty,tz);
1157
1158             /* Inner loop uses 128 flops */
1159         }
1160
1161         /* End of innermost loop */
1162
1163         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1164                                               f+i_coord_offset,fshift+i_shift_offset);
1165
1166         /* Increment number of inner iterations */
1167         inneriter                  += j_index_end - j_index_start;
1168
1169         /* Outer loop uses 27 flops */
1170     }
1171
1172     /* Increment number of outer iterations */
1173     outeriter        += nri;
1174
1175     /* Update outer/inner flops */
1176
1177     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*128);
1178 }