src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_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_ElecRF_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 122     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 123     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141         shX              = shiftvec[i_shift_offset+XX];
 142         shY              = shiftvec[i_shift_offset+YY];
 143         shZ              = shiftvec[i_shift_offset+ZZ];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 155         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 156         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 157         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 158         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 159         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 160         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 161         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 162         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 163         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 164         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 165         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 166
 167         fix0             = _mm_setzero_ps();
 168         fiy0             = _mm_setzero_ps();
 169         fiz0             = _mm_setzero_ps();
 170         fix1             = _mm_setzero_ps();
 171         fiy1             = _mm_setzero_ps();
 172         fiz1             = _mm_setzero_ps();
 173         fix2             = _mm_setzero_ps();
 174         fiy2             = _mm_setzero_ps();
 175         fiz2             = _mm_setzero_ps();
 176         fix3             = _mm_setzero_ps();
 177         fiy3             = _mm_setzero_ps();
 178         fiz3             = _mm_setzero_ps();
 179
 180         /* Reset potential sums */
 181         velecsum         = _mm_setzero_ps();
 182         vvdwsum          = _mm_setzero_ps();
 183
 184         /* Start inner kernel loop */
 185         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 186         {
 187
 188             /* Get j neighbor index, and coordinate index */
 189             jnrA             = jjnr[jidx];
 190             jnrB             = jjnr[jidx+1];
 191             jnrC             = jjnr[jidx+2];
 192             jnrD             = jjnr[jidx+3];
 193
 194             j_coord_offsetA  = DIM*jnrA;
 195             j_coord_offsetB  = DIM*jnrB;
 196             j_coord_offsetC  = DIM*jnrC;
 197             j_coord_offsetD  = DIM*jnrD;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               x+j_coord_offsetC,x+j_coord_offsetD,
 202                                               &jx0,&jy0,&jz0);
 203
 204             /* Calculate displacement vector */
 205             dx00             = _mm_sub_ps(ix0,jx0);
 206             dy00             = _mm_sub_ps(iy0,jy0);
 207             dz00             = _mm_sub_ps(iz0,jz0);
 208             dx10             = _mm_sub_ps(ix1,jx0);
 209             dy10             = _mm_sub_ps(iy1,jy0);
 210             dz10             = _mm_sub_ps(iz1,jz0);
 211             dx20             = _mm_sub_ps(ix2,jx0);
 212             dy20             = _mm_sub_ps(iy2,jy0);
 213             dz20             = _mm_sub_ps(iz2,jz0);
 214             dx30             = _mm_sub_ps(ix3,jx0);
 215             dy30             = _mm_sub_ps(iy3,jy0);
 216             dz30             = _mm_sub_ps(iz3,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 222             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 223
 224             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 225             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 226             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 227             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 228
 229             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 230             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 231             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 232
 233             /* Load parameters for j particles */
 234             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 235                                                               charge+jnrC+0,charge+jnrD+0);
 236             vdwjidx0A        = 2*vdwtype[jnrA+0];
 237             vdwjidx0B        = 2*vdwtype[jnrB+0];
 238             vdwjidx0C        = 2*vdwtype[jnrC+0];
 239             vdwjidx0D        = 2*vdwtype[jnrD+0];
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             r00              = _mm_mul_ps(rsq00,rinv00);
 246
 247             /* Compute parameters for interactions between i and j atoms */
 248             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 249                                          vdwparam+vdwioffset0+vdwjidx0B,
 250                                          vdwparam+vdwioffset0+vdwjidx0C,
 251                                          vdwparam+vdwioffset0+vdwjidx0D,
 252                                          &c6_00,&c12_00);
 253
 254             /* Calculate table index by multiplying r with table scale and truncate to integer */
 255             rt               = _mm_mul_ps(r00,vftabscale);
 256             vfitab           = _mm_cvttps_epi32(rt);
 257             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 258             vfitab           = _mm_slli_epi32(vfitab,3);
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 263             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 264             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 265             _MM_TRANSPOSE4_PS(Y,F,G,H);
 266             Heps             = _mm_mul_ps(vfeps,H);
 267             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 268             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 269             vvdw6            = _mm_mul_ps(c6_00,VV);
 270             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 271             fvdw6            = _mm_mul_ps(c6_00,FF);
 272
 273             /* CUBIC SPLINE TABLE REPULSION */
 274             vfitab           = _mm_add_epi32(vfitab,ifour);
 275             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Heps             = _mm_mul_ps(vfeps,H);
 281             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 282             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 283             vvdw12           = _mm_mul_ps(c12_00,VV);
 284             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 285             fvdw12           = _mm_mul_ps(c12_00,FF);
 286             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 287             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 291
 292             fscal            = fvdw;
 293
 294             /* Calculate temporary vectorial force */
 295             tx               = _mm_mul_ps(fscal,dx00);
 296             ty               = _mm_mul_ps(fscal,dy00);
 297             tz               = _mm_mul_ps(fscal,dz00);
 298
 299             /* Update vectorial force */
 300             fix0             = _mm_add_ps(fix0,tx);
 301             fiy0             = _mm_add_ps(fiy0,ty);
 302             fiz0             = _mm_add_ps(fiz0,tz);
 303
 304             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 305                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 306                                                    tx,ty,tz);
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq10             = _mm_mul_ps(iq1,jq0);
 314
 315             /* REACTION-FIELD ELECTROSTATICS */
 316             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 317             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             /* Calculate temporary vectorial force */
 325             tx               = _mm_mul_ps(fscal,dx10);
 326             ty               = _mm_mul_ps(fscal,dy10);
 327             tz               = _mm_mul_ps(fscal,dz10);
 328
 329             /* Update vectorial force */
 330             fix1             = _mm_add_ps(fix1,tx);
 331             fiy1             = _mm_add_ps(fiy1,ty);
 332             fiz1             = _mm_add_ps(fiz1,tz);
 333
 334             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 335                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 336                                                    tx,ty,tz);
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             /* Compute parameters for interactions between i and j atoms */
 343             qq20             = _mm_mul_ps(iq2,jq0);
 344
 345             /* REACTION-FIELD ELECTROSTATICS */
 346             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 347             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 348
 349             /* Update potential sum for this i atom from the interaction with this j atom. */
 350             velecsum         = _mm_add_ps(velecsum,velec);
 351
 352             fscal            = felec;
 353
 354             /* Calculate temporary vectorial force */
 355             tx               = _mm_mul_ps(fscal,dx20);
 356             ty               = _mm_mul_ps(fscal,dy20);
 357             tz               = _mm_mul_ps(fscal,dz20);
 358
 359             /* Update vectorial force */
 360             fix2             = _mm_add_ps(fix2,tx);
 361             fiy2             = _mm_add_ps(fiy2,ty);
 362             fiz2             = _mm_add_ps(fiz2,tz);
 363
 364             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 365                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 366                                                    tx,ty,tz);
 367
 368             /**************************
 369              * CALCULATE INTERACTIONS *
 370              **************************/
 371
 372             /* Compute parameters for interactions between i and j atoms */
 373             qq30             = _mm_mul_ps(iq3,jq0);
 374
 375             /* REACTION-FIELD ELECTROSTATICS */
 376             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 377             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 378
 379             /* Update potential sum for this i atom from the interaction with this j atom. */
 380             velecsum         = _mm_add_ps(velecsum,velec);
 381
 382             fscal            = felec;
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm_mul_ps(fscal,dx30);
 386             ty               = _mm_mul_ps(fscal,dy30);
 387             tz               = _mm_mul_ps(fscal,dz30);
 388
 389             /* Update vectorial force */
 390             fix3             = _mm_add_ps(fix3,tx);
 391             fiy3             = _mm_add_ps(fiy3,ty);
 392             fiz3             = _mm_add_ps(fiz3,tz);
 393
 394             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 395                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 396                                                    tx,ty,tz);
 397
 398             /* Inner loop uses 152 flops */
 399         }
 400
 401         if(jidx<j_index_end)
 402         {
 403
 404             /* Get j neighbor index, and coordinate index */
 405             jnrA             = jjnr[jidx];
 406             jnrB             = jjnr[jidx+1];
 407             jnrC             = jjnr[jidx+2];
 408             jnrD             = jjnr[jidx+3];
 409
 410             /* Sign of each element will be negative for non-real atoms.
 411              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 412              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 413              */
 414             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 415             jnrA       = (jnrA>=0) ? jnrA : 0;
 416             jnrB       = (jnrB>=0) ? jnrB : 0;
 417             jnrC       = (jnrC>=0) ? jnrC : 0;
 418             jnrD       = (jnrD>=0) ? jnrD : 0;
 419
 420             j_coord_offsetA  = DIM*jnrA;
 421             j_coord_offsetB  = DIM*jnrB;
 422             j_coord_offsetC  = DIM*jnrC;
 423             j_coord_offsetD  = DIM*jnrD;
 424
 425             /* load j atom coordinates */
 426             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 427                                               x+j_coord_offsetC,x+j_coord_offsetD,
 428                                               &jx0,&jy0,&jz0);
 429
 430             /* Calculate displacement vector */
 431             dx00             = _mm_sub_ps(ix0,jx0);
 432             dy00             = _mm_sub_ps(iy0,jy0);
 433             dz00             = _mm_sub_ps(iz0,jz0);
 434             dx10             = _mm_sub_ps(ix1,jx0);
 435             dy10             = _mm_sub_ps(iy1,jy0);
 436             dz10             = _mm_sub_ps(iz1,jz0);
 437             dx20             = _mm_sub_ps(ix2,jx0);
 438             dy20             = _mm_sub_ps(iy2,jy0);
 439             dz20             = _mm_sub_ps(iz2,jz0);
 440             dx30             = _mm_sub_ps(ix3,jx0);
 441             dy30             = _mm_sub_ps(iy3,jy0);
 442             dz30             = _mm_sub_ps(iz3,jz0);
 443
 444             /* Calculate squared distance and things based on it */
 445             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 446             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 447             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 448             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 449
 450             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 451             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 452             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 453             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 454
 455             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 456             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 457             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 458
 459             /* Load parameters for j particles */
 460             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 461                                                               charge+jnrC+0,charge+jnrD+0);
 462             vdwjidx0A        = 2*vdwtype[jnrA+0];
 463             vdwjidx0B        = 2*vdwtype[jnrB+0];
 464             vdwjidx0C        = 2*vdwtype[jnrC+0];
 465             vdwjidx0D        = 2*vdwtype[jnrD+0];
 466
 467             /**************************
 468              * CALCULATE INTERACTIONS *
 469              **************************/
 470
 471             r00              = _mm_mul_ps(rsq00,rinv00);
 472             r00              = _mm_andnot_ps(dummy_mask,r00);
 473
 474             /* Compute parameters for interactions between i and j atoms */
 475             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 476                                          vdwparam+vdwioffset0+vdwjidx0B,
 477                                          vdwparam+vdwioffset0+vdwjidx0C,
 478                                          vdwparam+vdwioffset0+vdwjidx0D,
 479                                          &c6_00,&c12_00);
 480
 481             /* Calculate table index by multiplying r with table scale and truncate to integer */
 482             rt               = _mm_mul_ps(r00,vftabscale);
 483             vfitab           = _mm_cvttps_epi32(rt);
 484             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 485             vfitab           = _mm_slli_epi32(vfitab,3);
 486
 487             /* CUBIC SPLINE TABLE DISPERSION */
 488             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 489             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 490             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 491             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 492             _MM_TRANSPOSE4_PS(Y,F,G,H);
 493             Heps             = _mm_mul_ps(vfeps,H);
 494             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 495             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 496             vvdw6            = _mm_mul_ps(c6_00,VV);
 497             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 498             fvdw6            = _mm_mul_ps(c6_00,FF);
 499
 500             /* CUBIC SPLINE TABLE REPULSION */
 501             vfitab           = _mm_add_epi32(vfitab,ifour);
 502             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 503             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 504             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 505             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 506             _MM_TRANSPOSE4_PS(Y,F,G,H);
 507             Heps             = _mm_mul_ps(vfeps,H);
 508             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 509             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 510             vvdw12           = _mm_mul_ps(c12_00,VV);
 511             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 512             fvdw12           = _mm_mul_ps(c12_00,FF);
 513             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 514             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 518             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 519
 520             fscal            = fvdw;
 521
 522             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 523
 524             /* Calculate temporary vectorial force */
 525             tx               = _mm_mul_ps(fscal,dx00);
 526             ty               = _mm_mul_ps(fscal,dy00);
 527             tz               = _mm_mul_ps(fscal,dz00);
 528
 529             /* Update vectorial force */
 530             fix0             = _mm_add_ps(fix0,tx);
 531             fiy0             = _mm_add_ps(fiy0,ty);
 532             fiz0             = _mm_add_ps(fiz0,tz);
 533
 534             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 535                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 536                                                    tx,ty,tz);
 537
 538             /**************************
 539              * CALCULATE INTERACTIONS *
 540              **************************/
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq10             = _mm_mul_ps(iq1,jq0);
 544
 545             /* REACTION-FIELD ELECTROSTATICS */
 546             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 547             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_andnot_ps(dummy_mask,velec);
 551             velecsum         = _mm_add_ps(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_ps(fscal,dx10);
 559             ty               = _mm_mul_ps(fscal,dy10);
 560             tz               = _mm_mul_ps(fscal,dz10);
 561
 562             /* Update vectorial force */
 563             fix1             = _mm_add_ps(fix1,tx);
 564             fiy1             = _mm_add_ps(fiy1,ty);
 565             fiz1             = _mm_add_ps(fiz1,tz);
 566
 567             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 568                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 569                                                    tx,ty,tz);
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq20             = _mm_mul_ps(iq2,jq0);
 577
 578             /* REACTION-FIELD ELECTROSTATICS */
 579             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 580             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_andnot_ps(dummy_mask,velec);
 584             velecsum         = _mm_add_ps(velecsum,velec);
 585
 586             fscal            = felec;
 587
 588             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 589
 590             /* Calculate temporary vectorial force */
 591             tx               = _mm_mul_ps(fscal,dx20);
 592             ty               = _mm_mul_ps(fscal,dy20);
 593             tz               = _mm_mul_ps(fscal,dz20);
 594
 595             /* Update vectorial force */
 596             fix2             = _mm_add_ps(fix2,tx);
 597             fiy2             = _mm_add_ps(fiy2,ty);
 598             fiz2             = _mm_add_ps(fiz2,tz);
 599
 600             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 601                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 602                                                    tx,ty,tz);
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq30             = _mm_mul_ps(iq3,jq0);
 610
 611             /* REACTION-FIELD ELECTROSTATICS */
 612             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 613             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 614
 615             /* Update potential sum for this i atom from the interaction with this j atom. */
 616             velec            = _mm_andnot_ps(dummy_mask,velec);
 617             velecsum         = _mm_add_ps(velecsum,velec);
 618
 619             fscal            = felec;
 620
 621             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_ps(fscal,dx30);
 625             ty               = _mm_mul_ps(fscal,dy30);
 626             tz               = _mm_mul_ps(fscal,dz30);
 627
 628             /* Update vectorial force */
 629             fix3             = _mm_add_ps(fix3,tx);
 630             fiy3             = _mm_add_ps(fiy3,ty);
 631             fiz3             = _mm_add_ps(fiz3,tz);
 632
 633             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 634                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 635                                                    tx,ty,tz);
 636
 637             /* Inner loop uses 153 flops */
 638         }
 639
 640         /* End of innermost loop */
 641
 642         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 643                                               f+i_coord_offset,fshift+i_shift_offset);
 644
 645         ggid                        = gid[iidx];
 646         /* Update potential energies */
 647         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 648         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 649
 650         /* Increment number of inner iterations */
 651         inneriter                  += j_index_end - j_index_start;
 652
 653         /* Outer loop uses 38 flops */
 654     }
 655
 656     /* Increment number of outer iterations */
 657     outeriter        += nri;
 658
 659     /* Update outer/inner flops */
 660
 661     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*153);
 662 }
 663 /*
 664  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 665  * Electrostatics interaction: ReactionField
 666  * VdW interaction:            CubicSplineTable
 667  * Geometry:                   Water4-Particle
 668  * Calculate force/pot:        Force
 669  */
 670 void
 671 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_sse2_single
 672                     (t_nblist * gmx_restrict                nlist,
 673                      rvec * gmx_restrict                    xx,
 674                      rvec * gmx_restrict                    ff,
 675                      t_forcerec * gmx_restrict              fr,
 676                      t_mdatoms * gmx_restrict               mdatoms,
 677                      nb_kernel_data_t * gmx_restrict        kernel_data,
 678                      t_nrnb * gmx_restrict                  nrnb)
 679 {
 680     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 681      * just 0 for non-waters.
 682      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 683      * jnr indices corresponding to data put in the four positions in the SIMD register.
 684      */
 685     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 686     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 687     int              jnrA,jnrB,jnrC,jnrD;
 688     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 689     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 690     real             shX,shY,shZ,rcutoff_scalar;
 691     real             *shiftvec,*fshift,*x,*f;
 692     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 693     int              vdwioffset0;
 694     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 695     int              vdwioffset1;
 696     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 697     int              vdwioffset2;
 698     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 699     int              vdwioffset3;
 700     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 701     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 702     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 703     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 704     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 705     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 706     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 707     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 708     real             *charge;
 709     int              nvdwtype;
 710     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 711     int              *vdwtype;
 712     real             *vdwparam;
 713     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 714     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 715     __m128i          vfitab;
 716     __m128i          ifour       = _mm_set1_epi32(4);
 717     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 718     real             *vftab;
 719     __m128           dummy_mask,cutoff_mask;
 720     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 721     __m128           one     = _mm_set1_ps(1.0);
 722     __m128           two     = _mm_set1_ps(2.0);
 723     x                = xx[0];
 724     f                = ff[0];
 725
 726     nri              = nlist->nri;
 727     iinr             = nlist->iinr;
 728     jindex           = nlist->jindex;
 729     jjnr             = nlist->jjnr;
 730     shiftidx         = nlist->shift;
 731     gid              = nlist->gid;
 732     shiftvec         = fr->shift_vec[0];
 733     fshift           = fr->fshift[0];
 734     facel            = _mm_set1_ps(fr->epsfac);
 735     charge           = mdatoms->chargeA;
 736     krf              = _mm_set1_ps(fr->ic->k_rf);
 737     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 738     crf              = _mm_set1_ps(fr->ic->c_rf);
 739     nvdwtype         = fr->ntype;
 740     vdwparam         = fr->nbfp;
 741     vdwtype          = mdatoms->typeA;
 742
 743     vftab            = kernel_data->table_vdw->data;
 744     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 745
 746     /* Setup water-specific parameters */
 747     inr              = nlist->iinr[0];
 748     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 749     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 750     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 751     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 752
 753     /* Avoid stupid compiler warnings */
 754     jnrA = jnrB = jnrC = jnrD = 0;
 755     j_coord_offsetA = 0;
 756     j_coord_offsetB = 0;
 757     j_coord_offsetC = 0;
 758     j_coord_offsetD = 0;
 759
 760     outeriter        = 0;
 761     inneriter        = 0;
 762
 763     /* Start outer loop over neighborlists */
 764     for(iidx=0; iidx<nri; iidx++)
 765     {
 766         /* Load shift vector for this list */
 767         i_shift_offset   = DIM*shiftidx[iidx];
 768         shX              = shiftvec[i_shift_offset+XX];
 769         shY              = shiftvec[i_shift_offset+YY];
 770         shZ              = shiftvec[i_shift_offset+ZZ];
 771
 772         /* Load limits for loop over neighbors */
 773         j_index_start    = jindex[iidx];
 774         j_index_end      = jindex[iidx+1];
 775
 776         /* Get outer coordinate index */
 777         inr              = iinr[iidx];
 778         i_coord_offset   = DIM*inr;
 779
 780         /* Load i particle coords and add shift vector */
 781         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 782         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 783         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 784         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 785         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 786         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 787         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 788         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 789         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 790         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 791         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 792         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 793
 794         fix0             = _mm_setzero_ps();
 795         fiy0             = _mm_setzero_ps();
 796         fiz0             = _mm_setzero_ps();
 797         fix1             = _mm_setzero_ps();
 798         fiy1             = _mm_setzero_ps();
 799         fiz1             = _mm_setzero_ps();
 800         fix2             = _mm_setzero_ps();
 801         fiy2             = _mm_setzero_ps();
 802         fiz2             = _mm_setzero_ps();
 803         fix3             = _mm_setzero_ps();
 804         fiy3             = _mm_setzero_ps();
 805         fiz3             = _mm_setzero_ps();
 806
 807         /* Start inner kernel loop */
 808         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 809         {
 810
 811             /* Get j neighbor index, and coordinate index */
 812             jnrA             = jjnr[jidx];
 813             jnrB             = jjnr[jidx+1];
 814             jnrC             = jjnr[jidx+2];
 815             jnrD             = jjnr[jidx+3];
 816
 817             j_coord_offsetA  = DIM*jnrA;
 818             j_coord_offsetB  = DIM*jnrB;
 819             j_coord_offsetC  = DIM*jnrC;
 820             j_coord_offsetD  = DIM*jnrD;
 821
 822             /* load j atom coordinates */
 823             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 824                                               x+j_coord_offsetC,x+j_coord_offsetD,
 825                                               &jx0,&jy0,&jz0);
 826
 827             /* Calculate displacement vector */
 828             dx00             = _mm_sub_ps(ix0,jx0);
 829             dy00             = _mm_sub_ps(iy0,jy0);
 830             dz00             = _mm_sub_ps(iz0,jz0);
 831             dx10             = _mm_sub_ps(ix1,jx0);
 832             dy10             = _mm_sub_ps(iy1,jy0);
 833             dz10             = _mm_sub_ps(iz1,jz0);
 834             dx20             = _mm_sub_ps(ix2,jx0);
 835             dy20             = _mm_sub_ps(iy2,jy0);
 836             dz20             = _mm_sub_ps(iz2,jz0);
 837             dx30             = _mm_sub_ps(ix3,jx0);
 838             dy30             = _mm_sub_ps(iy3,jy0);
 839             dz30             = _mm_sub_ps(iz3,jz0);
 840
 841             /* Calculate squared distance and things based on it */
 842             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 843             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 844             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 845             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 846
 847             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 848             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 849             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 850             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 851
 852             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 853             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 854             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 855
 856             /* Load parameters for j particles */
 857             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 858                                                               charge+jnrC+0,charge+jnrD+0);
 859             vdwjidx0A        = 2*vdwtype[jnrA+0];
 860             vdwjidx0B        = 2*vdwtype[jnrB+0];
 861             vdwjidx0C        = 2*vdwtype[jnrC+0];
 862             vdwjidx0D        = 2*vdwtype[jnrD+0];
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             r00              = _mm_mul_ps(rsq00,rinv00);
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 872                                          vdwparam+vdwioffset0+vdwjidx0B,
 873                                          vdwparam+vdwioffset0+vdwjidx0C,
 874                                          vdwparam+vdwioffset0+vdwjidx0D,
 875                                          &c6_00,&c12_00);
 876
 877             /* Calculate table index by multiplying r with table scale and truncate to integer */
 878             rt               = _mm_mul_ps(r00,vftabscale);
 879             vfitab           = _mm_cvttps_epi32(rt);
 880             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 881             vfitab           = _mm_slli_epi32(vfitab,3);
 882
 883             /* CUBIC SPLINE TABLE DISPERSION */
 884             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 885             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 886             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 887             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 888             _MM_TRANSPOSE4_PS(Y,F,G,H);
 889             Heps             = _mm_mul_ps(vfeps,H);
 890             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 891             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 892             fvdw6            = _mm_mul_ps(c6_00,FF);
 893
 894             /* CUBIC SPLINE TABLE REPULSION */
 895             vfitab           = _mm_add_epi32(vfitab,ifour);
 896             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 897             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 898             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 899             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 900             _MM_TRANSPOSE4_PS(Y,F,G,H);
 901             Heps             = _mm_mul_ps(vfeps,H);
 902             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 903             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 904             fvdw12           = _mm_mul_ps(c12_00,FF);
 905             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 906
 907             fscal            = fvdw;
 908
 909             /* Calculate temporary vectorial force */
 910             tx               = _mm_mul_ps(fscal,dx00);
 911             ty               = _mm_mul_ps(fscal,dy00);
 912             tz               = _mm_mul_ps(fscal,dz00);
 913
 914             /* Update vectorial force */
 915             fix0             = _mm_add_ps(fix0,tx);
 916             fiy0             = _mm_add_ps(fiy0,ty);
 917             fiz0             = _mm_add_ps(fiz0,tz);
 918
 919             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 920                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 921                                                    tx,ty,tz);
 922
 923             /**************************
 924              * CALCULATE INTERACTIONS *
 925              **************************/
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq10             = _mm_mul_ps(iq1,jq0);
 929
 930             /* REACTION-FIELD ELECTROSTATICS */
 931             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 932
 933             fscal            = felec;
 934
 935             /* Calculate temporary vectorial force */
 936             tx               = _mm_mul_ps(fscal,dx10);
 937             ty               = _mm_mul_ps(fscal,dy10);
 938             tz               = _mm_mul_ps(fscal,dz10);
 939
 940             /* Update vectorial force */
 941             fix1             = _mm_add_ps(fix1,tx);
 942             fiy1             = _mm_add_ps(fiy1,ty);
 943             fiz1             = _mm_add_ps(fiz1,tz);
 944
 945             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 946                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 947                                                    tx,ty,tz);
 948
 949             /**************************
 950              * CALCULATE INTERACTIONS *
 951              **************************/
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq20             = _mm_mul_ps(iq2,jq0);
 955
 956             /* REACTION-FIELD ELECTROSTATICS */
 957             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 958
 959             fscal            = felec;
 960
 961             /* Calculate temporary vectorial force */
 962             tx               = _mm_mul_ps(fscal,dx20);
 963             ty               = _mm_mul_ps(fscal,dy20);
 964             tz               = _mm_mul_ps(fscal,dz20);
 965
 966             /* Update vectorial force */
 967             fix2             = _mm_add_ps(fix2,tx);
 968             fiy2             = _mm_add_ps(fiy2,ty);
 969             fiz2             = _mm_add_ps(fiz2,tz);
 970
 971             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 972                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 973                                                    tx,ty,tz);
 974
 975             /**************************
 976              * CALCULATE INTERACTIONS *
 977              **************************/
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             qq30             = _mm_mul_ps(iq3,jq0);
 981
 982             /* REACTION-FIELD ELECTROSTATICS */
 983             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 984
 985             fscal            = felec;
 986
 987             /* Calculate temporary vectorial force */
 988             tx               = _mm_mul_ps(fscal,dx30);
 989             ty               = _mm_mul_ps(fscal,dy30);
 990             tz               = _mm_mul_ps(fscal,dz30);
 991
 992             /* Update vectorial force */
 993             fix3             = _mm_add_ps(fix3,tx);
 994             fiy3             = _mm_add_ps(fiy3,ty);
 995             fiz3             = _mm_add_ps(fiz3,tz);
 996
 997             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 998                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 999                                                    tx,ty,tz);
1000
1001             /* Inner loop uses 129 flops */
1002         }
1003
1004         if(jidx<j_index_end)
1005         {
1006
1007             /* Get j neighbor index, and coordinate index */
1008             jnrA             = jjnr[jidx];
1009             jnrB             = jjnr[jidx+1];
1010             jnrC             = jjnr[jidx+2];
1011             jnrD             = jjnr[jidx+3];
1012
1013             /* Sign of each element will be negative for non-real atoms.
1014              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1015              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1016              */
1017             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1018             jnrA       = (jnrA>=0) ? jnrA : 0;
1019             jnrB       = (jnrB>=0) ? jnrB : 0;
1020             jnrC       = (jnrC>=0) ? jnrC : 0;
1021             jnrD       = (jnrD>=0) ? jnrD : 0;
1022
1023             j_coord_offsetA  = DIM*jnrA;
1024             j_coord_offsetB  = DIM*jnrB;
1025             j_coord_offsetC  = DIM*jnrC;
1026             j_coord_offsetD  = DIM*jnrD;
1027
1028             /* load j atom coordinates */
1029             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1030                                               x+j_coord_offsetC,x+j_coord_offsetD,
1031                                               &jx0,&jy0,&jz0);
1032
1033             /* Calculate displacement vector */
1034             dx00             = _mm_sub_ps(ix0,jx0);
1035             dy00             = _mm_sub_ps(iy0,jy0);
1036             dz00             = _mm_sub_ps(iz0,jz0);
1037             dx10             = _mm_sub_ps(ix1,jx0);
1038             dy10             = _mm_sub_ps(iy1,jy0);
1039             dz10             = _mm_sub_ps(iz1,jz0);
1040             dx20             = _mm_sub_ps(ix2,jx0);
1041             dy20             = _mm_sub_ps(iy2,jy0);
1042             dz20             = _mm_sub_ps(iz2,jz0);
1043             dx30             = _mm_sub_ps(ix3,jx0);
1044             dy30             = _mm_sub_ps(iy3,jy0);
1045             dz30             = _mm_sub_ps(iz3,jz0);
1046
1047             /* Calculate squared distance and things based on it */
1048             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1049             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1050             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1051             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1052
1053             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1054             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1055             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1056             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1057
1058             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1059             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1060             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1061
1062             /* Load parameters for j particles */
1063             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1064                                                               charge+jnrC+0,charge+jnrD+0);
1065             vdwjidx0A        = 2*vdwtype[jnrA+0];
1066             vdwjidx0B        = 2*vdwtype[jnrB+0];
1067             vdwjidx0C        = 2*vdwtype[jnrC+0];
1068             vdwjidx0D        = 2*vdwtype[jnrD+0];
1069
1070             /**************************
1071              * CALCULATE INTERACTIONS *
1072              **************************/
1073
1074             r00              = _mm_mul_ps(rsq00,rinv00);
1075             r00              = _mm_andnot_ps(dummy_mask,r00);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1079                                          vdwparam+vdwioffset0+vdwjidx0B,
1080                                          vdwparam+vdwioffset0+vdwjidx0C,
1081                                          vdwparam+vdwioffset0+vdwjidx0D,
1082                                          &c6_00,&c12_00);
1083
1084             /* Calculate table index by multiplying r with table scale and truncate to integer */
1085             rt               = _mm_mul_ps(r00,vftabscale);
1086             vfitab           = _mm_cvttps_epi32(rt);
1087             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1088             vfitab           = _mm_slli_epi32(vfitab,3);
1089
1090             /* CUBIC SPLINE TABLE DISPERSION */
1091             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1092             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1093             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1094             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1095             _MM_TRANSPOSE4_PS(Y,F,G,H);
1096             Heps             = _mm_mul_ps(vfeps,H);
1097             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1098             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1099             fvdw6            = _mm_mul_ps(c6_00,FF);
1100
1101             /* CUBIC SPLINE TABLE REPULSION */
1102             vfitab           = _mm_add_epi32(vfitab,ifour);
1103             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1104             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1105             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1106             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1107             _MM_TRANSPOSE4_PS(Y,F,G,H);
1108             Heps             = _mm_mul_ps(vfeps,H);
1109             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1110             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1111             fvdw12           = _mm_mul_ps(c12_00,FF);
1112             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1113
1114             fscal            = fvdw;
1115
1116             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1117
1118             /* Calculate temporary vectorial force */
1119             tx               = _mm_mul_ps(fscal,dx00);
1120             ty               = _mm_mul_ps(fscal,dy00);
1121             tz               = _mm_mul_ps(fscal,dz00);
1122
1123             /* Update vectorial force */
1124             fix0             = _mm_add_ps(fix0,tx);
1125             fiy0             = _mm_add_ps(fiy0,ty);
1126             fiz0             = _mm_add_ps(fiz0,tz);
1127
1128             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1129                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1130                                                    tx,ty,tz);
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             /* Compute parameters for interactions between i and j atoms */
1137             qq10             = _mm_mul_ps(iq1,jq0);
1138
1139             /* REACTION-FIELD ELECTROSTATICS */
1140             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1141
1142             fscal            = felec;
1143
1144             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1145
1146             /* Calculate temporary vectorial force */
1147             tx               = _mm_mul_ps(fscal,dx10);
1148             ty               = _mm_mul_ps(fscal,dy10);
1149             tz               = _mm_mul_ps(fscal,dz10);
1150
1151             /* Update vectorial force */
1152             fix1             = _mm_add_ps(fix1,tx);
1153             fiy1             = _mm_add_ps(fiy1,ty);
1154             fiz1             = _mm_add_ps(fiz1,tz);
1155
1156             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1157                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1158                                                    tx,ty,tz);
1159
1160             /**************************
1161              * CALCULATE INTERACTIONS *
1162              **************************/
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq20             = _mm_mul_ps(iq2,jq0);
1166
1167             /* REACTION-FIELD ELECTROSTATICS */
1168             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174             /* Calculate temporary vectorial force */
1175             tx               = _mm_mul_ps(fscal,dx20);
1176             ty               = _mm_mul_ps(fscal,dy20);
1177             tz               = _mm_mul_ps(fscal,dz20);
1178
1179             /* Update vectorial force */
1180             fix2             = _mm_add_ps(fix2,tx);
1181             fiy2             = _mm_add_ps(fiy2,ty);
1182             fiz2             = _mm_add_ps(fiz2,tz);
1183
1184             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1185                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1186                                                    tx,ty,tz);
1187
1188             /**************************
1189              * CALCULATE INTERACTIONS *
1190              **************************/
1191
1192             /* Compute parameters for interactions between i and j atoms */
1193             qq30             = _mm_mul_ps(iq3,jq0);
1194
1195             /* REACTION-FIELD ELECTROSTATICS */
1196             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
1197
1198             fscal            = felec;
1199
1200             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1201
1202             /* Calculate temporary vectorial force */
1203             tx               = _mm_mul_ps(fscal,dx30);
1204             ty               = _mm_mul_ps(fscal,dy30);
1205             tz               = _mm_mul_ps(fscal,dz30);
1206
1207             /* Update vectorial force */
1208             fix3             = _mm_add_ps(fix3,tx);
1209             fiy3             = _mm_add_ps(fiy3,ty);
1210             fiz3             = _mm_add_ps(fiz3,tz);
1211
1212             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1213                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1214                                                    tx,ty,tz);
1215
1216             /* Inner loop uses 130 flops */
1217         }
1218
1219         /* End of innermost loop */
1220
1221         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1222                                               f+i_coord_offset,fshift+i_shift_offset);
1223
1224         /* Increment number of inner iterations */
1225         inneriter                  += j_index_end - j_index_start;
1226
1227         /* Outer loop uses 36 flops */
1228     }
1229
1230     /* Increment number of outer iterations */
1231     outeriter        += nri;
1232
1233     /* Update outer/inner flops */
1234
1235     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*130);
1236 }