src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  88     real             *vftab;
  89     __m128           dummy_mask,cutoff_mask;
  90     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  91     __m128           one     = _mm_set1_ps(1.0);
  92     __m128           two     = _mm_set1_ps(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_ps(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_ps(fr->ic->k_rf);
 107     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_ps(fr->ic->c_rf);
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 126     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143         shX              = shiftvec[i_shift_offset+XX];
 144         shY              = shiftvec[i_shift_offset+YY];
 145         shZ              = shiftvec[i_shift_offset+ZZ];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 157         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 158         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 159         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 160         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 161         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 162         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 163         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 164         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 165
 166         fix0             = _mm_setzero_ps();
 167         fiy0             = _mm_setzero_ps();
 168         fiz0             = _mm_setzero_ps();
 169         fix1             = _mm_setzero_ps();
 170         fiy1             = _mm_setzero_ps();
 171         fiz1             = _mm_setzero_ps();
 172         fix2             = _mm_setzero_ps();
 173         fiy2             = _mm_setzero_ps();
 174         fiz2             = _mm_setzero_ps();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_ps();
 178         vvdwsum          = _mm_setzero_ps();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             jnrC             = jjnr[jidx+2];
 188             jnrD             = jjnr[jidx+3];
 189
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192             j_coord_offsetC  = DIM*jnrC;
 193             j_coord_offsetD  = DIM*jnrD;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               x+j_coord_offsetC,x+j_coord_offsetD,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_ps(ix0,jx0);
 202             dy00             = _mm_sub_ps(iy0,jy0);
 203             dz00             = _mm_sub_ps(iz0,jz0);
 204             dx10             = _mm_sub_ps(ix1,jx0);
 205             dy10             = _mm_sub_ps(iy1,jy0);
 206             dz10             = _mm_sub_ps(iz1,jz0);
 207             dx20             = _mm_sub_ps(ix2,jx0);
 208             dy20             = _mm_sub_ps(iy2,jy0);
 209             dz20             = _mm_sub_ps(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 217             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 218             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 219
 220             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 226                                                               charge+jnrC+0,charge+jnrD+0);
 227             vdwjidx0A        = 2*vdwtype[jnrA+0];
 228             vdwjidx0B        = 2*vdwtype[jnrB+0];
 229             vdwjidx0C        = 2*vdwtype[jnrC+0];
 230             vdwjidx0D        = 2*vdwtype[jnrD+0];
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             if (gmx_mm_any_lt(rsq00,rcutoff2))
 237             {
 238
 239             r00              = _mm_mul_ps(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_ps(iq0,jq0);
 243             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 244                                          vdwparam+vdwioffset0+vdwjidx0B,
 245                                          vdwparam+vdwioffset0+vdwjidx0C,
 246                                          vdwparam+vdwioffset0+vdwjidx0D,
 247                                          &c6_00,&c12_00);
 248
 249             /* Calculate table index by multiplying r with table scale and truncate to integer */
 250             rt               = _mm_mul_ps(r00,vftabscale);
 251             vfitab           = _mm_cvttps_epi32(rt);
 252             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 253             vfitab           = _mm_slli_epi32(vfitab,3);
 254
 255             /* REACTION-FIELD ELECTROSTATICS */
 256             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 257             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 258
 259             /* CUBIC SPLINE TABLE DISPERSION */
 260             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 261             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 262             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 263             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 264             _MM_TRANSPOSE4_PS(Y,F,G,H);
 265             Heps             = _mm_mul_ps(vfeps,H);
 266             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 267             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 268             vvdw6            = _mm_mul_ps(c6_00,VV);
 269             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 270             fvdw6            = _mm_mul_ps(c6_00,FF);
 271
 272             /* CUBIC SPLINE TABLE REPULSION */
 273             vfitab           = _mm_add_epi32(vfitab,ifour);
 274             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 275             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 276             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 277             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 278             _MM_TRANSPOSE4_PS(Y,F,G,H);
 279             Heps             = _mm_mul_ps(vfeps,H);
 280             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 281             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 282             vvdw12           = _mm_mul_ps(c12_00,VV);
 283             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 284             fvdw12           = _mm_mul_ps(c12_00,FF);
 285             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 286             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 287
 288             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 289
 290             /* Update potential sum for this i atom from the interaction with this j atom. */
 291             velec            = _mm_and_ps(velec,cutoff_mask);
 292             velecsum         = _mm_add_ps(velecsum,velec);
 293             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 294             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 295
 296             fscal            = _mm_add_ps(felec,fvdw);
 297
 298             fscal            = _mm_and_ps(fscal,cutoff_mask);
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_ps(fscal,dx00);
 302             ty               = _mm_mul_ps(fscal,dy00);
 303             tz               = _mm_mul_ps(fscal,dz00);
 304
 305             /* Update vectorial force */
 306             fix0             = _mm_add_ps(fix0,tx);
 307             fiy0             = _mm_add_ps(fiy0,ty);
 308             fiz0             = _mm_add_ps(fiz0,tz);
 309
 310             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 311                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 312                                                    tx,ty,tz);
 313
 314             }
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             if (gmx_mm_any_lt(rsq10,rcutoff2))
 321             {
 322
 323             /* Compute parameters for interactions between i and j atoms */
 324             qq10             = _mm_mul_ps(iq1,jq0);
 325
 326             /* REACTION-FIELD ELECTROSTATICS */
 327             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 328             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 329
 330             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_and_ps(velec,cutoff_mask);
 334             velecsum         = _mm_add_ps(velecsum,velec);
 335
 336             fscal            = felec;
 337
 338             fscal            = _mm_and_ps(fscal,cutoff_mask);
 339
 340             /* Calculate temporary vectorial force */
 341             tx               = _mm_mul_ps(fscal,dx10);
 342             ty               = _mm_mul_ps(fscal,dy10);
 343             tz               = _mm_mul_ps(fscal,dz10);
 344
 345             /* Update vectorial force */
 346             fix1             = _mm_add_ps(fix1,tx);
 347             fiy1             = _mm_add_ps(fiy1,ty);
 348             fiz1             = _mm_add_ps(fiz1,tz);
 349
 350             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 351                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 352                                                    tx,ty,tz);
 353
 354             }
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             if (gmx_mm_any_lt(rsq20,rcutoff2))
 361             {
 362
 363             /* Compute parameters for interactions between i and j atoms */
 364             qq20             = _mm_mul_ps(iq2,jq0);
 365
 366             /* REACTION-FIELD ELECTROSTATICS */
 367             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 368             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 369
 370             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velec            = _mm_and_ps(velec,cutoff_mask);
 374             velecsum         = _mm_add_ps(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             fscal            = _mm_and_ps(fscal,cutoff_mask);
 379
 380             /* Calculate temporary vectorial force */
 381             tx               = _mm_mul_ps(fscal,dx20);
 382             ty               = _mm_mul_ps(fscal,dy20);
 383             tz               = _mm_mul_ps(fscal,dz20);
 384
 385             /* Update vectorial force */
 386             fix2             = _mm_add_ps(fix2,tx);
 387             fiy2             = _mm_add_ps(fiy2,ty);
 388             fiz2             = _mm_add_ps(fiz2,tz);
 389
 390             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 391                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 392                                                    tx,ty,tz);
 393
 394             }
 395
 396             /* Inner loop uses 144 flops */
 397         }
 398
 399         if(jidx<j_index_end)
 400         {
 401
 402             /* Get j neighbor index, and coordinate index */
 403             jnrA             = jjnr[jidx];
 404             jnrB             = jjnr[jidx+1];
 405             jnrC             = jjnr[jidx+2];
 406             jnrD             = jjnr[jidx+3];
 407
 408             /* Sign of each element will be negative for non-real atoms.
 409              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 410              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 411              */
 412             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 413             jnrA       = (jnrA>=0) ? jnrA : 0;
 414             jnrB       = (jnrB>=0) ? jnrB : 0;
 415             jnrC       = (jnrC>=0) ? jnrC : 0;
 416             jnrD       = (jnrD>=0) ? jnrD : 0;
 417
 418             j_coord_offsetA  = DIM*jnrA;
 419             j_coord_offsetB  = DIM*jnrB;
 420             j_coord_offsetC  = DIM*jnrC;
 421             j_coord_offsetD  = DIM*jnrD;
 422
 423             /* load j atom coordinates */
 424             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 425                                               x+j_coord_offsetC,x+j_coord_offsetD,
 426                                               &jx0,&jy0,&jz0);
 427
 428             /* Calculate displacement vector */
 429             dx00             = _mm_sub_ps(ix0,jx0);
 430             dy00             = _mm_sub_ps(iy0,jy0);
 431             dz00             = _mm_sub_ps(iz0,jz0);
 432             dx10             = _mm_sub_ps(ix1,jx0);
 433             dy10             = _mm_sub_ps(iy1,jy0);
 434             dz10             = _mm_sub_ps(iz1,jz0);
 435             dx20             = _mm_sub_ps(ix2,jx0);
 436             dy20             = _mm_sub_ps(iy2,jy0);
 437             dz20             = _mm_sub_ps(iz2,jz0);
 438
 439             /* Calculate squared distance and things based on it */
 440             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 441             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 442             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 443
 444             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 445             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 446             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 447
 448             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 449             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 450             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 451
 452             /* Load parameters for j particles */
 453             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 454                                                               charge+jnrC+0,charge+jnrD+0);
 455             vdwjidx0A        = 2*vdwtype[jnrA+0];
 456             vdwjidx0B        = 2*vdwtype[jnrB+0];
 457             vdwjidx0C        = 2*vdwtype[jnrC+0];
 458             vdwjidx0D        = 2*vdwtype[jnrD+0];
 459
 460             /**************************
 461              * CALCULATE INTERACTIONS *
 462              **************************/
 463
 464             if (gmx_mm_any_lt(rsq00,rcutoff2))
 465             {
 466
 467             r00              = _mm_mul_ps(rsq00,rinv00);
 468             r00              = _mm_andnot_ps(dummy_mask,r00);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq00             = _mm_mul_ps(iq0,jq0);
 472             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 473                                          vdwparam+vdwioffset0+vdwjidx0B,
 474                                          vdwparam+vdwioffset0+vdwjidx0C,
 475                                          vdwparam+vdwioffset0+vdwjidx0D,
 476                                          &c6_00,&c12_00);
 477
 478             /* Calculate table index by multiplying r with table scale and truncate to integer */
 479             rt               = _mm_mul_ps(r00,vftabscale);
 480             vfitab           = _mm_cvttps_epi32(rt);
 481             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 482             vfitab           = _mm_slli_epi32(vfitab,3);
 483
 484             /* REACTION-FIELD ELECTROSTATICS */
 485             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 486             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 487
 488             /* CUBIC SPLINE TABLE DISPERSION */
 489             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 490             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 491             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 492             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 493             _MM_TRANSPOSE4_PS(Y,F,G,H);
 494             Heps             = _mm_mul_ps(vfeps,H);
 495             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 496             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 497             vvdw6            = _mm_mul_ps(c6_00,VV);
 498             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 499             fvdw6            = _mm_mul_ps(c6_00,FF);
 500
 501             /* CUBIC SPLINE TABLE REPULSION */
 502             vfitab           = _mm_add_epi32(vfitab,ifour);
 503             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 504             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 505             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 506             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 507             _MM_TRANSPOSE4_PS(Y,F,G,H);
 508             Heps             = _mm_mul_ps(vfeps,H);
 509             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 510             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 511             vvdw12           = _mm_mul_ps(c12_00,VV);
 512             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 513             fvdw12           = _mm_mul_ps(c12_00,FF);
 514             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 515             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 516
 517             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 518
 519             /* Update potential sum for this i atom from the interaction with this j atom. */
 520             velec            = _mm_and_ps(velec,cutoff_mask);
 521             velec            = _mm_andnot_ps(dummy_mask,velec);
 522             velecsum         = _mm_add_ps(velecsum,velec);
 523             vvdw             = _mm_and_ps(vvdw,cutoff_mask);
 524             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 525             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 526
 527             fscal            = _mm_add_ps(felec,fvdw);
 528
 529             fscal            = _mm_and_ps(fscal,cutoff_mask);
 530
 531             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 532
 533             /* Calculate temporary vectorial force */
 534             tx               = _mm_mul_ps(fscal,dx00);
 535             ty               = _mm_mul_ps(fscal,dy00);
 536             tz               = _mm_mul_ps(fscal,dz00);
 537
 538             /* Update vectorial force */
 539             fix0             = _mm_add_ps(fix0,tx);
 540             fiy0             = _mm_add_ps(fiy0,ty);
 541             fiz0             = _mm_add_ps(fiz0,tz);
 542
 543             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 544                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 545                                                    tx,ty,tz);
 546
 547             }
 548
 549             /**************************
 550              * CALCULATE INTERACTIONS *
 551              **************************/
 552
 553             if (gmx_mm_any_lt(rsq10,rcutoff2))
 554             {
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq10             = _mm_mul_ps(iq1,jq0);
 558
 559             /* REACTION-FIELD ELECTROSTATICS */
 560             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 561             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 562
 563             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm_and_ps(velec,cutoff_mask);
 567             velec            = _mm_andnot_ps(dummy_mask,velec);
 568             velecsum         = _mm_add_ps(velecsum,velec);
 569
 570             fscal            = felec;
 571
 572             fscal            = _mm_and_ps(fscal,cutoff_mask);
 573
 574             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 575
 576             /* Calculate temporary vectorial force */
 577             tx               = _mm_mul_ps(fscal,dx10);
 578             ty               = _mm_mul_ps(fscal,dy10);
 579             tz               = _mm_mul_ps(fscal,dz10);
 580
 581             /* Update vectorial force */
 582             fix1             = _mm_add_ps(fix1,tx);
 583             fiy1             = _mm_add_ps(fiy1,ty);
 584             fiz1             = _mm_add_ps(fiz1,tz);
 585
 586             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 587                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 588                                                    tx,ty,tz);
 589
 590             }
 591
 592             /**************************
 593              * CALCULATE INTERACTIONS *
 594              **************************/
 595
 596             if (gmx_mm_any_lt(rsq20,rcutoff2))
 597             {
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq20             = _mm_mul_ps(iq2,jq0);
 601
 602             /* REACTION-FIELD ELECTROSTATICS */
 603             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 604             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 605
 606             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 607
 608             /* Update potential sum for this i atom from the interaction with this j atom. */
 609             velec            = _mm_and_ps(velec,cutoff_mask);
 610             velec            = _mm_andnot_ps(dummy_mask,velec);
 611             velecsum         = _mm_add_ps(velecsum,velec);
 612
 613             fscal            = felec;
 614
 615             fscal            = _mm_and_ps(fscal,cutoff_mask);
 616
 617             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 618
 619             /* Calculate temporary vectorial force */
 620             tx               = _mm_mul_ps(fscal,dx20);
 621             ty               = _mm_mul_ps(fscal,dy20);
 622             tz               = _mm_mul_ps(fscal,dz20);
 623
 624             /* Update vectorial force */
 625             fix2             = _mm_add_ps(fix2,tx);
 626             fiy2             = _mm_add_ps(fiy2,ty);
 627             fiz2             = _mm_add_ps(fiz2,tz);
 628
 629             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 630                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 631                                                    tx,ty,tz);
 632
 633             }
 634
 635             /* Inner loop uses 145 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         ggid                        = gid[iidx];
 644         /* Update potential energies */
 645         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 646         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 29 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*145);
 660 }
 661 /*
 662  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse2_single
 663  * Electrostatics interaction: ReactionField
 664  * VdW interaction:            CubicSplineTable
 665  * Geometry:                   Water3-Particle
 666  * Calculate force/pot:        Force
 667  */
 668 void
 669 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_sse2_single
 670                     (t_nblist * gmx_restrict                nlist,
 671                      rvec * gmx_restrict                    xx,
 672                      rvec * gmx_restrict                    ff,
 673                      t_forcerec * gmx_restrict              fr,
 674                      t_mdatoms * gmx_restrict               mdatoms,
 675                      nb_kernel_data_t * gmx_restrict        kernel_data,
 676                      t_nrnb * gmx_restrict                  nrnb)
 677 {
 678     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 679      * just 0 for non-waters.
 680      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 681      * jnr indices corresponding to data put in the four positions in the SIMD register.
 682      */
 683     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 684     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 685     int              jnrA,jnrB,jnrC,jnrD;
 686     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 687     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 688     real             shX,shY,shZ,rcutoff_scalar;
 689     real             *shiftvec,*fshift,*x,*f;
 690     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 691     int              vdwioffset0;
 692     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 693     int              vdwioffset1;
 694     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 695     int              vdwioffset2;
 696     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 697     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 698     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 699     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 700     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 703     real             *charge;
 704     int              nvdwtype;
 705     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 706     int              *vdwtype;
 707     real             *vdwparam;
 708     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 709     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 710     __m128i          vfitab;
 711     __m128i          ifour       = _mm_set1_epi32(4);
 712     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 713     real             *vftab;
 714     __m128           dummy_mask,cutoff_mask;
 715     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 716     __m128           one     = _mm_set1_ps(1.0);
 717     __m128           two     = _mm_set1_ps(2.0);
 718     x                = xx[0];
 719     f                = ff[0];
 720
 721     nri              = nlist->nri;
 722     iinr             = nlist->iinr;
 723     jindex           = nlist->jindex;
 724     jjnr             = nlist->jjnr;
 725     shiftidx         = nlist->shift;
 726     gid              = nlist->gid;
 727     shiftvec         = fr->shift_vec[0];
 728     fshift           = fr->fshift[0];
 729     facel            = _mm_set1_ps(fr->epsfac);
 730     charge           = mdatoms->chargeA;
 731     krf              = _mm_set1_ps(fr->ic->k_rf);
 732     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 733     crf              = _mm_set1_ps(fr->ic->c_rf);
 734     nvdwtype         = fr->ntype;
 735     vdwparam         = fr->nbfp;
 736     vdwtype          = mdatoms->typeA;
 737
 738     vftab            = kernel_data->table_vdw->data;
 739     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 740
 741     /* Setup water-specific parameters */
 742     inr              = nlist->iinr[0];
 743     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 744     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 745     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 746     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 747
 748     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 749     rcutoff_scalar   = fr->rcoulomb;
 750     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 751     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 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
 791         fix0             = _mm_setzero_ps();
 792         fiy0             = _mm_setzero_ps();
 793         fiz0             = _mm_setzero_ps();
 794         fix1             = _mm_setzero_ps();
 795         fiy1             = _mm_setzero_ps();
 796         fiz1             = _mm_setzero_ps();
 797         fix2             = _mm_setzero_ps();
 798         fiy2             = _mm_setzero_ps();
 799         fiz2             = _mm_setzero_ps();
 800
 801         /* Start inner kernel loop */
 802         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 803         {
 804
 805             /* Get j neighbor index, and coordinate index */
 806             jnrA             = jjnr[jidx];
 807             jnrB             = jjnr[jidx+1];
 808             jnrC             = jjnr[jidx+2];
 809             jnrD             = jjnr[jidx+3];
 810
 811             j_coord_offsetA  = DIM*jnrA;
 812             j_coord_offsetB  = DIM*jnrB;
 813             j_coord_offsetC  = DIM*jnrC;
 814             j_coord_offsetD  = DIM*jnrD;
 815
 816             /* load j atom coordinates */
 817             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 818                                               x+j_coord_offsetC,x+j_coord_offsetD,
 819                                               &jx0,&jy0,&jz0);
 820
 821             /* Calculate displacement vector */
 822             dx00             = _mm_sub_ps(ix0,jx0);
 823             dy00             = _mm_sub_ps(iy0,jy0);
 824             dz00             = _mm_sub_ps(iz0,jz0);
 825             dx10             = _mm_sub_ps(ix1,jx0);
 826             dy10             = _mm_sub_ps(iy1,jy0);
 827             dz10             = _mm_sub_ps(iz1,jz0);
 828             dx20             = _mm_sub_ps(ix2,jx0);
 829             dy20             = _mm_sub_ps(iy2,jy0);
 830             dz20             = _mm_sub_ps(iz2,jz0);
 831
 832             /* Calculate squared distance and things based on it */
 833             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 834             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 835             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 836
 837             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 838             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 839             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 840
 841             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 842             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 843             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 844
 845             /* Load parameters for j particles */
 846             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 847                                                               charge+jnrC+0,charge+jnrD+0);
 848             vdwjidx0A        = 2*vdwtype[jnrA+0];
 849             vdwjidx0B        = 2*vdwtype[jnrB+0];
 850             vdwjidx0C        = 2*vdwtype[jnrC+0];
 851             vdwjidx0D        = 2*vdwtype[jnrD+0];
 852
 853             /**************************
 854              * CALCULATE INTERACTIONS *
 855              **************************/
 856
 857             if (gmx_mm_any_lt(rsq00,rcutoff2))
 858             {
 859
 860             r00              = _mm_mul_ps(rsq00,rinv00);
 861
 862             /* Compute parameters for interactions between i and j atoms */
 863             qq00             = _mm_mul_ps(iq0,jq0);
 864             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 865                                          vdwparam+vdwioffset0+vdwjidx0B,
 866                                          vdwparam+vdwioffset0+vdwjidx0C,
 867                                          vdwparam+vdwioffset0+vdwjidx0D,
 868                                          &c6_00,&c12_00);
 869
 870             /* Calculate table index by multiplying r with table scale and truncate to integer */
 871             rt               = _mm_mul_ps(r00,vftabscale);
 872             vfitab           = _mm_cvttps_epi32(rt);
 873             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 874             vfitab           = _mm_slli_epi32(vfitab,3);
 875
 876             /* REACTION-FIELD ELECTROSTATICS */
 877             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 878
 879             /* CUBIC SPLINE TABLE DISPERSION */
 880             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 881             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 882             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 883             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 884             _MM_TRANSPOSE4_PS(Y,F,G,H);
 885             Heps             = _mm_mul_ps(vfeps,H);
 886             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 887             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 888             fvdw6            = _mm_mul_ps(c6_00,FF);
 889
 890             /* CUBIC SPLINE TABLE REPULSION */
 891             vfitab           = _mm_add_epi32(vfitab,ifour);
 892             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 893             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 894             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 895             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 896             _MM_TRANSPOSE4_PS(Y,F,G,H);
 897             Heps             = _mm_mul_ps(vfeps,H);
 898             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 899             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 900             fvdw12           = _mm_mul_ps(c12_00,FF);
 901             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 902
 903             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 904
 905             fscal            = _mm_add_ps(felec,fvdw);
 906
 907             fscal            = _mm_and_ps(fscal,cutoff_mask);
 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
 925             /**************************
 926              * CALCULATE INTERACTIONS *
 927              **************************/
 928
 929             if (gmx_mm_any_lt(rsq10,rcutoff2))
 930             {
 931
 932             /* Compute parameters for interactions between i and j atoms */
 933             qq10             = _mm_mul_ps(iq1,jq0);
 934
 935             /* REACTION-FIELD ELECTROSTATICS */
 936             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 937
 938             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 939
 940             fscal            = felec;
 941
 942             fscal            = _mm_and_ps(fscal,cutoff_mask);
 943
 944             /* Calculate temporary vectorial force */
 945             tx               = _mm_mul_ps(fscal,dx10);
 946             ty               = _mm_mul_ps(fscal,dy10);
 947             tz               = _mm_mul_ps(fscal,dz10);
 948
 949             /* Update vectorial force */
 950             fix1             = _mm_add_ps(fix1,tx);
 951             fiy1             = _mm_add_ps(fiy1,ty);
 952             fiz1             = _mm_add_ps(fiz1,tz);
 953
 954             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 955                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 956                                                    tx,ty,tz);
 957
 958             }
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             if (gmx_mm_any_lt(rsq20,rcutoff2))
 965             {
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq20             = _mm_mul_ps(iq2,jq0);
 969
 970             /* REACTION-FIELD ELECTROSTATICS */
 971             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 972
 973             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 974
 975             fscal            = felec;
 976
 977             fscal            = _mm_and_ps(fscal,cutoff_mask);
 978
 979             /* Calculate temporary vectorial force */
 980             tx               = _mm_mul_ps(fscal,dx20);
 981             ty               = _mm_mul_ps(fscal,dy20);
 982             tz               = _mm_mul_ps(fscal,dz20);
 983
 984             /* Update vectorial force */
 985             fix2             = _mm_add_ps(fix2,tx);
 986             fiy2             = _mm_add_ps(fiy2,ty);
 987             fiz2             = _mm_add_ps(fiz2,tz);
 988
 989             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 990                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 991                                                    tx,ty,tz);
 992
 993             }
 994
 995             /* Inner loop uses 117 flops */
 996         }
 997
 998         if(jidx<j_index_end)
 999         {
1000
1001             /* Get j neighbor index, and coordinate index */
1002             jnrA             = jjnr[jidx];
1003             jnrB             = jjnr[jidx+1];
1004             jnrC             = jjnr[jidx+2];
1005             jnrD             = jjnr[jidx+3];
1006
1007             /* Sign of each element will be negative for non-real atoms.
1008              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1009              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1010              */
1011             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1012             jnrA       = (jnrA>=0) ? jnrA : 0;
1013             jnrB       = (jnrB>=0) ? jnrB : 0;
1014             jnrC       = (jnrC>=0) ? jnrC : 0;
1015             jnrD       = (jnrD>=0) ? jnrD : 0;
1016
1017             j_coord_offsetA  = DIM*jnrA;
1018             j_coord_offsetB  = DIM*jnrB;
1019             j_coord_offsetC  = DIM*jnrC;
1020             j_coord_offsetD  = DIM*jnrD;
1021
1022             /* load j atom coordinates */
1023             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1024                                               x+j_coord_offsetC,x+j_coord_offsetD,
1025                                               &jx0,&jy0,&jz0);
1026
1027             /* Calculate displacement vector */
1028             dx00             = _mm_sub_ps(ix0,jx0);
1029             dy00             = _mm_sub_ps(iy0,jy0);
1030             dz00             = _mm_sub_ps(iz0,jz0);
1031             dx10             = _mm_sub_ps(ix1,jx0);
1032             dy10             = _mm_sub_ps(iy1,jy0);
1033             dz10             = _mm_sub_ps(iz1,jz0);
1034             dx20             = _mm_sub_ps(ix2,jx0);
1035             dy20             = _mm_sub_ps(iy2,jy0);
1036             dz20             = _mm_sub_ps(iz2,jz0);
1037
1038             /* Calculate squared distance and things based on it */
1039             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1040             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1041             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1042
1043             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1044             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1045             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1046
1047             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1048             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1049             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1050
1051             /* Load parameters for j particles */
1052             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1053                                                               charge+jnrC+0,charge+jnrD+0);
1054             vdwjidx0A        = 2*vdwtype[jnrA+0];
1055             vdwjidx0B        = 2*vdwtype[jnrB+0];
1056             vdwjidx0C        = 2*vdwtype[jnrC+0];
1057             vdwjidx0D        = 2*vdwtype[jnrD+0];
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             if (gmx_mm_any_lt(rsq00,rcutoff2))
1064             {
1065
1066             r00              = _mm_mul_ps(rsq00,rinv00);
1067             r00              = _mm_andnot_ps(dummy_mask,r00);
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq00             = _mm_mul_ps(iq0,jq0);
1071             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1072                                          vdwparam+vdwioffset0+vdwjidx0B,
1073                                          vdwparam+vdwioffset0+vdwjidx0C,
1074                                          vdwparam+vdwioffset0+vdwjidx0D,
1075                                          &c6_00,&c12_00);
1076
1077             /* Calculate table index by multiplying r with table scale and truncate to integer */
1078             rt               = _mm_mul_ps(r00,vftabscale);
1079             vfitab           = _mm_cvttps_epi32(rt);
1080             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1081             vfitab           = _mm_slli_epi32(vfitab,3);
1082
1083             /* REACTION-FIELD ELECTROSTATICS */
1084             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
1085
1086             /* CUBIC SPLINE TABLE DISPERSION */
1087             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1088             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1089             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1090             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1091             _MM_TRANSPOSE4_PS(Y,F,G,H);
1092             Heps             = _mm_mul_ps(vfeps,H);
1093             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1094             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1095             fvdw6            = _mm_mul_ps(c6_00,FF);
1096
1097             /* CUBIC SPLINE TABLE REPULSION */
1098             vfitab           = _mm_add_epi32(vfitab,ifour);
1099             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1100             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1101             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1102             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1103             _MM_TRANSPOSE4_PS(Y,F,G,H);
1104             Heps             = _mm_mul_ps(vfeps,H);
1105             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1106             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1107             fvdw12           = _mm_mul_ps(c12_00,FF);
1108             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1109
1110             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1111
1112             fscal            = _mm_add_ps(felec,fvdw);
1113
1114             fscal            = _mm_and_ps(fscal,cutoff_mask);
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
1134             /**************************
1135              * CALCULATE INTERACTIONS *
1136              **************************/
1137
1138             if (gmx_mm_any_lt(rsq10,rcutoff2))
1139             {
1140
1141             /* Compute parameters for interactions between i and j atoms */
1142             qq10             = _mm_mul_ps(iq1,jq0);
1143
1144             /* REACTION-FIELD ELECTROSTATICS */
1145             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
1146
1147             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1148
1149             fscal            = felec;
1150
1151             fscal            = _mm_and_ps(fscal,cutoff_mask);
1152
1153             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1154
1155             /* Calculate temporary vectorial force */
1156             tx               = _mm_mul_ps(fscal,dx10);
1157             ty               = _mm_mul_ps(fscal,dy10);
1158             tz               = _mm_mul_ps(fscal,dz10);
1159
1160             /* Update vectorial force */
1161             fix1             = _mm_add_ps(fix1,tx);
1162             fiy1             = _mm_add_ps(fiy1,ty);
1163             fiz1             = _mm_add_ps(fiz1,tz);
1164
1165             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1166                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1167                                                    tx,ty,tz);
1168
1169             }
1170
1171             /**************************
1172              * CALCULATE INTERACTIONS *
1173              **************************/
1174
1175             if (gmx_mm_any_lt(rsq20,rcutoff2))
1176             {
1177
1178             /* Compute parameters for interactions between i and j atoms */
1179             qq20             = _mm_mul_ps(iq2,jq0);
1180
1181             /* REACTION-FIELD ELECTROSTATICS */
1182             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
1183
1184             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1185
1186             fscal            = felec;
1187
1188             fscal            = _mm_and_ps(fscal,cutoff_mask);
1189
1190             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1191
1192             /* Calculate temporary vectorial force */
1193             tx               = _mm_mul_ps(fscal,dx20);
1194             ty               = _mm_mul_ps(fscal,dy20);
1195             tz               = _mm_mul_ps(fscal,dz20);
1196
1197             /* Update vectorial force */
1198             fix2             = _mm_add_ps(fix2,tx);
1199             fiy2             = _mm_add_ps(fiy2,ty);
1200             fiz2             = _mm_add_ps(fiz2,tz);
1201
1202             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1203                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1204                                                    tx,ty,tz);
1205
1206             }
1207
1208             /* Inner loop uses 118 flops */
1209         }
1210
1211         /* End of innermost loop */
1212
1213         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1214                                               f+i_coord_offset,fshift+i_shift_offset);
1215
1216         /* Increment number of inner iterations */
1217         inneriter                  += j_index_end - j_index_start;
1218
1219         /* Outer loop uses 27 flops */
1220     }
1221
1222     /* Increment number of outer iterations */
1223     outeriter        += nri;
1224
1225     /* Update outer/inner flops */
1226
1227     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*118);
1228 }