src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_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_ElecCoul_VdwCSTab_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  91     real             *vftab;
  92     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 119     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 120     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     /* Start outer loop over neighborlists */
 134     for(iidx=0; iidx<nri; iidx++)
 135     {
 136         /* Load shift vector for this list */
 137         i_shift_offset   = DIM*shiftidx[iidx];
 138         shX              = shiftvec[i_shift_offset+XX];
 139         shY              = shiftvec[i_shift_offset+YY];
 140         shZ              = shiftvec[i_shift_offset+ZZ];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 152         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 153         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 154         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 155         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 156         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 157         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 158         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 159         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 160         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 161         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 162         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 163
 164         fix0             = _mm_setzero_ps();
 165         fiy0             = _mm_setzero_ps();
 166         fiz0             = _mm_setzero_ps();
 167         fix1             = _mm_setzero_ps();
 168         fiy1             = _mm_setzero_ps();
 169         fiz1             = _mm_setzero_ps();
 170         fix2             = _mm_setzero_ps();
 171         fiy2             = _mm_setzero_ps();
 172         fiz2             = _mm_setzero_ps();
 173         fix3             = _mm_setzero_ps();
 174         fiy3             = _mm_setzero_ps();
 175         fiz3             = _mm_setzero_ps();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_ps();
 179         vvdwsum          = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_ps(ix0,jx0);
 203             dy00             = _mm_sub_ps(iy0,jy0);
 204             dz00             = _mm_sub_ps(iz0,jz0);
 205             dx10             = _mm_sub_ps(ix1,jx0);
 206             dy10             = _mm_sub_ps(iy1,jy0);
 207             dz10             = _mm_sub_ps(iz1,jz0);
 208             dx20             = _mm_sub_ps(ix2,jx0);
 209             dy20             = _mm_sub_ps(iy2,jy0);
 210             dz20             = _mm_sub_ps(iz2,jz0);
 211             dx30             = _mm_sub_ps(ix3,jx0);
 212             dy30             = _mm_sub_ps(iy3,jy0);
 213             dz30             = _mm_sub_ps(iz3,jz0);
 214
 215             /* Calculate squared distance and things based on it */
 216             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 217             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 218             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 219             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 220
 221             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 225
 226             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 227             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 228             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 232                                                               charge+jnrC+0,charge+jnrD+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235             vdwjidx0C        = 2*vdwtype[jnrC+0];
 236             vdwjidx0D        = 2*vdwtype[jnrD+0];
 237
 238             /**************************
 239              * CALCULATE INTERACTIONS *
 240              **************************/
 241
 242             r00              = _mm_mul_ps(rsq00,rinv00);
 243
 244             /* Compute parameters for interactions between i and j atoms */
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm_mul_ps(r00,vftabscale);
 253             vfitab           = _mm_cvttps_epi32(rt);
 254             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 255             vfitab           = _mm_slli_epi32(vfitab,3);
 256
 257             /* CUBIC SPLINE TABLE DISPERSION */
 258             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 260             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 261             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 262             _MM_TRANSPOSE4_PS(Y,F,G,H);
 263             Heps             = _mm_mul_ps(vfeps,H);
 264             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 265             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 266             vvdw6            = _mm_mul_ps(c6_00,VV);
 267             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 268             fvdw6            = _mm_mul_ps(c6_00,FF);
 269
 270             /* CUBIC SPLINE TABLE REPULSION */
 271             vfitab           = _mm_add_epi32(vfitab,ifour);
 272             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 273             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 274             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 275             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 276             _MM_TRANSPOSE4_PS(Y,F,G,H);
 277             Heps             = _mm_mul_ps(vfeps,H);
 278             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 279             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 280             vvdw12           = _mm_mul_ps(c12_00,VV);
 281             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 282             fvdw12           = _mm_mul_ps(c12_00,FF);
 283             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 284             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 288
 289             fscal            = fvdw;
 290
 291             /* Calculate temporary vectorial force */
 292             tx               = _mm_mul_ps(fscal,dx00);
 293             ty               = _mm_mul_ps(fscal,dy00);
 294             tz               = _mm_mul_ps(fscal,dz00);
 295
 296             /* Update vectorial force */
 297             fix0             = _mm_add_ps(fix0,tx);
 298             fiy0             = _mm_add_ps(fiy0,ty);
 299             fiz0             = _mm_add_ps(fiz0,tz);
 300
 301             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 302                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 303                                                    tx,ty,tz);
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             /* Compute parameters for interactions between i and j atoms */
 310             qq10             = _mm_mul_ps(iq1,jq0);
 311
 312             /* COULOMB ELECTROSTATICS */
 313             velec            = _mm_mul_ps(qq10,rinv10);
 314             felec            = _mm_mul_ps(velec,rinvsq10);
 315
 316             /* Update potential sum for this i atom from the interaction with this j atom. */
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm_mul_ps(fscal,dx10);
 323             ty               = _mm_mul_ps(fscal,dy10);
 324             tz               = _mm_mul_ps(fscal,dz10);
 325
 326             /* Update vectorial force */
 327             fix1             = _mm_add_ps(fix1,tx);
 328             fiy1             = _mm_add_ps(fiy1,ty);
 329             fiz1             = _mm_add_ps(fiz1,tz);
 330
 331             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 332                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 333                                                    tx,ty,tz);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq20             = _mm_mul_ps(iq2,jq0);
 341
 342             /* COULOMB ELECTROSTATICS */
 343             velec            = _mm_mul_ps(qq20,rinv20);
 344             felec            = _mm_mul_ps(velec,rinvsq20);
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_ps(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Calculate temporary vectorial force */
 352             tx               = _mm_mul_ps(fscal,dx20);
 353             ty               = _mm_mul_ps(fscal,dy20);
 354             tz               = _mm_mul_ps(fscal,dz20);
 355
 356             /* Update vectorial force */
 357             fix2             = _mm_add_ps(fix2,tx);
 358             fiy2             = _mm_add_ps(fiy2,ty);
 359             fiz2             = _mm_add_ps(fiz2,tz);
 360
 361             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 362                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 363                                                    tx,ty,tz);
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq30             = _mm_mul_ps(iq3,jq0);
 371
 372             /* COULOMB ELECTROSTATICS */
 373             velec            = _mm_mul_ps(qq30,rinv30);
 374             felec            = _mm_mul_ps(velec,rinvsq30);
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx30);
 383             ty               = _mm_mul_ps(fscal,dy30);
 384             tz               = _mm_mul_ps(fscal,dz30);
 385
 386             /* Update vectorial force */
 387             fix3             = _mm_add_ps(fix3,tx);
 388             fiy3             = _mm_add_ps(fiy3,ty);
 389             fiz3             = _mm_add_ps(fiz3,tz);
 390
 391             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 392                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 393                                                    tx,ty,tz);
 394
 395             /* Inner loop uses 140 flops */
 396         }
 397
 398         if(jidx<j_index_end)
 399         {
 400
 401             /* Get j neighbor index, and coordinate index */
 402             jnrA             = jjnr[jidx];
 403             jnrB             = jjnr[jidx+1];
 404             jnrC             = jjnr[jidx+2];
 405             jnrD             = jjnr[jidx+3];
 406
 407             /* Sign of each element will be negative for non-real atoms.
 408              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 409              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 410              */
 411             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 412             jnrA       = (jnrA>=0) ? jnrA : 0;
 413             jnrB       = (jnrB>=0) ? jnrB : 0;
 414             jnrC       = (jnrC>=0) ? jnrC : 0;
 415             jnrD       = (jnrD>=0) ? jnrD : 0;
 416
 417             j_coord_offsetA  = DIM*jnrA;
 418             j_coord_offsetB  = DIM*jnrB;
 419             j_coord_offsetC  = DIM*jnrC;
 420             j_coord_offsetD  = DIM*jnrD;
 421
 422             /* load j atom coordinates */
 423             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 424                                               x+j_coord_offsetC,x+j_coord_offsetD,
 425                                               &jx0,&jy0,&jz0);
 426
 427             /* Calculate displacement vector */
 428             dx00             = _mm_sub_ps(ix0,jx0);
 429             dy00             = _mm_sub_ps(iy0,jy0);
 430             dz00             = _mm_sub_ps(iz0,jz0);
 431             dx10             = _mm_sub_ps(ix1,jx0);
 432             dy10             = _mm_sub_ps(iy1,jy0);
 433             dz10             = _mm_sub_ps(iz1,jz0);
 434             dx20             = _mm_sub_ps(ix2,jx0);
 435             dy20             = _mm_sub_ps(iy2,jy0);
 436             dz20             = _mm_sub_ps(iz2,jz0);
 437             dx30             = _mm_sub_ps(ix3,jx0);
 438             dy30             = _mm_sub_ps(iy3,jy0);
 439             dz30             = _mm_sub_ps(iz3,jz0);
 440
 441             /* Calculate squared distance and things based on it */
 442             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 443             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 444             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 445             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 446
 447             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 448             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 449             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 450             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 451
 452             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 453             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 454             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 455
 456             /* Load parameters for j particles */
 457             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 458                                                               charge+jnrC+0,charge+jnrD+0);
 459             vdwjidx0A        = 2*vdwtype[jnrA+0];
 460             vdwjidx0B        = 2*vdwtype[jnrB+0];
 461             vdwjidx0C        = 2*vdwtype[jnrC+0];
 462             vdwjidx0D        = 2*vdwtype[jnrD+0];
 463
 464             /**************************
 465              * CALCULATE INTERACTIONS *
 466              **************************/
 467
 468             r00              = _mm_mul_ps(rsq00,rinv00);
 469             r00              = _mm_andnot_ps(dummy_mask,r00);
 470
 471             /* Compute parameters for interactions between i and j atoms */
 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             /* CUBIC SPLINE TABLE DISPERSION */
 485             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 486             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 487             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 488             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 489             _MM_TRANSPOSE4_PS(Y,F,G,H);
 490             Heps             = _mm_mul_ps(vfeps,H);
 491             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 492             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 493             vvdw6            = _mm_mul_ps(c6_00,VV);
 494             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 495             fvdw6            = _mm_mul_ps(c6_00,FF);
 496
 497             /* CUBIC SPLINE TABLE REPULSION */
 498             vfitab           = _mm_add_epi32(vfitab,ifour);
 499             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 500             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 501             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 502             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 503             _MM_TRANSPOSE4_PS(Y,F,G,H);
 504             Heps             = _mm_mul_ps(vfeps,H);
 505             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 506             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 507             vvdw12           = _mm_mul_ps(c12_00,VV);
 508             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 509             fvdw12           = _mm_mul_ps(c12_00,FF);
 510             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 511             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 515             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 516
 517             fscal            = fvdw;
 518
 519             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm_mul_ps(fscal,dx00);
 523             ty               = _mm_mul_ps(fscal,dy00);
 524             tz               = _mm_mul_ps(fscal,dz00);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_add_ps(fix0,tx);
 528             fiy0             = _mm_add_ps(fiy0,ty);
 529             fiz0             = _mm_add_ps(fiz0,tz);
 530
 531             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 532                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 533                                                    tx,ty,tz);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             /* Compute parameters for interactions between i and j atoms */
 540             qq10             = _mm_mul_ps(iq1,jq0);
 541
 542             /* COULOMB ELECTROSTATICS */
 543             velec            = _mm_mul_ps(qq10,rinv10);
 544             felec            = _mm_mul_ps(velec,rinvsq10);
 545
 546             /* Update potential sum for this i atom from the interaction with this j atom. */
 547             velec            = _mm_andnot_ps(dummy_mask,velec);
 548             velecsum         = _mm_add_ps(velecsum,velec);
 549
 550             fscal            = felec;
 551
 552             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 553
 554             /* Calculate temporary vectorial force */
 555             tx               = _mm_mul_ps(fscal,dx10);
 556             ty               = _mm_mul_ps(fscal,dy10);
 557             tz               = _mm_mul_ps(fscal,dz10);
 558
 559             /* Update vectorial force */
 560             fix1             = _mm_add_ps(fix1,tx);
 561             fiy1             = _mm_add_ps(fiy1,ty);
 562             fiz1             = _mm_add_ps(fiz1,tz);
 563
 564             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 565                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 566                                                    tx,ty,tz);
 567
 568             /**************************
 569              * CALCULATE INTERACTIONS *
 570              **************************/
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_ps(iq2,jq0);
 574
 575             /* COULOMB ELECTROSTATICS */
 576             velec            = _mm_mul_ps(qq20,rinv20);
 577             felec            = _mm_mul_ps(velec,rinvsq20);
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             velec            = _mm_andnot_ps(dummy_mask,velec);
 581             velecsum         = _mm_add_ps(velecsum,velec);
 582
 583             fscal            = felec;
 584
 585             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 586
 587             /* Calculate temporary vectorial force */
 588             tx               = _mm_mul_ps(fscal,dx20);
 589             ty               = _mm_mul_ps(fscal,dy20);
 590             tz               = _mm_mul_ps(fscal,dz20);
 591
 592             /* Update vectorial force */
 593             fix2             = _mm_add_ps(fix2,tx);
 594             fiy2             = _mm_add_ps(fiy2,ty);
 595             fiz2             = _mm_add_ps(fiz2,tz);
 596
 597             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 598                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 599                                                    tx,ty,tz);
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             /* Compute parameters for interactions between i and j atoms */
 606             qq30             = _mm_mul_ps(iq3,jq0);
 607
 608             /* COULOMB ELECTROSTATICS */
 609             velec            = _mm_mul_ps(qq30,rinv30);
 610             felec            = _mm_mul_ps(velec,rinvsq30);
 611
 612             /* Update potential sum for this i atom from the interaction with this j atom. */
 613             velec            = _mm_andnot_ps(dummy_mask,velec);
 614             velecsum         = _mm_add_ps(velecsum,velec);
 615
 616             fscal            = felec;
 617
 618             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 619
 620             /* Calculate temporary vectorial force */
 621             tx               = _mm_mul_ps(fscal,dx30);
 622             ty               = _mm_mul_ps(fscal,dy30);
 623             tz               = _mm_mul_ps(fscal,dz30);
 624
 625             /* Update vectorial force */
 626             fix3             = _mm_add_ps(fix3,tx);
 627             fiy3             = _mm_add_ps(fiy3,ty);
 628             fiz3             = _mm_add_ps(fiz3,tz);
 629
 630             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 631                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 632                                                    tx,ty,tz);
 633
 634             /* Inner loop uses 141 flops */
 635         }
 636
 637         /* End of innermost loop */
 638
 639         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 640                                               f+i_coord_offset,fshift+i_shift_offset);
 641
 642         ggid                        = gid[iidx];
 643         /* Update potential energies */
 644         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 645         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 646
 647         /* Increment number of inner iterations */
 648         inneriter                  += j_index_end - j_index_start;
 649
 650         /* Outer loop uses 38 flops */
 651     }
 652
 653     /* Increment number of outer iterations */
 654     outeriter        += nri;
 655
 656     /* Update outer/inner flops */
 657
 658     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*141);
 659 }
 660 /*
 661  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse2_single
 662  * Electrostatics interaction: Coulomb
 663  * VdW interaction:            CubicSplineTable
 664  * Geometry:                   Water4-Particle
 665  * Calculate force/pot:        Force
 666  */
 667 void
 668 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_sse2_single
 669                     (t_nblist * gmx_restrict                nlist,
 670                      rvec * gmx_restrict                    xx,
 671                      rvec * gmx_restrict                    ff,
 672                      t_forcerec * gmx_restrict              fr,
 673                      t_mdatoms * gmx_restrict               mdatoms,
 674                      nb_kernel_data_t * gmx_restrict        kernel_data,
 675                      t_nrnb * gmx_restrict                  nrnb)
 676 {
 677     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 678      * just 0 for non-waters.
 679      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 680      * jnr indices corresponding to data put in the four positions in the SIMD register.
 681      */
 682     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 683     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 684     int              jnrA,jnrB,jnrC,jnrD;
 685     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 686     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 687     real             shX,shY,shZ,rcutoff_scalar;
 688     real             *shiftvec,*fshift,*x,*f;
 689     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 690     int              vdwioffset0;
 691     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 692     int              vdwioffset1;
 693     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 694     int              vdwioffset2;
 695     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 696     int              vdwioffset3;
 697     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 698     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 699     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 700     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 701     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 702     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 703     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 704     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 705     real             *charge;
 706     int              nvdwtype;
 707     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 708     int              *vdwtype;
 709     real             *vdwparam;
 710     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 711     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 712     __m128i          vfitab;
 713     __m128i          ifour       = _mm_set1_epi32(4);
 714     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 715     real             *vftab;
 716     __m128           dummy_mask,cutoff_mask;
 717     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 718     __m128           one     = _mm_set1_ps(1.0);
 719     __m128           two     = _mm_set1_ps(2.0);
 720     x                = xx[0];
 721     f                = ff[0];
 722
 723     nri              = nlist->nri;
 724     iinr             = nlist->iinr;
 725     jindex           = nlist->jindex;
 726     jjnr             = nlist->jjnr;
 727     shiftidx         = nlist->shift;
 728     gid              = nlist->gid;
 729     shiftvec         = fr->shift_vec[0];
 730     fshift           = fr->fshift[0];
 731     facel            = _mm_set1_ps(fr->epsfac);
 732     charge           = mdatoms->chargeA;
 733     nvdwtype         = fr->ntype;
 734     vdwparam         = fr->nbfp;
 735     vdwtype          = mdatoms->typeA;
 736
 737     vftab            = kernel_data->table_vdw->data;
 738     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 739
 740     /* Setup water-specific parameters */
 741     inr              = nlist->iinr[0];
 742     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 743     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 744     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 745     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 746
 747     /* Avoid stupid compiler warnings */
 748     jnrA = jnrB = jnrC = jnrD = 0;
 749     j_coord_offsetA = 0;
 750     j_coord_offsetB = 0;
 751     j_coord_offsetC = 0;
 752     j_coord_offsetD = 0;
 753
 754     outeriter        = 0;
 755     inneriter        = 0;
 756
 757     /* Start outer loop over neighborlists */
 758     for(iidx=0; iidx<nri; iidx++)
 759     {
 760         /* Load shift vector for this list */
 761         i_shift_offset   = DIM*shiftidx[iidx];
 762         shX              = shiftvec[i_shift_offset+XX];
 763         shY              = shiftvec[i_shift_offset+YY];
 764         shZ              = shiftvec[i_shift_offset+ZZ];
 765
 766         /* Load limits for loop over neighbors */
 767         j_index_start    = jindex[iidx];
 768         j_index_end      = jindex[iidx+1];
 769
 770         /* Get outer coordinate index */
 771         inr              = iinr[iidx];
 772         i_coord_offset   = DIM*inr;
 773
 774         /* Load i particle coords and add shift vector */
 775         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 776         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 777         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 778         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 779         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 780         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 781         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 782         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 783         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 784         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 785         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 786         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 787
 788         fix0             = _mm_setzero_ps();
 789         fiy0             = _mm_setzero_ps();
 790         fiz0             = _mm_setzero_ps();
 791         fix1             = _mm_setzero_ps();
 792         fiy1             = _mm_setzero_ps();
 793         fiz1             = _mm_setzero_ps();
 794         fix2             = _mm_setzero_ps();
 795         fiy2             = _mm_setzero_ps();
 796         fiz2             = _mm_setzero_ps();
 797         fix3             = _mm_setzero_ps();
 798         fiy3             = _mm_setzero_ps();
 799         fiz3             = _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             dx30             = _mm_sub_ps(ix3,jx0);
 832             dy30             = _mm_sub_ps(iy3,jy0);
 833             dz30             = _mm_sub_ps(iz3,jz0);
 834
 835             /* Calculate squared distance and things based on it */
 836             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 837             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 838             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 839             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 840
 841             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 842             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 843             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 844             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 845
 846             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 847             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 848             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 849
 850             /* Load parameters for j particles */
 851             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 852                                                               charge+jnrC+0,charge+jnrD+0);
 853             vdwjidx0A        = 2*vdwtype[jnrA+0];
 854             vdwjidx0B        = 2*vdwtype[jnrB+0];
 855             vdwjidx0C        = 2*vdwtype[jnrC+0];
 856             vdwjidx0D        = 2*vdwtype[jnrD+0];
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r00              = _mm_mul_ps(rsq00,rinv00);
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 866                                          vdwparam+vdwioffset0+vdwjidx0B,
 867                                          vdwparam+vdwioffset0+vdwjidx0C,
 868                                          vdwparam+vdwioffset0+vdwjidx0D,
 869                                          &c6_00,&c12_00);
 870
 871             /* Calculate table index by multiplying r with table scale and truncate to integer */
 872             rt               = _mm_mul_ps(r00,vftabscale);
 873             vfitab           = _mm_cvttps_epi32(rt);
 874             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
 875             vfitab           = _mm_slli_epi32(vfitab,3);
 876
 877             /* CUBIC SPLINE TABLE DISPERSION */
 878             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 879             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 880             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 881             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 882             _MM_TRANSPOSE4_PS(Y,F,G,H);
 883             Heps             = _mm_mul_ps(vfeps,H);
 884             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 885             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 886             fvdw6            = _mm_mul_ps(c6_00,FF);
 887
 888             /* CUBIC SPLINE TABLE REPULSION */
 889             vfitab           = _mm_add_epi32(vfitab,ifour);
 890             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 891             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 892             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 893             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 894             _MM_TRANSPOSE4_PS(Y,F,G,H);
 895             Heps             = _mm_mul_ps(vfeps,H);
 896             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 897             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 898             fvdw12           = _mm_mul_ps(c12_00,FF);
 899             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 900
 901             fscal            = fvdw;
 902
 903             /* Calculate temporary vectorial force */
 904             tx               = _mm_mul_ps(fscal,dx00);
 905             ty               = _mm_mul_ps(fscal,dy00);
 906             tz               = _mm_mul_ps(fscal,dz00);
 907
 908             /* Update vectorial force */
 909             fix0             = _mm_add_ps(fix0,tx);
 910             fiy0             = _mm_add_ps(fiy0,ty);
 911             fiz0             = _mm_add_ps(fiz0,tz);
 912
 913             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 914                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 915                                                    tx,ty,tz);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq10             = _mm_mul_ps(iq1,jq0);
 923
 924             /* COULOMB ELECTROSTATICS */
 925             velec            = _mm_mul_ps(qq10,rinv10);
 926             felec            = _mm_mul_ps(velec,rinvsq10);
 927
 928             fscal            = felec;
 929
 930             /* Calculate temporary vectorial force */
 931             tx               = _mm_mul_ps(fscal,dx10);
 932             ty               = _mm_mul_ps(fscal,dy10);
 933             tz               = _mm_mul_ps(fscal,dz10);
 934
 935             /* Update vectorial force */
 936             fix1             = _mm_add_ps(fix1,tx);
 937             fiy1             = _mm_add_ps(fiy1,ty);
 938             fiz1             = _mm_add_ps(fiz1,tz);
 939
 940             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 941                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 942                                                    tx,ty,tz);
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             qq20             = _mm_mul_ps(iq2,jq0);
 950
 951             /* COULOMB ELECTROSTATICS */
 952             velec            = _mm_mul_ps(qq20,rinv20);
 953             felec            = _mm_mul_ps(velec,rinvsq20);
 954
 955             fscal            = felec;
 956
 957             /* Calculate temporary vectorial force */
 958             tx               = _mm_mul_ps(fscal,dx20);
 959             ty               = _mm_mul_ps(fscal,dy20);
 960             tz               = _mm_mul_ps(fscal,dz20);
 961
 962             /* Update vectorial force */
 963             fix2             = _mm_add_ps(fix2,tx);
 964             fiy2             = _mm_add_ps(fiy2,ty);
 965             fiz2             = _mm_add_ps(fiz2,tz);
 966
 967             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 968                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 969                                                    tx,ty,tz);
 970
 971             /**************************
 972              * CALCULATE INTERACTIONS *
 973              **************************/
 974
 975             /* Compute parameters for interactions between i and j atoms */
 976             qq30             = _mm_mul_ps(iq3,jq0);
 977
 978             /* COULOMB ELECTROSTATICS */
 979             velec            = _mm_mul_ps(qq30,rinv30);
 980             felec            = _mm_mul_ps(velec,rinvsq30);
 981
 982             fscal            = felec;
 983
 984             /* Calculate temporary vectorial force */
 985             tx               = _mm_mul_ps(fscal,dx30);
 986             ty               = _mm_mul_ps(fscal,dy30);
 987             tz               = _mm_mul_ps(fscal,dz30);
 988
 989             /* Update vectorial force */
 990             fix3             = _mm_add_ps(fix3,tx);
 991             fiy3             = _mm_add_ps(fiy3,ty);
 992             fiz3             = _mm_add_ps(fiz3,tz);
 993
 994             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 995                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 996                                                    tx,ty,tz);
 997
 998             /* Inner loop uses 129 flops */
 999         }
1000
1001         if(jidx<j_index_end)
1002         {
1003
1004             /* Get j neighbor index, and coordinate index */
1005             jnrA             = jjnr[jidx];
1006             jnrB             = jjnr[jidx+1];
1007             jnrC             = jjnr[jidx+2];
1008             jnrD             = jjnr[jidx+3];
1009
1010             /* Sign of each element will be negative for non-real atoms.
1011              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1012              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1013              */
1014             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1015             jnrA       = (jnrA>=0) ? jnrA : 0;
1016             jnrB       = (jnrB>=0) ? jnrB : 0;
1017             jnrC       = (jnrC>=0) ? jnrC : 0;
1018             jnrD       = (jnrD>=0) ? jnrD : 0;
1019
1020             j_coord_offsetA  = DIM*jnrA;
1021             j_coord_offsetB  = DIM*jnrB;
1022             j_coord_offsetC  = DIM*jnrC;
1023             j_coord_offsetD  = DIM*jnrD;
1024
1025             /* load j atom coordinates */
1026             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1027                                               x+j_coord_offsetC,x+j_coord_offsetD,
1028                                               &jx0,&jy0,&jz0);
1029
1030             /* Calculate displacement vector */
1031             dx00             = _mm_sub_ps(ix0,jx0);
1032             dy00             = _mm_sub_ps(iy0,jy0);
1033             dz00             = _mm_sub_ps(iz0,jz0);
1034             dx10             = _mm_sub_ps(ix1,jx0);
1035             dy10             = _mm_sub_ps(iy1,jy0);
1036             dz10             = _mm_sub_ps(iz1,jz0);
1037             dx20             = _mm_sub_ps(ix2,jx0);
1038             dy20             = _mm_sub_ps(iy2,jy0);
1039             dz20             = _mm_sub_ps(iz2,jz0);
1040             dx30             = _mm_sub_ps(ix3,jx0);
1041             dy30             = _mm_sub_ps(iy3,jy0);
1042             dz30             = _mm_sub_ps(iz3,jz0);
1043
1044             /* Calculate squared distance and things based on it */
1045             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1046             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1047             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1048             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1049
1050             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1051             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1052             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1053             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1054
1055             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1056             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1057             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1058
1059             /* Load parameters for j particles */
1060             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1061                                                               charge+jnrC+0,charge+jnrD+0);
1062             vdwjidx0A        = 2*vdwtype[jnrA+0];
1063             vdwjidx0B        = 2*vdwtype[jnrB+0];
1064             vdwjidx0C        = 2*vdwtype[jnrC+0];
1065             vdwjidx0D        = 2*vdwtype[jnrD+0];
1066
1067             /**************************
1068              * CALCULATE INTERACTIONS *
1069              **************************/
1070
1071             r00              = _mm_mul_ps(rsq00,rinv00);
1072             r00              = _mm_andnot_ps(dummy_mask,r00);
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1076                                          vdwparam+vdwioffset0+vdwjidx0B,
1077                                          vdwparam+vdwioffset0+vdwjidx0C,
1078                                          vdwparam+vdwioffset0+vdwjidx0D,
1079                                          &c6_00,&c12_00);
1080
1081             /* Calculate table index by multiplying r with table scale and truncate to integer */
1082             rt               = _mm_mul_ps(r00,vftabscale);
1083             vfitab           = _mm_cvttps_epi32(rt);
1084             vfeps            = _mm_sub_ps(rt,_mm_cvtepi32_ps(vfitab));
1085             vfitab           = _mm_slli_epi32(vfitab,3);
1086
1087             /* CUBIC SPLINE TABLE DISPERSION */
1088             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1089             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1090             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1091             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1092             _MM_TRANSPOSE4_PS(Y,F,G,H);
1093             Heps             = _mm_mul_ps(vfeps,H);
1094             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1095             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1096             fvdw6            = _mm_mul_ps(c6_00,FF);
1097
1098             /* CUBIC SPLINE TABLE REPULSION */
1099             vfitab           = _mm_add_epi32(vfitab,ifour);
1100             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1101             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1102             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1103             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1104             _MM_TRANSPOSE4_PS(Y,F,G,H);
1105             Heps             = _mm_mul_ps(vfeps,H);
1106             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1107             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1108             fvdw12           = _mm_mul_ps(c12_00,FF);
1109             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1110
1111             fscal            = fvdw;
1112
1113             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1114
1115             /* Calculate temporary vectorial force */
1116             tx               = _mm_mul_ps(fscal,dx00);
1117             ty               = _mm_mul_ps(fscal,dy00);
1118             tz               = _mm_mul_ps(fscal,dz00);
1119
1120             /* Update vectorial force */
1121             fix0             = _mm_add_ps(fix0,tx);
1122             fiy0             = _mm_add_ps(fiy0,ty);
1123             fiz0             = _mm_add_ps(fiz0,tz);
1124
1125             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1126                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1127                                                    tx,ty,tz);
1128
1129             /**************************
1130              * CALCULATE INTERACTIONS *
1131              **************************/
1132
1133             /* Compute parameters for interactions between i and j atoms */
1134             qq10             = _mm_mul_ps(iq1,jq0);
1135
1136             /* COULOMB ELECTROSTATICS */
1137             velec            = _mm_mul_ps(qq10,rinv10);
1138             felec            = _mm_mul_ps(velec,rinvsq10);
1139
1140             fscal            = felec;
1141
1142             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1143
1144             /* Calculate temporary vectorial force */
1145             tx               = _mm_mul_ps(fscal,dx10);
1146             ty               = _mm_mul_ps(fscal,dy10);
1147             tz               = _mm_mul_ps(fscal,dz10);
1148
1149             /* Update vectorial force */
1150             fix1             = _mm_add_ps(fix1,tx);
1151             fiy1             = _mm_add_ps(fiy1,ty);
1152             fiz1             = _mm_add_ps(fiz1,tz);
1153
1154             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1155                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1156                                                    tx,ty,tz);
1157
1158             /**************************
1159              * CALCULATE INTERACTIONS *
1160              **************************/
1161
1162             /* Compute parameters for interactions between i and j atoms */
1163             qq20             = _mm_mul_ps(iq2,jq0);
1164
1165             /* COULOMB ELECTROSTATICS */
1166             velec            = _mm_mul_ps(qq20,rinv20);
1167             felec            = _mm_mul_ps(velec,rinvsq20);
1168
1169             fscal            = felec;
1170
1171             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1172
1173             /* Calculate temporary vectorial force */
1174             tx               = _mm_mul_ps(fscal,dx20);
1175             ty               = _mm_mul_ps(fscal,dy20);
1176             tz               = _mm_mul_ps(fscal,dz20);
1177
1178             /* Update vectorial force */
1179             fix2             = _mm_add_ps(fix2,tx);
1180             fiy2             = _mm_add_ps(fiy2,ty);
1181             fiz2             = _mm_add_ps(fiz2,tz);
1182
1183             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1184                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1185                                                    tx,ty,tz);
1186
1187             /**************************
1188              * CALCULATE INTERACTIONS *
1189              **************************/
1190
1191             /* Compute parameters for interactions between i and j atoms */
1192             qq30             = _mm_mul_ps(iq3,jq0);
1193
1194             /* COULOMB ELECTROSTATICS */
1195             velec            = _mm_mul_ps(qq30,rinv30);
1196             felec            = _mm_mul_ps(velec,rinvsq30);
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 }