src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_VF_sse4_1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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     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     /* 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     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 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
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159         fix1             = _mm_setzero_ps();
 160         fiy1             = _mm_setzero_ps();
 161         fiz1             = _mm_setzero_ps();
 162         fix2             = _mm_setzero_ps();
 163         fiy2             = _mm_setzero_ps();
 164         fiz2             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168         vvdwsum          = _mm_setzero_ps();
 169
 170         /* Start inner kernel loop */
 171         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 172         {
 173
 174             /* Get j neighbor index, and coordinate index */
 175             jnrA             = jjnr[jidx];
 176             jnrB             = jjnr[jidx+1];
 177             jnrC             = jjnr[jidx+2];
 178             jnrD             = jjnr[jidx+3];
 179             j_coord_offsetA  = DIM*jnrA;
 180             j_coord_offsetB  = DIM*jnrB;
 181             j_coord_offsetC  = DIM*jnrC;
 182             j_coord_offsetD  = DIM*jnrD;
 183
 184             /* load j atom coordinates */
 185             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 186                                               x+j_coord_offsetC,x+j_coord_offsetD,
 187                                               &jx0,&jy0,&jz0);
 188
 189             /* Calculate displacement vector */
 190             dx00             = _mm_sub_ps(ix0,jx0);
 191             dy00             = _mm_sub_ps(iy0,jy0);
 192             dz00             = _mm_sub_ps(iz0,jz0);
 193             dx10             = _mm_sub_ps(ix1,jx0);
 194             dy10             = _mm_sub_ps(iy1,jy0);
 195             dz10             = _mm_sub_ps(iz1,jz0);
 196             dx20             = _mm_sub_ps(ix2,jx0);
 197             dy20             = _mm_sub_ps(iy2,jy0);
 198             dz20             = _mm_sub_ps(iz2,jz0);
 199
 200             /* Calculate squared distance and things based on it */
 201             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 202             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 203             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 204
 205             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 206             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 207             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 208
 209             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 210             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 211             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 212
 213             /* Load parameters for j particles */
 214             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 215                                                               charge+jnrC+0,charge+jnrD+0);
 216             vdwjidx0A        = 2*vdwtype[jnrA+0];
 217             vdwjidx0B        = 2*vdwtype[jnrB+0];
 218             vdwjidx0C        = 2*vdwtype[jnrC+0];
 219             vdwjidx0D        = 2*vdwtype[jnrD+0];
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_ps(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             qq00             = _mm_mul_ps(iq0,jq0);
 229             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 230                                          vdwparam+vdwioffset0+vdwjidx0B,
 231                                          vdwparam+vdwioffset0+vdwjidx0C,
 232                                          vdwparam+vdwioffset0+vdwjidx0D,
 233                                          &c6_00,&c12_00);
 234
 235             /* Calculate table index by multiplying r with table scale and truncate to integer */
 236             rt               = _mm_mul_ps(r00,vftabscale);
 237             vfitab           = _mm_cvttps_epi32(rt);
 238             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 239             vfitab           = _mm_slli_epi32(vfitab,3);
 240
 241             /* COULOMB ELECTROSTATICS */
 242             velec            = _mm_mul_ps(qq00,rinv00);
 243             felec            = _mm_mul_ps(velec,rinvsq00);
 244
 245             /* CUBIC SPLINE TABLE DISPERSION */
 246             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 247             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 248             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 249             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 250             _MM_TRANSPOSE4_PS(Y,F,G,H);
 251             Heps             = _mm_mul_ps(vfeps,H);
 252             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 253             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 254             vvdw6            = _mm_mul_ps(c6_00,VV);
 255             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 256             fvdw6            = _mm_mul_ps(c6_00,FF);
 257
 258             /* CUBIC SPLINE TABLE REPULSION */
 259             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdw12           = _mm_mul_ps(c12_00,VV);
 269             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 270             fvdw12           = _mm_mul_ps(c12_00,FF);
 271             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 272             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 277
 278             fscal            = _mm_add_ps(felec,fvdw);
 279
 280             /* Calculate temporary vectorial force */
 281             tx               = _mm_mul_ps(fscal,dx00);
 282             ty               = _mm_mul_ps(fscal,dy00);
 283             tz               = _mm_mul_ps(fscal,dz00);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm_add_ps(fix0,tx);
 287             fiy0             = _mm_add_ps(fiy0,ty);
 288             fiz0             = _mm_add_ps(fiz0,tz);
 289
 290             fjptrA             = f+j_coord_offsetA;
 291             fjptrB             = f+j_coord_offsetB;
 292             fjptrC             = f+j_coord_offsetC;
 293             fjptrD             = f+j_coord_offsetD;
 294             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             /* Compute parameters for interactions between i and j atoms */
 301             qq10             = _mm_mul_ps(iq1,jq0);
 302
 303             /* COULOMB ELECTROSTATICS */
 304             velec            = _mm_mul_ps(qq10,rinv10);
 305             felec            = _mm_mul_ps(velec,rinvsq10);
 306
 307             /* Update potential sum for this i atom from the interaction with this j atom. */
 308             velecsum         = _mm_add_ps(velecsum,velec);
 309
 310             fscal            = felec;
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_ps(fscal,dx10);
 314             ty               = _mm_mul_ps(fscal,dy10);
 315             tz               = _mm_mul_ps(fscal,dz10);
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_add_ps(fix1,tx);
 319             fiy1             = _mm_add_ps(fiy1,ty);
 320             fiz1             = _mm_add_ps(fiz1,tz);
 321
 322             fjptrA             = f+j_coord_offsetA;
 323             fjptrB             = f+j_coord_offsetB;
 324             fjptrC             = f+j_coord_offsetC;
 325             fjptrD             = f+j_coord_offsetD;
 326             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             /* Compute parameters for interactions between i and j atoms */
 333             qq20             = _mm_mul_ps(iq2,jq0);
 334
 335             /* COULOMB ELECTROSTATICS */
 336             velec            = _mm_mul_ps(qq20,rinv20);
 337             felec            = _mm_mul_ps(velec,rinvsq20);
 338
 339             /* Update potential sum for this i atom from the interaction with this j atom. */
 340             velecsum         = _mm_add_ps(velecsum,velec);
 341
 342             fscal            = felec;
 343
 344             /* Calculate temporary vectorial force */
 345             tx               = _mm_mul_ps(fscal,dx20);
 346             ty               = _mm_mul_ps(fscal,dy20);
 347             tz               = _mm_mul_ps(fscal,dz20);
 348
 349             /* Update vectorial force */
 350             fix2             = _mm_add_ps(fix2,tx);
 351             fiy2             = _mm_add_ps(fiy2,ty);
 352             fiz2             = _mm_add_ps(fiz2,tz);
 353
 354             fjptrA             = f+j_coord_offsetA;
 355             fjptrB             = f+j_coord_offsetB;
 356             fjptrC             = f+j_coord_offsetC;
 357             fjptrD             = f+j_coord_offsetD;
 358             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 359
 360             /* Inner loop uses 119 flops */
 361         }
 362
 363         if(jidx<j_index_end)
 364         {
 365
 366             /* Get j neighbor index, and coordinate index */
 367             jnrlistA         = jjnr[jidx];
 368             jnrlistB         = jjnr[jidx+1];
 369             jnrlistC         = jjnr[jidx+2];
 370             jnrlistD         = jjnr[jidx+3];
 371             /* Sign of each element will be negative for non-real atoms.
 372              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 373              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 374              */
 375             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 376             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 377             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 378             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 379             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 380             j_coord_offsetA  = DIM*jnrA;
 381             j_coord_offsetB  = DIM*jnrB;
 382             j_coord_offsetC  = DIM*jnrC;
 383             j_coord_offsetD  = DIM*jnrD;
 384
 385             /* load j atom coordinates */
 386             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 387                                               x+j_coord_offsetC,x+j_coord_offsetD,
 388                                               &jx0,&jy0,&jz0);
 389
 390             /* Calculate displacement vector */
 391             dx00             = _mm_sub_ps(ix0,jx0);
 392             dy00             = _mm_sub_ps(iy0,jy0);
 393             dz00             = _mm_sub_ps(iz0,jz0);
 394             dx10             = _mm_sub_ps(ix1,jx0);
 395             dy10             = _mm_sub_ps(iy1,jy0);
 396             dz10             = _mm_sub_ps(iz1,jz0);
 397             dx20             = _mm_sub_ps(ix2,jx0);
 398             dy20             = _mm_sub_ps(iy2,jy0);
 399             dz20             = _mm_sub_ps(iz2,jz0);
 400
 401             /* Calculate squared distance and things based on it */
 402             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 403             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 404             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 405
 406             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 407             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 408             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 409
 410             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 411             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 412             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 413
 414             /* Load parameters for j particles */
 415             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 416                                                               charge+jnrC+0,charge+jnrD+0);
 417             vdwjidx0A        = 2*vdwtype[jnrA+0];
 418             vdwjidx0B        = 2*vdwtype[jnrB+0];
 419             vdwjidx0C        = 2*vdwtype[jnrC+0];
 420             vdwjidx0D        = 2*vdwtype[jnrD+0];
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r00              = _mm_mul_ps(rsq00,rinv00);
 427             r00              = _mm_andnot_ps(dummy_mask,r00);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq00             = _mm_mul_ps(iq0,jq0);
 431             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 432                                          vdwparam+vdwioffset0+vdwjidx0B,
 433                                          vdwparam+vdwioffset0+vdwjidx0C,
 434                                          vdwparam+vdwioffset0+vdwjidx0D,
 435                                          &c6_00,&c12_00);
 436
 437             /* Calculate table index by multiplying r with table scale and truncate to integer */
 438             rt               = _mm_mul_ps(r00,vftabscale);
 439             vfitab           = _mm_cvttps_epi32(rt);
 440             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 441             vfitab           = _mm_slli_epi32(vfitab,3);
 442
 443             /* COULOMB ELECTROSTATICS */
 444             velec            = _mm_mul_ps(qq00,rinv00);
 445             felec            = _mm_mul_ps(velec,rinvsq00);
 446
 447             /* CUBIC SPLINE TABLE DISPERSION */
 448             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 449             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 450             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 451             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 452             _MM_TRANSPOSE4_PS(Y,F,G,H);
 453             Heps             = _mm_mul_ps(vfeps,H);
 454             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 455             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 456             vvdw6            = _mm_mul_ps(c6_00,VV);
 457             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 458             fvdw6            = _mm_mul_ps(c6_00,FF);
 459
 460             /* CUBIC SPLINE TABLE REPULSION */
 461             vfitab           = _mm_add_epi32(vfitab,ifour);
 462             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 463             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 464             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 465             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 466             _MM_TRANSPOSE4_PS(Y,F,G,H);
 467             Heps             = _mm_mul_ps(vfeps,H);
 468             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 469             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 470             vvdw12           = _mm_mul_ps(c12_00,VV);
 471             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 472             fvdw12           = _mm_mul_ps(c12_00,FF);
 473             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 474             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 475
 476             /* Update potential sum for this i atom from the interaction with this j atom. */
 477             velec            = _mm_andnot_ps(dummy_mask,velec);
 478             velecsum         = _mm_add_ps(velecsum,velec);
 479             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 480             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 481
 482             fscal            = _mm_add_ps(felec,fvdw);
 483
 484             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 485
 486             /* Calculate temporary vectorial force */
 487             tx               = _mm_mul_ps(fscal,dx00);
 488             ty               = _mm_mul_ps(fscal,dy00);
 489             tz               = _mm_mul_ps(fscal,dz00);
 490
 491             /* Update vectorial force */
 492             fix0             = _mm_add_ps(fix0,tx);
 493             fiy0             = _mm_add_ps(fiy0,ty);
 494             fiz0             = _mm_add_ps(fiz0,tz);
 495
 496             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 497             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 498             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 499             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 500             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             /* Compute parameters for interactions between i and j atoms */
 507             qq10             = _mm_mul_ps(iq1,jq0);
 508
 509             /* COULOMB ELECTROSTATICS */
 510             velec            = _mm_mul_ps(qq10,rinv10);
 511             felec            = _mm_mul_ps(velec,rinvsq10);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm_andnot_ps(dummy_mask,velec);
 515             velecsum         = _mm_add_ps(velecsum,velec);
 516
 517             fscal            = felec;
 518
 519             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 520
 521             /* Calculate temporary vectorial force */
 522             tx               = _mm_mul_ps(fscal,dx10);
 523             ty               = _mm_mul_ps(fscal,dy10);
 524             tz               = _mm_mul_ps(fscal,dz10);
 525
 526             /* Update vectorial force */
 527             fix1             = _mm_add_ps(fix1,tx);
 528             fiy1             = _mm_add_ps(fiy1,ty);
 529             fiz1             = _mm_add_ps(fiz1,tz);
 530
 531             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 532             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 533             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 534             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 535             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 536
 537             /**************************
 538              * CALCULATE INTERACTIONS *
 539              **************************/
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq20             = _mm_mul_ps(iq2,jq0);
 543
 544             /* COULOMB ELECTROSTATICS */
 545             velec            = _mm_mul_ps(qq20,rinv20);
 546             felec            = _mm_mul_ps(velec,rinvsq20);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556             /* Calculate temporary vectorial force */
 557             tx               = _mm_mul_ps(fscal,dx20);
 558             ty               = _mm_mul_ps(fscal,dy20);
 559             tz               = _mm_mul_ps(fscal,dz20);
 560
 561             /* Update vectorial force */
 562             fix2             = _mm_add_ps(fix2,tx);
 563             fiy2             = _mm_add_ps(fiy2,ty);
 564             fiz2             = _mm_add_ps(fiz2,tz);
 565
 566             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 567             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 568             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 569             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 570             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 571
 572             /* Inner loop uses 120 flops */
 573         }
 574
 575         /* End of innermost loop */
 576
 577         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 578                                               f+i_coord_offset,fshift+i_shift_offset);
 579
 580         ggid                        = gid[iidx];
 581         /* Update potential energies */
 582         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 583         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 584
 585         /* Increment number of inner iterations */
 586         inneriter                  += j_index_end - j_index_start;
 587
 588         /* Outer loop uses 20 flops */
 589     }
 590
 591     /* Increment number of outer iterations */
 592     outeriter        += nri;
 593
 594     /* Update outer/inner flops */
 595
 596     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*120);
 597 }
 598 /*
 599  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 600  * Electrostatics interaction: Coulomb
 601  * VdW interaction:            CubicSplineTable
 602  * Geometry:                   Water3-Particle
 603  * Calculate force/pot:        Force
 604  */
 605 void
 606 nb_kernel_ElecCoul_VdwCSTab_GeomW3P1_F_sse4_1_single
 607                     (t_nblist * gmx_restrict                nlist,
 608                      rvec * gmx_restrict                    xx,
 609                      rvec * gmx_restrict                    ff,
 610                      t_forcerec * gmx_restrict              fr,
 611                      t_mdatoms * gmx_restrict               mdatoms,
 612                      nb_kernel_data_t * gmx_restrict        kernel_data,
 613                      t_nrnb * gmx_restrict                  nrnb)
 614 {
 615     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 616      * just 0 for non-waters.
 617      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 618      * jnr indices corresponding to data put in the four positions in the SIMD register.
 619      */
 620     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 621     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 622     int              jnrA,jnrB,jnrC,jnrD;
 623     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 624     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 625     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 626     real             rcutoff_scalar;
 627     real             *shiftvec,*fshift,*x,*f;
 628     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 629     real             scratch[4*DIM];
 630     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 631     int              vdwioffset0;
 632     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 633     int              vdwioffset1;
 634     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 635     int              vdwioffset2;
 636     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 637     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 638     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 639     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 640     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 641     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 642     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 643     real             *charge;
 644     int              nvdwtype;
 645     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 646     int              *vdwtype;
 647     real             *vdwparam;
 648     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 649     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 650     __m128i          vfitab;
 651     __m128i          ifour       = _mm_set1_epi32(4);
 652     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 653     real             *vftab;
 654     __m128           dummy_mask,cutoff_mask;
 655     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 656     __m128           one     = _mm_set1_ps(1.0);
 657     __m128           two     = _mm_set1_ps(2.0);
 658     x                = xx[0];
 659     f                = ff[0];
 660
 661     nri              = nlist->nri;
 662     iinr             = nlist->iinr;
 663     jindex           = nlist->jindex;
 664     jjnr             = nlist->jjnr;
 665     shiftidx         = nlist->shift;
 666     gid              = nlist->gid;
 667     shiftvec         = fr->shift_vec[0];
 668     fshift           = fr->fshift[0];
 669     facel            = _mm_set1_ps(fr->epsfac);
 670     charge           = mdatoms->chargeA;
 671     nvdwtype         = fr->ntype;
 672     vdwparam         = fr->nbfp;
 673     vdwtype          = mdatoms->typeA;
 674
 675     vftab            = kernel_data->table_vdw->data;
 676     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 677
 678     /* Setup water-specific parameters */
 679     inr              = nlist->iinr[0];
 680     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 681     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 682     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 683     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 684
 685     /* Avoid stupid compiler warnings */
 686     jnrA = jnrB = jnrC = jnrD = 0;
 687     j_coord_offsetA = 0;
 688     j_coord_offsetB = 0;
 689     j_coord_offsetC = 0;
 690     j_coord_offsetD = 0;
 691
 692     outeriter        = 0;
 693     inneriter        = 0;
 694
 695     for(iidx=0;iidx<4*DIM;iidx++)
 696     {
 697         scratch[iidx] = 0.0;
 698     }
 699
 700     /* Start outer loop over neighborlists */
 701     for(iidx=0; iidx<nri; iidx++)
 702     {
 703         /* Load shift vector for this list */
 704         i_shift_offset   = DIM*shiftidx[iidx];
 705
 706         /* Load limits for loop over neighbors */
 707         j_index_start    = jindex[iidx];
 708         j_index_end      = jindex[iidx+1];
 709
 710         /* Get outer coordinate index */
 711         inr              = iinr[iidx];
 712         i_coord_offset   = DIM*inr;
 713
 714         /* Load i particle coords and add shift vector */
 715         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 716                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 717
 718         fix0             = _mm_setzero_ps();
 719         fiy0             = _mm_setzero_ps();
 720         fiz0             = _mm_setzero_ps();
 721         fix1             = _mm_setzero_ps();
 722         fiy1             = _mm_setzero_ps();
 723         fiz1             = _mm_setzero_ps();
 724         fix2             = _mm_setzero_ps();
 725         fiy2             = _mm_setzero_ps();
 726         fiz2             = _mm_setzero_ps();
 727
 728         /* Start inner kernel loop */
 729         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 730         {
 731
 732             /* Get j neighbor index, and coordinate index */
 733             jnrA             = jjnr[jidx];
 734             jnrB             = jjnr[jidx+1];
 735             jnrC             = jjnr[jidx+2];
 736             jnrD             = jjnr[jidx+3];
 737             j_coord_offsetA  = DIM*jnrA;
 738             j_coord_offsetB  = DIM*jnrB;
 739             j_coord_offsetC  = DIM*jnrC;
 740             j_coord_offsetD  = DIM*jnrD;
 741
 742             /* load j atom coordinates */
 743             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 744                                               x+j_coord_offsetC,x+j_coord_offsetD,
 745                                               &jx0,&jy0,&jz0);
 746
 747             /* Calculate displacement vector */
 748             dx00             = _mm_sub_ps(ix0,jx0);
 749             dy00             = _mm_sub_ps(iy0,jy0);
 750             dz00             = _mm_sub_ps(iz0,jz0);
 751             dx10             = _mm_sub_ps(ix1,jx0);
 752             dy10             = _mm_sub_ps(iy1,jy0);
 753             dz10             = _mm_sub_ps(iz1,jz0);
 754             dx20             = _mm_sub_ps(ix2,jx0);
 755             dy20             = _mm_sub_ps(iy2,jy0);
 756             dz20             = _mm_sub_ps(iz2,jz0);
 757
 758             /* Calculate squared distance and things based on it */
 759             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 760             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 761             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 762
 763             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 764             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 765             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 766
 767             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 768             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 769             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 770
 771             /* Load parameters for j particles */
 772             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 773                                                               charge+jnrC+0,charge+jnrD+0);
 774             vdwjidx0A        = 2*vdwtype[jnrA+0];
 775             vdwjidx0B        = 2*vdwtype[jnrB+0];
 776             vdwjidx0C        = 2*vdwtype[jnrC+0];
 777             vdwjidx0D        = 2*vdwtype[jnrD+0];
 778
 779             /**************************
 780              * CALCULATE INTERACTIONS *
 781              **************************/
 782
 783             r00              = _mm_mul_ps(rsq00,rinv00);
 784
 785             /* Compute parameters for interactions between i and j atoms */
 786             qq00             = _mm_mul_ps(iq0,jq0);
 787             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 788                                          vdwparam+vdwioffset0+vdwjidx0B,
 789                                          vdwparam+vdwioffset0+vdwjidx0C,
 790                                          vdwparam+vdwioffset0+vdwjidx0D,
 791                                          &c6_00,&c12_00);
 792
 793             /* Calculate table index by multiplying r with table scale and truncate to integer */
 794             rt               = _mm_mul_ps(r00,vftabscale);
 795             vfitab           = _mm_cvttps_epi32(rt);
 796             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 797             vfitab           = _mm_slli_epi32(vfitab,3);
 798
 799             /* COULOMB ELECTROSTATICS */
 800             velec            = _mm_mul_ps(qq00,rinv00);
 801             felec            = _mm_mul_ps(velec,rinvsq00);
 802
 803             /* CUBIC SPLINE TABLE DISPERSION */
 804             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 805             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 806             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 807             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 808             _MM_TRANSPOSE4_PS(Y,F,G,H);
 809             Heps             = _mm_mul_ps(vfeps,H);
 810             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 811             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 812             fvdw6            = _mm_mul_ps(c6_00,FF);
 813
 814             /* CUBIC SPLINE TABLE REPULSION */
 815             vfitab           = _mm_add_epi32(vfitab,ifour);
 816             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 817             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 818             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 819             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 820             _MM_TRANSPOSE4_PS(Y,F,G,H);
 821             Heps             = _mm_mul_ps(vfeps,H);
 822             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 823             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 824             fvdw12           = _mm_mul_ps(c12_00,FF);
 825             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 826
 827             fscal            = _mm_add_ps(felec,fvdw);
 828
 829             /* Calculate temporary vectorial force */
 830             tx               = _mm_mul_ps(fscal,dx00);
 831             ty               = _mm_mul_ps(fscal,dy00);
 832             tz               = _mm_mul_ps(fscal,dz00);
 833
 834             /* Update vectorial force */
 835             fix0             = _mm_add_ps(fix0,tx);
 836             fiy0             = _mm_add_ps(fiy0,ty);
 837             fiz0             = _mm_add_ps(fiz0,tz);
 838
 839             fjptrA             = f+j_coord_offsetA;
 840             fjptrB             = f+j_coord_offsetB;
 841             fjptrC             = f+j_coord_offsetC;
 842             fjptrD             = f+j_coord_offsetD;
 843             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq10             = _mm_mul_ps(iq1,jq0);
 851
 852             /* COULOMB ELECTROSTATICS */
 853             velec            = _mm_mul_ps(qq10,rinv10);
 854             felec            = _mm_mul_ps(velec,rinvsq10);
 855
 856             fscal            = felec;
 857
 858             /* Calculate temporary vectorial force */
 859             tx               = _mm_mul_ps(fscal,dx10);
 860             ty               = _mm_mul_ps(fscal,dy10);
 861             tz               = _mm_mul_ps(fscal,dz10);
 862
 863             /* Update vectorial force */
 864             fix1             = _mm_add_ps(fix1,tx);
 865             fiy1             = _mm_add_ps(fiy1,ty);
 866             fiz1             = _mm_add_ps(fiz1,tz);
 867
 868             fjptrA             = f+j_coord_offsetA;
 869             fjptrB             = f+j_coord_offsetB;
 870             fjptrC             = f+j_coord_offsetC;
 871             fjptrD             = f+j_coord_offsetD;
 872             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 873
 874             /**************************
 875              * CALCULATE INTERACTIONS *
 876              **************************/
 877
 878             /* Compute parameters for interactions between i and j atoms */
 879             qq20             = _mm_mul_ps(iq2,jq0);
 880
 881             /* COULOMB ELECTROSTATICS */
 882             velec            = _mm_mul_ps(qq20,rinv20);
 883             felec            = _mm_mul_ps(velec,rinvsq20);
 884
 885             fscal            = felec;
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_ps(fscal,dx20);
 889             ty               = _mm_mul_ps(fscal,dy20);
 890             tz               = _mm_mul_ps(fscal,dz20);
 891
 892             /* Update vectorial force */
 893             fix2             = _mm_add_ps(fix2,tx);
 894             fiy2             = _mm_add_ps(fiy2,ty);
 895             fiz2             = _mm_add_ps(fiz2,tz);
 896
 897             fjptrA             = f+j_coord_offsetA;
 898             fjptrB             = f+j_coord_offsetB;
 899             fjptrC             = f+j_coord_offsetC;
 900             fjptrD             = f+j_coord_offsetD;
 901             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 902
 903             /* Inner loop uses 108 flops */
 904         }
 905
 906         if(jidx<j_index_end)
 907         {
 908
 909             /* Get j neighbor index, and coordinate index */
 910             jnrlistA         = jjnr[jidx];
 911             jnrlistB         = jjnr[jidx+1];
 912             jnrlistC         = jjnr[jidx+2];
 913             jnrlistD         = jjnr[jidx+3];
 914             /* Sign of each element will be negative for non-real atoms.
 915              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 916              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 917              */
 918             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 919             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 920             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 921             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 922             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 923             j_coord_offsetA  = DIM*jnrA;
 924             j_coord_offsetB  = DIM*jnrB;
 925             j_coord_offsetC  = DIM*jnrC;
 926             j_coord_offsetD  = DIM*jnrD;
 927
 928             /* load j atom coordinates */
 929             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 930                                               x+j_coord_offsetC,x+j_coord_offsetD,
 931                                               &jx0,&jy0,&jz0);
 932
 933             /* Calculate displacement vector */
 934             dx00             = _mm_sub_ps(ix0,jx0);
 935             dy00             = _mm_sub_ps(iy0,jy0);
 936             dz00             = _mm_sub_ps(iz0,jz0);
 937             dx10             = _mm_sub_ps(ix1,jx0);
 938             dy10             = _mm_sub_ps(iy1,jy0);
 939             dz10             = _mm_sub_ps(iz1,jz0);
 940             dx20             = _mm_sub_ps(ix2,jx0);
 941             dy20             = _mm_sub_ps(iy2,jy0);
 942             dz20             = _mm_sub_ps(iz2,jz0);
 943
 944             /* Calculate squared distance and things based on it */
 945             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 946             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 947             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 948
 949             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 950             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 951             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 952
 953             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 954             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 955             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 956
 957             /* Load parameters for j particles */
 958             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 959                                                               charge+jnrC+0,charge+jnrD+0);
 960             vdwjidx0A        = 2*vdwtype[jnrA+0];
 961             vdwjidx0B        = 2*vdwtype[jnrB+0];
 962             vdwjidx0C        = 2*vdwtype[jnrC+0];
 963             vdwjidx0D        = 2*vdwtype[jnrD+0];
 964
 965             /**************************
 966              * CALCULATE INTERACTIONS *
 967              **************************/
 968
 969             r00              = _mm_mul_ps(rsq00,rinv00);
 970             r00              = _mm_andnot_ps(dummy_mask,r00);
 971
 972             /* Compute parameters for interactions between i and j atoms */
 973             qq00             = _mm_mul_ps(iq0,jq0);
 974             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 975                                          vdwparam+vdwioffset0+vdwjidx0B,
 976                                          vdwparam+vdwioffset0+vdwjidx0C,
 977                                          vdwparam+vdwioffset0+vdwjidx0D,
 978                                          &c6_00,&c12_00);
 979
 980             /* Calculate table index by multiplying r with table scale and truncate to integer */
 981             rt               = _mm_mul_ps(r00,vftabscale);
 982             vfitab           = _mm_cvttps_epi32(rt);
 983             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 984             vfitab           = _mm_slli_epi32(vfitab,3);
 985
 986             /* COULOMB ELECTROSTATICS */
 987             velec            = _mm_mul_ps(qq00,rinv00);
 988             felec            = _mm_mul_ps(velec,rinvsq00);
 989
 990             /* CUBIC SPLINE TABLE DISPERSION */
 991             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 992             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 993             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 994             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 995             _MM_TRANSPOSE4_PS(Y,F,G,H);
 996             Heps             = _mm_mul_ps(vfeps,H);
 997             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 998             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 999             fvdw6            = _mm_mul_ps(c6_00,FF);
1000
1001             /* CUBIC SPLINE TABLE REPULSION */
1002             vfitab           = _mm_add_epi32(vfitab,ifour);
1003             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1004             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1005             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1006             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1007             _MM_TRANSPOSE4_PS(Y,F,G,H);
1008             Heps             = _mm_mul_ps(vfeps,H);
1009             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1010             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1011             fvdw12           = _mm_mul_ps(c12_00,FF);
1012             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1013
1014             fscal            = _mm_add_ps(felec,fvdw);
1015
1016             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm_mul_ps(fscal,dx00);
1020             ty               = _mm_mul_ps(fscal,dy00);
1021             tz               = _mm_mul_ps(fscal,dz00);
1022
1023             /* Update vectorial force */
1024             fix0             = _mm_add_ps(fix0,tx);
1025             fiy0             = _mm_add_ps(fiy0,ty);
1026             fiz0             = _mm_add_ps(fiz0,tz);
1027
1028             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1029             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1030             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1031             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1032             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             /* Compute parameters for interactions between i and j atoms */
1039             qq10             = _mm_mul_ps(iq1,jq0);
1040
1041             /* COULOMB ELECTROSTATICS */
1042             velec            = _mm_mul_ps(qq10,rinv10);
1043             felec            = _mm_mul_ps(velec,rinvsq10);
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm_mul_ps(fscal,dx10);
1051             ty               = _mm_mul_ps(fscal,dy10);
1052             tz               = _mm_mul_ps(fscal,dz10);
1053
1054             /* Update vectorial force */
1055             fix1             = _mm_add_ps(fix1,tx);
1056             fiy1             = _mm_add_ps(fiy1,ty);
1057             fiz1             = _mm_add_ps(fiz1,tz);
1058
1059             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1060             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1061             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1062             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1063             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq20             = _mm_mul_ps(iq2,jq0);
1071
1072             /* COULOMB ELECTROSTATICS */
1073             velec            = _mm_mul_ps(qq20,rinv20);
1074             felec            = _mm_mul_ps(velec,rinvsq20);
1075
1076             fscal            = felec;
1077
1078             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1079
1080             /* Calculate temporary vectorial force */
1081             tx               = _mm_mul_ps(fscal,dx20);
1082             ty               = _mm_mul_ps(fscal,dy20);
1083             tz               = _mm_mul_ps(fscal,dz20);
1084
1085             /* Update vectorial force */
1086             fix2             = _mm_add_ps(fix2,tx);
1087             fiy2             = _mm_add_ps(fiy2,ty);
1088             fiz2             = _mm_add_ps(fiz2,tz);
1089
1090             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1091             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1092             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1093             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1094             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1095
1096             /* Inner loop uses 109 flops */
1097         }
1098
1099         /* End of innermost loop */
1100
1101         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1102                                               f+i_coord_offset,fshift+i_shift_offset);
1103
1104         /* Increment number of inner iterations */
1105         inneriter                  += j_index_end - j_index_start;
1106
1107         /* Outer loop uses 18 flops */
1108     }
1109
1110     /* Increment number of outer iterations */
1111     outeriter        += nri;
1112
1113     /* Update outer/inner flops */
1114
1115     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*109);
1116 }