src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwNone_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_ElecCSTab_VdwNone_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwNone_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     __m128i          vfitab;
  83     __m128i          ifour       = _mm_set1_epi32(4);
  84     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
  85     real             *vftab;
  86     __m128           dummy_mask,cutoff_mask;
  87     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  88     __m128           one     = _mm_set1_ps(1.0);
  89     __m128           two     = _mm_set1_ps(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_ps(fr->epsfac);
 102     charge           = mdatoms->chargeA;
 103
 104     vftab            = kernel_data->table_elec->data;
 105     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 106
 107     /* Setup water-specific parameters */
 108     inr              = nlist->iinr[0];
 109     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 110     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 111     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = jnrC = jnrD = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117     j_coord_offsetC = 0;
 118     j_coord_offsetD = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     for(iidx=0;iidx<4*DIM;iidx++)
 124     {
 125         scratch[iidx] = 0.0;
 126     }
 127
 128     /* Start outer loop over neighborlists */
 129     for(iidx=0; iidx<nri; iidx++)
 130     {
 131         /* Load shift vector for this list */
 132         i_shift_offset   = DIM*shiftidx[iidx];
 133
 134         /* Load limits for loop over neighbors */
 135         j_index_start    = jindex[iidx];
 136         j_index_end      = jindex[iidx+1];
 137
 138         /* Get outer coordinate index */
 139         inr              = iinr[iidx];
 140         i_coord_offset   = DIM*inr;
 141
 142         /* Load i particle coords and add shift vector */
 143         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 144                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 145
 146         fix0             = _mm_setzero_ps();
 147         fiy0             = _mm_setzero_ps();
 148         fiz0             = _mm_setzero_ps();
 149         fix1             = _mm_setzero_ps();
 150         fiy1             = _mm_setzero_ps();
 151         fiz1             = _mm_setzero_ps();
 152         fix2             = _mm_setzero_ps();
 153         fiy2             = _mm_setzero_ps();
 154         fiz2             = _mm_setzero_ps();
 155
 156         /* Reset potential sums */
 157         velecsum         = _mm_setzero_ps();
 158
 159         /* Start inner kernel loop */
 160         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 161         {
 162
 163             /* Get j neighbor index, and coordinate index */
 164             jnrA             = jjnr[jidx];
 165             jnrB             = jjnr[jidx+1];
 166             jnrC             = jjnr[jidx+2];
 167             jnrD             = jjnr[jidx+3];
 168             j_coord_offsetA  = DIM*jnrA;
 169             j_coord_offsetB  = DIM*jnrB;
 170             j_coord_offsetC  = DIM*jnrC;
 171             j_coord_offsetD  = DIM*jnrD;
 172
 173             /* load j atom coordinates */
 174             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 175                                               x+j_coord_offsetC,x+j_coord_offsetD,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_ps(ix0,jx0);
 180             dy00             = _mm_sub_ps(iy0,jy0);
 181             dz00             = _mm_sub_ps(iz0,jz0);
 182             dx10             = _mm_sub_ps(ix1,jx0);
 183             dy10             = _mm_sub_ps(iy1,jy0);
 184             dz10             = _mm_sub_ps(iz1,jz0);
 185             dx20             = _mm_sub_ps(ix2,jx0);
 186             dy20             = _mm_sub_ps(iy2,jy0);
 187             dz20             = _mm_sub_ps(iz2,jz0);
 188
 189             /* Calculate squared distance and things based on it */
 190             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 191             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 192             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 193
 194             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 195             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 196             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201
 202             fjx0             = _mm_setzero_ps();
 203             fjy0             = _mm_setzero_ps();
 204             fjz0             = _mm_setzero_ps();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             r00              = _mm_mul_ps(rsq00,rinv00);
 211
 212             /* Compute parameters for interactions between i and j atoms */
 213             qq00             = _mm_mul_ps(iq0,jq0);
 214
 215             /* Calculate table index by multiplying r with table scale and truncate to integer */
 216             rt               = _mm_mul_ps(r00,vftabscale);
 217             vfitab           = _mm_cvttps_epi32(rt);
 218             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 219             vfitab           = _mm_slli_epi32(vfitab,2);
 220
 221             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 222             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 223             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 224             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 225             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 226             _MM_TRANSPOSE4_PS(Y,F,G,H);
 227             Heps             = _mm_mul_ps(vfeps,H);
 228             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 229             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 230             velec            = _mm_mul_ps(qq00,VV);
 231             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 232             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 233
 234             /* Update potential sum for this i atom from the interaction with this j atom. */
 235             velecsum         = _mm_add_ps(velecsum,velec);
 236
 237             fscal            = felec;
 238
 239             /* Calculate temporary vectorial force */
 240             tx               = _mm_mul_ps(fscal,dx00);
 241             ty               = _mm_mul_ps(fscal,dy00);
 242             tz               = _mm_mul_ps(fscal,dz00);
 243
 244             /* Update vectorial force */
 245             fix0             = _mm_add_ps(fix0,tx);
 246             fiy0             = _mm_add_ps(fiy0,ty);
 247             fiz0             = _mm_add_ps(fiz0,tz);
 248
 249             fjx0             = _mm_add_ps(fjx0,tx);
 250             fjy0             = _mm_add_ps(fjy0,ty);
 251             fjz0             = _mm_add_ps(fjz0,tz);
 252
 253             /**************************
 254              * CALCULATE INTERACTIONS *
 255              **************************/
 256
 257             r10              = _mm_mul_ps(rsq10,rinv10);
 258
 259             /* Compute parameters for interactions between i and j atoms */
 260             qq10             = _mm_mul_ps(iq1,jq0);
 261
 262             /* Calculate table index by multiplying r with table scale and truncate to integer */
 263             rt               = _mm_mul_ps(r10,vftabscale);
 264             vfitab           = _mm_cvttps_epi32(rt);
 265             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 266             vfitab           = _mm_slli_epi32(vfitab,2);
 267
 268             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 269             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 270             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 271             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 272             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 273             _MM_TRANSPOSE4_PS(Y,F,G,H);
 274             Heps             = _mm_mul_ps(vfeps,H);
 275             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 276             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 277             velec            = _mm_mul_ps(qq10,VV);
 278             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 279             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx10);
 288             ty               = _mm_mul_ps(fscal,dy10);
 289             tz               = _mm_mul_ps(fscal,dz10);
 290
 291             /* Update vectorial force */
 292             fix1             = _mm_add_ps(fix1,tx);
 293             fiy1             = _mm_add_ps(fiy1,ty);
 294             fiz1             = _mm_add_ps(fiz1,tz);
 295
 296             fjx0             = _mm_add_ps(fjx0,tx);
 297             fjy0             = _mm_add_ps(fjy0,ty);
 298             fjz0             = _mm_add_ps(fjz0,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r20              = _mm_mul_ps(rsq20,rinv20);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq20             = _mm_mul_ps(iq2,jq0);
 308
 309             /* Calculate table index by multiplying r with table scale and truncate to integer */
 310             rt               = _mm_mul_ps(r20,vftabscale);
 311             vfitab           = _mm_cvttps_epi32(rt);
 312             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 313             vfitab           = _mm_slli_epi32(vfitab,2);
 314
 315             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 316             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 317             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 318             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 319             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 320             _MM_TRANSPOSE4_PS(Y,F,G,H);
 321             Heps             = _mm_mul_ps(vfeps,H);
 322             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 323             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 324             velec            = _mm_mul_ps(qq20,VV);
 325             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 326             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 327
 328             /* Update potential sum for this i atom from the interaction with this j atom. */
 329             velecsum         = _mm_add_ps(velecsum,velec);
 330
 331             fscal            = felec;
 332
 333             /* Calculate temporary vectorial force */
 334             tx               = _mm_mul_ps(fscal,dx20);
 335             ty               = _mm_mul_ps(fscal,dy20);
 336             tz               = _mm_mul_ps(fscal,dz20);
 337
 338             /* Update vectorial force */
 339             fix2             = _mm_add_ps(fix2,tx);
 340             fiy2             = _mm_add_ps(fiy2,ty);
 341             fiz2             = _mm_add_ps(fiz2,tz);
 342
 343             fjx0             = _mm_add_ps(fjx0,tx);
 344             fjy0             = _mm_add_ps(fjy0,ty);
 345             fjz0             = _mm_add_ps(fjz0,tz);
 346
 347             fjptrA             = f+j_coord_offsetA;
 348             fjptrB             = f+j_coord_offsetB;
 349             fjptrC             = f+j_coord_offsetC;
 350             fjptrD             = f+j_coord_offsetD;
 351
 352             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 353
 354             /* Inner loop uses 129 flops */
 355         }
 356
 357         if(jidx<j_index_end)
 358         {
 359
 360             /* Get j neighbor index, and coordinate index */
 361             jnrlistA         = jjnr[jidx];
 362             jnrlistB         = jjnr[jidx+1];
 363             jnrlistC         = jjnr[jidx+2];
 364             jnrlistD         = jjnr[jidx+3];
 365             /* Sign of each element will be negative for non-real atoms.
 366              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 367              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 368              */
 369             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 370             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 371             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 372             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 373             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 374             j_coord_offsetA  = DIM*jnrA;
 375             j_coord_offsetB  = DIM*jnrB;
 376             j_coord_offsetC  = DIM*jnrC;
 377             j_coord_offsetD  = DIM*jnrD;
 378
 379             /* load j atom coordinates */
 380             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 381                                               x+j_coord_offsetC,x+j_coord_offsetD,
 382                                               &jx0,&jy0,&jz0);
 383
 384             /* Calculate displacement vector */
 385             dx00             = _mm_sub_ps(ix0,jx0);
 386             dy00             = _mm_sub_ps(iy0,jy0);
 387             dz00             = _mm_sub_ps(iz0,jz0);
 388             dx10             = _mm_sub_ps(ix1,jx0);
 389             dy10             = _mm_sub_ps(iy1,jy0);
 390             dz10             = _mm_sub_ps(iz1,jz0);
 391             dx20             = _mm_sub_ps(ix2,jx0);
 392             dy20             = _mm_sub_ps(iy2,jy0);
 393             dz20             = _mm_sub_ps(iz2,jz0);
 394
 395             /* Calculate squared distance and things based on it */
 396             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 397             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 398             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 399
 400             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 401             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 402             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 403
 404             /* Load parameters for j particles */
 405             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 406                                                               charge+jnrC+0,charge+jnrD+0);
 407
 408             fjx0             = _mm_setzero_ps();
 409             fjy0             = _mm_setzero_ps();
 410             fjz0             = _mm_setzero_ps();
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             r00              = _mm_mul_ps(rsq00,rinv00);
 417             r00              = _mm_andnot_ps(dummy_mask,r00);
 418
 419             /* Compute parameters for interactions between i and j atoms */
 420             qq00             = _mm_mul_ps(iq0,jq0);
 421
 422             /* Calculate table index by multiplying r with table scale and truncate to integer */
 423             rt               = _mm_mul_ps(r00,vftabscale);
 424             vfitab           = _mm_cvttps_epi32(rt);
 425             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 426             vfitab           = _mm_slli_epi32(vfitab,2);
 427
 428             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 429             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 430             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 431             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 432             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 433             _MM_TRANSPOSE4_PS(Y,F,G,H);
 434             Heps             = _mm_mul_ps(vfeps,H);
 435             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 436             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 437             velec            = _mm_mul_ps(qq00,VV);
 438             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 439             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 440
 441             /* Update potential sum for this i atom from the interaction with this j atom. */
 442             velec            = _mm_andnot_ps(dummy_mask,velec);
 443             velecsum         = _mm_add_ps(velecsum,velec);
 444
 445             fscal            = felec;
 446
 447             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 448
 449             /* Calculate temporary vectorial force */
 450             tx               = _mm_mul_ps(fscal,dx00);
 451             ty               = _mm_mul_ps(fscal,dy00);
 452             tz               = _mm_mul_ps(fscal,dz00);
 453
 454             /* Update vectorial force */
 455             fix0             = _mm_add_ps(fix0,tx);
 456             fiy0             = _mm_add_ps(fiy0,ty);
 457             fiz0             = _mm_add_ps(fiz0,tz);
 458
 459             fjx0             = _mm_add_ps(fjx0,tx);
 460             fjy0             = _mm_add_ps(fjy0,ty);
 461             fjz0             = _mm_add_ps(fjz0,tz);
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             r10              = _mm_mul_ps(rsq10,rinv10);
 468             r10              = _mm_andnot_ps(dummy_mask,r10);
 469
 470             /* Compute parameters for interactions between i and j atoms */
 471             qq10             = _mm_mul_ps(iq1,jq0);
 472
 473             /* Calculate table index by multiplying r with table scale and truncate to integer */
 474             rt               = _mm_mul_ps(r10,vftabscale);
 475             vfitab           = _mm_cvttps_epi32(rt);
 476             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 477             vfitab           = _mm_slli_epi32(vfitab,2);
 478
 479             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 480             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 481             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 482             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 483             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 484             _MM_TRANSPOSE4_PS(Y,F,G,H);
 485             Heps             = _mm_mul_ps(vfeps,H);
 486             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 487             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 488             velec            = _mm_mul_ps(qq10,VV);
 489             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 490             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 491
 492             /* Update potential sum for this i atom from the interaction with this j atom. */
 493             velec            = _mm_andnot_ps(dummy_mask,velec);
 494             velecsum         = _mm_add_ps(velecsum,velec);
 495
 496             fscal            = felec;
 497
 498             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 499
 500             /* Calculate temporary vectorial force */
 501             tx               = _mm_mul_ps(fscal,dx10);
 502             ty               = _mm_mul_ps(fscal,dy10);
 503             tz               = _mm_mul_ps(fscal,dz10);
 504
 505             /* Update vectorial force */
 506             fix1             = _mm_add_ps(fix1,tx);
 507             fiy1             = _mm_add_ps(fiy1,ty);
 508             fiz1             = _mm_add_ps(fiz1,tz);
 509
 510             fjx0             = _mm_add_ps(fjx0,tx);
 511             fjy0             = _mm_add_ps(fjy0,ty);
 512             fjz0             = _mm_add_ps(fjz0,tz);
 513
 514             /**************************
 515              * CALCULATE INTERACTIONS *
 516              **************************/
 517
 518             r20              = _mm_mul_ps(rsq20,rinv20);
 519             r20              = _mm_andnot_ps(dummy_mask,r20);
 520
 521             /* Compute parameters for interactions between i and j atoms */
 522             qq20             = _mm_mul_ps(iq2,jq0);
 523
 524             /* Calculate table index by multiplying r with table scale and truncate to integer */
 525             rt               = _mm_mul_ps(r20,vftabscale);
 526             vfitab           = _mm_cvttps_epi32(rt);
 527             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 528             vfitab           = _mm_slli_epi32(vfitab,2);
 529
 530             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 531             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 532             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 533             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 534             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 535             _MM_TRANSPOSE4_PS(Y,F,G,H);
 536             Heps             = _mm_mul_ps(vfeps,H);
 537             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 538             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 539             velec            = _mm_mul_ps(qq20,VV);
 540             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 541             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 542
 543             /* Update potential sum for this i atom from the interaction with this j atom. */
 544             velec            = _mm_andnot_ps(dummy_mask,velec);
 545             velecsum         = _mm_add_ps(velecsum,velec);
 546
 547             fscal            = felec;
 548
 549             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 550
 551             /* Calculate temporary vectorial force */
 552             tx               = _mm_mul_ps(fscal,dx20);
 553             ty               = _mm_mul_ps(fscal,dy20);
 554             tz               = _mm_mul_ps(fscal,dz20);
 555
 556             /* Update vectorial force */
 557             fix2             = _mm_add_ps(fix2,tx);
 558             fiy2             = _mm_add_ps(fiy2,ty);
 559             fiz2             = _mm_add_ps(fiz2,tz);
 560
 561             fjx0             = _mm_add_ps(fjx0,tx);
 562             fjy0             = _mm_add_ps(fjy0,ty);
 563             fjz0             = _mm_add_ps(fjz0,tz);
 564
 565             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 566             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 567             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 568             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 569
 570             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 571
 572             /* Inner loop uses 132 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
 584         /* Increment number of inner iterations */
 585         inneriter                  += j_index_end - j_index_start;
 586
 587         /* Outer loop uses 19 flops */
 588     }
 589
 590     /* Increment number of outer iterations */
 591     outeriter        += nri;
 592
 593     /* Update outer/inner flops */
 594
 595     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*132);
 596 }
 597 /*
 598  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse4_1_single
 599  * Electrostatics interaction: CubicSplineTable
 600  * VdW interaction:            None
 601  * Geometry:                   Water3-Particle
 602  * Calculate force/pot:        Force
 603  */
 604 void
 605 nb_kernel_ElecCSTab_VdwNone_GeomW3P1_F_sse4_1_single
 606                     (t_nblist * gmx_restrict                nlist,
 607                      rvec * gmx_restrict                    xx,
 608                      rvec * gmx_restrict                    ff,
 609                      t_forcerec * gmx_restrict              fr,
 610                      t_mdatoms * gmx_restrict               mdatoms,
 611                      nb_kernel_data_t * gmx_restrict        kernel_data,
 612                      t_nrnb * gmx_restrict                  nrnb)
 613 {
 614     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 615      * just 0 for non-waters.
 616      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 617      * jnr indices corresponding to data put in the four positions in the SIMD register.
 618      */
 619     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 620     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 621     int              jnrA,jnrB,jnrC,jnrD;
 622     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 623     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 624     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 625     real             rcutoff_scalar;
 626     real             *shiftvec,*fshift,*x,*f;
 627     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 628     real             scratch[4*DIM];
 629     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 630     int              vdwioffset0;
 631     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 632     int              vdwioffset1;
 633     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 634     int              vdwioffset2;
 635     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 636     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 637     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 638     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 639     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 640     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 641     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 642     real             *charge;
 643     __m128i          vfitab;
 644     __m128i          ifour       = _mm_set1_epi32(4);
 645     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 646     real             *vftab;
 647     __m128           dummy_mask,cutoff_mask;
 648     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 649     __m128           one     = _mm_set1_ps(1.0);
 650     __m128           two     = _mm_set1_ps(2.0);
 651     x                = xx[0];
 652     f                = ff[0];
 653
 654     nri              = nlist->nri;
 655     iinr             = nlist->iinr;
 656     jindex           = nlist->jindex;
 657     jjnr             = nlist->jjnr;
 658     shiftidx         = nlist->shift;
 659     gid              = nlist->gid;
 660     shiftvec         = fr->shift_vec[0];
 661     fshift           = fr->fshift[0];
 662     facel            = _mm_set1_ps(fr->epsfac);
 663     charge           = mdatoms->chargeA;
 664
 665     vftab            = kernel_data->table_elec->data;
 666     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 667
 668     /* Setup water-specific parameters */
 669     inr              = nlist->iinr[0];
 670     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 671     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 672     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 673
 674     /* Avoid stupid compiler warnings */
 675     jnrA = jnrB = jnrC = jnrD = 0;
 676     j_coord_offsetA = 0;
 677     j_coord_offsetB = 0;
 678     j_coord_offsetC = 0;
 679     j_coord_offsetD = 0;
 680
 681     outeriter        = 0;
 682     inneriter        = 0;
 683
 684     for(iidx=0;iidx<4*DIM;iidx++)
 685     {
 686         scratch[iidx] = 0.0;
 687     }
 688
 689     /* Start outer loop over neighborlists */
 690     for(iidx=0; iidx<nri; iidx++)
 691     {
 692         /* Load shift vector for this list */
 693         i_shift_offset   = DIM*shiftidx[iidx];
 694
 695         /* Load limits for loop over neighbors */
 696         j_index_start    = jindex[iidx];
 697         j_index_end      = jindex[iidx+1];
 698
 699         /* Get outer coordinate index */
 700         inr              = iinr[iidx];
 701         i_coord_offset   = DIM*inr;
 702
 703         /* Load i particle coords and add shift vector */
 704         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 705                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 706
 707         fix0             = _mm_setzero_ps();
 708         fiy0             = _mm_setzero_ps();
 709         fiz0             = _mm_setzero_ps();
 710         fix1             = _mm_setzero_ps();
 711         fiy1             = _mm_setzero_ps();
 712         fiz1             = _mm_setzero_ps();
 713         fix2             = _mm_setzero_ps();
 714         fiy2             = _mm_setzero_ps();
 715         fiz2             = _mm_setzero_ps();
 716
 717         /* Start inner kernel loop */
 718         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 719         {
 720
 721             /* Get j neighbor index, and coordinate index */
 722             jnrA             = jjnr[jidx];
 723             jnrB             = jjnr[jidx+1];
 724             jnrC             = jjnr[jidx+2];
 725             jnrD             = jjnr[jidx+3];
 726             j_coord_offsetA  = DIM*jnrA;
 727             j_coord_offsetB  = DIM*jnrB;
 728             j_coord_offsetC  = DIM*jnrC;
 729             j_coord_offsetD  = DIM*jnrD;
 730
 731             /* load j atom coordinates */
 732             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 733                                               x+j_coord_offsetC,x+j_coord_offsetD,
 734                                               &jx0,&jy0,&jz0);
 735
 736             /* Calculate displacement vector */
 737             dx00             = _mm_sub_ps(ix0,jx0);
 738             dy00             = _mm_sub_ps(iy0,jy0);
 739             dz00             = _mm_sub_ps(iz0,jz0);
 740             dx10             = _mm_sub_ps(ix1,jx0);
 741             dy10             = _mm_sub_ps(iy1,jy0);
 742             dz10             = _mm_sub_ps(iz1,jz0);
 743             dx20             = _mm_sub_ps(ix2,jx0);
 744             dy20             = _mm_sub_ps(iy2,jy0);
 745             dz20             = _mm_sub_ps(iz2,jz0);
 746
 747             /* Calculate squared distance and things based on it */
 748             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 749             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 750             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 751
 752             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 753             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 754             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 755
 756             /* Load parameters for j particles */
 757             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 758                                                               charge+jnrC+0,charge+jnrD+0);
 759
 760             fjx0             = _mm_setzero_ps();
 761             fjy0             = _mm_setzero_ps();
 762             fjz0             = _mm_setzero_ps();
 763
 764             /**************************
 765              * CALCULATE INTERACTIONS *
 766              **************************/
 767
 768             r00              = _mm_mul_ps(rsq00,rinv00);
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq00             = _mm_mul_ps(iq0,jq0);
 772
 773             /* Calculate table index by multiplying r with table scale and truncate to integer */
 774             rt               = _mm_mul_ps(r00,vftabscale);
 775             vfitab           = _mm_cvttps_epi32(rt);
 776             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 777             vfitab           = _mm_slli_epi32(vfitab,2);
 778
 779             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 780             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 781             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 782             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 783             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 784             _MM_TRANSPOSE4_PS(Y,F,G,H);
 785             Heps             = _mm_mul_ps(vfeps,H);
 786             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 787             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 788             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 789
 790             fscal            = felec;
 791
 792             /* Calculate temporary vectorial force */
 793             tx               = _mm_mul_ps(fscal,dx00);
 794             ty               = _mm_mul_ps(fscal,dy00);
 795             tz               = _mm_mul_ps(fscal,dz00);
 796
 797             /* Update vectorial force */
 798             fix0             = _mm_add_ps(fix0,tx);
 799             fiy0             = _mm_add_ps(fiy0,ty);
 800             fiz0             = _mm_add_ps(fiz0,tz);
 801
 802             fjx0             = _mm_add_ps(fjx0,tx);
 803             fjy0             = _mm_add_ps(fjy0,ty);
 804             fjz0             = _mm_add_ps(fjz0,tz);
 805
 806             /**************************
 807              * CALCULATE INTERACTIONS *
 808              **************************/
 809
 810             r10              = _mm_mul_ps(rsq10,rinv10);
 811
 812             /* Compute parameters for interactions between i and j atoms */
 813             qq10             = _mm_mul_ps(iq1,jq0);
 814
 815             /* Calculate table index by multiplying r with table scale and truncate to integer */
 816             rt               = _mm_mul_ps(r10,vftabscale);
 817             vfitab           = _mm_cvttps_epi32(rt);
 818             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 819             vfitab           = _mm_slli_epi32(vfitab,2);
 820
 821             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 822             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 823             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 824             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 825             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 826             _MM_TRANSPOSE4_PS(Y,F,G,H);
 827             Heps             = _mm_mul_ps(vfeps,H);
 828             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 829             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 830             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 831
 832             fscal            = felec;
 833
 834             /* Calculate temporary vectorial force */
 835             tx               = _mm_mul_ps(fscal,dx10);
 836             ty               = _mm_mul_ps(fscal,dy10);
 837             tz               = _mm_mul_ps(fscal,dz10);
 838
 839             /* Update vectorial force */
 840             fix1             = _mm_add_ps(fix1,tx);
 841             fiy1             = _mm_add_ps(fiy1,ty);
 842             fiz1             = _mm_add_ps(fiz1,tz);
 843
 844             fjx0             = _mm_add_ps(fjx0,tx);
 845             fjy0             = _mm_add_ps(fjy0,ty);
 846             fjz0             = _mm_add_ps(fjz0,tz);
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             r20              = _mm_mul_ps(rsq20,rinv20);
 853
 854             /* Compute parameters for interactions between i and j atoms */
 855             qq20             = _mm_mul_ps(iq2,jq0);
 856
 857             /* Calculate table index by multiplying r with table scale and truncate to integer */
 858             rt               = _mm_mul_ps(r20,vftabscale);
 859             vfitab           = _mm_cvttps_epi32(rt);
 860             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 861             vfitab           = _mm_slli_epi32(vfitab,2);
 862
 863             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 864             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 865             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 866             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 867             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 868             _MM_TRANSPOSE4_PS(Y,F,G,H);
 869             Heps             = _mm_mul_ps(vfeps,H);
 870             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 871             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 872             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 873
 874             fscal            = felec;
 875
 876             /* Calculate temporary vectorial force */
 877             tx               = _mm_mul_ps(fscal,dx20);
 878             ty               = _mm_mul_ps(fscal,dy20);
 879             tz               = _mm_mul_ps(fscal,dz20);
 880
 881             /* Update vectorial force */
 882             fix2             = _mm_add_ps(fix2,tx);
 883             fiy2             = _mm_add_ps(fiy2,ty);
 884             fiz2             = _mm_add_ps(fiz2,tz);
 885
 886             fjx0             = _mm_add_ps(fjx0,tx);
 887             fjy0             = _mm_add_ps(fjy0,ty);
 888             fjz0             = _mm_add_ps(fjz0,tz);
 889
 890             fjptrA             = f+j_coord_offsetA;
 891             fjptrB             = f+j_coord_offsetB;
 892             fjptrC             = f+j_coord_offsetC;
 893             fjptrD             = f+j_coord_offsetD;
 894
 895             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 896
 897             /* Inner loop uses 117 flops */
 898         }
 899
 900         if(jidx<j_index_end)
 901         {
 902
 903             /* Get j neighbor index, and coordinate index */
 904             jnrlistA         = jjnr[jidx];
 905             jnrlistB         = jjnr[jidx+1];
 906             jnrlistC         = jjnr[jidx+2];
 907             jnrlistD         = jjnr[jidx+3];
 908             /* Sign of each element will be negative for non-real atoms.
 909              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 910              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 911              */
 912             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 913             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 914             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 915             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 916             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 917             j_coord_offsetA  = DIM*jnrA;
 918             j_coord_offsetB  = DIM*jnrB;
 919             j_coord_offsetC  = DIM*jnrC;
 920             j_coord_offsetD  = DIM*jnrD;
 921
 922             /* load j atom coordinates */
 923             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 924                                               x+j_coord_offsetC,x+j_coord_offsetD,
 925                                               &jx0,&jy0,&jz0);
 926
 927             /* Calculate displacement vector */
 928             dx00             = _mm_sub_ps(ix0,jx0);
 929             dy00             = _mm_sub_ps(iy0,jy0);
 930             dz00             = _mm_sub_ps(iz0,jz0);
 931             dx10             = _mm_sub_ps(ix1,jx0);
 932             dy10             = _mm_sub_ps(iy1,jy0);
 933             dz10             = _mm_sub_ps(iz1,jz0);
 934             dx20             = _mm_sub_ps(ix2,jx0);
 935             dy20             = _mm_sub_ps(iy2,jy0);
 936             dz20             = _mm_sub_ps(iz2,jz0);
 937
 938             /* Calculate squared distance and things based on it */
 939             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 940             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 941             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 942
 943             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 944             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 945             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 946
 947             /* Load parameters for j particles */
 948             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 949                                                               charge+jnrC+0,charge+jnrD+0);
 950
 951             fjx0             = _mm_setzero_ps();
 952             fjy0             = _mm_setzero_ps();
 953             fjz0             = _mm_setzero_ps();
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r00              = _mm_mul_ps(rsq00,rinv00);
 960             r00              = _mm_andnot_ps(dummy_mask,r00);
 961
 962             /* Compute parameters for interactions between i and j atoms */
 963             qq00             = _mm_mul_ps(iq0,jq0);
 964
 965             /* Calculate table index by multiplying r with table scale and truncate to integer */
 966             rt               = _mm_mul_ps(r00,vftabscale);
 967             vfitab           = _mm_cvttps_epi32(rt);
 968             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 969             vfitab           = _mm_slli_epi32(vfitab,2);
 970
 971             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 972             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 973             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 974             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 975             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 976             _MM_TRANSPOSE4_PS(Y,F,G,H);
 977             Heps             = _mm_mul_ps(vfeps,H);
 978             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 979             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 980             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm_mul_ps(fscal,dx00);
 988             ty               = _mm_mul_ps(fscal,dy00);
 989             tz               = _mm_mul_ps(fscal,dz00);
 990
 991             /* Update vectorial force */
 992             fix0             = _mm_add_ps(fix0,tx);
 993             fiy0             = _mm_add_ps(fiy0,ty);
 994             fiz0             = _mm_add_ps(fiz0,tz);
 995
 996             fjx0             = _mm_add_ps(fjx0,tx);
 997             fjy0             = _mm_add_ps(fjy0,ty);
 998             fjz0             = _mm_add_ps(fjz0,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r10              = _mm_mul_ps(rsq10,rinv10);
1005             r10              = _mm_andnot_ps(dummy_mask,r10);
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq10             = _mm_mul_ps(iq1,jq0);
1009
1010             /* Calculate table index by multiplying r with table scale and truncate to integer */
1011             rt               = _mm_mul_ps(r10,vftabscale);
1012             vfitab           = _mm_cvttps_epi32(rt);
1013             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1014             vfitab           = _mm_slli_epi32(vfitab,2);
1015
1016             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1017             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1018             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1019             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1020             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1021             _MM_TRANSPOSE4_PS(Y,F,G,H);
1022             Heps             = _mm_mul_ps(vfeps,H);
1023             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1024             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1025             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1026
1027             fscal            = felec;
1028
1029             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1030
1031             /* Calculate temporary vectorial force */
1032             tx               = _mm_mul_ps(fscal,dx10);
1033             ty               = _mm_mul_ps(fscal,dy10);
1034             tz               = _mm_mul_ps(fscal,dz10);
1035
1036             /* Update vectorial force */
1037             fix1             = _mm_add_ps(fix1,tx);
1038             fiy1             = _mm_add_ps(fiy1,ty);
1039             fiz1             = _mm_add_ps(fiz1,tz);
1040
1041             fjx0             = _mm_add_ps(fjx0,tx);
1042             fjy0             = _mm_add_ps(fjy0,ty);
1043             fjz0             = _mm_add_ps(fjz0,tz);
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             r20              = _mm_mul_ps(rsq20,rinv20);
1050             r20              = _mm_andnot_ps(dummy_mask,r20);
1051
1052             /* Compute parameters for interactions between i and j atoms */
1053             qq20             = _mm_mul_ps(iq2,jq0);
1054
1055             /* Calculate table index by multiplying r with table scale and truncate to integer */
1056             rt               = _mm_mul_ps(r20,vftabscale);
1057             vfitab           = _mm_cvttps_epi32(rt);
1058             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1059             vfitab           = _mm_slli_epi32(vfitab,2);
1060
1061             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1062             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1063             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1064             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1065             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1066             _MM_TRANSPOSE4_PS(Y,F,G,H);
1067             Heps             = _mm_mul_ps(vfeps,H);
1068             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1069             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1070             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1071
1072             fscal            = felec;
1073
1074             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1075
1076             /* Calculate temporary vectorial force */
1077             tx               = _mm_mul_ps(fscal,dx20);
1078             ty               = _mm_mul_ps(fscal,dy20);
1079             tz               = _mm_mul_ps(fscal,dz20);
1080
1081             /* Update vectorial force */
1082             fix2             = _mm_add_ps(fix2,tx);
1083             fiy2             = _mm_add_ps(fiy2,ty);
1084             fiz2             = _mm_add_ps(fiz2,tz);
1085
1086             fjx0             = _mm_add_ps(fjx0,tx);
1087             fjy0             = _mm_add_ps(fjy0,ty);
1088             fjz0             = _mm_add_ps(fjz0,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
1095             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1096
1097             /* Inner loop uses 120 flops */
1098         }
1099
1100         /* End of innermost loop */
1101
1102         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1103                                               f+i_coord_offset,fshift+i_shift_offset);
1104
1105         /* Increment number of inner iterations */
1106         inneriter                  += j_index_end - j_index_start;
1107
1108         /* Outer loop uses 18 flops */
1109     }
1110
1111     /* Increment number of outer iterations */
1112     outeriter        += nri;
1113
1114     /* Update outer/inner flops */
1115
1116     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1117 }