src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecCSTab_VdwLJ_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_VdwLJ_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_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_elec->data;
 114     vftabscale       = _mm_set1_ps(kernel_data->table_elec->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
 211             /* Load parameters for j particles */
 212             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 213                                                               charge+jnrC+0,charge+jnrD+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216             vdwjidx0C        = 2*vdwtype[jnrC+0];
 217             vdwjidx0D        = 2*vdwtype[jnrD+0];
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             r00              = _mm_mul_ps(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_ps(iq0,jq0);
 227             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 228                                          vdwparam+vdwioffset0+vdwjidx0B,
 229                                          vdwparam+vdwioffset0+vdwjidx0C,
 230                                          vdwparam+vdwioffset0+vdwjidx0D,
 231                                          &c6_00,&c12_00);
 232
 233             /* Calculate table index by multiplying r with table scale and truncate to integer */
 234             rt               = _mm_mul_ps(r00,vftabscale);
 235             vfitab           = _mm_cvttps_epi32(rt);
 236             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 237             vfitab           = _mm_slli_epi32(vfitab,2);
 238
 239             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 240             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 241             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 242             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 243             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 244             _MM_TRANSPOSE4_PS(Y,F,G,H);
 245             Heps             = _mm_mul_ps(vfeps,H);
 246             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 247             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 248             velec            = _mm_mul_ps(qq00,VV);
 249             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 250             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 251
 252             /* LENNARD-JONES DISPERSION/REPULSION */
 253
 254             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 255             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 256             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 257             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 258             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velecsum         = _mm_add_ps(velecsum,velec);
 262             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 263
 264             fscal            = _mm_add_ps(felec,fvdw);
 265
 266             /* Calculate temporary vectorial force */
 267             tx               = _mm_mul_ps(fscal,dx00);
 268             ty               = _mm_mul_ps(fscal,dy00);
 269             tz               = _mm_mul_ps(fscal,dz00);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_add_ps(fix0,tx);
 273             fiy0             = _mm_add_ps(fiy0,ty);
 274             fiz0             = _mm_add_ps(fiz0,tz);
 275
 276             fjptrA             = f+j_coord_offsetA;
 277             fjptrB             = f+j_coord_offsetB;
 278             fjptrC             = f+j_coord_offsetC;
 279             fjptrD             = f+j_coord_offsetD;
 280             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             r10              = _mm_mul_ps(rsq10,rinv10);
 287
 288             /* Compute parameters for interactions between i and j atoms */
 289             qq10             = _mm_mul_ps(iq1,jq0);
 290
 291             /* Calculate table index by multiplying r with table scale and truncate to integer */
 292             rt               = _mm_mul_ps(r10,vftabscale);
 293             vfitab           = _mm_cvttps_epi32(rt);
 294             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 295             vfitab           = _mm_slli_epi32(vfitab,2);
 296
 297             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 298             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 299             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 300             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 301             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 302             _MM_TRANSPOSE4_PS(Y,F,G,H);
 303             Heps             = _mm_mul_ps(vfeps,H);
 304             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 305             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 306             velec            = _mm_mul_ps(qq10,VV);
 307             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 308             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm_add_ps(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_ps(fscal,dx10);
 317             ty               = _mm_mul_ps(fscal,dy10);
 318             tz               = _mm_mul_ps(fscal,dz10);
 319
 320             /* Update vectorial force */
 321             fix1             = _mm_add_ps(fix1,tx);
 322             fiy1             = _mm_add_ps(fiy1,ty);
 323             fiz1             = _mm_add_ps(fiz1,tz);
 324
 325             fjptrA             = f+j_coord_offsetA;
 326             fjptrB             = f+j_coord_offsetB;
 327             fjptrC             = f+j_coord_offsetC;
 328             fjptrD             = f+j_coord_offsetD;
 329             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             r20              = _mm_mul_ps(rsq20,rinv20);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_ps(iq2,jq0);
 339
 340             /* Calculate table index by multiplying r with table scale and truncate to integer */
 341             rt               = _mm_mul_ps(r20,vftabscale);
 342             vfitab           = _mm_cvttps_epi32(rt);
 343             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 344             vfitab           = _mm_slli_epi32(vfitab,2);
 345
 346             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 347             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 348             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 349             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 350             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 351             _MM_TRANSPOSE4_PS(Y,F,G,H);
 352             Heps             = _mm_mul_ps(vfeps,H);
 353             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 354             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 355             velec            = _mm_mul_ps(qq20,VV);
 356             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 357             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 358
 359             /* Update potential sum for this i atom from the interaction with this j atom. */
 360             velecsum         = _mm_add_ps(velecsum,velec);
 361
 362             fscal            = felec;
 363
 364             /* Calculate temporary vectorial force */
 365             tx               = _mm_mul_ps(fscal,dx20);
 366             ty               = _mm_mul_ps(fscal,dy20);
 367             tz               = _mm_mul_ps(fscal,dz20);
 368
 369             /* Update vectorial force */
 370             fix2             = _mm_add_ps(fix2,tx);
 371             fiy2             = _mm_add_ps(fiy2,ty);
 372             fiz2             = _mm_add_ps(fiz2,tz);
 373
 374             fjptrA             = f+j_coord_offsetA;
 375             fjptrB             = f+j_coord_offsetB;
 376             fjptrC             = f+j_coord_offsetC;
 377             fjptrD             = f+j_coord_offsetD;
 378             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 379
 380             /* Inner loop uses 142 flops */
 381         }
 382
 383         if(jidx<j_index_end)
 384         {
 385
 386             /* Get j neighbor index, and coordinate index */
 387             jnrlistA         = jjnr[jidx];
 388             jnrlistB         = jjnr[jidx+1];
 389             jnrlistC         = jjnr[jidx+2];
 390             jnrlistD         = jjnr[jidx+3];
 391             /* Sign of each element will be negative for non-real atoms.
 392              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 393              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 394              */
 395             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 396             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 397             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 398             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 399             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 400             j_coord_offsetA  = DIM*jnrA;
 401             j_coord_offsetB  = DIM*jnrB;
 402             j_coord_offsetC  = DIM*jnrC;
 403             j_coord_offsetD  = DIM*jnrD;
 404
 405             /* load j atom coordinates */
 406             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 407                                               x+j_coord_offsetC,x+j_coord_offsetD,
 408                                               &jx0,&jy0,&jz0);
 409
 410             /* Calculate displacement vector */
 411             dx00             = _mm_sub_ps(ix0,jx0);
 412             dy00             = _mm_sub_ps(iy0,jy0);
 413             dz00             = _mm_sub_ps(iz0,jz0);
 414             dx10             = _mm_sub_ps(ix1,jx0);
 415             dy10             = _mm_sub_ps(iy1,jy0);
 416             dz10             = _mm_sub_ps(iz1,jz0);
 417             dx20             = _mm_sub_ps(ix2,jx0);
 418             dy20             = _mm_sub_ps(iy2,jy0);
 419             dz20             = _mm_sub_ps(iz2,jz0);
 420
 421             /* Calculate squared distance and things based on it */
 422             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 423             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 424             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 425
 426             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 427             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 428             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 429
 430             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 431
 432             /* Load parameters for j particles */
 433             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 434                                                               charge+jnrC+0,charge+jnrD+0);
 435             vdwjidx0A        = 2*vdwtype[jnrA+0];
 436             vdwjidx0B        = 2*vdwtype[jnrB+0];
 437             vdwjidx0C        = 2*vdwtype[jnrC+0];
 438             vdwjidx0D        = 2*vdwtype[jnrD+0];
 439
 440             /**************************
 441              * CALCULATE INTERACTIONS *
 442              **************************/
 443
 444             r00              = _mm_mul_ps(rsq00,rinv00);
 445             r00              = _mm_andnot_ps(dummy_mask,r00);
 446
 447             /* Compute parameters for interactions between i and j atoms */
 448             qq00             = _mm_mul_ps(iq0,jq0);
 449             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 450                                          vdwparam+vdwioffset0+vdwjidx0B,
 451                                          vdwparam+vdwioffset0+vdwjidx0C,
 452                                          vdwparam+vdwioffset0+vdwjidx0D,
 453                                          &c6_00,&c12_00);
 454
 455             /* Calculate table index by multiplying r with table scale and truncate to integer */
 456             rt               = _mm_mul_ps(r00,vftabscale);
 457             vfitab           = _mm_cvttps_epi32(rt);
 458             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 459             vfitab           = _mm_slli_epi32(vfitab,2);
 460
 461             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             velec            = _mm_mul_ps(qq00,VV);
 471             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 472             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 473
 474             /* LENNARD-JONES DISPERSION/REPULSION */
 475
 476             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 477             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 478             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 479             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 480             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 481
 482             /* Update potential sum for this i atom from the interaction with this j atom. */
 483             velec            = _mm_andnot_ps(dummy_mask,velec);
 484             velecsum         = _mm_add_ps(velecsum,velec);
 485             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 486             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 487
 488             fscal            = _mm_add_ps(felec,fvdw);
 489
 490             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 491
 492             /* Calculate temporary vectorial force */
 493             tx               = _mm_mul_ps(fscal,dx00);
 494             ty               = _mm_mul_ps(fscal,dy00);
 495             tz               = _mm_mul_ps(fscal,dz00);
 496
 497             /* Update vectorial force */
 498             fix0             = _mm_add_ps(fix0,tx);
 499             fiy0             = _mm_add_ps(fiy0,ty);
 500             fiz0             = _mm_add_ps(fiz0,tz);
 501
 502             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 503             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 504             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 505             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 506             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 507
 508             /**************************
 509              * CALCULATE INTERACTIONS *
 510              **************************/
 511
 512             r10              = _mm_mul_ps(rsq10,rinv10);
 513             r10              = _mm_andnot_ps(dummy_mask,r10);
 514
 515             /* Compute parameters for interactions between i and j atoms */
 516             qq10             = _mm_mul_ps(iq1,jq0);
 517
 518             /* Calculate table index by multiplying r with table scale and truncate to integer */
 519             rt               = _mm_mul_ps(r10,vftabscale);
 520             vfitab           = _mm_cvttps_epi32(rt);
 521             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 522             vfitab           = _mm_slli_epi32(vfitab,2);
 523
 524             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 525             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 526             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 527             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 528             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 529             _MM_TRANSPOSE4_PS(Y,F,G,H);
 530             Heps             = _mm_mul_ps(vfeps,H);
 531             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 532             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 533             velec            = _mm_mul_ps(qq10,VV);
 534             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 535             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545             /* Calculate temporary vectorial force */
 546             tx               = _mm_mul_ps(fscal,dx10);
 547             ty               = _mm_mul_ps(fscal,dy10);
 548             tz               = _mm_mul_ps(fscal,dz10);
 549
 550             /* Update vectorial force */
 551             fix1             = _mm_add_ps(fix1,tx);
 552             fiy1             = _mm_add_ps(fiy1,ty);
 553             fiz1             = _mm_add_ps(fiz1,tz);
 554
 555             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 556             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 557             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 558             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             r20              = _mm_mul_ps(rsq20,rinv20);
 566             r20              = _mm_andnot_ps(dummy_mask,r20);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq20             = _mm_mul_ps(iq2,jq0);
 570
 571             /* Calculate table index by multiplying r with table scale and truncate to integer */
 572             rt               = _mm_mul_ps(r20,vftabscale);
 573             vfitab           = _mm_cvttps_epi32(rt);
 574             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 575             vfitab           = _mm_slli_epi32(vfitab,2);
 576
 577             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 578             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 579             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 580             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 581             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 582             _MM_TRANSPOSE4_PS(Y,F,G,H);
 583             Heps             = _mm_mul_ps(vfeps,H);
 584             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 585             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 586             velec            = _mm_mul_ps(qq20,VV);
 587             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 588             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 589
 590             /* Update potential sum for this i atom from the interaction with this j atom. */
 591             velec            = _mm_andnot_ps(dummy_mask,velec);
 592             velecsum         = _mm_add_ps(velecsum,velec);
 593
 594             fscal            = felec;
 595
 596             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 597
 598             /* Calculate temporary vectorial force */
 599             tx               = _mm_mul_ps(fscal,dx20);
 600             ty               = _mm_mul_ps(fscal,dy20);
 601             tz               = _mm_mul_ps(fscal,dz20);
 602
 603             /* Update vectorial force */
 604             fix2             = _mm_add_ps(fix2,tx);
 605             fiy2             = _mm_add_ps(fiy2,ty);
 606             fiz2             = _mm_add_ps(fiz2,tz);
 607
 608             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 609             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 610             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 611             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 612             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 613
 614             /* Inner loop uses 145 flops */
 615         }
 616
 617         /* End of innermost loop */
 618
 619         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 620                                               f+i_coord_offset,fshift+i_shift_offset);
 621
 622         ggid                        = gid[iidx];
 623         /* Update potential energies */
 624         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 625         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 626
 627         /* Increment number of inner iterations */
 628         inneriter                  += j_index_end - j_index_start;
 629
 630         /* Outer loop uses 20 flops */
 631     }
 632
 633     /* Increment number of outer iterations */
 634     outeriter        += nri;
 635
 636     /* Update outer/inner flops */
 637
 638     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*145);
 639 }
 640 /*
 641  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse4_1_single
 642  * Electrostatics interaction: CubicSplineTable
 643  * VdW interaction:            LennardJones
 644  * Geometry:                   Water3-Particle
 645  * Calculate force/pot:        Force
 646  */
 647 void
 648 nb_kernel_ElecCSTab_VdwLJ_GeomW3P1_F_sse4_1_single
 649                     (t_nblist * gmx_restrict                nlist,
 650                      rvec * gmx_restrict                    xx,
 651                      rvec * gmx_restrict                    ff,
 652                      t_forcerec * gmx_restrict              fr,
 653                      t_mdatoms * gmx_restrict               mdatoms,
 654                      nb_kernel_data_t * gmx_restrict        kernel_data,
 655                      t_nrnb * gmx_restrict                  nrnb)
 656 {
 657     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 658      * just 0 for non-waters.
 659      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 660      * jnr indices corresponding to data put in the four positions in the SIMD register.
 661      */
 662     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 663     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 664     int              jnrA,jnrB,jnrC,jnrD;
 665     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 666     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 667     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 668     real             rcutoff_scalar;
 669     real             *shiftvec,*fshift,*x,*f;
 670     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 671     real             scratch[4*DIM];
 672     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 673     int              vdwioffset0;
 674     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 675     int              vdwioffset1;
 676     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 677     int              vdwioffset2;
 678     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 679     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 680     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 681     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 682     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 683     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 684     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 685     real             *charge;
 686     int              nvdwtype;
 687     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 688     int              *vdwtype;
 689     real             *vdwparam;
 690     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 691     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 692     __m128i          vfitab;
 693     __m128i          ifour       = _mm_set1_epi32(4);
 694     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 695     real             *vftab;
 696     __m128           dummy_mask,cutoff_mask;
 697     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 698     __m128           one     = _mm_set1_ps(1.0);
 699     __m128           two     = _mm_set1_ps(2.0);
 700     x                = xx[0];
 701     f                = ff[0];
 702
 703     nri              = nlist->nri;
 704     iinr             = nlist->iinr;
 705     jindex           = nlist->jindex;
 706     jjnr             = nlist->jjnr;
 707     shiftidx         = nlist->shift;
 708     gid              = nlist->gid;
 709     shiftvec         = fr->shift_vec[0];
 710     fshift           = fr->fshift[0];
 711     facel            = _mm_set1_ps(fr->epsfac);
 712     charge           = mdatoms->chargeA;
 713     nvdwtype         = fr->ntype;
 714     vdwparam         = fr->nbfp;
 715     vdwtype          = mdatoms->typeA;
 716
 717     vftab            = kernel_data->table_elec->data;
 718     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 719
 720     /* Setup water-specific parameters */
 721     inr              = nlist->iinr[0];
 722     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 723     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 724     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 725     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 726
 727     /* Avoid stupid compiler warnings */
 728     jnrA = jnrB = jnrC = jnrD = 0;
 729     j_coord_offsetA = 0;
 730     j_coord_offsetB = 0;
 731     j_coord_offsetC = 0;
 732     j_coord_offsetD = 0;
 733
 734     outeriter        = 0;
 735     inneriter        = 0;
 736
 737     for(iidx=0;iidx<4*DIM;iidx++)
 738     {
 739         scratch[iidx] = 0.0;
 740     }
 741
 742     /* Start outer loop over neighborlists */
 743     for(iidx=0; iidx<nri; iidx++)
 744     {
 745         /* Load shift vector for this list */
 746         i_shift_offset   = DIM*shiftidx[iidx];
 747
 748         /* Load limits for loop over neighbors */
 749         j_index_start    = jindex[iidx];
 750         j_index_end      = jindex[iidx+1];
 751
 752         /* Get outer coordinate index */
 753         inr              = iinr[iidx];
 754         i_coord_offset   = DIM*inr;
 755
 756         /* Load i particle coords and add shift vector */
 757         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 758                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 759
 760         fix0             = _mm_setzero_ps();
 761         fiy0             = _mm_setzero_ps();
 762         fiz0             = _mm_setzero_ps();
 763         fix1             = _mm_setzero_ps();
 764         fiy1             = _mm_setzero_ps();
 765         fiz1             = _mm_setzero_ps();
 766         fix2             = _mm_setzero_ps();
 767         fiy2             = _mm_setzero_ps();
 768         fiz2             = _mm_setzero_ps();
 769
 770         /* Start inner kernel loop */
 771         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 772         {
 773
 774             /* Get j neighbor index, and coordinate index */
 775             jnrA             = jjnr[jidx];
 776             jnrB             = jjnr[jidx+1];
 777             jnrC             = jjnr[jidx+2];
 778             jnrD             = jjnr[jidx+3];
 779             j_coord_offsetA  = DIM*jnrA;
 780             j_coord_offsetB  = DIM*jnrB;
 781             j_coord_offsetC  = DIM*jnrC;
 782             j_coord_offsetD  = DIM*jnrD;
 783
 784             /* load j atom coordinates */
 785             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 786                                               x+j_coord_offsetC,x+j_coord_offsetD,
 787                                               &jx0,&jy0,&jz0);
 788
 789             /* Calculate displacement vector */
 790             dx00             = _mm_sub_ps(ix0,jx0);
 791             dy00             = _mm_sub_ps(iy0,jy0);
 792             dz00             = _mm_sub_ps(iz0,jz0);
 793             dx10             = _mm_sub_ps(ix1,jx0);
 794             dy10             = _mm_sub_ps(iy1,jy0);
 795             dz10             = _mm_sub_ps(iz1,jz0);
 796             dx20             = _mm_sub_ps(ix2,jx0);
 797             dy20             = _mm_sub_ps(iy2,jy0);
 798             dz20             = _mm_sub_ps(iz2,jz0);
 799
 800             /* Calculate squared distance and things based on it */
 801             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 802             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 803             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 804
 805             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 806             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 807             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 808
 809             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 810
 811             /* Load parameters for j particles */
 812             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 813                                                               charge+jnrC+0,charge+jnrD+0);
 814             vdwjidx0A        = 2*vdwtype[jnrA+0];
 815             vdwjidx0B        = 2*vdwtype[jnrB+0];
 816             vdwjidx0C        = 2*vdwtype[jnrC+0];
 817             vdwjidx0D        = 2*vdwtype[jnrD+0];
 818
 819             /**************************
 820              * CALCULATE INTERACTIONS *
 821              **************************/
 822
 823             r00              = _mm_mul_ps(rsq00,rinv00);
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq00             = _mm_mul_ps(iq0,jq0);
 827             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 828                                          vdwparam+vdwioffset0+vdwjidx0B,
 829                                          vdwparam+vdwioffset0+vdwjidx0C,
 830                                          vdwparam+vdwioffset0+vdwjidx0D,
 831                                          &c6_00,&c12_00);
 832
 833             /* Calculate table index by multiplying r with table scale and truncate to integer */
 834             rt               = _mm_mul_ps(r00,vftabscale);
 835             vfitab           = _mm_cvttps_epi32(rt);
 836             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 837             vfitab           = _mm_slli_epi32(vfitab,2);
 838
 839             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 840             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 841             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 842             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 843             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 844             _MM_TRANSPOSE4_PS(Y,F,G,H);
 845             Heps             = _mm_mul_ps(vfeps,H);
 846             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 847             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 848             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
 849
 850             /* LENNARD-JONES DISPERSION/REPULSION */
 851
 852             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 853             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 854
 855             fscal            = _mm_add_ps(felec,fvdw);
 856
 857             /* Calculate temporary vectorial force */
 858             tx               = _mm_mul_ps(fscal,dx00);
 859             ty               = _mm_mul_ps(fscal,dy00);
 860             tz               = _mm_mul_ps(fscal,dz00);
 861
 862             /* Update vectorial force */
 863             fix0             = _mm_add_ps(fix0,tx);
 864             fiy0             = _mm_add_ps(fiy0,ty);
 865             fiz0             = _mm_add_ps(fiz0,tz);
 866
 867             fjptrA             = f+j_coord_offsetA;
 868             fjptrB             = f+j_coord_offsetB;
 869             fjptrC             = f+j_coord_offsetC;
 870             fjptrD             = f+j_coord_offsetD;
 871             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 872
 873             /**************************
 874              * CALCULATE INTERACTIONS *
 875              **************************/
 876
 877             r10              = _mm_mul_ps(rsq10,rinv10);
 878
 879             /* Compute parameters for interactions between i and j atoms */
 880             qq10             = _mm_mul_ps(iq1,jq0);
 881
 882             /* Calculate table index by multiplying r with table scale and truncate to integer */
 883             rt               = _mm_mul_ps(r10,vftabscale);
 884             vfitab           = _mm_cvttps_epi32(rt);
 885             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 886             vfitab           = _mm_slli_epi32(vfitab,2);
 887
 888             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 889             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 890             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 891             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 892             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 893             _MM_TRANSPOSE4_PS(Y,F,G,H);
 894             Heps             = _mm_mul_ps(vfeps,H);
 895             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 896             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 897             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 898
 899             fscal            = felec;
 900
 901             /* Calculate temporary vectorial force */
 902             tx               = _mm_mul_ps(fscal,dx10);
 903             ty               = _mm_mul_ps(fscal,dy10);
 904             tz               = _mm_mul_ps(fscal,dz10);
 905
 906             /* Update vectorial force */
 907             fix1             = _mm_add_ps(fix1,tx);
 908             fiy1             = _mm_add_ps(fiy1,ty);
 909             fiz1             = _mm_add_ps(fiz1,tz);
 910
 911             fjptrA             = f+j_coord_offsetA;
 912             fjptrB             = f+j_coord_offsetB;
 913             fjptrC             = f+j_coord_offsetC;
 914             fjptrD             = f+j_coord_offsetD;
 915             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 916
 917             /**************************
 918              * CALCULATE INTERACTIONS *
 919              **************************/
 920
 921             r20              = _mm_mul_ps(rsq20,rinv20);
 922
 923             /* Compute parameters for interactions between i and j atoms */
 924             qq20             = _mm_mul_ps(iq2,jq0);
 925
 926             /* Calculate table index by multiplying r with table scale and truncate to integer */
 927             rt               = _mm_mul_ps(r20,vftabscale);
 928             vfitab           = _mm_cvttps_epi32(rt);
 929             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 930             vfitab           = _mm_slli_epi32(vfitab,2);
 931
 932             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 933             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 934             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 935             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 936             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 937             _MM_TRANSPOSE4_PS(Y,F,G,H);
 938             Heps             = _mm_mul_ps(vfeps,H);
 939             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 940             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 941             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 942
 943             fscal            = felec;
 944
 945             /* Calculate temporary vectorial force */
 946             tx               = _mm_mul_ps(fscal,dx20);
 947             ty               = _mm_mul_ps(fscal,dy20);
 948             tz               = _mm_mul_ps(fscal,dz20);
 949
 950             /* Update vectorial force */
 951             fix2             = _mm_add_ps(fix2,tx);
 952             fiy2             = _mm_add_ps(fiy2,ty);
 953             fiz2             = _mm_add_ps(fiz2,tz);
 954
 955             fjptrA             = f+j_coord_offsetA;
 956             fjptrB             = f+j_coord_offsetB;
 957             fjptrC             = f+j_coord_offsetC;
 958             fjptrD             = f+j_coord_offsetD;
 959             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 960
 961             /* Inner loop uses 125 flops */
 962         }
 963
 964         if(jidx<j_index_end)
 965         {
 966
 967             /* Get j neighbor index, and coordinate index */
 968             jnrlistA         = jjnr[jidx];
 969             jnrlistB         = jjnr[jidx+1];
 970             jnrlistC         = jjnr[jidx+2];
 971             jnrlistD         = jjnr[jidx+3];
 972             /* Sign of each element will be negative for non-real atoms.
 973              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 974              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 975              */
 976             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 977             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 978             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 979             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 980             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 981             j_coord_offsetA  = DIM*jnrA;
 982             j_coord_offsetB  = DIM*jnrB;
 983             j_coord_offsetC  = DIM*jnrC;
 984             j_coord_offsetD  = DIM*jnrD;
 985
 986             /* load j atom coordinates */
 987             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 988                                               x+j_coord_offsetC,x+j_coord_offsetD,
 989                                               &jx0,&jy0,&jz0);
 990
 991             /* Calculate displacement vector */
 992             dx00             = _mm_sub_ps(ix0,jx0);
 993             dy00             = _mm_sub_ps(iy0,jy0);
 994             dz00             = _mm_sub_ps(iz0,jz0);
 995             dx10             = _mm_sub_ps(ix1,jx0);
 996             dy10             = _mm_sub_ps(iy1,jy0);
 997             dz10             = _mm_sub_ps(iz1,jz0);
 998             dx20             = _mm_sub_ps(ix2,jx0);
 999             dy20             = _mm_sub_ps(iy2,jy0);
1000             dz20             = _mm_sub_ps(iz2,jz0);
1001
1002             /* Calculate squared distance and things based on it */
1003             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1004             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1005             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1006
1007             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1008             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1009             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1010
1011             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1012
1013             /* Load parameters for j particles */
1014             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1015                                                               charge+jnrC+0,charge+jnrD+0);
1016             vdwjidx0A        = 2*vdwtype[jnrA+0];
1017             vdwjidx0B        = 2*vdwtype[jnrB+0];
1018             vdwjidx0C        = 2*vdwtype[jnrC+0];
1019             vdwjidx0D        = 2*vdwtype[jnrD+0];
1020
1021             /**************************
1022              * CALCULATE INTERACTIONS *
1023              **************************/
1024
1025             r00              = _mm_mul_ps(rsq00,rinv00);
1026             r00              = _mm_andnot_ps(dummy_mask,r00);
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq00             = _mm_mul_ps(iq0,jq0);
1030             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1031                                          vdwparam+vdwioffset0+vdwjidx0B,
1032                                          vdwparam+vdwioffset0+vdwjidx0C,
1033                                          vdwparam+vdwioffset0+vdwjidx0D,
1034                                          &c6_00,&c12_00);
1035
1036             /* Calculate table index by multiplying r with table scale and truncate to integer */
1037             rt               = _mm_mul_ps(r00,vftabscale);
1038             vfitab           = _mm_cvttps_epi32(rt);
1039             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1040             vfitab           = _mm_slli_epi32(vfitab,2);
1041
1042             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1043             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1044             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1045             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1046             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1047             _MM_TRANSPOSE4_PS(Y,F,G,H);
1048             Heps             = _mm_mul_ps(vfeps,H);
1049             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1050             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1051             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq00,FF),_mm_mul_ps(vftabscale,rinv00)));
1052
1053             /* LENNARD-JONES DISPERSION/REPULSION */
1054
1055             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1056             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1057
1058             fscal            = _mm_add_ps(felec,fvdw);
1059
1060             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1061
1062             /* Calculate temporary vectorial force */
1063             tx               = _mm_mul_ps(fscal,dx00);
1064             ty               = _mm_mul_ps(fscal,dy00);
1065             tz               = _mm_mul_ps(fscal,dz00);
1066
1067             /* Update vectorial force */
1068             fix0             = _mm_add_ps(fix0,tx);
1069             fiy0             = _mm_add_ps(fiy0,ty);
1070             fiz0             = _mm_add_ps(fiz0,tz);
1071
1072             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1073             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1074             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1075             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1076             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1077
1078             /**************************
1079              * CALCULATE INTERACTIONS *
1080              **************************/
1081
1082             r10              = _mm_mul_ps(rsq10,rinv10);
1083             r10              = _mm_andnot_ps(dummy_mask,r10);
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq10             = _mm_mul_ps(iq1,jq0);
1087
1088             /* Calculate table index by multiplying r with table scale and truncate to integer */
1089             rt               = _mm_mul_ps(r10,vftabscale);
1090             vfitab           = _mm_cvttps_epi32(rt);
1091             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1092             vfitab           = _mm_slli_epi32(vfitab,2);
1093
1094             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1095             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1096             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1097             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1098             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1099             _MM_TRANSPOSE4_PS(Y,F,G,H);
1100             Heps             = _mm_mul_ps(vfeps,H);
1101             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1102             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1103             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1104
1105             fscal            = felec;
1106
1107             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1108
1109             /* Calculate temporary vectorial force */
1110             tx               = _mm_mul_ps(fscal,dx10);
1111             ty               = _mm_mul_ps(fscal,dy10);
1112             tz               = _mm_mul_ps(fscal,dz10);
1113
1114             /* Update vectorial force */
1115             fix1             = _mm_add_ps(fix1,tx);
1116             fiy1             = _mm_add_ps(fiy1,ty);
1117             fiz1             = _mm_add_ps(fiz1,tz);
1118
1119             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1120             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1121             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1122             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1123             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1124
1125             /**************************
1126              * CALCULATE INTERACTIONS *
1127              **************************/
1128
1129             r20              = _mm_mul_ps(rsq20,rinv20);
1130             r20              = _mm_andnot_ps(dummy_mask,r20);
1131
1132             /* Compute parameters for interactions between i and j atoms */
1133             qq20             = _mm_mul_ps(iq2,jq0);
1134
1135             /* Calculate table index by multiplying r with table scale and truncate to integer */
1136             rt               = _mm_mul_ps(r20,vftabscale);
1137             vfitab           = _mm_cvttps_epi32(rt);
1138             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1139             vfitab           = _mm_slli_epi32(vfitab,2);
1140
1141             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1142             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1143             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1144             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1145             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1146             _MM_TRANSPOSE4_PS(Y,F,G,H);
1147             Heps             = _mm_mul_ps(vfeps,H);
1148             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1149             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1150             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1151
1152             fscal            = felec;
1153
1154             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1155
1156             /* Calculate temporary vectorial force */
1157             tx               = _mm_mul_ps(fscal,dx20);
1158             ty               = _mm_mul_ps(fscal,dy20);
1159             tz               = _mm_mul_ps(fscal,dz20);
1160
1161             /* Update vectorial force */
1162             fix2             = _mm_add_ps(fix2,tx);
1163             fiy2             = _mm_add_ps(fiy2,ty);
1164             fiz2             = _mm_add_ps(fiz2,tz);
1165
1166             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1167             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1168             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1169             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1170             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1171
1172             /* Inner loop uses 128 flops */
1173         }
1174
1175         /* End of innermost loop */
1176
1177         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1178                                               f+i_coord_offset,fshift+i_shift_offset);
1179
1180         /* Increment number of inner iterations */
1181         inneriter                  += j_index_end - j_index_start;
1182
1183         /* Outer loop uses 18 flops */
1184     }
1185
1186     /* Increment number of outer iterations */
1187     outeriter        += nri;
1188
1189     /* Update outer/inner flops */
1190
1191     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*128);
1192 }