src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_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_ElecEw_VdwLJ_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_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          ewitab;
  89     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 120     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 121     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 122     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 156
 157         fix0             = _mm_setzero_ps();
 158         fiy0             = _mm_setzero_ps();
 159         fiz0             = _mm_setzero_ps();
 160         fix1             = _mm_setzero_ps();
 161         fiy1             = _mm_setzero_ps();
 162         fiz1             = _mm_setzero_ps();
 163         fix2             = _mm_setzero_ps();
 164         fiy2             = _mm_setzero_ps();
 165         fiz2             = _mm_setzero_ps();
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm_setzero_ps();
 169         vvdwsum          = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               x+j_coord_offsetC,x+j_coord_offsetD,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_ps(ix0,jx0);
 192             dy00             = _mm_sub_ps(iy0,jy0);
 193             dz00             = _mm_sub_ps(iz0,jz0);
 194             dx10             = _mm_sub_ps(ix1,jx0);
 195             dy10             = _mm_sub_ps(iy1,jy0);
 196             dz10             = _mm_sub_ps(iz1,jz0);
 197             dx20             = _mm_sub_ps(ix2,jx0);
 198             dy20             = _mm_sub_ps(iy2,jy0);
 199             dz20             = _mm_sub_ps(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 207             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 208             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 209
 210             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 211             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 212             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219             vdwjidx0C        = 2*vdwtype[jnrC+0];
 220             vdwjidx0D        = 2*vdwtype[jnrD+0];
 221
 222             /**************************
 223              * CALCULATE INTERACTIONS *
 224              **************************/
 225
 226             r00              = _mm_mul_ps(rsq00,rinv00);
 227
 228             /* Compute parameters for interactions between i and j atoms */
 229             qq00             = _mm_mul_ps(iq0,jq0);
 230             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 231                                          vdwparam+vdwioffset0+vdwjidx0B,
 232                                          vdwparam+vdwioffset0+vdwjidx0C,
 233                                          vdwparam+vdwioffset0+vdwjidx0D,
 234                                          &c6_00,&c12_00);
 235
 236             /* EWALD ELECTROSTATICS */
 237
 238             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 239             ewrt             = _mm_mul_ps(r00,ewtabscale);
 240             ewitab           = _mm_cvttps_epi32(ewrt);
 241             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 242             ewitab           = _mm_slli_epi32(ewitab,2);
 243             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 244             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 245             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 246             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 247             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 248             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 249             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 250             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 251             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 252
 253             /* LENNARD-JONES DISPERSION/REPULSION */
 254
 255             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 256             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 257             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 258             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 259             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm_add_ps(velecsum,velec);
 263             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 264
 265             fscal            = _mm_add_ps(felec,fvdw);
 266
 267             /* Calculate temporary vectorial force */
 268             tx               = _mm_mul_ps(fscal,dx00);
 269             ty               = _mm_mul_ps(fscal,dy00);
 270             tz               = _mm_mul_ps(fscal,dz00);
 271
 272             /* Update vectorial force */
 273             fix0             = _mm_add_ps(fix0,tx);
 274             fiy0             = _mm_add_ps(fiy0,ty);
 275             fiz0             = _mm_add_ps(fiz0,tz);
 276
 277             fjptrA             = f+j_coord_offsetA;
 278             fjptrB             = f+j_coord_offsetB;
 279             fjptrC             = f+j_coord_offsetC;
 280             fjptrD             = f+j_coord_offsetD;
 281             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 282
 283             /**************************
 284              * CALCULATE INTERACTIONS *
 285              **************************/
 286
 287             r10              = _mm_mul_ps(rsq10,rinv10);
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_ps(iq1,jq0);
 291
 292             /* EWALD ELECTROSTATICS */
 293
 294             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 295             ewrt             = _mm_mul_ps(r10,ewtabscale);
 296             ewitab           = _mm_cvttps_epi32(ewrt);
 297             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 298             ewitab           = _mm_slli_epi32(ewitab,2);
 299             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 300             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 301             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 302             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 303             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 304             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 305             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 306             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 307             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 308
 309             /* Update potential sum for this i atom from the interaction with this j atom. */
 310             velecsum         = _mm_add_ps(velecsum,velec);
 311
 312             fscal            = felec;
 313
 314             /* Calculate temporary vectorial force */
 315             tx               = _mm_mul_ps(fscal,dx10);
 316             ty               = _mm_mul_ps(fscal,dy10);
 317             tz               = _mm_mul_ps(fscal,dz10);
 318
 319             /* Update vectorial force */
 320             fix1             = _mm_add_ps(fix1,tx);
 321             fiy1             = _mm_add_ps(fiy1,ty);
 322             fiz1             = _mm_add_ps(fiz1,tz);
 323
 324             fjptrA             = f+j_coord_offsetA;
 325             fjptrB             = f+j_coord_offsetB;
 326             fjptrC             = f+j_coord_offsetC;
 327             fjptrD             = f+j_coord_offsetD;
 328             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             r20              = _mm_mul_ps(rsq20,rinv20);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* EWALD ELECTROSTATICS */
 340
 341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 342             ewrt             = _mm_mul_ps(r20,ewtabscale);
 343             ewitab           = _mm_cvttps_epi32(ewrt);
 344             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 345             ewitab           = _mm_slli_epi32(ewitab,2);
 346             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 347             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 348             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 349             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 350             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 351             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 352             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 353             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 354             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm_mul_ps(fscal,dx20);
 363             ty               = _mm_mul_ps(fscal,dy20);
 364             tz               = _mm_mul_ps(fscal,dz20);
 365
 366             /* Update vectorial force */
 367             fix2             = _mm_add_ps(fix2,tx);
 368             fiy2             = _mm_add_ps(fiy2,ty);
 369             fiz2             = _mm_add_ps(fiz2,tz);
 370
 371             fjptrA             = f+j_coord_offsetA;
 372             fjptrB             = f+j_coord_offsetB;
 373             fjptrC             = f+j_coord_offsetC;
 374             fjptrD             = f+j_coord_offsetD;
 375             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 376
 377             /* Inner loop uses 135 flops */
 378         }
 379
 380         if(jidx<j_index_end)
 381         {
 382
 383             /* Get j neighbor index, and coordinate index */
 384             jnrlistA         = jjnr[jidx];
 385             jnrlistB         = jjnr[jidx+1];
 386             jnrlistC         = jjnr[jidx+2];
 387             jnrlistD         = jjnr[jidx+3];
 388             /* Sign of each element will be negative for non-real atoms.
 389              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 390              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 391              */
 392             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 393             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 394             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 395             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 396             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 397             j_coord_offsetA  = DIM*jnrA;
 398             j_coord_offsetB  = DIM*jnrB;
 399             j_coord_offsetC  = DIM*jnrC;
 400             j_coord_offsetD  = DIM*jnrD;
 401
 402             /* load j atom coordinates */
 403             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 404                                               x+j_coord_offsetC,x+j_coord_offsetD,
 405                                               &jx0,&jy0,&jz0);
 406
 407             /* Calculate displacement vector */
 408             dx00             = _mm_sub_ps(ix0,jx0);
 409             dy00             = _mm_sub_ps(iy0,jy0);
 410             dz00             = _mm_sub_ps(iz0,jz0);
 411             dx10             = _mm_sub_ps(ix1,jx0);
 412             dy10             = _mm_sub_ps(iy1,jy0);
 413             dz10             = _mm_sub_ps(iz1,jz0);
 414             dx20             = _mm_sub_ps(ix2,jx0);
 415             dy20             = _mm_sub_ps(iy2,jy0);
 416             dz20             = _mm_sub_ps(iz2,jz0);
 417
 418             /* Calculate squared distance and things based on it */
 419             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 420             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 421             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 422
 423             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 424             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 425             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 426
 427             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 428             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 429             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 430
 431             /* Load parameters for j particles */
 432             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 433                                                               charge+jnrC+0,charge+jnrD+0);
 434             vdwjidx0A        = 2*vdwtype[jnrA+0];
 435             vdwjidx0B        = 2*vdwtype[jnrB+0];
 436             vdwjidx0C        = 2*vdwtype[jnrC+0];
 437             vdwjidx0D        = 2*vdwtype[jnrD+0];
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             r00              = _mm_mul_ps(rsq00,rinv00);
 444             r00              = _mm_andnot_ps(dummy_mask,r00);
 445
 446             /* Compute parameters for interactions between i and j atoms */
 447             qq00             = _mm_mul_ps(iq0,jq0);
 448             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 449                                          vdwparam+vdwioffset0+vdwjidx0B,
 450                                          vdwparam+vdwioffset0+vdwjidx0C,
 451                                          vdwparam+vdwioffset0+vdwjidx0D,
 452                                          &c6_00,&c12_00);
 453
 454             /* EWALD ELECTROSTATICS */
 455
 456             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 457             ewrt             = _mm_mul_ps(r00,ewtabscale);
 458             ewitab           = _mm_cvttps_epi32(ewrt);
 459             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 460             ewitab           = _mm_slli_epi32(ewitab,2);
 461             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 462             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 463             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 464             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 465             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 466             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 467             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 468             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 469             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 470
 471             /* LENNARD-JONES DISPERSION/REPULSION */
 472
 473             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 474             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 475             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 476             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 477             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm_andnot_ps(dummy_mask,velec);
 481             velecsum         = _mm_add_ps(velecsum,velec);
 482             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 483             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 484
 485             fscal            = _mm_add_ps(felec,fvdw);
 486
 487             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm_mul_ps(fscal,dx00);
 491             ty               = _mm_mul_ps(fscal,dy00);
 492             tz               = _mm_mul_ps(fscal,dz00);
 493
 494             /* Update vectorial force */
 495             fix0             = _mm_add_ps(fix0,tx);
 496             fiy0             = _mm_add_ps(fiy0,ty);
 497             fiz0             = _mm_add_ps(fiz0,tz);
 498
 499             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 500             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 501             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 502             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 503             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 504
 505             /**************************
 506              * CALCULATE INTERACTIONS *
 507              **************************/
 508
 509             r10              = _mm_mul_ps(rsq10,rinv10);
 510             r10              = _mm_andnot_ps(dummy_mask,r10);
 511
 512             /* Compute parameters for interactions between i and j atoms */
 513             qq10             = _mm_mul_ps(iq1,jq0);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 518             ewrt             = _mm_mul_ps(r10,ewtabscale);
 519             ewitab           = _mm_cvttps_epi32(ewrt);
 520             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 521             ewitab           = _mm_slli_epi32(ewitab,2);
 522             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 523             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 524             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 525             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 526             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 527             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 528             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 529             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 530             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 531
 532             /* Update potential sum for this i atom from the interaction with this j atom. */
 533             velec            = _mm_andnot_ps(dummy_mask,velec);
 534             velecsum         = _mm_add_ps(velecsum,velec);
 535
 536             fscal            = felec;
 537
 538             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 539
 540             /* Calculate temporary vectorial force */
 541             tx               = _mm_mul_ps(fscal,dx10);
 542             ty               = _mm_mul_ps(fscal,dy10);
 543             tz               = _mm_mul_ps(fscal,dz10);
 544
 545             /* Update vectorial force */
 546             fix1             = _mm_add_ps(fix1,tx);
 547             fiy1             = _mm_add_ps(fiy1,ty);
 548             fiz1             = _mm_add_ps(fiz1,tz);
 549
 550             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 551             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 552             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 553             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 554             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 555
 556             /**************************
 557              * CALCULATE INTERACTIONS *
 558              **************************/
 559
 560             r20              = _mm_mul_ps(rsq20,rinv20);
 561             r20              = _mm_andnot_ps(dummy_mask,r20);
 562
 563             /* Compute parameters for interactions between i and j atoms */
 564             qq20             = _mm_mul_ps(iq2,jq0);
 565
 566             /* EWALD ELECTROSTATICS */
 567
 568             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 569             ewrt             = _mm_mul_ps(r20,ewtabscale);
 570             ewitab           = _mm_cvttps_epi32(ewrt);
 571             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 572             ewitab           = _mm_slli_epi32(ewitab,2);
 573             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 574             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 575             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 576             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 577             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 578             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 579             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 580             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 581             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 582
 583             /* Update potential sum for this i atom from the interaction with this j atom. */
 584             velec            = _mm_andnot_ps(dummy_mask,velec);
 585             velecsum         = _mm_add_ps(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm_mul_ps(fscal,dx20);
 593             ty               = _mm_mul_ps(fscal,dy20);
 594             tz               = _mm_mul_ps(fscal,dz20);
 595
 596             /* Update vectorial force */
 597             fix2             = _mm_add_ps(fix2,tx);
 598             fiy2             = _mm_add_ps(fiy2,ty);
 599             fiz2             = _mm_add_ps(fiz2,tz);
 600
 601             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 602             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 603             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 604             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 605             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 606
 607             /* Inner loop uses 138 flops */
 608         }
 609
 610         /* End of innermost loop */
 611
 612         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 613                                               f+i_coord_offset,fshift+i_shift_offset);
 614
 615         ggid                        = gid[iidx];
 616         /* Update potential energies */
 617         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 618         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 619
 620         /* Increment number of inner iterations */
 621         inneriter                  += j_index_end - j_index_start;
 622
 623         /* Outer loop uses 20 flops */
 624     }
 625
 626     /* Increment number of outer iterations */
 627     outeriter        += nri;
 628
 629     /* Update outer/inner flops */
 630
 631     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 632 }
 633 /*
 634  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_single
 635  * Electrostatics interaction: Ewald
 636  * VdW interaction:            LennardJones
 637  * Geometry:                   Water3-Particle
 638  * Calculate force/pot:        Force
 639  */
 640 void
 641 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_single
 642                     (t_nblist * gmx_restrict                nlist,
 643                      rvec * gmx_restrict                    xx,
 644                      rvec * gmx_restrict                    ff,
 645                      t_forcerec * gmx_restrict              fr,
 646                      t_mdatoms * gmx_restrict               mdatoms,
 647                      nb_kernel_data_t * gmx_restrict        kernel_data,
 648                      t_nrnb * gmx_restrict                  nrnb)
 649 {
 650     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 651      * just 0 for non-waters.
 652      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 653      * jnr indices corresponding to data put in the four positions in the SIMD register.
 654      */
 655     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 656     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 657     int              jnrA,jnrB,jnrC,jnrD;
 658     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 659     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 664     real             scratch[4*DIM];
 665     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 666     int              vdwioffset0;
 667     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 668     int              vdwioffset1;
 669     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 670     int              vdwioffset2;
 671     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 672     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 673     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 674     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 675     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 676     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 677     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 678     real             *charge;
 679     int              nvdwtype;
 680     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 681     int              *vdwtype;
 682     real             *vdwparam;
 683     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 684     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 685     __m128i          ewitab;
 686     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 687     real             *ewtab;
 688     __m128           dummy_mask,cutoff_mask;
 689     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 690     __m128           one     = _mm_set1_ps(1.0);
 691     __m128           two     = _mm_set1_ps(2.0);
 692     x                = xx[0];
 693     f                = ff[0];
 694
 695     nri              = nlist->nri;
 696     iinr             = nlist->iinr;
 697     jindex           = nlist->jindex;
 698     jjnr             = nlist->jjnr;
 699     shiftidx         = nlist->shift;
 700     gid              = nlist->gid;
 701     shiftvec         = fr->shift_vec[0];
 702     fshift           = fr->fshift[0];
 703     facel            = _mm_set1_ps(fr->epsfac);
 704     charge           = mdatoms->chargeA;
 705     nvdwtype         = fr->ntype;
 706     vdwparam         = fr->nbfp;
 707     vdwtype          = mdatoms->typeA;
 708
 709     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 710     ewtab            = fr->ic->tabq_coul_F;
 711     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 712     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 713
 714     /* Setup water-specific parameters */
 715     inr              = nlist->iinr[0];
 716     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 717     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 718     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 719     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 720
 721     /* Avoid stupid compiler warnings */
 722     jnrA = jnrB = jnrC = jnrD = 0;
 723     j_coord_offsetA = 0;
 724     j_coord_offsetB = 0;
 725     j_coord_offsetC = 0;
 726     j_coord_offsetD = 0;
 727
 728     outeriter        = 0;
 729     inneriter        = 0;
 730
 731     for(iidx=0;iidx<4*DIM;iidx++)
 732     {
 733         scratch[iidx] = 0.0;
 734     }
 735
 736     /* Start outer loop over neighborlists */
 737     for(iidx=0; iidx<nri; iidx++)
 738     {
 739         /* Load shift vector for this list */
 740         i_shift_offset   = DIM*shiftidx[iidx];
 741
 742         /* Load limits for loop over neighbors */
 743         j_index_start    = jindex[iidx];
 744         j_index_end      = jindex[iidx+1];
 745
 746         /* Get outer coordinate index */
 747         inr              = iinr[iidx];
 748         i_coord_offset   = DIM*inr;
 749
 750         /* Load i particle coords and add shift vector */
 751         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 752                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 753
 754         fix0             = _mm_setzero_ps();
 755         fiy0             = _mm_setzero_ps();
 756         fiz0             = _mm_setzero_ps();
 757         fix1             = _mm_setzero_ps();
 758         fiy1             = _mm_setzero_ps();
 759         fiz1             = _mm_setzero_ps();
 760         fix2             = _mm_setzero_ps();
 761         fiy2             = _mm_setzero_ps();
 762         fiz2             = _mm_setzero_ps();
 763
 764         /* Start inner kernel loop */
 765         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 766         {
 767
 768             /* Get j neighbor index, and coordinate index */
 769             jnrA             = jjnr[jidx];
 770             jnrB             = jjnr[jidx+1];
 771             jnrC             = jjnr[jidx+2];
 772             jnrD             = jjnr[jidx+3];
 773             j_coord_offsetA  = DIM*jnrA;
 774             j_coord_offsetB  = DIM*jnrB;
 775             j_coord_offsetC  = DIM*jnrC;
 776             j_coord_offsetD  = DIM*jnrD;
 777
 778             /* load j atom coordinates */
 779             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 780                                               x+j_coord_offsetC,x+j_coord_offsetD,
 781                                               &jx0,&jy0,&jz0);
 782
 783             /* Calculate displacement vector */
 784             dx00             = _mm_sub_ps(ix0,jx0);
 785             dy00             = _mm_sub_ps(iy0,jy0);
 786             dz00             = _mm_sub_ps(iz0,jz0);
 787             dx10             = _mm_sub_ps(ix1,jx0);
 788             dy10             = _mm_sub_ps(iy1,jy0);
 789             dz10             = _mm_sub_ps(iz1,jz0);
 790             dx20             = _mm_sub_ps(ix2,jx0);
 791             dy20             = _mm_sub_ps(iy2,jy0);
 792             dz20             = _mm_sub_ps(iz2,jz0);
 793
 794             /* Calculate squared distance and things based on it */
 795             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 796             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 797             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 798
 799             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 800             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 801             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 802
 803             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 804             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 805             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 806
 807             /* Load parameters for j particles */
 808             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 809                                                               charge+jnrC+0,charge+jnrD+0);
 810             vdwjidx0A        = 2*vdwtype[jnrA+0];
 811             vdwjidx0B        = 2*vdwtype[jnrB+0];
 812             vdwjidx0C        = 2*vdwtype[jnrC+0];
 813             vdwjidx0D        = 2*vdwtype[jnrD+0];
 814
 815             /**************************
 816              * CALCULATE INTERACTIONS *
 817              **************************/
 818
 819             r00              = _mm_mul_ps(rsq00,rinv00);
 820
 821             /* Compute parameters for interactions between i and j atoms */
 822             qq00             = _mm_mul_ps(iq0,jq0);
 823             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 824                                          vdwparam+vdwioffset0+vdwjidx0B,
 825                                          vdwparam+vdwioffset0+vdwjidx0C,
 826                                          vdwparam+vdwioffset0+vdwjidx0D,
 827                                          &c6_00,&c12_00);
 828
 829             /* EWALD ELECTROSTATICS */
 830
 831             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 832             ewrt             = _mm_mul_ps(r00,ewtabscale);
 833             ewitab           = _mm_cvttps_epi32(ewrt);
 834             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 835             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 836                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 837                                          &ewtabF,&ewtabFn);
 838             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 839             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 840
 841             /* LENNARD-JONES DISPERSION/REPULSION */
 842
 843             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 844             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 845
 846             fscal            = _mm_add_ps(felec,fvdw);
 847
 848             /* Calculate temporary vectorial force */
 849             tx               = _mm_mul_ps(fscal,dx00);
 850             ty               = _mm_mul_ps(fscal,dy00);
 851             tz               = _mm_mul_ps(fscal,dz00);
 852
 853             /* Update vectorial force */
 854             fix0             = _mm_add_ps(fix0,tx);
 855             fiy0             = _mm_add_ps(fiy0,ty);
 856             fiz0             = _mm_add_ps(fiz0,tz);
 857
 858             fjptrA             = f+j_coord_offsetA;
 859             fjptrB             = f+j_coord_offsetB;
 860             fjptrC             = f+j_coord_offsetC;
 861             fjptrD             = f+j_coord_offsetD;
 862             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 863
 864             /**************************
 865              * CALCULATE INTERACTIONS *
 866              **************************/
 867
 868             r10              = _mm_mul_ps(rsq10,rinv10);
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq10             = _mm_mul_ps(iq1,jq0);
 872
 873             /* EWALD ELECTROSTATICS */
 874
 875             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 876             ewrt             = _mm_mul_ps(r10,ewtabscale);
 877             ewitab           = _mm_cvttps_epi32(ewrt);
 878             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 879             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 880                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 881                                          &ewtabF,&ewtabFn);
 882             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 883             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 884
 885             fscal            = felec;
 886
 887             /* Calculate temporary vectorial force */
 888             tx               = _mm_mul_ps(fscal,dx10);
 889             ty               = _mm_mul_ps(fscal,dy10);
 890             tz               = _mm_mul_ps(fscal,dz10);
 891
 892             /* Update vectorial force */
 893             fix1             = _mm_add_ps(fix1,tx);
 894             fiy1             = _mm_add_ps(fiy1,ty);
 895             fiz1             = _mm_add_ps(fiz1,tz);
 896
 897             fjptrA             = f+j_coord_offsetA;
 898             fjptrB             = f+j_coord_offsetB;
 899             fjptrC             = f+j_coord_offsetC;
 900             fjptrD             = f+j_coord_offsetD;
 901             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 902
 903             /**************************
 904              * CALCULATE INTERACTIONS *
 905              **************************/
 906
 907             r20              = _mm_mul_ps(rsq20,rinv20);
 908
 909             /* Compute parameters for interactions between i and j atoms */
 910             qq20             = _mm_mul_ps(iq2,jq0);
 911
 912             /* EWALD ELECTROSTATICS */
 913
 914             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 915             ewrt             = _mm_mul_ps(r20,ewtabscale);
 916             ewitab           = _mm_cvttps_epi32(ewrt);
 917             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 918             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 919                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 920                                          &ewtabF,&ewtabFn);
 921             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 922             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 923
 924             fscal            = felec;
 925
 926             /* Calculate temporary vectorial force */
 927             tx               = _mm_mul_ps(fscal,dx20);
 928             ty               = _mm_mul_ps(fscal,dy20);
 929             tz               = _mm_mul_ps(fscal,dz20);
 930
 931             /* Update vectorial force */
 932             fix2             = _mm_add_ps(fix2,tx);
 933             fiy2             = _mm_add_ps(fiy2,ty);
 934             fiz2             = _mm_add_ps(fiz2,tz);
 935
 936             fjptrA             = f+j_coord_offsetA;
 937             fjptrB             = f+j_coord_offsetB;
 938             fjptrC             = f+j_coord_offsetC;
 939             fjptrD             = f+j_coord_offsetD;
 940             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 941
 942             /* Inner loop uses 115 flops */
 943         }
 944
 945         if(jidx<j_index_end)
 946         {
 947
 948             /* Get j neighbor index, and coordinate index */
 949             jnrlistA         = jjnr[jidx];
 950             jnrlistB         = jjnr[jidx+1];
 951             jnrlistC         = jjnr[jidx+2];
 952             jnrlistD         = jjnr[jidx+3];
 953             /* Sign of each element will be negative for non-real atoms.
 954              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 955              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 956              */
 957             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 958             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 959             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 960             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 961             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 962             j_coord_offsetA  = DIM*jnrA;
 963             j_coord_offsetB  = DIM*jnrB;
 964             j_coord_offsetC  = DIM*jnrC;
 965             j_coord_offsetD  = DIM*jnrD;
 966
 967             /* load j atom coordinates */
 968             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 969                                               x+j_coord_offsetC,x+j_coord_offsetD,
 970                                               &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm_sub_ps(ix0,jx0);
 974             dy00             = _mm_sub_ps(iy0,jy0);
 975             dz00             = _mm_sub_ps(iz0,jz0);
 976             dx10             = _mm_sub_ps(ix1,jx0);
 977             dy10             = _mm_sub_ps(iy1,jy0);
 978             dz10             = _mm_sub_ps(iz1,jz0);
 979             dx20             = _mm_sub_ps(ix2,jx0);
 980             dy20             = _mm_sub_ps(iy2,jy0);
 981             dz20             = _mm_sub_ps(iz2,jz0);
 982
 983             /* Calculate squared distance and things based on it */
 984             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 985             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 986             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 987
 988             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 989             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 990             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 991
 992             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 993             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 994             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 995
 996             /* Load parameters for j particles */
 997             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 998                                                               charge+jnrC+0,charge+jnrD+0);
 999             vdwjidx0A        = 2*vdwtype[jnrA+0];
1000             vdwjidx0B        = 2*vdwtype[jnrB+0];
1001             vdwjidx0C        = 2*vdwtype[jnrC+0];
1002             vdwjidx0D        = 2*vdwtype[jnrD+0];
1003
1004             /**************************
1005              * CALCULATE INTERACTIONS *
1006              **************************/
1007
1008             r00              = _mm_mul_ps(rsq00,rinv00);
1009             r00              = _mm_andnot_ps(dummy_mask,r00);
1010
1011             /* Compute parameters for interactions between i and j atoms */
1012             qq00             = _mm_mul_ps(iq0,jq0);
1013             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1014                                          vdwparam+vdwioffset0+vdwjidx0B,
1015                                          vdwparam+vdwioffset0+vdwjidx0C,
1016                                          vdwparam+vdwioffset0+vdwjidx0D,
1017                                          &c6_00,&c12_00);
1018
1019             /* EWALD ELECTROSTATICS */
1020
1021             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1022             ewrt             = _mm_mul_ps(r00,ewtabscale);
1023             ewitab           = _mm_cvttps_epi32(ewrt);
1024             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1025             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1026                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1027                                          &ewtabF,&ewtabFn);
1028             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1029             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1030
1031             /* LENNARD-JONES DISPERSION/REPULSION */
1032
1033             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1034             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1035
1036             fscal            = _mm_add_ps(felec,fvdw);
1037
1038             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1039
1040             /* Calculate temporary vectorial force */
1041             tx               = _mm_mul_ps(fscal,dx00);
1042             ty               = _mm_mul_ps(fscal,dy00);
1043             tz               = _mm_mul_ps(fscal,dz00);
1044
1045             /* Update vectorial force */
1046             fix0             = _mm_add_ps(fix0,tx);
1047             fiy0             = _mm_add_ps(fiy0,ty);
1048             fiz0             = _mm_add_ps(fiz0,tz);
1049
1050             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1051             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1052             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1053             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1054             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             r10              = _mm_mul_ps(rsq10,rinv10);
1061             r10              = _mm_andnot_ps(dummy_mask,r10);
1062
1063             /* Compute parameters for interactions between i and j atoms */
1064             qq10             = _mm_mul_ps(iq1,jq0);
1065
1066             /* EWALD ELECTROSTATICS */
1067
1068             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1069             ewrt             = _mm_mul_ps(r10,ewtabscale);
1070             ewitab           = _mm_cvttps_epi32(ewrt);
1071             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1072             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1073                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1074                                          &ewtabF,&ewtabFn);
1075             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1076             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1077
1078             fscal            = felec;
1079
1080             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1081
1082             /* Calculate temporary vectorial force */
1083             tx               = _mm_mul_ps(fscal,dx10);
1084             ty               = _mm_mul_ps(fscal,dy10);
1085             tz               = _mm_mul_ps(fscal,dz10);
1086
1087             /* Update vectorial force */
1088             fix1             = _mm_add_ps(fix1,tx);
1089             fiy1             = _mm_add_ps(fiy1,ty);
1090             fiz1             = _mm_add_ps(fiz1,tz);
1091
1092             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1093             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1094             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1095             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1096             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1097
1098             /**************************
1099              * CALCULATE INTERACTIONS *
1100              **************************/
1101
1102             r20              = _mm_mul_ps(rsq20,rinv20);
1103             r20              = _mm_andnot_ps(dummy_mask,r20);
1104
1105             /* Compute parameters for interactions between i and j atoms */
1106             qq20             = _mm_mul_ps(iq2,jq0);
1107
1108             /* EWALD ELECTROSTATICS */
1109
1110             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1111             ewrt             = _mm_mul_ps(r20,ewtabscale);
1112             ewitab           = _mm_cvttps_epi32(ewrt);
1113             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1114             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1115                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1116                                          &ewtabF,&ewtabFn);
1117             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1118             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1119
1120             fscal            = felec;
1121
1122             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1123
1124             /* Calculate temporary vectorial force */
1125             tx               = _mm_mul_ps(fscal,dx20);
1126             ty               = _mm_mul_ps(fscal,dy20);
1127             tz               = _mm_mul_ps(fscal,dz20);
1128
1129             /* Update vectorial force */
1130             fix2             = _mm_add_ps(fix2,tx);
1131             fiy2             = _mm_add_ps(fiy2,ty);
1132             fiz2             = _mm_add_ps(fiz2,tz);
1133
1134             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1135             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1136             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1137             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1138             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
1139
1140             /* Inner loop uses 118 flops */
1141         }
1142
1143         /* End of innermost loop */
1144
1145         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1146                                               f+i_coord_offset,fshift+i_shift_offset);
1147
1148         /* Increment number of inner iterations */
1149         inneriter                  += j_index_end - j_index_start;
1150
1151         /* Outer loop uses 18 flops */
1152     }
1153
1154     /* Increment number of outer iterations */
1155     outeriter        += nri;
1156
1157     /* Update outer/inner flops */
1158
1159     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1160 }