src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSw_VdwNone_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128i          ewitab;
  80     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  83     real             rswitch_scalar,d_scalar;
  84     __m128           dummy_mask,cutoff_mask;
  85     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  86     __m128           one     = _mm_set1_ps(1.0);
  87     __m128           two     = _mm_set1_ps(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm_set1_ps(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101
 102     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 103     ewtab            = fr->ic->tabq_coul_FDV0;
 104     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 105     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 106
 107     /* Setup water-specific parameters */
 108     inr              = nlist->iinr[0];
 109     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 110     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 111     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 112
 113     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 114     rcutoff_scalar   = fr->rcoulomb;
 115     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 116     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 117
 118     rswitch_scalar   = fr->rcoulomb_switch;
 119     rswitch          = _mm_set1_ps(rswitch_scalar);
 120     /* Setup switch parameters */
 121     d_scalar         = rcutoff_scalar-rswitch_scalar;
 122     d                = _mm_set1_ps(d_scalar);
 123     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 124     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 125     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 126     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 127     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 128     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 129
 130     /* Avoid stupid compiler warnings */
 131     jnrA = jnrB = jnrC = jnrD = 0;
 132     j_coord_offsetA = 0;
 133     j_coord_offsetB = 0;
 134     j_coord_offsetC = 0;
 135     j_coord_offsetD = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145         shX              = shiftvec[i_shift_offset+XX];
 146         shY              = shiftvec[i_shift_offset+YY];
 147         shZ              = shiftvec[i_shift_offset+ZZ];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 159         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 160         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 161         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 162         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 163         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 164         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 165         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 166         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 167
 168         fix1             = _mm_setzero_ps();
 169         fiy1             = _mm_setzero_ps();
 170         fiz1             = _mm_setzero_ps();
 171         fix2             = _mm_setzero_ps();
 172         fiy2             = _mm_setzero_ps();
 173         fiz2             = _mm_setzero_ps();
 174         fix3             = _mm_setzero_ps();
 175         fiy3             = _mm_setzero_ps();
 176         fiz3             = _mm_setzero_ps();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190
 191             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx10             = _mm_sub_ps(ix1,jx0);
 203             dy10             = _mm_sub_ps(iy1,jy0);
 204             dz10             = _mm_sub_ps(iz1,jz0);
 205             dx20             = _mm_sub_ps(ix2,jx0);
 206             dy20             = _mm_sub_ps(iy2,jy0);
 207             dz20             = _mm_sub_ps(iz2,jz0);
 208             dx30             = _mm_sub_ps(ix3,jx0);
 209             dy30             = _mm_sub_ps(iy3,jy0);
 210             dz30             = _mm_sub_ps(iz3,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 215             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 216
 217             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 218             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 219             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 220
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq10,rcutoff2))
 234             {
 235
 236             r10              = _mm_mul_ps(rsq10,rinv10);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq10             = _mm_mul_ps(iq1,jq0);
 240
 241             /* EWALD ELECTROSTATICS */
 242
 243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 244             ewrt             = _mm_mul_ps(r10,ewtabscale);
 245             ewitab           = _mm_cvttps_epi32(ewrt);
 246             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 247             ewitab           = _mm_slli_epi32(ewitab,2);
 248             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 249             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 250             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 251             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 252             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 253             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 254             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 255             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 256             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 257
 258             d                = _mm_sub_ps(r10,rswitch);
 259             d                = _mm_max_ps(d,_mm_setzero_ps());
 260             d2               = _mm_mul_ps(d,d);
 261             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 262
 263             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 264
 265             /* Evaluate switch function */
 266             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 267             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 268             velec            = _mm_mul_ps(velec,sw);
 269             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             velec            = _mm_and_ps(velec,cutoff_mask);
 273             velecsum         = _mm_add_ps(velecsum,velec);
 274
 275             fscal            = felec;
 276
 277             fscal            = _mm_and_ps(fscal,cutoff_mask);
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm_mul_ps(fscal,dx10);
 281             ty               = _mm_mul_ps(fscal,dy10);
 282             tz               = _mm_mul_ps(fscal,dz10);
 283
 284             /* Update vectorial force */
 285             fix1             = _mm_add_ps(fix1,tx);
 286             fiy1             = _mm_add_ps(fiy1,ty);
 287             fiz1             = _mm_add_ps(fiz1,tz);
 288
 289             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 290                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 291                                                    tx,ty,tz);
 292
 293             }
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             if (gmx_mm_any_lt(rsq20,rcutoff2))
 300             {
 301
 302             r20              = _mm_mul_ps(rsq20,rinv20);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq20             = _mm_mul_ps(iq2,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_ps(r20,ewtabscale);
 311             ewitab           = _mm_cvttps_epi32(ewrt);
 312             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 313             ewitab           = _mm_slli_epi32(ewitab,2);
 314             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 315             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 316             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 317             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 318             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 319             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 320             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 321             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 322             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 323
 324             d                = _mm_sub_ps(r20,rswitch);
 325             d                = _mm_max_ps(d,_mm_setzero_ps());
 326             d2               = _mm_mul_ps(d,d);
 327             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 328
 329             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 330
 331             /* Evaluate switch function */
 332             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 333             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 334             velec            = _mm_mul_ps(velec,sw);
 335             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velec            = _mm_and_ps(velec,cutoff_mask);
 339             velecsum         = _mm_add_ps(velecsum,velec);
 340
 341             fscal            = felec;
 342
 343             fscal            = _mm_and_ps(fscal,cutoff_mask);
 344
 345             /* Calculate temporary vectorial force */
 346             tx               = _mm_mul_ps(fscal,dx20);
 347             ty               = _mm_mul_ps(fscal,dy20);
 348             tz               = _mm_mul_ps(fscal,dz20);
 349
 350             /* Update vectorial force */
 351             fix2             = _mm_add_ps(fix2,tx);
 352             fiy2             = _mm_add_ps(fiy2,ty);
 353             fiz2             = _mm_add_ps(fiz2,tz);
 354
 355             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 356                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 357                                                    tx,ty,tz);
 358
 359             }
 360
 361             /**************************
 362              * CALCULATE INTERACTIONS *
 363              **************************/
 364
 365             if (gmx_mm_any_lt(rsq30,rcutoff2))
 366             {
 367
 368             r30              = _mm_mul_ps(rsq30,rinv30);
 369
 370             /* Compute parameters for interactions between i and j atoms */
 371             qq30             = _mm_mul_ps(iq3,jq0);
 372
 373             /* EWALD ELECTROSTATICS */
 374
 375             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 376             ewrt             = _mm_mul_ps(r30,ewtabscale);
 377             ewitab           = _mm_cvttps_epi32(ewrt);
 378             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 379             ewitab           = _mm_slli_epi32(ewitab,2);
 380             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 381             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 382             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 383             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 384             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 385             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 386             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 387             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 388             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 389
 390             d                = _mm_sub_ps(r30,rswitch);
 391             d                = _mm_max_ps(d,_mm_setzero_ps());
 392             d2               = _mm_mul_ps(d,d);
 393             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 394
 395             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 396
 397             /* Evaluate switch function */
 398             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 399             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 400             velec            = _mm_mul_ps(velec,sw);
 401             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 402
 403             /* Update potential sum for this i atom from the interaction with this j atom. */
 404             velec            = _mm_and_ps(velec,cutoff_mask);
 405             velecsum         = _mm_add_ps(velecsum,velec);
 406
 407             fscal            = felec;
 408
 409             fscal            = _mm_and_ps(fscal,cutoff_mask);
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm_mul_ps(fscal,dx30);
 413             ty               = _mm_mul_ps(fscal,dy30);
 414             tz               = _mm_mul_ps(fscal,dz30);
 415
 416             /* Update vectorial force */
 417             fix3             = _mm_add_ps(fix3,tx);
 418             fiy3             = _mm_add_ps(fiy3,ty);
 419             fiz3             = _mm_add_ps(fiz3,tz);
 420
 421             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 422                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 423                                                    tx,ty,tz);
 424
 425             }
 426
 427             /* Inner loop uses 195 flops */
 428         }
 429
 430         if(jidx<j_index_end)
 431         {
 432
 433             /* Get j neighbor index, and coordinate index */
 434             jnrA             = jjnr[jidx];
 435             jnrB             = jjnr[jidx+1];
 436             jnrC             = jjnr[jidx+2];
 437             jnrD             = jjnr[jidx+3];
 438
 439             /* Sign of each element will be negative for non-real atoms.
 440              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 441              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 442              */
 443             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 444             jnrA       = (jnrA>=0) ? jnrA : 0;
 445             jnrB       = (jnrB>=0) ? jnrB : 0;
 446             jnrC       = (jnrC>=0) ? jnrC : 0;
 447             jnrD       = (jnrD>=0) ? jnrD : 0;
 448
 449             j_coord_offsetA  = DIM*jnrA;
 450             j_coord_offsetB  = DIM*jnrB;
 451             j_coord_offsetC  = DIM*jnrC;
 452             j_coord_offsetD  = DIM*jnrD;
 453
 454             /* load j atom coordinates */
 455             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 456                                               x+j_coord_offsetC,x+j_coord_offsetD,
 457                                               &jx0,&jy0,&jz0);
 458
 459             /* Calculate displacement vector */
 460             dx10             = _mm_sub_ps(ix1,jx0);
 461             dy10             = _mm_sub_ps(iy1,jy0);
 462             dz10             = _mm_sub_ps(iz1,jz0);
 463             dx20             = _mm_sub_ps(ix2,jx0);
 464             dy20             = _mm_sub_ps(iy2,jy0);
 465             dz20             = _mm_sub_ps(iz2,jz0);
 466             dx30             = _mm_sub_ps(ix3,jx0);
 467             dy30             = _mm_sub_ps(iy3,jy0);
 468             dz30             = _mm_sub_ps(iz3,jz0);
 469
 470             /* Calculate squared distance and things based on it */
 471             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 472             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 473             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 474
 475             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 476             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 477             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 478
 479             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 480             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 481             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 482
 483             /* Load parameters for j particles */
 484             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 485                                                               charge+jnrC+0,charge+jnrD+0);
 486
 487             /**************************
 488              * CALCULATE INTERACTIONS *
 489              **************************/
 490
 491             if (gmx_mm_any_lt(rsq10,rcutoff2))
 492             {
 493
 494             r10              = _mm_mul_ps(rsq10,rinv10);
 495             r10              = _mm_andnot_ps(dummy_mask,r10);
 496
 497             /* Compute parameters for interactions between i and j atoms */
 498             qq10             = _mm_mul_ps(iq1,jq0);
 499
 500             /* EWALD ELECTROSTATICS */
 501
 502             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 503             ewrt             = _mm_mul_ps(r10,ewtabscale);
 504             ewitab           = _mm_cvttps_epi32(ewrt);
 505             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 506             ewitab           = _mm_slli_epi32(ewitab,2);
 507             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 508             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 509             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 510             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 511             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 512             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 513             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 514             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 515             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 516
 517             d                = _mm_sub_ps(r10,rswitch);
 518             d                = _mm_max_ps(d,_mm_setzero_ps());
 519             d2               = _mm_mul_ps(d,d);
 520             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 521
 522             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 523
 524             /* Evaluate switch function */
 525             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 526             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 527             velec            = _mm_mul_ps(velec,sw);
 528             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_and_ps(velec,cutoff_mask);
 532             velec            = _mm_andnot_ps(dummy_mask,velec);
 533             velecsum         = _mm_add_ps(velecsum,velec);
 534
 535             fscal            = felec;
 536
 537             fscal            = _mm_and_ps(fscal,cutoff_mask);
 538
 539             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 540
 541             /* Calculate temporary vectorial force */
 542             tx               = _mm_mul_ps(fscal,dx10);
 543             ty               = _mm_mul_ps(fscal,dy10);
 544             tz               = _mm_mul_ps(fscal,dz10);
 545
 546             /* Update vectorial force */
 547             fix1             = _mm_add_ps(fix1,tx);
 548             fiy1             = _mm_add_ps(fiy1,ty);
 549             fiz1             = _mm_add_ps(fiz1,tz);
 550
 551             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 552                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 553                                                    tx,ty,tz);
 554
 555             }
 556
 557             /**************************
 558              * CALCULATE INTERACTIONS *
 559              **************************/
 560
 561             if (gmx_mm_any_lt(rsq20,rcutoff2))
 562             {
 563
 564             r20              = _mm_mul_ps(rsq20,rinv20);
 565             r20              = _mm_andnot_ps(dummy_mask,r20);
 566
 567             /* Compute parameters for interactions between i and j atoms */
 568             qq20             = _mm_mul_ps(iq2,jq0);
 569
 570             /* EWALD ELECTROSTATICS */
 571
 572             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 573             ewrt             = _mm_mul_ps(r20,ewtabscale);
 574             ewitab           = _mm_cvttps_epi32(ewrt);
 575             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 576             ewitab           = _mm_slli_epi32(ewitab,2);
 577             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 578             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 579             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 580             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 581             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 582             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 583             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 584             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 585             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 586
 587             d                = _mm_sub_ps(r20,rswitch);
 588             d                = _mm_max_ps(d,_mm_setzero_ps());
 589             d2               = _mm_mul_ps(d,d);
 590             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 591
 592             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 593
 594             /* Evaluate switch function */
 595             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 596             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
 597             velec            = _mm_mul_ps(velec,sw);
 598             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 599
 600             /* Update potential sum for this i atom from the interaction with this j atom. */
 601             velec            = _mm_and_ps(velec,cutoff_mask);
 602             velec            = _mm_andnot_ps(dummy_mask,velec);
 603             velecsum         = _mm_add_ps(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm_and_ps(fscal,cutoff_mask);
 608
 609             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 610
 611             /* Calculate temporary vectorial force */
 612             tx               = _mm_mul_ps(fscal,dx20);
 613             ty               = _mm_mul_ps(fscal,dy20);
 614             tz               = _mm_mul_ps(fscal,dz20);
 615
 616             /* Update vectorial force */
 617             fix2             = _mm_add_ps(fix2,tx);
 618             fiy2             = _mm_add_ps(fiy2,ty);
 619             fiz2             = _mm_add_ps(fiz2,tz);
 620
 621             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 622                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 623                                                    tx,ty,tz);
 624
 625             }
 626
 627             /**************************
 628              * CALCULATE INTERACTIONS *
 629              **************************/
 630
 631             if (gmx_mm_any_lt(rsq30,rcutoff2))
 632             {
 633
 634             r30              = _mm_mul_ps(rsq30,rinv30);
 635             r30              = _mm_andnot_ps(dummy_mask,r30);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq30             = _mm_mul_ps(iq3,jq0);
 639
 640             /* EWALD ELECTROSTATICS */
 641
 642             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 643             ewrt             = _mm_mul_ps(r30,ewtabscale);
 644             ewitab           = _mm_cvttps_epi32(ewrt);
 645             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 646             ewitab           = _mm_slli_epi32(ewitab,2);
 647             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 648             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 649             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 650             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 651             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 652             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 653             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 654             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 655             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 656
 657             d                = _mm_sub_ps(r30,rswitch);
 658             d                = _mm_max_ps(d,_mm_setzero_ps());
 659             d2               = _mm_mul_ps(d,d);
 660             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 661
 662             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 663
 664             /* Evaluate switch function */
 665             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 666             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
 667             velec            = _mm_mul_ps(velec,sw);
 668             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 669
 670             /* Update potential sum for this i atom from the interaction with this j atom. */
 671             velec            = _mm_and_ps(velec,cutoff_mask);
 672             velec            = _mm_andnot_ps(dummy_mask,velec);
 673             velecsum         = _mm_add_ps(velecsum,velec);
 674
 675             fscal            = felec;
 676
 677             fscal            = _mm_and_ps(fscal,cutoff_mask);
 678
 679             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 680
 681             /* Calculate temporary vectorial force */
 682             tx               = _mm_mul_ps(fscal,dx30);
 683             ty               = _mm_mul_ps(fscal,dy30);
 684             tz               = _mm_mul_ps(fscal,dz30);
 685
 686             /* Update vectorial force */
 687             fix3             = _mm_add_ps(fix3,tx);
 688             fiy3             = _mm_add_ps(fiy3,ty);
 689             fiz3             = _mm_add_ps(fiz3,tz);
 690
 691             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 692                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 693                                                    tx,ty,tz);
 694
 695             }
 696
 697             /* Inner loop uses 198 flops */
 698         }
 699
 700         /* End of innermost loop */
 701
 702         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 703                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 704
 705         ggid                        = gid[iidx];
 706         /* Update potential energies */
 707         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 708
 709         /* Increment number of inner iterations */
 710         inneriter                  += j_index_end - j_index_start;
 711
 712         /* Outer loop uses 28 flops */
 713     }
 714
 715     /* Increment number of outer iterations */
 716     outeriter        += nri;
 717
 718     /* Update outer/inner flops */
 719
 720     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*28 + inneriter*198);
 721 }
 722 /*
 723  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse2_single
 724  * Electrostatics interaction: Ewald
 725  * VdW interaction:            None
 726  * Geometry:                   Water4-Particle
 727  * Calculate force/pot:        Force
 728  */
 729 void
 730 nb_kernel_ElecEwSw_VdwNone_GeomW4P1_F_sse2_single
 731                     (t_nblist * gmx_restrict                nlist,
 732                      rvec * gmx_restrict                    xx,
 733                      rvec * gmx_restrict                    ff,
 734                      t_forcerec * gmx_restrict              fr,
 735                      t_mdatoms * gmx_restrict               mdatoms,
 736                      nb_kernel_data_t * gmx_restrict        kernel_data,
 737                      t_nrnb * gmx_restrict                  nrnb)
 738 {
 739     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 740      * just 0 for non-waters.
 741      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 742      * jnr indices corresponding to data put in the four positions in the SIMD register.
 743      */
 744     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 745     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 746     int              jnrA,jnrB,jnrC,jnrD;
 747     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 748     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 749     real             shX,shY,shZ,rcutoff_scalar;
 750     real             *shiftvec,*fshift,*x,*f;
 751     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 752     int              vdwioffset1;
 753     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 754     int              vdwioffset2;
 755     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 756     int              vdwioffset3;
 757     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 758     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 759     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 760     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 761     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 762     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 763     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 764     real             *charge;
 765     __m128i          ewitab;
 766     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 767     real             *ewtab;
 768     __m128           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 769     real             rswitch_scalar,d_scalar;
 770     __m128           dummy_mask,cutoff_mask;
 771     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 772     __m128           one     = _mm_set1_ps(1.0);
 773     __m128           two     = _mm_set1_ps(2.0);
 774     x                = xx[0];
 775     f                = ff[0];
 776
 777     nri              = nlist->nri;
 778     iinr             = nlist->iinr;
 779     jindex           = nlist->jindex;
 780     jjnr             = nlist->jjnr;
 781     shiftidx         = nlist->shift;
 782     gid              = nlist->gid;
 783     shiftvec         = fr->shift_vec[0];
 784     fshift           = fr->fshift[0];
 785     facel            = _mm_set1_ps(fr->epsfac);
 786     charge           = mdatoms->chargeA;
 787
 788     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 789     ewtab            = fr->ic->tabq_coul_FDV0;
 790     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 791     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 792
 793     /* Setup water-specific parameters */
 794     inr              = nlist->iinr[0];
 795     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 796     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 797     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 798
 799     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 800     rcutoff_scalar   = fr->rcoulomb;
 801     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 802     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 803
 804     rswitch_scalar   = fr->rcoulomb_switch;
 805     rswitch          = _mm_set1_ps(rswitch_scalar);
 806     /* Setup switch parameters */
 807     d_scalar         = rcutoff_scalar-rswitch_scalar;
 808     d                = _mm_set1_ps(d_scalar);
 809     swV3             = _mm_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 810     swV4             = _mm_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 811     swV5             = _mm_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 812     swF2             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 813     swF3             = _mm_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 814     swF4             = _mm_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 815
 816     /* Avoid stupid compiler warnings */
 817     jnrA = jnrB = jnrC = jnrD = 0;
 818     j_coord_offsetA = 0;
 819     j_coord_offsetB = 0;
 820     j_coord_offsetC = 0;
 821     j_coord_offsetD = 0;
 822
 823     outeriter        = 0;
 824     inneriter        = 0;
 825
 826     /* Start outer loop over neighborlists */
 827     for(iidx=0; iidx<nri; iidx++)
 828     {
 829         /* Load shift vector for this list */
 830         i_shift_offset   = DIM*shiftidx[iidx];
 831         shX              = shiftvec[i_shift_offset+XX];
 832         shY              = shiftvec[i_shift_offset+YY];
 833         shZ              = shiftvec[i_shift_offset+ZZ];
 834
 835         /* Load limits for loop over neighbors */
 836         j_index_start    = jindex[iidx];
 837         j_index_end      = jindex[iidx+1];
 838
 839         /* Get outer coordinate index */
 840         inr              = iinr[iidx];
 841         i_coord_offset   = DIM*inr;
 842
 843         /* Load i particle coords and add shift vector */
 844         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 845         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 846         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 847         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 848         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 849         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 850         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 851         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 852         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 853
 854         fix1             = _mm_setzero_ps();
 855         fiy1             = _mm_setzero_ps();
 856         fiz1             = _mm_setzero_ps();
 857         fix2             = _mm_setzero_ps();
 858         fiy2             = _mm_setzero_ps();
 859         fiz2             = _mm_setzero_ps();
 860         fix3             = _mm_setzero_ps();
 861         fiy3             = _mm_setzero_ps();
 862         fiz3             = _mm_setzero_ps();
 863
 864         /* Start inner kernel loop */
 865         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 866         {
 867
 868             /* Get j neighbor index, and coordinate index */
 869             jnrA             = jjnr[jidx];
 870             jnrB             = jjnr[jidx+1];
 871             jnrC             = jjnr[jidx+2];
 872             jnrD             = jjnr[jidx+3];
 873
 874             j_coord_offsetA  = DIM*jnrA;
 875             j_coord_offsetB  = DIM*jnrB;
 876             j_coord_offsetC  = DIM*jnrC;
 877             j_coord_offsetD  = DIM*jnrD;
 878
 879             /* load j atom coordinates */
 880             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 881                                               x+j_coord_offsetC,x+j_coord_offsetD,
 882                                               &jx0,&jy0,&jz0);
 883
 884             /* Calculate displacement vector */
 885             dx10             = _mm_sub_ps(ix1,jx0);
 886             dy10             = _mm_sub_ps(iy1,jy0);
 887             dz10             = _mm_sub_ps(iz1,jz0);
 888             dx20             = _mm_sub_ps(ix2,jx0);
 889             dy20             = _mm_sub_ps(iy2,jy0);
 890             dz20             = _mm_sub_ps(iz2,jz0);
 891             dx30             = _mm_sub_ps(ix3,jx0);
 892             dy30             = _mm_sub_ps(iy3,jy0);
 893             dz30             = _mm_sub_ps(iz3,jz0);
 894
 895             /* Calculate squared distance and things based on it */
 896             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 897             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 898             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 899
 900             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 901             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 902             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 903
 904             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 905             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 906             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 907
 908             /* Load parameters for j particles */
 909             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 910                                                               charge+jnrC+0,charge+jnrD+0);
 911
 912             /**************************
 913              * CALCULATE INTERACTIONS *
 914              **************************/
 915
 916             if (gmx_mm_any_lt(rsq10,rcutoff2))
 917             {
 918
 919             r10              = _mm_mul_ps(rsq10,rinv10);
 920
 921             /* Compute parameters for interactions between i and j atoms */
 922             qq10             = _mm_mul_ps(iq1,jq0);
 923
 924             /* EWALD ELECTROSTATICS */
 925
 926             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 927             ewrt             = _mm_mul_ps(r10,ewtabscale);
 928             ewitab           = _mm_cvttps_epi32(ewrt);
 929             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 930             ewitab           = _mm_slli_epi32(ewitab,2);
 931             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 932             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 933             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 934             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 935             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 936             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 937             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 938             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 939             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 940
 941             d                = _mm_sub_ps(r10,rswitch);
 942             d                = _mm_max_ps(d,_mm_setzero_ps());
 943             d2               = _mm_mul_ps(d,d);
 944             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
 945
 946             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
 947
 948             /* Evaluate switch function */
 949             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 950             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
 951             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 952
 953             fscal            = felec;
 954
 955             fscal            = _mm_and_ps(fscal,cutoff_mask);
 956
 957             /* Calculate temporary vectorial force */
 958             tx               = _mm_mul_ps(fscal,dx10);
 959             ty               = _mm_mul_ps(fscal,dy10);
 960             tz               = _mm_mul_ps(fscal,dz10);
 961
 962             /* Update vectorial force */
 963             fix1             = _mm_add_ps(fix1,tx);
 964             fiy1             = _mm_add_ps(fiy1,ty);
 965             fiz1             = _mm_add_ps(fiz1,tz);
 966
 967             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 968                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 969                                                    tx,ty,tz);
 970
 971             }
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             if (gmx_mm_any_lt(rsq20,rcutoff2))
 978             {
 979
 980             r20              = _mm_mul_ps(rsq20,rinv20);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq20             = _mm_mul_ps(iq2,jq0);
 984
 985             /* EWALD ELECTROSTATICS */
 986
 987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 988             ewrt             = _mm_mul_ps(r20,ewtabscale);
 989             ewitab           = _mm_cvttps_epi32(ewrt);
 990             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 991             ewitab           = _mm_slli_epi32(ewitab,2);
 992             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 993             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 994             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 995             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 996             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 997             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 998             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 999             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1000             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1001
1002             d                = _mm_sub_ps(r20,rswitch);
1003             d                = _mm_max_ps(d,_mm_setzero_ps());
1004             d2               = _mm_mul_ps(d,d);
1005             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1006
1007             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1008
1009             /* Evaluate switch function */
1010             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1011             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1012             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1013
1014             fscal            = felec;
1015
1016             fscal            = _mm_and_ps(fscal,cutoff_mask);
1017
1018             /* Calculate temporary vectorial force */
1019             tx               = _mm_mul_ps(fscal,dx20);
1020             ty               = _mm_mul_ps(fscal,dy20);
1021             tz               = _mm_mul_ps(fscal,dz20);
1022
1023             /* Update vectorial force */
1024             fix2             = _mm_add_ps(fix2,tx);
1025             fiy2             = _mm_add_ps(fiy2,ty);
1026             fiz2             = _mm_add_ps(fiz2,tz);
1027
1028             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1029                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1030                                                    tx,ty,tz);
1031
1032             }
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             if (gmx_mm_any_lt(rsq30,rcutoff2))
1039             {
1040
1041             r30              = _mm_mul_ps(rsq30,rinv30);
1042
1043             /* Compute parameters for interactions between i and j atoms */
1044             qq30             = _mm_mul_ps(iq3,jq0);
1045
1046             /* EWALD ELECTROSTATICS */
1047
1048             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1049             ewrt             = _mm_mul_ps(r30,ewtabscale);
1050             ewitab           = _mm_cvttps_epi32(ewrt);
1051             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1052             ewitab           = _mm_slli_epi32(ewitab,2);
1053             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1054             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1055             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1056             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1057             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1058             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1059             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1060             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1061             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1062
1063             d                = _mm_sub_ps(r30,rswitch);
1064             d                = _mm_max_ps(d,_mm_setzero_ps());
1065             d2               = _mm_mul_ps(d,d);
1066             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1067
1068             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1069
1070             /* Evaluate switch function */
1071             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1072             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1073             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1074
1075             fscal            = felec;
1076
1077             fscal            = _mm_and_ps(fscal,cutoff_mask);
1078
1079             /* Calculate temporary vectorial force */
1080             tx               = _mm_mul_ps(fscal,dx30);
1081             ty               = _mm_mul_ps(fscal,dy30);
1082             tz               = _mm_mul_ps(fscal,dz30);
1083
1084             /* Update vectorial force */
1085             fix3             = _mm_add_ps(fix3,tx);
1086             fiy3             = _mm_add_ps(fiy3,ty);
1087             fiz3             = _mm_add_ps(fiz3,tz);
1088
1089             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1090                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1091                                                    tx,ty,tz);
1092
1093             }
1094
1095             /* Inner loop uses 186 flops */
1096         }
1097
1098         if(jidx<j_index_end)
1099         {
1100
1101             /* Get j neighbor index, and coordinate index */
1102             jnrA             = jjnr[jidx];
1103             jnrB             = jjnr[jidx+1];
1104             jnrC             = jjnr[jidx+2];
1105             jnrD             = jjnr[jidx+3];
1106
1107             /* Sign of each element will be negative for non-real atoms.
1108              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1109              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1110              */
1111             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1112             jnrA       = (jnrA>=0) ? jnrA : 0;
1113             jnrB       = (jnrB>=0) ? jnrB : 0;
1114             jnrC       = (jnrC>=0) ? jnrC : 0;
1115             jnrD       = (jnrD>=0) ? jnrD : 0;
1116
1117             j_coord_offsetA  = DIM*jnrA;
1118             j_coord_offsetB  = DIM*jnrB;
1119             j_coord_offsetC  = DIM*jnrC;
1120             j_coord_offsetD  = DIM*jnrD;
1121
1122             /* load j atom coordinates */
1123             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1124                                               x+j_coord_offsetC,x+j_coord_offsetD,
1125                                               &jx0,&jy0,&jz0);
1126
1127             /* Calculate displacement vector */
1128             dx10             = _mm_sub_ps(ix1,jx0);
1129             dy10             = _mm_sub_ps(iy1,jy0);
1130             dz10             = _mm_sub_ps(iz1,jz0);
1131             dx20             = _mm_sub_ps(ix2,jx0);
1132             dy20             = _mm_sub_ps(iy2,jy0);
1133             dz20             = _mm_sub_ps(iz2,jz0);
1134             dx30             = _mm_sub_ps(ix3,jx0);
1135             dy30             = _mm_sub_ps(iy3,jy0);
1136             dz30             = _mm_sub_ps(iz3,jz0);
1137
1138             /* Calculate squared distance and things based on it */
1139             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1140             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1141             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1142
1143             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1144             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1145             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1146
1147             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1148             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1149             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1150
1151             /* Load parameters for j particles */
1152             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1153                                                               charge+jnrC+0,charge+jnrD+0);
1154
1155             /**************************
1156              * CALCULATE INTERACTIONS *
1157              **************************/
1158
1159             if (gmx_mm_any_lt(rsq10,rcutoff2))
1160             {
1161
1162             r10              = _mm_mul_ps(rsq10,rinv10);
1163             r10              = _mm_andnot_ps(dummy_mask,r10);
1164
1165             /* Compute parameters for interactions between i and j atoms */
1166             qq10             = _mm_mul_ps(iq1,jq0);
1167
1168             /* EWALD ELECTROSTATICS */
1169
1170             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171             ewrt             = _mm_mul_ps(r10,ewtabscale);
1172             ewitab           = _mm_cvttps_epi32(ewrt);
1173             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1174             ewitab           = _mm_slli_epi32(ewitab,2);
1175             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1176             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1177             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1178             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1179             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1180             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1181             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1182             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
1183             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1184
1185             d                = _mm_sub_ps(r10,rswitch);
1186             d                = _mm_max_ps(d,_mm_setzero_ps());
1187             d2               = _mm_mul_ps(d,d);
1188             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1189
1190             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1191
1192             /* Evaluate switch function */
1193             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1194             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv10,_mm_mul_ps(velec,dsw)) );
1195             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1196
1197             fscal            = felec;
1198
1199             fscal            = _mm_and_ps(fscal,cutoff_mask);
1200
1201             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1202
1203             /* Calculate temporary vectorial force */
1204             tx               = _mm_mul_ps(fscal,dx10);
1205             ty               = _mm_mul_ps(fscal,dy10);
1206             tz               = _mm_mul_ps(fscal,dz10);
1207
1208             /* Update vectorial force */
1209             fix1             = _mm_add_ps(fix1,tx);
1210             fiy1             = _mm_add_ps(fiy1,ty);
1211             fiz1             = _mm_add_ps(fiz1,tz);
1212
1213             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1214                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1215                                                    tx,ty,tz);
1216
1217             }
1218
1219             /**************************
1220              * CALCULATE INTERACTIONS *
1221              **************************/
1222
1223             if (gmx_mm_any_lt(rsq20,rcutoff2))
1224             {
1225
1226             r20              = _mm_mul_ps(rsq20,rinv20);
1227             r20              = _mm_andnot_ps(dummy_mask,r20);
1228
1229             /* Compute parameters for interactions between i and j atoms */
1230             qq20             = _mm_mul_ps(iq2,jq0);
1231
1232             /* EWALD ELECTROSTATICS */
1233
1234             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1235             ewrt             = _mm_mul_ps(r20,ewtabscale);
1236             ewitab           = _mm_cvttps_epi32(ewrt);
1237             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1238             ewitab           = _mm_slli_epi32(ewitab,2);
1239             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1240             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1241             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1242             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1243             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1244             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1245             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1246             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
1247             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1248
1249             d                = _mm_sub_ps(r20,rswitch);
1250             d                = _mm_max_ps(d,_mm_setzero_ps());
1251             d2               = _mm_mul_ps(d,d);
1252             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1253
1254             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1255
1256             /* Evaluate switch function */
1257             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1258             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv20,_mm_mul_ps(velec,dsw)) );
1259             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1260
1261             fscal            = felec;
1262
1263             fscal            = _mm_and_ps(fscal,cutoff_mask);
1264
1265             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1266
1267             /* Calculate temporary vectorial force */
1268             tx               = _mm_mul_ps(fscal,dx20);
1269             ty               = _mm_mul_ps(fscal,dy20);
1270             tz               = _mm_mul_ps(fscal,dz20);
1271
1272             /* Update vectorial force */
1273             fix2             = _mm_add_ps(fix2,tx);
1274             fiy2             = _mm_add_ps(fiy2,ty);
1275             fiz2             = _mm_add_ps(fiz2,tz);
1276
1277             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1278                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1279                                                    tx,ty,tz);
1280
1281             }
1282
1283             /**************************
1284              * CALCULATE INTERACTIONS *
1285              **************************/
1286
1287             if (gmx_mm_any_lt(rsq30,rcutoff2))
1288             {
1289
1290             r30              = _mm_mul_ps(rsq30,rinv30);
1291             r30              = _mm_andnot_ps(dummy_mask,r30);
1292
1293             /* Compute parameters for interactions between i and j atoms */
1294             qq30             = _mm_mul_ps(iq3,jq0);
1295
1296             /* EWALD ELECTROSTATICS */
1297
1298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1299             ewrt             = _mm_mul_ps(r30,ewtabscale);
1300             ewitab           = _mm_cvttps_epi32(ewrt);
1301             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1302             ewitab           = _mm_slli_epi32(ewitab,2);
1303             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
1304             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
1305             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
1306             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
1307             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
1308             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
1309             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
1310             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
1311             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1312
1313             d                = _mm_sub_ps(r30,rswitch);
1314             d                = _mm_max_ps(d,_mm_setzero_ps());
1315             d2               = _mm_mul_ps(d,d);
1316             sw               = _mm_add_ps(one,_mm_mul_ps(d2,_mm_mul_ps(d,_mm_add_ps(swV3,_mm_mul_ps(d,_mm_add_ps(swV4,_mm_mul_ps(d,swV5)))))));
1317
1318             dsw              = _mm_mul_ps(d2,_mm_add_ps(swF2,_mm_mul_ps(d,_mm_add_ps(swF3,_mm_mul_ps(d,swF4)))));
1319
1320             /* Evaluate switch function */
1321             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1322             felec            = _mm_sub_ps( _mm_mul_ps(felec,sw) , _mm_mul_ps(rinv30,_mm_mul_ps(velec,dsw)) );
1323             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1324
1325             fscal            = felec;
1326
1327             fscal            = _mm_and_ps(fscal,cutoff_mask);
1328
1329             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1330
1331             /* Calculate temporary vectorial force */
1332             tx               = _mm_mul_ps(fscal,dx30);
1333             ty               = _mm_mul_ps(fscal,dy30);
1334             tz               = _mm_mul_ps(fscal,dz30);
1335
1336             /* Update vectorial force */
1337             fix3             = _mm_add_ps(fix3,tx);
1338             fiy3             = _mm_add_ps(fiy3,ty);
1339             fiz3             = _mm_add_ps(fiz3,tz);
1340
1341             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1342                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1343                                                    tx,ty,tz);
1344
1345             }
1346
1347             /* Inner loop uses 189 flops */
1348         }
1349
1350         /* End of innermost loop */
1351
1352         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1353                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1354
1355         /* Increment number of inner iterations */
1356         inneriter                  += j_index_end - j_index_start;
1357
1358         /* Outer loop uses 27 flops */
1359     }
1360
1361     /* Increment number of outer iterations */
1362     outeriter        += nri;
1363
1364     /* Update outer/inner flops */
1365
1366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*27 + inneriter*189);
1367 }