src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double 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_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_VF_avx_128_fma_double
  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 refer to j loop unrolling done with SSE double precision, e.g. for the two 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;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128i          ewitab;
  86     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  87     real             *ewtab;
  88     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  89     real             rswitch_scalar,d_scalar;
  90     __m128d          dummy_mask,cutoff_mask;
  91     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  92     __m128d          one     = _mm_set1_pd(1.0);
  93     __m128d          two     = _mm_set1_pd(2.0);
  94     x                = xx[0];
  95     f                = ff[0];
  96
  97     nri              = nlist->nri;
  98     iinr             = nlist->iinr;
  99     jindex           = nlist->jindex;
 100     jjnr             = nlist->jjnr;
 101     shiftidx         = nlist->shift;
 102     gid              = nlist->gid;
 103     shiftvec         = fr->shift_vec[0];
 104     fshift           = fr->fshift[0];
 105     facel            = _mm_set1_pd(fr->epsfac);
 106     charge           = mdatoms->chargeA;
 107     nvdwtype         = fr->ntype;
 108     vdwparam         = fr->nbfp;
 109     vdwtype          = mdatoms->typeA;
 110
 111     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 112     ewtab            = fr->ic->tabq_coul_FDV0;
 113     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 114     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 115
 116     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 119     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 120     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 126     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 127
 128     rswitch_scalar   = fr->rcoulomb_switch;
 129     rswitch          = _mm_set1_pd(rswitch_scalar);
 130     /* Setup switch parameters */
 131     d_scalar         = rcutoff_scalar-rswitch_scalar;
 132     d                = _mm_set1_pd(d_scalar);
 133     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 134     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 135     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 136     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 137     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139
 140     /* Avoid stupid compiler warnings */
 141     jnrA = jnrB = 0;
 142     j_coord_offsetA = 0;
 143     j_coord_offsetB = 0;
 144
 145     outeriter        = 0;
 146     inneriter        = 0;
 147
 148     /* Start outer loop over neighborlists */
 149     for(iidx=0; iidx<nri; iidx++)
 150     {
 151         /* Load shift vector for this list */
 152         i_shift_offset   = DIM*shiftidx[iidx];
 153
 154         /* Load limits for loop over neighbors */
 155         j_index_start    = jindex[iidx];
 156         j_index_end      = jindex[iidx+1];
 157
 158         /* Get outer coordinate index */
 159         inr              = iinr[iidx];
 160         i_coord_offset   = DIM*inr;
 161
 162         /* Load i particle coords and add shift vector */
 163         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 164                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 165
 166         fix0             = _mm_setzero_pd();
 167         fiy0             = _mm_setzero_pd();
 168         fiz0             = _mm_setzero_pd();
 169         fix1             = _mm_setzero_pd();
 170         fiy1             = _mm_setzero_pd();
 171         fiz1             = _mm_setzero_pd();
 172         fix2             = _mm_setzero_pd();
 173         fiy2             = _mm_setzero_pd();
 174         fiz2             = _mm_setzero_pd();
 175
 176         /* Reset potential sums */
 177         velecsum         = _mm_setzero_pd();
 178         vvdwsum          = _mm_setzero_pd();
 179
 180         /* Start inner kernel loop */
 181         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 182         {
 183
 184             /* Get j neighbor index, and coordinate index */
 185             jnrA             = jjnr[jidx];
 186             jnrB             = jjnr[jidx+1];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               &jx0,&jy0,&jz0);
 193
 194             /* Calculate displacement vector */
 195             dx00             = _mm_sub_pd(ix0,jx0);
 196             dy00             = _mm_sub_pd(iy0,jy0);
 197             dz00             = _mm_sub_pd(iz0,jz0);
 198             dx10             = _mm_sub_pd(ix1,jx0);
 199             dy10             = _mm_sub_pd(iy1,jy0);
 200             dz10             = _mm_sub_pd(iz1,jz0);
 201             dx20             = _mm_sub_pd(ix2,jx0);
 202             dy20             = _mm_sub_pd(iy2,jy0);
 203             dz20             = _mm_sub_pd(iz2,jz0);
 204
 205             /* Calculate squared distance and things based on it */
 206             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 207             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 208             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 209
 210             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 211             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 212             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 213
 214             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 215             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 216             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 217
 218             /* Load parameters for j particles */
 219             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 220             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222
 223             fjx0             = _mm_setzero_pd();
 224             fjy0             = _mm_setzero_pd();
 225             fjz0             = _mm_setzero_pd();
 226
 227             /**************************
 228              * CALCULATE INTERACTIONS *
 229              **************************/
 230
 231             if (gmx_mm_any_lt(rsq00,rcutoff2))
 232             {
 233
 234             r00              = _mm_mul_pd(rsq00,rinv00);
 235
 236             /* Compute parameters for interactions between i and j atoms */
 237             qq00             = _mm_mul_pd(iq0,jq0);
 238             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 240
 241             /* EWALD ELECTROSTATICS */
 242
 243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 244             ewrt             = _mm_mul_pd(r00,ewtabscale);
 245             ewitab           = _mm_cvttpd_epi32(ewrt);
 246 #ifdef __XOP__
 247             eweps            = _mm_frcz_pd(ewrt);
 248 #else
 249             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 250 #endif
 251             twoeweps         = _mm_add_pd(eweps,eweps);
 252             ewitab           = _mm_slli_epi32(ewitab,2);
 253             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 254             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 255             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 256             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 257             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 258             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 259             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 260             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 261             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 262             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 263
 264             /* LENNARD-JONES DISPERSION/REPULSION */
 265
 266             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 267             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 268             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 269             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 270             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 271
 272             d                = _mm_sub_pd(r00,rswitch);
 273             d                = _mm_max_pd(d,_mm_setzero_pd());
 274             d2               = _mm_mul_pd(d,d);
 275             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 276
 277             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 278
 279             /* Evaluate switch function */
 280             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 281             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 282             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 283             velec            = _mm_mul_pd(velec,sw);
 284             vvdw             = _mm_mul_pd(vvdw,sw);
 285             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 286
 287             /* Update potential sum for this i atom from the interaction with this j atom. */
 288             velec            = _mm_and_pd(velec,cutoff_mask);
 289             velecsum         = _mm_add_pd(velecsum,velec);
 290             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = _mm_add_pd(felec,fvdw);
 294
 295             fscal            = _mm_and_pd(fscal,cutoff_mask);
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 299             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 300             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 301
 302             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 303             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 304             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 305
 306             }
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             if (gmx_mm_any_lt(rsq10,rcutoff2))
 313             {
 314
 315             r10              = _mm_mul_pd(rsq10,rinv10);
 316
 317             /* Compute parameters for interactions between i and j atoms */
 318             qq10             = _mm_mul_pd(iq1,jq0);
 319
 320             /* EWALD ELECTROSTATICS */
 321
 322             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 323             ewrt             = _mm_mul_pd(r10,ewtabscale);
 324             ewitab           = _mm_cvttpd_epi32(ewrt);
 325 #ifdef __XOP__
 326             eweps            = _mm_frcz_pd(ewrt);
 327 #else
 328             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 329 #endif
 330             twoeweps         = _mm_add_pd(eweps,eweps);
 331             ewitab           = _mm_slli_epi32(ewitab,2);
 332             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 333             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 334             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 335             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 336             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 337             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 338             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 339             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 340             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 341             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 342
 343             d                = _mm_sub_pd(r10,rswitch);
 344             d                = _mm_max_pd(d,_mm_setzero_pd());
 345             d2               = _mm_mul_pd(d,d);
 346             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 347
 348             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 349
 350             /* Evaluate switch function */
 351             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 352             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 353             velec            = _mm_mul_pd(velec,sw);
 354             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velec            = _mm_and_pd(velec,cutoff_mask);
 358             velecsum         = _mm_add_pd(velecsum,velec);
 359
 360             fscal            = felec;
 361
 362             fscal            = _mm_and_pd(fscal,cutoff_mask);
 363
 364             /* Update vectorial force */
 365             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 366             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 367             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 368
 369             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 370             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 371             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 372
 373             }
 374
 375             /**************************
 376              * CALCULATE INTERACTIONS *
 377              **************************/
 378
 379             if (gmx_mm_any_lt(rsq20,rcutoff2))
 380             {
 381
 382             r20              = _mm_mul_pd(rsq20,rinv20);
 383
 384             /* Compute parameters for interactions between i and j atoms */
 385             qq20             = _mm_mul_pd(iq2,jq0);
 386
 387             /* EWALD ELECTROSTATICS */
 388
 389             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 390             ewrt             = _mm_mul_pd(r20,ewtabscale);
 391             ewitab           = _mm_cvttpd_epi32(ewrt);
 392 #ifdef __XOP__
 393             eweps            = _mm_frcz_pd(ewrt);
 394 #else
 395             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 396 #endif
 397             twoeweps         = _mm_add_pd(eweps,eweps);
 398             ewitab           = _mm_slli_epi32(ewitab,2);
 399             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 400             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 401             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 402             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 403             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 404             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 405             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 406             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 407             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 408             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 409
 410             d                = _mm_sub_pd(r20,rswitch);
 411             d                = _mm_max_pd(d,_mm_setzero_pd());
 412             d2               = _mm_mul_pd(d,d);
 413             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 414
 415             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 416
 417             /* Evaluate switch function */
 418             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 419             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 420             velec            = _mm_mul_pd(velec,sw);
 421             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velec            = _mm_and_pd(velec,cutoff_mask);
 425             velecsum         = _mm_add_pd(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             fscal            = _mm_and_pd(fscal,cutoff_mask);
 430
 431             /* Update vectorial force */
 432             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 433             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 434             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 435
 436             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 437             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 438             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 439
 440             }
 441
 442             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 443
 444             /* Inner loop uses 225 flops */
 445         }
 446
 447         if(jidx<j_index_end)
 448         {
 449
 450             jnrA             = jjnr[jidx];
 451             j_coord_offsetA  = DIM*jnrA;
 452
 453             /* load j atom coordinates */
 454             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 455                                               &jx0,&jy0,&jz0);
 456
 457             /* Calculate displacement vector */
 458             dx00             = _mm_sub_pd(ix0,jx0);
 459             dy00             = _mm_sub_pd(iy0,jy0);
 460             dz00             = _mm_sub_pd(iz0,jz0);
 461             dx10             = _mm_sub_pd(ix1,jx0);
 462             dy10             = _mm_sub_pd(iy1,jy0);
 463             dz10             = _mm_sub_pd(iz1,jz0);
 464             dx20             = _mm_sub_pd(ix2,jx0);
 465             dy20             = _mm_sub_pd(iy2,jy0);
 466             dz20             = _mm_sub_pd(iz2,jz0);
 467
 468             /* Calculate squared distance and things based on it */
 469             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 470             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 471             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 472
 473             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 474             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 475             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 476
 477             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 478             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 479             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 480
 481             /* Load parameters for j particles */
 482             jq0              = _mm_load_sd(charge+jnrA+0);
 483             vdwjidx0A        = 2*vdwtype[jnrA+0];
 484
 485             fjx0             = _mm_setzero_pd();
 486             fjy0             = _mm_setzero_pd();
 487             fjz0             = _mm_setzero_pd();
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             if (gmx_mm_any_lt(rsq00,rcutoff2))
 494             {
 495
 496             r00              = _mm_mul_pd(rsq00,rinv00);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq00             = _mm_mul_pd(iq0,jq0);
 500             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 501
 502             /* EWALD ELECTROSTATICS */
 503
 504             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 505             ewrt             = _mm_mul_pd(r00,ewtabscale);
 506             ewitab           = _mm_cvttpd_epi32(ewrt);
 507 #ifdef __XOP__
 508             eweps            = _mm_frcz_pd(ewrt);
 509 #else
 510             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 511 #endif
 512             twoeweps         = _mm_add_pd(eweps,eweps);
 513             ewitab           = _mm_slli_epi32(ewitab,2);
 514             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 515             ewtabD           = _mm_setzero_pd();
 516             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 517             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 518             ewtabFn          = _mm_setzero_pd();
 519             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 520             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 521             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 522             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 523             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 524
 525             /* LENNARD-JONES DISPERSION/REPULSION */
 526
 527             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 528             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 529             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 530             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 531             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 532
 533             d                = _mm_sub_pd(r00,rswitch);
 534             d                = _mm_max_pd(d,_mm_setzero_pd());
 535             d2               = _mm_mul_pd(d,d);
 536             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 537
 538             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 539
 540             /* Evaluate switch function */
 541             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 542             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 543             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 544             velec            = _mm_mul_pd(velec,sw);
 545             vvdw             = _mm_mul_pd(vvdw,sw);
 546             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_and_pd(velec,cutoff_mask);
 550             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 551             velecsum         = _mm_add_pd(velecsum,velec);
 552             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 553             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 554             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 555
 556             fscal            = _mm_add_pd(felec,fvdw);
 557
 558             fscal            = _mm_and_pd(fscal,cutoff_mask);
 559
 560             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 564             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 565             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 566
 567             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 568             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 569             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 570
 571             }
 572
 573             /**************************
 574              * CALCULATE INTERACTIONS *
 575              **************************/
 576
 577             if (gmx_mm_any_lt(rsq10,rcutoff2))
 578             {
 579
 580             r10              = _mm_mul_pd(rsq10,rinv10);
 581
 582             /* Compute parameters for interactions between i and j atoms */
 583             qq10             = _mm_mul_pd(iq1,jq0);
 584
 585             /* EWALD ELECTROSTATICS */
 586
 587             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 588             ewrt             = _mm_mul_pd(r10,ewtabscale);
 589             ewitab           = _mm_cvttpd_epi32(ewrt);
 590 #ifdef __XOP__
 591             eweps            = _mm_frcz_pd(ewrt);
 592 #else
 593             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 594 #endif
 595             twoeweps         = _mm_add_pd(eweps,eweps);
 596             ewitab           = _mm_slli_epi32(ewitab,2);
 597             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 598             ewtabD           = _mm_setzero_pd();
 599             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 600             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 601             ewtabFn          = _mm_setzero_pd();
 602             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 603             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 604             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 605             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 606             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 607
 608             d                = _mm_sub_pd(r10,rswitch);
 609             d                = _mm_max_pd(d,_mm_setzero_pd());
 610             d2               = _mm_mul_pd(d,d);
 611             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 612
 613             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 614
 615             /* Evaluate switch function */
 616             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 617             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 618             velec            = _mm_mul_pd(velec,sw);
 619             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 620
 621             /* Update potential sum for this i atom from the interaction with this j atom. */
 622             velec            = _mm_and_pd(velec,cutoff_mask);
 623             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 624             velecsum         = _mm_add_pd(velecsum,velec);
 625
 626             fscal            = felec;
 627
 628             fscal            = _mm_and_pd(fscal,cutoff_mask);
 629
 630             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 631
 632             /* Update vectorial force */
 633             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 634             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 635             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 636
 637             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 638             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 639             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 640
 641             }
 642
 643             /**************************
 644              * CALCULATE INTERACTIONS *
 645              **************************/
 646
 647             if (gmx_mm_any_lt(rsq20,rcutoff2))
 648             {
 649
 650             r20              = _mm_mul_pd(rsq20,rinv20);
 651
 652             /* Compute parameters for interactions between i and j atoms */
 653             qq20             = _mm_mul_pd(iq2,jq0);
 654
 655             /* EWALD ELECTROSTATICS */
 656
 657             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 658             ewrt             = _mm_mul_pd(r20,ewtabscale);
 659             ewitab           = _mm_cvttpd_epi32(ewrt);
 660 #ifdef __XOP__
 661             eweps            = _mm_frcz_pd(ewrt);
 662 #else
 663             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 664 #endif
 665             twoeweps         = _mm_add_pd(eweps,eweps);
 666             ewitab           = _mm_slli_epi32(ewitab,2);
 667             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 668             ewtabD           = _mm_setzero_pd();
 669             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 670             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 671             ewtabFn          = _mm_setzero_pd();
 672             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 673             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 674             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 675             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 676             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 677
 678             d                = _mm_sub_pd(r20,rswitch);
 679             d                = _mm_max_pd(d,_mm_setzero_pd());
 680             d2               = _mm_mul_pd(d,d);
 681             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 682
 683             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 684
 685             /* Evaluate switch function */
 686             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 687             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 688             velec            = _mm_mul_pd(velec,sw);
 689             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 690
 691             /* Update potential sum for this i atom from the interaction with this j atom. */
 692             velec            = _mm_and_pd(velec,cutoff_mask);
 693             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 694             velecsum         = _mm_add_pd(velecsum,velec);
 695
 696             fscal            = felec;
 697
 698             fscal            = _mm_and_pd(fscal,cutoff_mask);
 699
 700             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 701
 702             /* Update vectorial force */
 703             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 704             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 705             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 706
 707             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 708             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 709             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 710
 711             }
 712
 713             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 714
 715             /* Inner loop uses 225 flops */
 716         }
 717
 718         /* End of innermost loop */
 719
 720         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 721                                               f+i_coord_offset,fshift+i_shift_offset);
 722
 723         ggid                        = gid[iidx];
 724         /* Update potential energies */
 725         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 726         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 727
 728         /* Increment number of inner iterations */
 729         inneriter                  += j_index_end - j_index_start;
 730
 731         /* Outer loop uses 20 flops */
 732     }
 733
 734     /* Increment number of outer iterations */
 735     outeriter        += nri;
 736
 737     /* Update outer/inner flops */
 738
 739     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*225);
 740 }
 741 /*
 742  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 743  * Electrostatics interaction: Ewald
 744  * VdW interaction:            LennardJones
 745  * Geometry:                   Water3-Particle
 746  * Calculate force/pot:        Force
 747  */
 748 void
 749 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 750                     (t_nblist * gmx_restrict                nlist,
 751                      rvec * gmx_restrict                    xx,
 752                      rvec * gmx_restrict                    ff,
 753                      t_forcerec * gmx_restrict              fr,
 754                      t_mdatoms * gmx_restrict               mdatoms,
 755                      nb_kernel_data_t * gmx_restrict        kernel_data,
 756                      t_nrnb * gmx_restrict                  nrnb)
 757 {
 758     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 759      * just 0 for non-waters.
 760      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 761      * jnr indices corresponding to data put in the four positions in the SIMD register.
 762      */
 763     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 764     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 765     int              jnrA,jnrB;
 766     int              j_coord_offsetA,j_coord_offsetB;
 767     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 768     real             rcutoff_scalar;
 769     real             *shiftvec,*fshift,*x,*f;
 770     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 771     int              vdwioffset0;
 772     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 773     int              vdwioffset1;
 774     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 775     int              vdwioffset2;
 776     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 777     int              vdwjidx0A,vdwjidx0B;
 778     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 779     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 780     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 781     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 782     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 783     real             *charge;
 784     int              nvdwtype;
 785     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 786     int              *vdwtype;
 787     real             *vdwparam;
 788     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 789     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 790     __m128i          ewitab;
 791     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 792     real             *ewtab;
 793     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 794     real             rswitch_scalar,d_scalar;
 795     __m128d          dummy_mask,cutoff_mask;
 796     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 797     __m128d          one     = _mm_set1_pd(1.0);
 798     __m128d          two     = _mm_set1_pd(2.0);
 799     x                = xx[0];
 800     f                = ff[0];
 801
 802     nri              = nlist->nri;
 803     iinr             = nlist->iinr;
 804     jindex           = nlist->jindex;
 805     jjnr             = nlist->jjnr;
 806     shiftidx         = nlist->shift;
 807     gid              = nlist->gid;
 808     shiftvec         = fr->shift_vec[0];
 809     fshift           = fr->fshift[0];
 810     facel            = _mm_set1_pd(fr->epsfac);
 811     charge           = mdatoms->chargeA;
 812     nvdwtype         = fr->ntype;
 813     vdwparam         = fr->nbfp;
 814     vdwtype          = mdatoms->typeA;
 815
 816     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 817     ewtab            = fr->ic->tabq_coul_FDV0;
 818     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 819     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 820
 821     /* Setup water-specific parameters */
 822     inr              = nlist->iinr[0];
 823     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 824     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 825     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 826     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 827
 828     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 829     rcutoff_scalar   = fr->rcoulomb;
 830     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 831     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 832
 833     rswitch_scalar   = fr->rcoulomb_switch;
 834     rswitch          = _mm_set1_pd(rswitch_scalar);
 835     /* Setup switch parameters */
 836     d_scalar         = rcutoff_scalar-rswitch_scalar;
 837     d                = _mm_set1_pd(d_scalar);
 838     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 839     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 840     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 841     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 842     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 843     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 844
 845     /* Avoid stupid compiler warnings */
 846     jnrA = jnrB = 0;
 847     j_coord_offsetA = 0;
 848     j_coord_offsetB = 0;
 849
 850     outeriter        = 0;
 851     inneriter        = 0;
 852
 853     /* Start outer loop over neighborlists */
 854     for(iidx=0; iidx<nri; iidx++)
 855     {
 856         /* Load shift vector for this list */
 857         i_shift_offset   = DIM*shiftidx[iidx];
 858
 859         /* Load limits for loop over neighbors */
 860         j_index_start    = jindex[iidx];
 861         j_index_end      = jindex[iidx+1];
 862
 863         /* Get outer coordinate index */
 864         inr              = iinr[iidx];
 865         i_coord_offset   = DIM*inr;
 866
 867         /* Load i particle coords and add shift vector */
 868         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 869                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 870
 871         fix0             = _mm_setzero_pd();
 872         fiy0             = _mm_setzero_pd();
 873         fiz0             = _mm_setzero_pd();
 874         fix1             = _mm_setzero_pd();
 875         fiy1             = _mm_setzero_pd();
 876         fiz1             = _mm_setzero_pd();
 877         fix2             = _mm_setzero_pd();
 878         fiy2             = _mm_setzero_pd();
 879         fiz2             = _mm_setzero_pd();
 880
 881         /* Start inner kernel loop */
 882         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 883         {
 884
 885             /* Get j neighbor index, and coordinate index */
 886             jnrA             = jjnr[jidx];
 887             jnrB             = jjnr[jidx+1];
 888             j_coord_offsetA  = DIM*jnrA;
 889             j_coord_offsetB  = DIM*jnrB;
 890
 891             /* load j atom coordinates */
 892             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 893                                               &jx0,&jy0,&jz0);
 894
 895             /* Calculate displacement vector */
 896             dx00             = _mm_sub_pd(ix0,jx0);
 897             dy00             = _mm_sub_pd(iy0,jy0);
 898             dz00             = _mm_sub_pd(iz0,jz0);
 899             dx10             = _mm_sub_pd(ix1,jx0);
 900             dy10             = _mm_sub_pd(iy1,jy0);
 901             dz10             = _mm_sub_pd(iz1,jz0);
 902             dx20             = _mm_sub_pd(ix2,jx0);
 903             dy20             = _mm_sub_pd(iy2,jy0);
 904             dz20             = _mm_sub_pd(iz2,jz0);
 905
 906             /* Calculate squared distance and things based on it */
 907             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 908             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 909             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 910
 911             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 912             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 913             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 914
 915             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 916             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 917             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 918
 919             /* Load parameters for j particles */
 920             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 921             vdwjidx0A        = 2*vdwtype[jnrA+0];
 922             vdwjidx0B        = 2*vdwtype[jnrB+0];
 923
 924             fjx0             = _mm_setzero_pd();
 925             fjy0             = _mm_setzero_pd();
 926             fjz0             = _mm_setzero_pd();
 927
 928             /**************************
 929              * CALCULATE INTERACTIONS *
 930              **************************/
 931
 932             if (gmx_mm_any_lt(rsq00,rcutoff2))
 933             {
 934
 935             r00              = _mm_mul_pd(rsq00,rinv00);
 936
 937             /* Compute parameters for interactions between i and j atoms */
 938             qq00             = _mm_mul_pd(iq0,jq0);
 939             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 940                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 941
 942             /* EWALD ELECTROSTATICS */
 943
 944             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 945             ewrt             = _mm_mul_pd(r00,ewtabscale);
 946             ewitab           = _mm_cvttpd_epi32(ewrt);
 947 #ifdef __XOP__
 948             eweps            = _mm_frcz_pd(ewrt);
 949 #else
 950             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 951 #endif
 952             twoeweps         = _mm_add_pd(eweps,eweps);
 953             ewitab           = _mm_slli_epi32(ewitab,2);
 954             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 955             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 956             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 957             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 958             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 959             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 960             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 961             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 962             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 963             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 964
 965             /* LENNARD-JONES DISPERSION/REPULSION */
 966
 967             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 968             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 969             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 970             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 971             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 972
 973             d                = _mm_sub_pd(r00,rswitch);
 974             d                = _mm_max_pd(d,_mm_setzero_pd());
 975             d2               = _mm_mul_pd(d,d);
 976             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 977
 978             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 979
 980             /* Evaluate switch function */
 981             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 982             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 983             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 984             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 985
 986             fscal            = _mm_add_pd(felec,fvdw);
 987
 988             fscal            = _mm_and_pd(fscal,cutoff_mask);
 989
 990             /* Update vectorial force */
 991             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 992             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 993             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 994
 995             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 996             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 997             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 998
 999             }
1000
1001             /**************************
1002              * CALCULATE INTERACTIONS *
1003              **************************/
1004
1005             if (gmx_mm_any_lt(rsq10,rcutoff2))
1006             {
1007
1008             r10              = _mm_mul_pd(rsq10,rinv10);
1009
1010             /* Compute parameters for interactions between i and j atoms */
1011             qq10             = _mm_mul_pd(iq1,jq0);
1012
1013             /* EWALD ELECTROSTATICS */
1014
1015             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1016             ewrt             = _mm_mul_pd(r10,ewtabscale);
1017             ewitab           = _mm_cvttpd_epi32(ewrt);
1018 #ifdef __XOP__
1019             eweps            = _mm_frcz_pd(ewrt);
1020 #else
1021             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1022 #endif
1023             twoeweps         = _mm_add_pd(eweps,eweps);
1024             ewitab           = _mm_slli_epi32(ewitab,2);
1025             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1026             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1027             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1028             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1029             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1030             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1031             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1032             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1033             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1034             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1035
1036             d                = _mm_sub_pd(r10,rswitch);
1037             d                = _mm_max_pd(d,_mm_setzero_pd());
1038             d2               = _mm_mul_pd(d,d);
1039             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1040
1041             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1042
1043             /* Evaluate switch function */
1044             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1045             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1046             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_and_pd(fscal,cutoff_mask);
1051
1052             /* Update vectorial force */
1053             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1054             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1055             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1056
1057             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1058             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1059             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1060
1061             }
1062
1063             /**************************
1064              * CALCULATE INTERACTIONS *
1065              **************************/
1066
1067             if (gmx_mm_any_lt(rsq20,rcutoff2))
1068             {
1069
1070             r20              = _mm_mul_pd(rsq20,rinv20);
1071
1072             /* Compute parameters for interactions between i and j atoms */
1073             qq20             = _mm_mul_pd(iq2,jq0);
1074
1075             /* EWALD ELECTROSTATICS */
1076
1077             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1078             ewrt             = _mm_mul_pd(r20,ewtabscale);
1079             ewitab           = _mm_cvttpd_epi32(ewrt);
1080 #ifdef __XOP__
1081             eweps            = _mm_frcz_pd(ewrt);
1082 #else
1083             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1084 #endif
1085             twoeweps         = _mm_add_pd(eweps,eweps);
1086             ewitab           = _mm_slli_epi32(ewitab,2);
1087             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1088             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1089             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1090             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1091             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1092             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1093             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1094             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1095             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1096             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1097
1098             d                = _mm_sub_pd(r20,rswitch);
1099             d                = _mm_max_pd(d,_mm_setzero_pd());
1100             d2               = _mm_mul_pd(d,d);
1101             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1102
1103             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1104
1105             /* Evaluate switch function */
1106             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1107             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1108             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1109
1110             fscal            = felec;
1111
1112             fscal            = _mm_and_pd(fscal,cutoff_mask);
1113
1114             /* Update vectorial force */
1115             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1116             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1117             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1118
1119             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1120             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1121             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1122
1123             }
1124
1125             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1126
1127             /* Inner loop uses 213 flops */
1128         }
1129
1130         if(jidx<j_index_end)
1131         {
1132
1133             jnrA             = jjnr[jidx];
1134             j_coord_offsetA  = DIM*jnrA;
1135
1136             /* load j atom coordinates */
1137             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1138                                               &jx0,&jy0,&jz0);
1139
1140             /* Calculate displacement vector */
1141             dx00             = _mm_sub_pd(ix0,jx0);
1142             dy00             = _mm_sub_pd(iy0,jy0);
1143             dz00             = _mm_sub_pd(iz0,jz0);
1144             dx10             = _mm_sub_pd(ix1,jx0);
1145             dy10             = _mm_sub_pd(iy1,jy0);
1146             dz10             = _mm_sub_pd(iz1,jz0);
1147             dx20             = _mm_sub_pd(ix2,jx0);
1148             dy20             = _mm_sub_pd(iy2,jy0);
1149             dz20             = _mm_sub_pd(iz2,jz0);
1150
1151             /* Calculate squared distance and things based on it */
1152             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1153             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1154             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1155
1156             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1157             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1158             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1159
1160             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1161             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1162             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1163
1164             /* Load parameters for j particles */
1165             jq0              = _mm_load_sd(charge+jnrA+0);
1166             vdwjidx0A        = 2*vdwtype[jnrA+0];
1167
1168             fjx0             = _mm_setzero_pd();
1169             fjy0             = _mm_setzero_pd();
1170             fjz0             = _mm_setzero_pd();
1171
1172             /**************************
1173              * CALCULATE INTERACTIONS *
1174              **************************/
1175
1176             if (gmx_mm_any_lt(rsq00,rcutoff2))
1177             {
1178
1179             r00              = _mm_mul_pd(rsq00,rinv00);
1180
1181             /* Compute parameters for interactions between i and j atoms */
1182             qq00             = _mm_mul_pd(iq0,jq0);
1183             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1184
1185             /* EWALD ELECTROSTATICS */
1186
1187             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1188             ewrt             = _mm_mul_pd(r00,ewtabscale);
1189             ewitab           = _mm_cvttpd_epi32(ewrt);
1190 #ifdef __XOP__
1191             eweps            = _mm_frcz_pd(ewrt);
1192 #else
1193             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1194 #endif
1195             twoeweps         = _mm_add_pd(eweps,eweps);
1196             ewitab           = _mm_slli_epi32(ewitab,2);
1197             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1198             ewtabD           = _mm_setzero_pd();
1199             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1200             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1201             ewtabFn          = _mm_setzero_pd();
1202             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1203             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1204             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1205             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1206             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1207
1208             /* LENNARD-JONES DISPERSION/REPULSION */
1209
1210             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1211             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1212             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1213             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1214             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1215
1216             d                = _mm_sub_pd(r00,rswitch);
1217             d                = _mm_max_pd(d,_mm_setzero_pd());
1218             d2               = _mm_mul_pd(d,d);
1219             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1220
1221             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1222
1223             /* Evaluate switch function */
1224             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1225             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1226             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1227             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1228
1229             fscal            = _mm_add_pd(felec,fvdw);
1230
1231             fscal            = _mm_and_pd(fscal,cutoff_mask);
1232
1233             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1234
1235             /* Update vectorial force */
1236             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1237             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1238             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1239
1240             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1241             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1242             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1243
1244             }
1245
1246             /**************************
1247              * CALCULATE INTERACTIONS *
1248              **************************/
1249
1250             if (gmx_mm_any_lt(rsq10,rcutoff2))
1251             {
1252
1253             r10              = _mm_mul_pd(rsq10,rinv10);
1254
1255             /* Compute parameters for interactions between i and j atoms */
1256             qq10             = _mm_mul_pd(iq1,jq0);
1257
1258             /* EWALD ELECTROSTATICS */
1259
1260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1261             ewrt             = _mm_mul_pd(r10,ewtabscale);
1262             ewitab           = _mm_cvttpd_epi32(ewrt);
1263 #ifdef __XOP__
1264             eweps            = _mm_frcz_pd(ewrt);
1265 #else
1266             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1267 #endif
1268             twoeweps         = _mm_add_pd(eweps,eweps);
1269             ewitab           = _mm_slli_epi32(ewitab,2);
1270             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1271             ewtabD           = _mm_setzero_pd();
1272             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1273             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1274             ewtabFn          = _mm_setzero_pd();
1275             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1276             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1277             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1278             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1279             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1280
1281             d                = _mm_sub_pd(r10,rswitch);
1282             d                = _mm_max_pd(d,_mm_setzero_pd());
1283             d2               = _mm_mul_pd(d,d);
1284             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1285
1286             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1287
1288             /* Evaluate switch function */
1289             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1290             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1291             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1292
1293             fscal            = felec;
1294
1295             fscal            = _mm_and_pd(fscal,cutoff_mask);
1296
1297             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1298
1299             /* Update vectorial force */
1300             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1301             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1302             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1303
1304             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1305             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1306             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1307
1308             }
1309
1310             /**************************
1311              * CALCULATE INTERACTIONS *
1312              **************************/
1313
1314             if (gmx_mm_any_lt(rsq20,rcutoff2))
1315             {
1316
1317             r20              = _mm_mul_pd(rsq20,rinv20);
1318
1319             /* Compute parameters for interactions between i and j atoms */
1320             qq20             = _mm_mul_pd(iq2,jq0);
1321
1322             /* EWALD ELECTROSTATICS */
1323
1324             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1325             ewrt             = _mm_mul_pd(r20,ewtabscale);
1326             ewitab           = _mm_cvttpd_epi32(ewrt);
1327 #ifdef __XOP__
1328             eweps            = _mm_frcz_pd(ewrt);
1329 #else
1330             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1331 #endif
1332             twoeweps         = _mm_add_pd(eweps,eweps);
1333             ewitab           = _mm_slli_epi32(ewitab,2);
1334             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1335             ewtabD           = _mm_setzero_pd();
1336             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1337             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1338             ewtabFn          = _mm_setzero_pd();
1339             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1340             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1341             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1342             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1343             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1344
1345             d                = _mm_sub_pd(r20,rswitch);
1346             d                = _mm_max_pd(d,_mm_setzero_pd());
1347             d2               = _mm_mul_pd(d,d);
1348             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1349
1350             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1351
1352             /* Evaluate switch function */
1353             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1354             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1355             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1356
1357             fscal            = felec;
1358
1359             fscal            = _mm_and_pd(fscal,cutoff_mask);
1360
1361             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1362
1363             /* Update vectorial force */
1364             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1365             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1366             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1367
1368             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1369             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1370             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1371
1372             }
1373
1374             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1375
1376             /* Inner loop uses 213 flops */
1377         }
1378
1379         /* End of innermost loop */
1380
1381         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1382                                               f+i_coord_offset,fshift+i_shift_offset);
1383
1384         /* Increment number of inner iterations */
1385         inneriter                  += j_index_end - j_index_start;
1386
1387         /* Outer loop uses 18 flops */
1388     }
1389
1390     /* Increment number of outer iterations */
1391     outeriter        += nri;
1392
1393     /* Update outer/inner flops */
1394
1395     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*213);
1396 }