src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_avx_128_fma_double.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017,2018, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_128_fma_double kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_avx_128_fma_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_128_fma_double
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB;
  74     int              j_coord_offsetA,j_coord_offsetB;
  75     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  76     real             rcutoff_scalar;
  77     real             *shiftvec,*fshift,*x,*f;
  78     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  79     int              vdwioffset0;
  80     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  81     int              vdwioffset1;
  82     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  83     int              vdwioffset2;
  84     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  85     int              vdwioffset3;
  86     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  87     int              vdwjidx0A,vdwjidx0B;
  88     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  89     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  90     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     int              nvdwtype;
  96     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  97     int              *vdwtype;
  98     real             *vdwparam;
  99     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 100     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 101     __m128i          ewitab;
 102     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 103     real             *ewtab;
 104     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 105     real             rswitch_scalar,d_scalar;
 106     __m128d          dummy_mask,cutoff_mask;
 107     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 108     __m128d          one     = _mm_set1_pd(1.0);
 109     __m128d          two     = _mm_set1_pd(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm_set1_pd(fr->ic->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 128     ewtab            = fr->ic->tabq_coul_FDV0;
 129     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 130     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 131
 132     /* Setup water-specific parameters */
 133     inr              = nlist->iinr[0];
 134     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 135     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 136     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 137     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 138
 139     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 140     rcutoff_scalar   = fr->ic->rcoulomb;
 141     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 142     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 143
 144     rswitch_scalar   = fr->ic->rcoulomb_switch;
 145     rswitch          = _mm_set1_pd(rswitch_scalar);
 146     /* Setup switch parameters */
 147     d_scalar         = rcutoff_scalar-rswitch_scalar;
 148     d                = _mm_set1_pd(d_scalar);
 149     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 150     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 153     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 154     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 155
 156     /* Avoid stupid compiler warnings */
 157     jnrA = jnrB = 0;
 158     j_coord_offsetA = 0;
 159     j_coord_offsetB = 0;
 160
 161     outeriter        = 0;
 162     inneriter        = 0;
 163
 164     /* Start outer loop over neighborlists */
 165     for(iidx=0; iidx<nri; iidx++)
 166     {
 167         /* Load shift vector for this list */
 168         i_shift_offset   = DIM*shiftidx[iidx];
 169
 170         /* Load limits for loop over neighbors */
 171         j_index_start    = jindex[iidx];
 172         j_index_end      = jindex[iidx+1];
 173
 174         /* Get outer coordinate index */
 175         inr              = iinr[iidx];
 176         i_coord_offset   = DIM*inr;
 177
 178         /* Load i particle coords and add shift vector */
 179         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 180                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 181
 182         fix0             = _mm_setzero_pd();
 183         fiy0             = _mm_setzero_pd();
 184         fiz0             = _mm_setzero_pd();
 185         fix1             = _mm_setzero_pd();
 186         fiy1             = _mm_setzero_pd();
 187         fiz1             = _mm_setzero_pd();
 188         fix2             = _mm_setzero_pd();
 189         fiy2             = _mm_setzero_pd();
 190         fiz2             = _mm_setzero_pd();
 191         fix3             = _mm_setzero_pd();
 192         fiy3             = _mm_setzero_pd();
 193         fiz3             = _mm_setzero_pd();
 194
 195         /* Reset potential sums */
 196         velecsum         = _mm_setzero_pd();
 197         vvdwsum          = _mm_setzero_pd();
 198
 199         /* Start inner kernel loop */
 200         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 201         {
 202
 203             /* Get j neighbor index, and coordinate index */
 204             jnrA             = jjnr[jidx];
 205             jnrB             = jjnr[jidx+1];
 206             j_coord_offsetA  = DIM*jnrA;
 207             j_coord_offsetB  = DIM*jnrB;
 208
 209             /* load j atom coordinates */
 210             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_pd(ix0,jx0);
 215             dy00             = _mm_sub_pd(iy0,jy0);
 216             dz00             = _mm_sub_pd(iz0,jz0);
 217             dx10             = _mm_sub_pd(ix1,jx0);
 218             dy10             = _mm_sub_pd(iy1,jy0);
 219             dz10             = _mm_sub_pd(iz1,jz0);
 220             dx20             = _mm_sub_pd(ix2,jx0);
 221             dy20             = _mm_sub_pd(iy2,jy0);
 222             dz20             = _mm_sub_pd(iz2,jz0);
 223             dx30             = _mm_sub_pd(ix3,jx0);
 224             dy30             = _mm_sub_pd(iy3,jy0);
 225             dz30             = _mm_sub_pd(iz3,jz0);
 226
 227             /* Calculate squared distance and things based on it */
 228             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 229             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 230             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 231             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 232
 233             rinv00           = avx128fma_invsqrt_d(rsq00);
 234             rinv10           = avx128fma_invsqrt_d(rsq10);
 235             rinv20           = avx128fma_invsqrt_d(rsq20);
 236             rinv30           = avx128fma_invsqrt_d(rsq30);
 237
 238             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 239             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 240             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 241             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 242
 243             /* Load parameters for j particles */
 244             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 245             vdwjidx0A        = 2*vdwtype[jnrA+0];
 246             vdwjidx0B        = 2*vdwtype[jnrB+0];
 247
 248             fjx0             = _mm_setzero_pd();
 249             fjy0             = _mm_setzero_pd();
 250             fjz0             = _mm_setzero_pd();
 251
 252             /**************************
 253              * CALCULATE INTERACTIONS *
 254              **************************/
 255
 256             if (gmx_mm_any_lt(rsq00,rcutoff2))
 257             {
 258
 259             r00              = _mm_mul_pd(rsq00,rinv00);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 263                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 264
 265             /* LENNARD-JONES DISPERSION/REPULSION */
 266
 267             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 268             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 269             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 270             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 271             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 272
 273             d                = _mm_sub_pd(r00,rswitch);
 274             d                = _mm_max_pd(d,_mm_setzero_pd());
 275             d2               = _mm_mul_pd(d,d);
 276             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 277
 278             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 279
 280             /* Evaluate switch function */
 281             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 282             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 283             vvdw             = _mm_mul_pd(vvdw,sw);
 284             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 285
 286             /* Update potential sum for this i atom from the interaction with this j atom. */
 287             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 288             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 289
 290             fscal            = fvdw;
 291
 292             fscal            = _mm_and_pd(fscal,cutoff_mask);
 293
 294             /* Update vectorial force */
 295             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 296             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 297             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 298
 299             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 300             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 301             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 302
 303             }
 304
 305             /**************************
 306              * CALCULATE INTERACTIONS *
 307              **************************/
 308
 309             if (gmx_mm_any_lt(rsq10,rcutoff2))
 310             {
 311
 312             r10              = _mm_mul_pd(rsq10,rinv10);
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq10             = _mm_mul_pd(iq1,jq0);
 316
 317             /* EWALD ELECTROSTATICS */
 318
 319             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 320             ewrt             = _mm_mul_pd(r10,ewtabscale);
 321             ewitab           = _mm_cvttpd_epi32(ewrt);
 322 #ifdef __XOP__
 323             eweps            = _mm_frcz_pd(ewrt);
 324 #else
 325             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 326 #endif
 327             twoeweps         = _mm_add_pd(eweps,eweps);
 328             ewitab           = _mm_slli_epi32(ewitab,2);
 329             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 330             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 331             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 332             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 333             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 334             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 335             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 336             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 337             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 338             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 339
 340             d                = _mm_sub_pd(r10,rswitch);
 341             d                = _mm_max_pd(d,_mm_setzero_pd());
 342             d2               = _mm_mul_pd(d,d);
 343             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 344
 345             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 346
 347             /* Evaluate switch function */
 348             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 349             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 350             velec            = _mm_mul_pd(velec,sw);
 351             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm_and_pd(velec,cutoff_mask);
 355             velecsum         = _mm_add_pd(velecsum,velec);
 356
 357             fscal            = felec;
 358
 359             fscal            = _mm_and_pd(fscal,cutoff_mask);
 360
 361             /* Update vectorial force */
 362             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 363             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 364             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 365
 366             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 367             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 368             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 369
 370             }
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             if (gmx_mm_any_lt(rsq20,rcutoff2))
 377             {
 378
 379             r20              = _mm_mul_pd(rsq20,rinv20);
 380
 381             /* Compute parameters for interactions between i and j atoms */
 382             qq20             = _mm_mul_pd(iq2,jq0);
 383
 384             /* EWALD ELECTROSTATICS */
 385
 386             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 387             ewrt             = _mm_mul_pd(r20,ewtabscale);
 388             ewitab           = _mm_cvttpd_epi32(ewrt);
 389 #ifdef __XOP__
 390             eweps            = _mm_frcz_pd(ewrt);
 391 #else
 392             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 393 #endif
 394             twoeweps         = _mm_add_pd(eweps,eweps);
 395             ewitab           = _mm_slli_epi32(ewitab,2);
 396             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 397             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 398             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 399             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 400             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 401             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 402             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 403             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 404             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 405             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 406
 407             d                = _mm_sub_pd(r20,rswitch);
 408             d                = _mm_max_pd(d,_mm_setzero_pd());
 409             d2               = _mm_mul_pd(d,d);
 410             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 411
 412             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 413
 414             /* Evaluate switch function */
 415             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 416             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 417             velec            = _mm_mul_pd(velec,sw);
 418             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 419
 420             /* Update potential sum for this i atom from the interaction with this j atom. */
 421             velec            = _mm_and_pd(velec,cutoff_mask);
 422             velecsum         = _mm_add_pd(velecsum,velec);
 423
 424             fscal            = felec;
 425
 426             fscal            = _mm_and_pd(fscal,cutoff_mask);
 427
 428             /* Update vectorial force */
 429             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 430             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 431             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 432
 433             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 434             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 435             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 436
 437             }
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             if (gmx_mm_any_lt(rsq30,rcutoff2))
 444             {
 445
 446             r30              = _mm_mul_pd(rsq30,rinv30);
 447
 448             /* Compute parameters for interactions between i and j atoms */
 449             qq30             = _mm_mul_pd(iq3,jq0);
 450
 451             /* EWALD ELECTROSTATICS */
 452
 453             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 454             ewrt             = _mm_mul_pd(r30,ewtabscale);
 455             ewitab           = _mm_cvttpd_epi32(ewrt);
 456 #ifdef __XOP__
 457             eweps            = _mm_frcz_pd(ewrt);
 458 #else
 459             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 460 #endif
 461             twoeweps         = _mm_add_pd(eweps,eweps);
 462             ewitab           = _mm_slli_epi32(ewitab,2);
 463             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 464             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 465             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 466             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 467             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 468             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 469             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 470             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 471             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 472             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 473
 474             d                = _mm_sub_pd(r30,rswitch);
 475             d                = _mm_max_pd(d,_mm_setzero_pd());
 476             d2               = _mm_mul_pd(d,d);
 477             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 478
 479             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 480
 481             /* Evaluate switch function */
 482             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 483             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 484             velec            = _mm_mul_pd(velec,sw);
 485             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 486
 487             /* Update potential sum for this i atom from the interaction with this j atom. */
 488             velec            = _mm_and_pd(velec,cutoff_mask);
 489             velecsum         = _mm_add_pd(velecsum,velec);
 490
 491             fscal            = felec;
 492
 493             fscal            = _mm_and_pd(fscal,cutoff_mask);
 494
 495             /* Update vectorial force */
 496             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 497             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 498             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 499
 500             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 501             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 502             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 503
 504             }
 505
 506             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 507
 508             /* Inner loop uses 269 flops */
 509         }
 510
 511         if(jidx<j_index_end)
 512         {
 513
 514             jnrA             = jjnr[jidx];
 515             j_coord_offsetA  = DIM*jnrA;
 516
 517             /* load j atom coordinates */
 518             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 519                                               &jx0,&jy0,&jz0);
 520
 521             /* Calculate displacement vector */
 522             dx00             = _mm_sub_pd(ix0,jx0);
 523             dy00             = _mm_sub_pd(iy0,jy0);
 524             dz00             = _mm_sub_pd(iz0,jz0);
 525             dx10             = _mm_sub_pd(ix1,jx0);
 526             dy10             = _mm_sub_pd(iy1,jy0);
 527             dz10             = _mm_sub_pd(iz1,jz0);
 528             dx20             = _mm_sub_pd(ix2,jx0);
 529             dy20             = _mm_sub_pd(iy2,jy0);
 530             dz20             = _mm_sub_pd(iz2,jz0);
 531             dx30             = _mm_sub_pd(ix3,jx0);
 532             dy30             = _mm_sub_pd(iy3,jy0);
 533             dz30             = _mm_sub_pd(iz3,jz0);
 534
 535             /* Calculate squared distance and things based on it */
 536             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 537             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 538             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 539             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 540
 541             rinv00           = avx128fma_invsqrt_d(rsq00);
 542             rinv10           = avx128fma_invsqrt_d(rsq10);
 543             rinv20           = avx128fma_invsqrt_d(rsq20);
 544             rinv30           = avx128fma_invsqrt_d(rsq30);
 545
 546             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 547             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 548             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 549             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 550
 551             /* Load parameters for j particles */
 552             jq0              = _mm_load_sd(charge+jnrA+0);
 553             vdwjidx0A        = 2*vdwtype[jnrA+0];
 554
 555             fjx0             = _mm_setzero_pd();
 556             fjy0             = _mm_setzero_pd();
 557             fjz0             = _mm_setzero_pd();
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             if (gmx_mm_any_lt(rsq00,rcutoff2))
 564             {
 565
 566             r00              = _mm_mul_pd(rsq00,rinv00);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 570
 571             /* LENNARD-JONES DISPERSION/REPULSION */
 572
 573             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 574             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 575             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 576             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 577             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 578
 579             d                = _mm_sub_pd(r00,rswitch);
 580             d                = _mm_max_pd(d,_mm_setzero_pd());
 581             d2               = _mm_mul_pd(d,d);
 582             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 583
 584             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 585
 586             /* Evaluate switch function */
 587             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 588             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 589             vvdw             = _mm_mul_pd(vvdw,sw);
 590             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 591
 592             /* Update potential sum for this i atom from the interaction with this j atom. */
 593             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 594             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 595             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 596
 597             fscal            = fvdw;
 598
 599             fscal            = _mm_and_pd(fscal,cutoff_mask);
 600
 601             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 602
 603             /* Update vectorial force */
 604             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 605             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 606             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 607
 608             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 609             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 610             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 611
 612             }
 613
 614             /**************************
 615              * CALCULATE INTERACTIONS *
 616              **************************/
 617
 618             if (gmx_mm_any_lt(rsq10,rcutoff2))
 619             {
 620
 621             r10              = _mm_mul_pd(rsq10,rinv10);
 622
 623             /* Compute parameters for interactions between i and j atoms */
 624             qq10             = _mm_mul_pd(iq1,jq0);
 625
 626             /* EWALD ELECTROSTATICS */
 627
 628             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 629             ewrt             = _mm_mul_pd(r10,ewtabscale);
 630             ewitab           = _mm_cvttpd_epi32(ewrt);
 631 #ifdef __XOP__
 632             eweps            = _mm_frcz_pd(ewrt);
 633 #else
 634             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 635 #endif
 636             twoeweps         = _mm_add_pd(eweps,eweps);
 637             ewitab           = _mm_slli_epi32(ewitab,2);
 638             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 639             ewtabD           = _mm_setzero_pd();
 640             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 641             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 642             ewtabFn          = _mm_setzero_pd();
 643             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 644             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 645             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 646             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 647             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 648
 649             d                = _mm_sub_pd(r10,rswitch);
 650             d                = _mm_max_pd(d,_mm_setzero_pd());
 651             d2               = _mm_mul_pd(d,d);
 652             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 653
 654             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 655
 656             /* Evaluate switch function */
 657             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 658             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 659             velec            = _mm_mul_pd(velec,sw);
 660             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 661
 662             /* Update potential sum for this i atom from the interaction with this j atom. */
 663             velec            = _mm_and_pd(velec,cutoff_mask);
 664             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 665             velecsum         = _mm_add_pd(velecsum,velec);
 666
 667             fscal            = felec;
 668
 669             fscal            = _mm_and_pd(fscal,cutoff_mask);
 670
 671             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 672
 673             /* Update vectorial force */
 674             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 675             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 676             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 677
 678             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 679             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 680             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 681
 682             }
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             if (gmx_mm_any_lt(rsq20,rcutoff2))
 689             {
 690
 691             r20              = _mm_mul_pd(rsq20,rinv20);
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq20             = _mm_mul_pd(iq2,jq0);
 695
 696             /* EWALD ELECTROSTATICS */
 697
 698             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 699             ewrt             = _mm_mul_pd(r20,ewtabscale);
 700             ewitab           = _mm_cvttpd_epi32(ewrt);
 701 #ifdef __XOP__
 702             eweps            = _mm_frcz_pd(ewrt);
 703 #else
 704             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 705 #endif
 706             twoeweps         = _mm_add_pd(eweps,eweps);
 707             ewitab           = _mm_slli_epi32(ewitab,2);
 708             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 709             ewtabD           = _mm_setzero_pd();
 710             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 711             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 712             ewtabFn          = _mm_setzero_pd();
 713             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 714             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 715             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 716             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 717             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 718
 719             d                = _mm_sub_pd(r20,rswitch);
 720             d                = _mm_max_pd(d,_mm_setzero_pd());
 721             d2               = _mm_mul_pd(d,d);
 722             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 723
 724             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 725
 726             /* Evaluate switch function */
 727             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 728             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 729             velec            = _mm_mul_pd(velec,sw);
 730             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 731
 732             /* Update potential sum for this i atom from the interaction with this j atom. */
 733             velec            = _mm_and_pd(velec,cutoff_mask);
 734             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 735             velecsum         = _mm_add_pd(velecsum,velec);
 736
 737             fscal            = felec;
 738
 739             fscal            = _mm_and_pd(fscal,cutoff_mask);
 740
 741             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 742
 743             /* Update vectorial force */
 744             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 745             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 746             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 747
 748             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 749             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 750             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 751
 752             }
 753
 754             /**************************
 755              * CALCULATE INTERACTIONS *
 756              **************************/
 757
 758             if (gmx_mm_any_lt(rsq30,rcutoff2))
 759             {
 760
 761             r30              = _mm_mul_pd(rsq30,rinv30);
 762
 763             /* Compute parameters for interactions between i and j atoms */
 764             qq30             = _mm_mul_pd(iq3,jq0);
 765
 766             /* EWALD ELECTROSTATICS */
 767
 768             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 769             ewrt             = _mm_mul_pd(r30,ewtabscale);
 770             ewitab           = _mm_cvttpd_epi32(ewrt);
 771 #ifdef __XOP__
 772             eweps            = _mm_frcz_pd(ewrt);
 773 #else
 774             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 775 #endif
 776             twoeweps         = _mm_add_pd(eweps,eweps);
 777             ewitab           = _mm_slli_epi32(ewitab,2);
 778             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 779             ewtabD           = _mm_setzero_pd();
 780             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 781             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 782             ewtabFn          = _mm_setzero_pd();
 783             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 784             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 785             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 786             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 787             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 788
 789             d                = _mm_sub_pd(r30,rswitch);
 790             d                = _mm_max_pd(d,_mm_setzero_pd());
 791             d2               = _mm_mul_pd(d,d);
 792             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 793
 794             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 795
 796             /* Evaluate switch function */
 797             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 798             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 799             velec            = _mm_mul_pd(velec,sw);
 800             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 801
 802             /* Update potential sum for this i atom from the interaction with this j atom. */
 803             velec            = _mm_and_pd(velec,cutoff_mask);
 804             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 805             velecsum         = _mm_add_pd(velecsum,velec);
 806
 807             fscal            = felec;
 808
 809             fscal            = _mm_and_pd(fscal,cutoff_mask);
 810
 811             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 812
 813             /* Update vectorial force */
 814             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 815             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 816             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 817
 818             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 819             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 820             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 821
 822             }
 823
 824             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 825
 826             /* Inner loop uses 269 flops */
 827         }
 828
 829         /* End of innermost loop */
 830
 831         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 832                                               f+i_coord_offset,fshift+i_shift_offset);
 833
 834         ggid                        = gid[iidx];
 835         /* Update potential energies */
 836         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 837         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 838
 839         /* Increment number of inner iterations */
 840         inneriter                  += j_index_end - j_index_start;
 841
 842         /* Outer loop uses 26 flops */
 843     }
 844
 845     /* Increment number of outer iterations */
 846     outeriter        += nri;
 847
 848     /* Update outer/inner flops */
 849
 850     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*269);
 851 }
 852 /*
 853  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 854  * Electrostatics interaction: Ewald
 855  * VdW interaction:            LennardJones
 856  * Geometry:                   Water4-Particle
 857  * Calculate force/pot:        Force
 858  */
 859 void
 860 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 861                     (t_nblist                    * gmx_restrict       nlist,
 862                      rvec                        * gmx_restrict          xx,
 863                      rvec                        * gmx_restrict          ff,
 864                      struct t_forcerec           * gmx_restrict          fr,
 865                      t_mdatoms                   * gmx_restrict     mdatoms,
 866                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 867                      t_nrnb                      * gmx_restrict        nrnb)
 868 {
 869     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 870      * just 0 for non-waters.
 871      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 872      * jnr indices corresponding to data put in the four positions in the SIMD register.
 873      */
 874     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 875     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 876     int              jnrA,jnrB;
 877     int              j_coord_offsetA,j_coord_offsetB;
 878     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 879     real             rcutoff_scalar;
 880     real             *shiftvec,*fshift,*x,*f;
 881     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 882     int              vdwioffset0;
 883     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 884     int              vdwioffset1;
 885     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 886     int              vdwioffset2;
 887     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 888     int              vdwioffset3;
 889     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 890     int              vdwjidx0A,vdwjidx0B;
 891     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 892     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 893     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 894     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 895     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 896     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 897     real             *charge;
 898     int              nvdwtype;
 899     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 900     int              *vdwtype;
 901     real             *vdwparam;
 902     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 903     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 904     __m128i          ewitab;
 905     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 906     real             *ewtab;
 907     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 908     real             rswitch_scalar,d_scalar;
 909     __m128d          dummy_mask,cutoff_mask;
 910     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 911     __m128d          one     = _mm_set1_pd(1.0);
 912     __m128d          two     = _mm_set1_pd(2.0);
 913     x                = xx[0];
 914     f                = ff[0];
 915
 916     nri              = nlist->nri;
 917     iinr             = nlist->iinr;
 918     jindex           = nlist->jindex;
 919     jjnr             = nlist->jjnr;
 920     shiftidx         = nlist->shift;
 921     gid              = nlist->gid;
 922     shiftvec         = fr->shift_vec[0];
 923     fshift           = fr->fshift[0];
 924     facel            = _mm_set1_pd(fr->ic->epsfac);
 925     charge           = mdatoms->chargeA;
 926     nvdwtype         = fr->ntype;
 927     vdwparam         = fr->nbfp;
 928     vdwtype          = mdatoms->typeA;
 929
 930     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 931     ewtab            = fr->ic->tabq_coul_FDV0;
 932     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 933     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 934
 935     /* Setup water-specific parameters */
 936     inr              = nlist->iinr[0];
 937     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 938     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 939     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 940     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 941
 942     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 943     rcutoff_scalar   = fr->ic->rcoulomb;
 944     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 945     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 946
 947     rswitch_scalar   = fr->ic->rcoulomb_switch;
 948     rswitch          = _mm_set1_pd(rswitch_scalar);
 949     /* Setup switch parameters */
 950     d_scalar         = rcutoff_scalar-rswitch_scalar;
 951     d                = _mm_set1_pd(d_scalar);
 952     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 953     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 954     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 955     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 956     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 957     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 958
 959     /* Avoid stupid compiler warnings */
 960     jnrA = jnrB = 0;
 961     j_coord_offsetA = 0;
 962     j_coord_offsetB = 0;
 963
 964     outeriter        = 0;
 965     inneriter        = 0;
 966
 967     /* Start outer loop over neighborlists */
 968     for(iidx=0; iidx<nri; iidx++)
 969     {
 970         /* Load shift vector for this list */
 971         i_shift_offset   = DIM*shiftidx[iidx];
 972
 973         /* Load limits for loop over neighbors */
 974         j_index_start    = jindex[iidx];
 975         j_index_end      = jindex[iidx+1];
 976
 977         /* Get outer coordinate index */
 978         inr              = iinr[iidx];
 979         i_coord_offset   = DIM*inr;
 980
 981         /* Load i particle coords and add shift vector */
 982         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 983                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 984
 985         fix0             = _mm_setzero_pd();
 986         fiy0             = _mm_setzero_pd();
 987         fiz0             = _mm_setzero_pd();
 988         fix1             = _mm_setzero_pd();
 989         fiy1             = _mm_setzero_pd();
 990         fiz1             = _mm_setzero_pd();
 991         fix2             = _mm_setzero_pd();
 992         fiy2             = _mm_setzero_pd();
 993         fiz2             = _mm_setzero_pd();
 994         fix3             = _mm_setzero_pd();
 995         fiy3             = _mm_setzero_pd();
 996         fiz3             = _mm_setzero_pd();
 997
 998         /* Start inner kernel loop */
 999         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
1000         {
1001
1002             /* Get j neighbor index, and coordinate index */
1003             jnrA             = jjnr[jidx];
1004             jnrB             = jjnr[jidx+1];
1005             j_coord_offsetA  = DIM*jnrA;
1006             j_coord_offsetB  = DIM*jnrB;
1007
1008             /* load j atom coordinates */
1009             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
1010                                               &jx0,&jy0,&jz0);
1011
1012             /* Calculate displacement vector */
1013             dx00             = _mm_sub_pd(ix0,jx0);
1014             dy00             = _mm_sub_pd(iy0,jy0);
1015             dz00             = _mm_sub_pd(iz0,jz0);
1016             dx10             = _mm_sub_pd(ix1,jx0);
1017             dy10             = _mm_sub_pd(iy1,jy0);
1018             dz10             = _mm_sub_pd(iz1,jz0);
1019             dx20             = _mm_sub_pd(ix2,jx0);
1020             dy20             = _mm_sub_pd(iy2,jy0);
1021             dz20             = _mm_sub_pd(iz2,jz0);
1022             dx30             = _mm_sub_pd(ix3,jx0);
1023             dy30             = _mm_sub_pd(iy3,jy0);
1024             dz30             = _mm_sub_pd(iz3,jz0);
1025
1026             /* Calculate squared distance and things based on it */
1027             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1028             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1029             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1030             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1031
1032             rinv00           = avx128fma_invsqrt_d(rsq00);
1033             rinv10           = avx128fma_invsqrt_d(rsq10);
1034             rinv20           = avx128fma_invsqrt_d(rsq20);
1035             rinv30           = avx128fma_invsqrt_d(rsq30);
1036
1037             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1038             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1039             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1040             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1041
1042             /* Load parameters for j particles */
1043             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
1044             vdwjidx0A        = 2*vdwtype[jnrA+0];
1045             vdwjidx0B        = 2*vdwtype[jnrB+0];
1046
1047             fjx0             = _mm_setzero_pd();
1048             fjy0             = _mm_setzero_pd();
1049             fjz0             = _mm_setzero_pd();
1050
1051             /**************************
1052              * CALCULATE INTERACTIONS *
1053              **************************/
1054
1055             if (gmx_mm_any_lt(rsq00,rcutoff2))
1056             {
1057
1058             r00              = _mm_mul_pd(rsq00,rinv00);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1062                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1063
1064             /* LENNARD-JONES DISPERSION/REPULSION */
1065
1066             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1067             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1068             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1069             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1070             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1071
1072             d                = _mm_sub_pd(r00,rswitch);
1073             d                = _mm_max_pd(d,_mm_setzero_pd());
1074             d2               = _mm_mul_pd(d,d);
1075             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1076
1077             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1078
1079             /* Evaluate switch function */
1080             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1081             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1082             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1083
1084             fscal            = fvdw;
1085
1086             fscal            = _mm_and_pd(fscal,cutoff_mask);
1087
1088             /* Update vectorial force */
1089             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1090             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1091             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1092
1093             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1094             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1095             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1096
1097             }
1098
1099             /**************************
1100              * CALCULATE INTERACTIONS *
1101              **************************/
1102
1103             if (gmx_mm_any_lt(rsq10,rcutoff2))
1104             {
1105
1106             r10              = _mm_mul_pd(rsq10,rinv10);
1107
1108             /* Compute parameters for interactions between i and j atoms */
1109             qq10             = _mm_mul_pd(iq1,jq0);
1110
1111             /* EWALD ELECTROSTATICS */
1112
1113             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1114             ewrt             = _mm_mul_pd(r10,ewtabscale);
1115             ewitab           = _mm_cvttpd_epi32(ewrt);
1116 #ifdef __XOP__
1117             eweps            = _mm_frcz_pd(ewrt);
1118 #else
1119             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1120 #endif
1121             twoeweps         = _mm_add_pd(eweps,eweps);
1122             ewitab           = _mm_slli_epi32(ewitab,2);
1123             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1124             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1125             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1126             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1127             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1128             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1129             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1130             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1131             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1132             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1133
1134             d                = _mm_sub_pd(r10,rswitch);
1135             d                = _mm_max_pd(d,_mm_setzero_pd());
1136             d2               = _mm_mul_pd(d,d);
1137             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1138
1139             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1140
1141             /* Evaluate switch function */
1142             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1143             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1144             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm_and_pd(fscal,cutoff_mask);
1149
1150             /* Update vectorial force */
1151             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1152             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1153             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1154
1155             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1156             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1157             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1158
1159             }
1160
1161             /**************************
1162              * CALCULATE INTERACTIONS *
1163              **************************/
1164
1165             if (gmx_mm_any_lt(rsq20,rcutoff2))
1166             {
1167
1168             r20              = _mm_mul_pd(rsq20,rinv20);
1169
1170             /* Compute parameters for interactions between i and j atoms */
1171             qq20             = _mm_mul_pd(iq2,jq0);
1172
1173             /* EWALD ELECTROSTATICS */
1174
1175             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1176             ewrt             = _mm_mul_pd(r20,ewtabscale);
1177             ewitab           = _mm_cvttpd_epi32(ewrt);
1178 #ifdef __XOP__
1179             eweps            = _mm_frcz_pd(ewrt);
1180 #else
1181             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1182 #endif
1183             twoeweps         = _mm_add_pd(eweps,eweps);
1184             ewitab           = _mm_slli_epi32(ewitab,2);
1185             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1186             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1187             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1188             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1189             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1190             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1191             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1192             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1193             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1194             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1195
1196             d                = _mm_sub_pd(r20,rswitch);
1197             d                = _mm_max_pd(d,_mm_setzero_pd());
1198             d2               = _mm_mul_pd(d,d);
1199             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1200
1201             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1202
1203             /* Evaluate switch function */
1204             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1205             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1206             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1207
1208             fscal            = felec;
1209
1210             fscal            = _mm_and_pd(fscal,cutoff_mask);
1211
1212             /* Update vectorial force */
1213             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1214             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1215             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1216
1217             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1218             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1219             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1220
1221             }
1222
1223             /**************************
1224              * CALCULATE INTERACTIONS *
1225              **************************/
1226
1227             if (gmx_mm_any_lt(rsq30,rcutoff2))
1228             {
1229
1230             r30              = _mm_mul_pd(rsq30,rinv30);
1231
1232             /* Compute parameters for interactions between i and j atoms */
1233             qq30             = _mm_mul_pd(iq3,jq0);
1234
1235             /* EWALD ELECTROSTATICS */
1236
1237             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1238             ewrt             = _mm_mul_pd(r30,ewtabscale);
1239             ewitab           = _mm_cvttpd_epi32(ewrt);
1240 #ifdef __XOP__
1241             eweps            = _mm_frcz_pd(ewrt);
1242 #else
1243             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1244 #endif
1245             twoeweps         = _mm_add_pd(eweps,eweps);
1246             ewitab           = _mm_slli_epi32(ewitab,2);
1247             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1248             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1249             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1250             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1251             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1252             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1253             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1254             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1255             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1256             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1257
1258             d                = _mm_sub_pd(r30,rswitch);
1259             d                = _mm_max_pd(d,_mm_setzero_pd());
1260             d2               = _mm_mul_pd(d,d);
1261             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1262
1263             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1264
1265             /* Evaluate switch function */
1266             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1267             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1268             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1269
1270             fscal            = felec;
1271
1272             fscal            = _mm_and_pd(fscal,cutoff_mask);
1273
1274             /* Update vectorial force */
1275             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1276             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1277             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1278
1279             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1280             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1281             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1282
1283             }
1284
1285             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1286
1287             /* Inner loop uses 257 flops */
1288         }
1289
1290         if(jidx<j_index_end)
1291         {
1292
1293             jnrA             = jjnr[jidx];
1294             j_coord_offsetA  = DIM*jnrA;
1295
1296             /* load j atom coordinates */
1297             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1298                                               &jx0,&jy0,&jz0);
1299
1300             /* Calculate displacement vector */
1301             dx00             = _mm_sub_pd(ix0,jx0);
1302             dy00             = _mm_sub_pd(iy0,jy0);
1303             dz00             = _mm_sub_pd(iz0,jz0);
1304             dx10             = _mm_sub_pd(ix1,jx0);
1305             dy10             = _mm_sub_pd(iy1,jy0);
1306             dz10             = _mm_sub_pd(iz1,jz0);
1307             dx20             = _mm_sub_pd(ix2,jx0);
1308             dy20             = _mm_sub_pd(iy2,jy0);
1309             dz20             = _mm_sub_pd(iz2,jz0);
1310             dx30             = _mm_sub_pd(ix3,jx0);
1311             dy30             = _mm_sub_pd(iy3,jy0);
1312             dz30             = _mm_sub_pd(iz3,jz0);
1313
1314             /* Calculate squared distance and things based on it */
1315             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1316             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1317             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1318             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1319
1320             rinv00           = avx128fma_invsqrt_d(rsq00);
1321             rinv10           = avx128fma_invsqrt_d(rsq10);
1322             rinv20           = avx128fma_invsqrt_d(rsq20);
1323             rinv30           = avx128fma_invsqrt_d(rsq30);
1324
1325             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1326             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1327             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1328             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1329
1330             /* Load parameters for j particles */
1331             jq0              = _mm_load_sd(charge+jnrA+0);
1332             vdwjidx0A        = 2*vdwtype[jnrA+0];
1333
1334             fjx0             = _mm_setzero_pd();
1335             fjy0             = _mm_setzero_pd();
1336             fjz0             = _mm_setzero_pd();
1337
1338             /**************************
1339              * CALCULATE INTERACTIONS *
1340              **************************/
1341
1342             if (gmx_mm_any_lt(rsq00,rcutoff2))
1343             {
1344
1345             r00              = _mm_mul_pd(rsq00,rinv00);
1346
1347             /* Compute parameters for interactions between i and j atoms */
1348             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1349
1350             /* LENNARD-JONES DISPERSION/REPULSION */
1351
1352             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1353             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1354             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1355             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1356             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1357
1358             d                = _mm_sub_pd(r00,rswitch);
1359             d                = _mm_max_pd(d,_mm_setzero_pd());
1360             d2               = _mm_mul_pd(d,d);
1361             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1362
1363             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1364
1365             /* Evaluate switch function */
1366             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1367             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1368             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1369
1370             fscal            = fvdw;
1371
1372             fscal            = _mm_and_pd(fscal,cutoff_mask);
1373
1374             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1375
1376             /* Update vectorial force */
1377             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1378             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1379             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1380
1381             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1382             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1383             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1384
1385             }
1386
1387             /**************************
1388              * CALCULATE INTERACTIONS *
1389              **************************/
1390
1391             if (gmx_mm_any_lt(rsq10,rcutoff2))
1392             {
1393
1394             r10              = _mm_mul_pd(rsq10,rinv10);
1395
1396             /* Compute parameters for interactions between i and j atoms */
1397             qq10             = _mm_mul_pd(iq1,jq0);
1398
1399             /* EWALD ELECTROSTATICS */
1400
1401             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1402             ewrt             = _mm_mul_pd(r10,ewtabscale);
1403             ewitab           = _mm_cvttpd_epi32(ewrt);
1404 #ifdef __XOP__
1405             eweps            = _mm_frcz_pd(ewrt);
1406 #else
1407             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1408 #endif
1409             twoeweps         = _mm_add_pd(eweps,eweps);
1410             ewitab           = _mm_slli_epi32(ewitab,2);
1411             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1412             ewtabD           = _mm_setzero_pd();
1413             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1414             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1415             ewtabFn          = _mm_setzero_pd();
1416             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1417             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1418             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1419             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1420             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1421
1422             d                = _mm_sub_pd(r10,rswitch);
1423             d                = _mm_max_pd(d,_mm_setzero_pd());
1424             d2               = _mm_mul_pd(d,d);
1425             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1426
1427             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1428
1429             /* Evaluate switch function */
1430             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1431             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1432             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1433
1434             fscal            = felec;
1435
1436             fscal            = _mm_and_pd(fscal,cutoff_mask);
1437
1438             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1439
1440             /* Update vectorial force */
1441             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1442             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1443             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1444
1445             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1446             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1447             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1448
1449             }
1450
1451             /**************************
1452              * CALCULATE INTERACTIONS *
1453              **************************/
1454
1455             if (gmx_mm_any_lt(rsq20,rcutoff2))
1456             {
1457
1458             r20              = _mm_mul_pd(rsq20,rinv20);
1459
1460             /* Compute parameters for interactions between i and j atoms */
1461             qq20             = _mm_mul_pd(iq2,jq0);
1462
1463             /* EWALD ELECTROSTATICS */
1464
1465             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1466             ewrt             = _mm_mul_pd(r20,ewtabscale);
1467             ewitab           = _mm_cvttpd_epi32(ewrt);
1468 #ifdef __XOP__
1469             eweps            = _mm_frcz_pd(ewrt);
1470 #else
1471             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1472 #endif
1473             twoeweps         = _mm_add_pd(eweps,eweps);
1474             ewitab           = _mm_slli_epi32(ewitab,2);
1475             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1476             ewtabD           = _mm_setzero_pd();
1477             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1478             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1479             ewtabFn          = _mm_setzero_pd();
1480             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1481             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1482             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1483             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1484             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1485
1486             d                = _mm_sub_pd(r20,rswitch);
1487             d                = _mm_max_pd(d,_mm_setzero_pd());
1488             d2               = _mm_mul_pd(d,d);
1489             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1490
1491             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1492
1493             /* Evaluate switch function */
1494             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1495             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1496             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1497
1498             fscal            = felec;
1499
1500             fscal            = _mm_and_pd(fscal,cutoff_mask);
1501
1502             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1503
1504             /* Update vectorial force */
1505             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1506             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1507             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1508
1509             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1510             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1511             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1512
1513             }
1514
1515             /**************************
1516              * CALCULATE INTERACTIONS *
1517              **************************/
1518
1519             if (gmx_mm_any_lt(rsq30,rcutoff2))
1520             {
1521
1522             r30              = _mm_mul_pd(rsq30,rinv30);
1523
1524             /* Compute parameters for interactions between i and j atoms */
1525             qq30             = _mm_mul_pd(iq3,jq0);
1526
1527             /* EWALD ELECTROSTATICS */
1528
1529             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1530             ewrt             = _mm_mul_pd(r30,ewtabscale);
1531             ewitab           = _mm_cvttpd_epi32(ewrt);
1532 #ifdef __XOP__
1533             eweps            = _mm_frcz_pd(ewrt);
1534 #else
1535             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1536 #endif
1537             twoeweps         = _mm_add_pd(eweps,eweps);
1538             ewitab           = _mm_slli_epi32(ewitab,2);
1539             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1540             ewtabD           = _mm_setzero_pd();
1541             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1542             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1543             ewtabFn          = _mm_setzero_pd();
1544             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1545             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1546             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1547             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1548             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1549
1550             d                = _mm_sub_pd(r30,rswitch);
1551             d                = _mm_max_pd(d,_mm_setzero_pd());
1552             d2               = _mm_mul_pd(d,d);
1553             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1554
1555             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1556
1557             /* Evaluate switch function */
1558             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1559             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1560             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1561
1562             fscal            = felec;
1563
1564             fscal            = _mm_and_pd(fscal,cutoff_mask);
1565
1566             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1567
1568             /* Update vectorial force */
1569             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1570             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1571             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1572
1573             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1574             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1575             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1576
1577             }
1578
1579             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1580
1581             /* Inner loop uses 257 flops */
1582         }
1583
1584         /* End of innermost loop */
1585
1586         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1587                                               f+i_coord_offset,fshift+i_shift_offset);
1588
1589         /* Increment number of inner iterations */
1590         inneriter                  += j_index_end - j_index_start;
1591
1592         /* Outer loop uses 24 flops */
1593     }
1594
1595     /* Increment number of outer iterations */
1596     outeriter        += nri;
1597
1598     /* Update outer/inner flops */
1599
1600     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*257);
1601 }