src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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_GeomW3P1_VF_avx_128_fma_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            LennardJones
  53  * Geometry:                   Water3-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_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              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  88     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  89     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     int              nvdwtype;
  93     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  94     int              *vdwtype;
  95     real             *vdwparam;
  96     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  97     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  98     __m128i          ewitab;
  99     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 102     real             rswitch_scalar,d_scalar;
 103     __m128d          dummy_mask,cutoff_mask;
 104     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 105     __m128d          one     = _mm_set1_pd(1.0);
 106     __m128d          two     = _mm_set1_pd(2.0);
 107     x                = xx[0];
 108     f                = ff[0];
 109
 110     nri              = nlist->nri;
 111     iinr             = nlist->iinr;
 112     jindex           = nlist->jindex;
 113     jjnr             = nlist->jjnr;
 114     shiftidx         = nlist->shift;
 115     gid              = nlist->gid;
 116     shiftvec         = fr->shift_vec[0];
 117     fshift           = fr->fshift[0];
 118     facel            = _mm_set1_pd(fr->ic->epsfac);
 119     charge           = mdatoms->chargeA;
 120     nvdwtype         = fr->ntype;
 121     vdwparam         = fr->nbfp;
 122     vdwtype          = mdatoms->typeA;
 123
 124     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 125     ewtab            = fr->ic->tabq_coul_FDV0;
 126     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 127     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 128
 129     /* Setup water-specific parameters */
 130     inr              = nlist->iinr[0];
 131     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 132     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 133     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 134     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->ic->rcoulomb;
 138     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 139     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 140
 141     rswitch_scalar   = fr->ic->rcoulomb_switch;
 142     rswitch          = _mm_set1_pd(rswitch_scalar);
 143     /* Setup switch parameters */
 144     d_scalar         = rcutoff_scalar-rswitch_scalar;
 145     d                = _mm_set1_pd(d_scalar);
 146     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 147     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 148     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 150     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157
 158     outeriter        = 0;
 159     inneriter        = 0;
 160
 161     /* Start outer loop over neighborlists */
 162     for(iidx=0; iidx<nri; iidx++)
 163     {
 164         /* Load shift vector for this list */
 165         i_shift_offset   = DIM*shiftidx[iidx];
 166
 167         /* Load limits for loop over neighbors */
 168         j_index_start    = jindex[iidx];
 169         j_index_end      = jindex[iidx+1];
 170
 171         /* Get outer coordinate index */
 172         inr              = iinr[iidx];
 173         i_coord_offset   = DIM*inr;
 174
 175         /* Load i particle coords and add shift vector */
 176         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 177                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 178
 179         fix0             = _mm_setzero_pd();
 180         fiy0             = _mm_setzero_pd();
 181         fiz0             = _mm_setzero_pd();
 182         fix1             = _mm_setzero_pd();
 183         fiy1             = _mm_setzero_pd();
 184         fiz1             = _mm_setzero_pd();
 185         fix2             = _mm_setzero_pd();
 186         fiy2             = _mm_setzero_pd();
 187         fiz2             = _mm_setzero_pd();
 188
 189         /* Reset potential sums */
 190         velecsum         = _mm_setzero_pd();
 191         vvdwsum          = _mm_setzero_pd();
 192
 193         /* Start inner kernel loop */
 194         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 195         {
 196
 197             /* Get j neighbor index, and coordinate index */
 198             jnrA             = jjnr[jidx];
 199             jnrB             = jjnr[jidx+1];
 200             j_coord_offsetA  = DIM*jnrA;
 201             j_coord_offsetB  = DIM*jnrB;
 202
 203             /* load j atom coordinates */
 204             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 205                                               &jx0,&jy0,&jz0);
 206
 207             /* Calculate displacement vector */
 208             dx00             = _mm_sub_pd(ix0,jx0);
 209             dy00             = _mm_sub_pd(iy0,jy0);
 210             dz00             = _mm_sub_pd(iz0,jz0);
 211             dx10             = _mm_sub_pd(ix1,jx0);
 212             dy10             = _mm_sub_pd(iy1,jy0);
 213             dz10             = _mm_sub_pd(iz1,jz0);
 214             dx20             = _mm_sub_pd(ix2,jx0);
 215             dy20             = _mm_sub_pd(iy2,jy0);
 216             dz20             = _mm_sub_pd(iz2,jz0);
 217
 218             /* Calculate squared distance and things based on it */
 219             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 220             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 221             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 222
 223             rinv00           = avx128fma_invsqrt_d(rsq00);
 224             rinv10           = avx128fma_invsqrt_d(rsq10);
 225             rinv20           = avx128fma_invsqrt_d(rsq20);
 226
 227             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 228             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 233             vdwjidx0A        = 2*vdwtype[jnrA+0];
 234             vdwjidx0B        = 2*vdwtype[jnrB+0];
 235
 236             fjx0             = _mm_setzero_pd();
 237             fjy0             = _mm_setzero_pd();
 238             fjz0             = _mm_setzero_pd();
 239
 240             /**************************
 241              * CALCULATE INTERACTIONS *
 242              **************************/
 243
 244             if (gmx_mm_any_lt(rsq00,rcutoff2))
 245             {
 246
 247             r00              = _mm_mul_pd(rsq00,rinv00);
 248
 249             /* Compute parameters for interactions between i and j atoms */
 250             qq00             = _mm_mul_pd(iq0,jq0);
 251             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 252                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 253
 254             /* EWALD ELECTROSTATICS */
 255
 256             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 257             ewrt             = _mm_mul_pd(r00,ewtabscale);
 258             ewitab           = _mm_cvttpd_epi32(ewrt);
 259 #ifdef __XOP__
 260             eweps            = _mm_frcz_pd(ewrt);
 261 #else
 262             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 263 #endif
 264             twoeweps         = _mm_add_pd(eweps,eweps);
 265             ewitab           = _mm_slli_epi32(ewitab,2);
 266             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 267             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 268             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 269             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 270             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 271             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 272             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 273             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 274             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 275             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 276
 277             /* LENNARD-JONES DISPERSION/REPULSION */
 278
 279             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 280             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 281             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 282             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 283             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 284
 285             d                = _mm_sub_pd(r00,rswitch);
 286             d                = _mm_max_pd(d,_mm_setzero_pd());
 287             d2               = _mm_mul_pd(d,d);
 288             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 289
 290             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 291
 292             /* Evaluate switch function */
 293             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 294             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 295             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 296             velec            = _mm_mul_pd(velec,sw);
 297             vvdw             = _mm_mul_pd(vvdw,sw);
 298             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 299
 300             /* Update potential sum for this i atom from the interaction with this j atom. */
 301             velec            = _mm_and_pd(velec,cutoff_mask);
 302             velecsum         = _mm_add_pd(velecsum,velec);
 303             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 304             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 305
 306             fscal            = _mm_add_pd(felec,fvdw);
 307
 308             fscal            = _mm_and_pd(fscal,cutoff_mask);
 309
 310             /* Update vectorial force */
 311             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 312             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 313             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 314
 315             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 316             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 317             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 318
 319             }
 320
 321             /**************************
 322              * CALCULATE INTERACTIONS *
 323              **************************/
 324
 325             if (gmx_mm_any_lt(rsq10,rcutoff2))
 326             {
 327
 328             r10              = _mm_mul_pd(rsq10,rinv10);
 329
 330             /* Compute parameters for interactions between i and j atoms */
 331             qq10             = _mm_mul_pd(iq1,jq0);
 332
 333             /* EWALD ELECTROSTATICS */
 334
 335             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 336             ewrt             = _mm_mul_pd(r10,ewtabscale);
 337             ewitab           = _mm_cvttpd_epi32(ewrt);
 338 #ifdef __XOP__
 339             eweps            = _mm_frcz_pd(ewrt);
 340 #else
 341             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 342 #endif
 343             twoeweps         = _mm_add_pd(eweps,eweps);
 344             ewitab           = _mm_slli_epi32(ewitab,2);
 345             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 346             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 347             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 348             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 349             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 350             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 351             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 352             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 353             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 354             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 355
 356             d                = _mm_sub_pd(r10,rswitch);
 357             d                = _mm_max_pd(d,_mm_setzero_pd());
 358             d2               = _mm_mul_pd(d,d);
 359             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 360
 361             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 362
 363             /* Evaluate switch function */
 364             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 365             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 366             velec            = _mm_mul_pd(velec,sw);
 367             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 368
 369             /* Update potential sum for this i atom from the interaction with this j atom. */
 370             velec            = _mm_and_pd(velec,cutoff_mask);
 371             velecsum         = _mm_add_pd(velecsum,velec);
 372
 373             fscal            = felec;
 374
 375             fscal            = _mm_and_pd(fscal,cutoff_mask);
 376
 377             /* Update vectorial force */
 378             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 379             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 380             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 381
 382             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 383             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 384             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 385
 386             }
 387
 388             /**************************
 389              * CALCULATE INTERACTIONS *
 390              **************************/
 391
 392             if (gmx_mm_any_lt(rsq20,rcutoff2))
 393             {
 394
 395             r20              = _mm_mul_pd(rsq20,rinv20);
 396
 397             /* Compute parameters for interactions between i and j atoms */
 398             qq20             = _mm_mul_pd(iq2,jq0);
 399
 400             /* EWALD ELECTROSTATICS */
 401
 402             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 403             ewrt             = _mm_mul_pd(r20,ewtabscale);
 404             ewitab           = _mm_cvttpd_epi32(ewrt);
 405 #ifdef __XOP__
 406             eweps            = _mm_frcz_pd(ewrt);
 407 #else
 408             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 409 #endif
 410             twoeweps         = _mm_add_pd(eweps,eweps);
 411             ewitab           = _mm_slli_epi32(ewitab,2);
 412             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 413             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 414             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 415             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 416             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 417             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 418             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 419             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 420             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 421             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 422
 423             d                = _mm_sub_pd(r20,rswitch);
 424             d                = _mm_max_pd(d,_mm_setzero_pd());
 425             d2               = _mm_mul_pd(d,d);
 426             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 427
 428             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 429
 430             /* Evaluate switch function */
 431             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 432             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 433             velec            = _mm_mul_pd(velec,sw);
 434             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 435
 436             /* Update potential sum for this i atom from the interaction with this j atom. */
 437             velec            = _mm_and_pd(velec,cutoff_mask);
 438             velecsum         = _mm_add_pd(velecsum,velec);
 439
 440             fscal            = felec;
 441
 442             fscal            = _mm_and_pd(fscal,cutoff_mask);
 443
 444             /* Update vectorial force */
 445             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 446             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 447             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 448
 449             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 450             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 451             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 452
 453             }
 454
 455             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 456
 457             /* Inner loop uses 225 flops */
 458         }
 459
 460         if(jidx<j_index_end)
 461         {
 462
 463             jnrA             = jjnr[jidx];
 464             j_coord_offsetA  = DIM*jnrA;
 465
 466             /* load j atom coordinates */
 467             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 468                                               &jx0,&jy0,&jz0);
 469
 470             /* Calculate displacement vector */
 471             dx00             = _mm_sub_pd(ix0,jx0);
 472             dy00             = _mm_sub_pd(iy0,jy0);
 473             dz00             = _mm_sub_pd(iz0,jz0);
 474             dx10             = _mm_sub_pd(ix1,jx0);
 475             dy10             = _mm_sub_pd(iy1,jy0);
 476             dz10             = _mm_sub_pd(iz1,jz0);
 477             dx20             = _mm_sub_pd(ix2,jx0);
 478             dy20             = _mm_sub_pd(iy2,jy0);
 479             dz20             = _mm_sub_pd(iz2,jz0);
 480
 481             /* Calculate squared distance and things based on it */
 482             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 483             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 484             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 485
 486             rinv00           = avx128fma_invsqrt_d(rsq00);
 487             rinv10           = avx128fma_invsqrt_d(rsq10);
 488             rinv20           = avx128fma_invsqrt_d(rsq20);
 489
 490             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 491             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 492             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 493
 494             /* Load parameters for j particles */
 495             jq0              = _mm_load_sd(charge+jnrA+0);
 496             vdwjidx0A        = 2*vdwtype[jnrA+0];
 497
 498             fjx0             = _mm_setzero_pd();
 499             fjy0             = _mm_setzero_pd();
 500             fjz0             = _mm_setzero_pd();
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             if (gmx_mm_any_lt(rsq00,rcutoff2))
 507             {
 508
 509             r00              = _mm_mul_pd(rsq00,rinv00);
 510
 511             /* Compute parameters for interactions between i and j atoms */
 512             qq00             = _mm_mul_pd(iq0,jq0);
 513             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 514
 515             /* EWALD ELECTROSTATICS */
 516
 517             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 518             ewrt             = _mm_mul_pd(r00,ewtabscale);
 519             ewitab           = _mm_cvttpd_epi32(ewrt);
 520 #ifdef __XOP__
 521             eweps            = _mm_frcz_pd(ewrt);
 522 #else
 523             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 524 #endif
 525             twoeweps         = _mm_add_pd(eweps,eweps);
 526             ewitab           = _mm_slli_epi32(ewitab,2);
 527             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 528             ewtabD           = _mm_setzero_pd();
 529             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 530             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 531             ewtabFn          = _mm_setzero_pd();
 532             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 533             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 534             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 535             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 536             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 537
 538             /* LENNARD-JONES DISPERSION/REPULSION */
 539
 540             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 541             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 542             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 543             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 544             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 545
 546             d                = _mm_sub_pd(r00,rswitch);
 547             d                = _mm_max_pd(d,_mm_setzero_pd());
 548             d2               = _mm_mul_pd(d,d);
 549             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 550
 551             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 552
 553             /* Evaluate switch function */
 554             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 555             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 556             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 557             velec            = _mm_mul_pd(velec,sw);
 558             vvdw             = _mm_mul_pd(vvdw,sw);
 559             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 560
 561             /* Update potential sum for this i atom from the interaction with this j atom. */
 562             velec            = _mm_and_pd(velec,cutoff_mask);
 563             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 564             velecsum         = _mm_add_pd(velecsum,velec);
 565             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 566             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 567             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 568
 569             fscal            = _mm_add_pd(felec,fvdw);
 570
 571             fscal            = _mm_and_pd(fscal,cutoff_mask);
 572
 573             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 574
 575             /* Update vectorial force */
 576             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 577             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 578             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 579
 580             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 581             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 582             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 583
 584             }
 585
 586             /**************************
 587              * CALCULATE INTERACTIONS *
 588              **************************/
 589
 590             if (gmx_mm_any_lt(rsq10,rcutoff2))
 591             {
 592
 593             r10              = _mm_mul_pd(rsq10,rinv10);
 594
 595             /* Compute parameters for interactions between i and j atoms */
 596             qq10             = _mm_mul_pd(iq1,jq0);
 597
 598             /* EWALD ELECTROSTATICS */
 599
 600             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 601             ewrt             = _mm_mul_pd(r10,ewtabscale);
 602             ewitab           = _mm_cvttpd_epi32(ewrt);
 603 #ifdef __XOP__
 604             eweps            = _mm_frcz_pd(ewrt);
 605 #else
 606             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 607 #endif
 608             twoeweps         = _mm_add_pd(eweps,eweps);
 609             ewitab           = _mm_slli_epi32(ewitab,2);
 610             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 611             ewtabD           = _mm_setzero_pd();
 612             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 613             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 614             ewtabFn          = _mm_setzero_pd();
 615             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 616             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 617             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 618             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 619             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 620
 621             d                = _mm_sub_pd(r10,rswitch);
 622             d                = _mm_max_pd(d,_mm_setzero_pd());
 623             d2               = _mm_mul_pd(d,d);
 624             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 625
 626             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 627
 628             /* Evaluate switch function */
 629             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 630             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 631             velec            = _mm_mul_pd(velec,sw);
 632             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 633
 634             /* Update potential sum for this i atom from the interaction with this j atom. */
 635             velec            = _mm_and_pd(velec,cutoff_mask);
 636             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 637             velecsum         = _mm_add_pd(velecsum,velec);
 638
 639             fscal            = felec;
 640
 641             fscal            = _mm_and_pd(fscal,cutoff_mask);
 642
 643             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 644
 645             /* Update vectorial force */
 646             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 647             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 648             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 649
 650             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 651             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 652             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 653
 654             }
 655
 656             /**************************
 657              * CALCULATE INTERACTIONS *
 658              **************************/
 659
 660             if (gmx_mm_any_lt(rsq20,rcutoff2))
 661             {
 662
 663             r20              = _mm_mul_pd(rsq20,rinv20);
 664
 665             /* Compute parameters for interactions between i and j atoms */
 666             qq20             = _mm_mul_pd(iq2,jq0);
 667
 668             /* EWALD ELECTROSTATICS */
 669
 670             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 671             ewrt             = _mm_mul_pd(r20,ewtabscale);
 672             ewitab           = _mm_cvttpd_epi32(ewrt);
 673 #ifdef __XOP__
 674             eweps            = _mm_frcz_pd(ewrt);
 675 #else
 676             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 677 #endif
 678             twoeweps         = _mm_add_pd(eweps,eweps);
 679             ewitab           = _mm_slli_epi32(ewitab,2);
 680             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 681             ewtabD           = _mm_setzero_pd();
 682             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 683             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 684             ewtabFn          = _mm_setzero_pd();
 685             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 686             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 687             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 688             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 689             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 690
 691             d                = _mm_sub_pd(r20,rswitch);
 692             d                = _mm_max_pd(d,_mm_setzero_pd());
 693             d2               = _mm_mul_pd(d,d);
 694             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 695
 696             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 697
 698             /* Evaluate switch function */
 699             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 700             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 701             velec            = _mm_mul_pd(velec,sw);
 702             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 703
 704             /* Update potential sum for this i atom from the interaction with this j atom. */
 705             velec            = _mm_and_pd(velec,cutoff_mask);
 706             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 707             velecsum         = _mm_add_pd(velecsum,velec);
 708
 709             fscal            = felec;
 710
 711             fscal            = _mm_and_pd(fscal,cutoff_mask);
 712
 713             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 714
 715             /* Update vectorial force */
 716             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 717             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 718             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 719
 720             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 721             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 722             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 723
 724             }
 725
 726             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 727
 728             /* Inner loop uses 225 flops */
 729         }
 730
 731         /* End of innermost loop */
 732
 733         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 734                                               f+i_coord_offset,fshift+i_shift_offset);
 735
 736         ggid                        = gid[iidx];
 737         /* Update potential energies */
 738         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 739         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 740
 741         /* Increment number of inner iterations */
 742         inneriter                  += j_index_end - j_index_start;
 743
 744         /* Outer loop uses 20 flops */
 745     }
 746
 747     /* Increment number of outer iterations */
 748     outeriter        += nri;
 749
 750     /* Update outer/inner flops */
 751
 752     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*225);
 753 }
 754 /*
 755  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 756  * Electrostatics interaction: Ewald
 757  * VdW interaction:            LennardJones
 758  * Geometry:                   Water3-Particle
 759  * Calculate force/pot:        Force
 760  */
 761 void
 762 nb_kernel_ElecEwSw_VdwLJSw_GeomW3P1_F_avx_128_fma_double
 763                     (t_nblist                    * gmx_restrict       nlist,
 764                      rvec                        * gmx_restrict          xx,
 765                      rvec                        * gmx_restrict          ff,
 766                      struct t_forcerec           * gmx_restrict          fr,
 767                      t_mdatoms                   * gmx_restrict     mdatoms,
 768                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 769                      t_nrnb                      * gmx_restrict        nrnb)
 770 {
 771     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 772      * just 0 for non-waters.
 773      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 774      * jnr indices corresponding to data put in the four positions in the SIMD register.
 775      */
 776     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 777     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 778     int              jnrA,jnrB;
 779     int              j_coord_offsetA,j_coord_offsetB;
 780     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 781     real             rcutoff_scalar;
 782     real             *shiftvec,*fshift,*x,*f;
 783     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 784     int              vdwioffset0;
 785     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 786     int              vdwioffset1;
 787     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 788     int              vdwioffset2;
 789     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 790     int              vdwjidx0A,vdwjidx0B;
 791     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 792     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 793     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 794     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 795     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 796     real             *charge;
 797     int              nvdwtype;
 798     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 799     int              *vdwtype;
 800     real             *vdwparam;
 801     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 802     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 803     __m128i          ewitab;
 804     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 805     real             *ewtab;
 806     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 807     real             rswitch_scalar,d_scalar;
 808     __m128d          dummy_mask,cutoff_mask;
 809     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 810     __m128d          one     = _mm_set1_pd(1.0);
 811     __m128d          two     = _mm_set1_pd(2.0);
 812     x                = xx[0];
 813     f                = ff[0];
 814
 815     nri              = nlist->nri;
 816     iinr             = nlist->iinr;
 817     jindex           = nlist->jindex;
 818     jjnr             = nlist->jjnr;
 819     shiftidx         = nlist->shift;
 820     gid              = nlist->gid;
 821     shiftvec         = fr->shift_vec[0];
 822     fshift           = fr->fshift[0];
 823     facel            = _mm_set1_pd(fr->ic->epsfac);
 824     charge           = mdatoms->chargeA;
 825     nvdwtype         = fr->ntype;
 826     vdwparam         = fr->nbfp;
 827     vdwtype          = mdatoms->typeA;
 828
 829     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 830     ewtab            = fr->ic->tabq_coul_FDV0;
 831     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 832     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 833
 834     /* Setup water-specific parameters */
 835     inr              = nlist->iinr[0];
 836     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 837     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 838     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 839     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 840
 841     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 842     rcutoff_scalar   = fr->ic->rcoulomb;
 843     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 844     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 845
 846     rswitch_scalar   = fr->ic->rcoulomb_switch;
 847     rswitch          = _mm_set1_pd(rswitch_scalar);
 848     /* Setup switch parameters */
 849     d_scalar         = rcutoff_scalar-rswitch_scalar;
 850     d                = _mm_set1_pd(d_scalar);
 851     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 852     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 853     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 854     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 855     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 856     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 857
 858     /* Avoid stupid compiler warnings */
 859     jnrA = jnrB = 0;
 860     j_coord_offsetA = 0;
 861     j_coord_offsetB = 0;
 862
 863     outeriter        = 0;
 864     inneriter        = 0;
 865
 866     /* Start outer loop over neighborlists */
 867     for(iidx=0; iidx<nri; iidx++)
 868     {
 869         /* Load shift vector for this list */
 870         i_shift_offset   = DIM*shiftidx[iidx];
 871
 872         /* Load limits for loop over neighbors */
 873         j_index_start    = jindex[iidx];
 874         j_index_end      = jindex[iidx+1];
 875
 876         /* Get outer coordinate index */
 877         inr              = iinr[iidx];
 878         i_coord_offset   = DIM*inr;
 879
 880         /* Load i particle coords and add shift vector */
 881         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 882                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 883
 884         fix0             = _mm_setzero_pd();
 885         fiy0             = _mm_setzero_pd();
 886         fiz0             = _mm_setzero_pd();
 887         fix1             = _mm_setzero_pd();
 888         fiy1             = _mm_setzero_pd();
 889         fiz1             = _mm_setzero_pd();
 890         fix2             = _mm_setzero_pd();
 891         fiy2             = _mm_setzero_pd();
 892         fiz2             = _mm_setzero_pd();
 893
 894         /* Start inner kernel loop */
 895         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 896         {
 897
 898             /* Get j neighbor index, and coordinate index */
 899             jnrA             = jjnr[jidx];
 900             jnrB             = jjnr[jidx+1];
 901             j_coord_offsetA  = DIM*jnrA;
 902             j_coord_offsetB  = DIM*jnrB;
 903
 904             /* load j atom coordinates */
 905             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 906                                               &jx0,&jy0,&jz0);
 907
 908             /* Calculate displacement vector */
 909             dx00             = _mm_sub_pd(ix0,jx0);
 910             dy00             = _mm_sub_pd(iy0,jy0);
 911             dz00             = _mm_sub_pd(iz0,jz0);
 912             dx10             = _mm_sub_pd(ix1,jx0);
 913             dy10             = _mm_sub_pd(iy1,jy0);
 914             dz10             = _mm_sub_pd(iz1,jz0);
 915             dx20             = _mm_sub_pd(ix2,jx0);
 916             dy20             = _mm_sub_pd(iy2,jy0);
 917             dz20             = _mm_sub_pd(iz2,jz0);
 918
 919             /* Calculate squared distance and things based on it */
 920             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 921             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 922             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 923
 924             rinv00           = avx128fma_invsqrt_d(rsq00);
 925             rinv10           = avx128fma_invsqrt_d(rsq10);
 926             rinv20           = avx128fma_invsqrt_d(rsq20);
 927
 928             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 929             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 930             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 931
 932             /* Load parameters for j particles */
 933             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 934             vdwjidx0A        = 2*vdwtype[jnrA+0];
 935             vdwjidx0B        = 2*vdwtype[jnrB+0];
 936
 937             fjx0             = _mm_setzero_pd();
 938             fjy0             = _mm_setzero_pd();
 939             fjz0             = _mm_setzero_pd();
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             if (gmx_mm_any_lt(rsq00,rcutoff2))
 946             {
 947
 948             r00              = _mm_mul_pd(rsq00,rinv00);
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq00             = _mm_mul_pd(iq0,jq0);
 952             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 953                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 954
 955             /* EWALD ELECTROSTATICS */
 956
 957             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 958             ewrt             = _mm_mul_pd(r00,ewtabscale);
 959             ewitab           = _mm_cvttpd_epi32(ewrt);
 960 #ifdef __XOP__
 961             eweps            = _mm_frcz_pd(ewrt);
 962 #else
 963             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 964 #endif
 965             twoeweps         = _mm_add_pd(eweps,eweps);
 966             ewitab           = _mm_slli_epi32(ewitab,2);
 967             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 968             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 969             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 970             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 971             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 972             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 973             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 974             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 975             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 976             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 977
 978             /* LENNARD-JONES DISPERSION/REPULSION */
 979
 980             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 981             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 982             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 983             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 984             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 985
 986             d                = _mm_sub_pd(r00,rswitch);
 987             d                = _mm_max_pd(d,_mm_setzero_pd());
 988             d2               = _mm_mul_pd(d,d);
 989             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 990
 991             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 992
 993             /* Evaluate switch function */
 994             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 995             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 996             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 997             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 998
 999             fscal            = _mm_add_pd(felec,fvdw);
1000
1001             fscal            = _mm_and_pd(fscal,cutoff_mask);
1002
1003             /* Update vectorial force */
1004             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1005             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1006             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1007
1008             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1009             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1010             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1011
1012             }
1013
1014             /**************************
1015              * CALCULATE INTERACTIONS *
1016              **************************/
1017
1018             if (gmx_mm_any_lt(rsq10,rcutoff2))
1019             {
1020
1021             r10              = _mm_mul_pd(rsq10,rinv10);
1022
1023             /* Compute parameters for interactions between i and j atoms */
1024             qq10             = _mm_mul_pd(iq1,jq0);
1025
1026             /* EWALD ELECTROSTATICS */
1027
1028             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1029             ewrt             = _mm_mul_pd(r10,ewtabscale);
1030             ewitab           = _mm_cvttpd_epi32(ewrt);
1031 #ifdef __XOP__
1032             eweps            = _mm_frcz_pd(ewrt);
1033 #else
1034             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1035 #endif
1036             twoeweps         = _mm_add_pd(eweps,eweps);
1037             ewitab           = _mm_slli_epi32(ewitab,2);
1038             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1039             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1040             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1041             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1042             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1043             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1044             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1045             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1046             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1047             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1048
1049             d                = _mm_sub_pd(r10,rswitch);
1050             d                = _mm_max_pd(d,_mm_setzero_pd());
1051             d2               = _mm_mul_pd(d,d);
1052             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1053
1054             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1055
1056             /* Evaluate switch function */
1057             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1058             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1059             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1060
1061             fscal            = felec;
1062
1063             fscal            = _mm_and_pd(fscal,cutoff_mask);
1064
1065             /* Update vectorial force */
1066             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1067             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1068             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1069
1070             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1071             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1072             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1073
1074             }
1075
1076             /**************************
1077              * CALCULATE INTERACTIONS *
1078              **************************/
1079
1080             if (gmx_mm_any_lt(rsq20,rcutoff2))
1081             {
1082
1083             r20              = _mm_mul_pd(rsq20,rinv20);
1084
1085             /* Compute parameters for interactions between i and j atoms */
1086             qq20             = _mm_mul_pd(iq2,jq0);
1087
1088             /* EWALD ELECTROSTATICS */
1089
1090             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1091             ewrt             = _mm_mul_pd(r20,ewtabscale);
1092             ewitab           = _mm_cvttpd_epi32(ewrt);
1093 #ifdef __XOP__
1094             eweps            = _mm_frcz_pd(ewrt);
1095 #else
1096             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1097 #endif
1098             twoeweps         = _mm_add_pd(eweps,eweps);
1099             ewitab           = _mm_slli_epi32(ewitab,2);
1100             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1101             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1102             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1103             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1104             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1105             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1106             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1107             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1108             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1109             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1110
1111             d                = _mm_sub_pd(r20,rswitch);
1112             d                = _mm_max_pd(d,_mm_setzero_pd());
1113             d2               = _mm_mul_pd(d,d);
1114             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1115
1116             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1117
1118             /* Evaluate switch function */
1119             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1120             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1121             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1122
1123             fscal            = felec;
1124
1125             fscal            = _mm_and_pd(fscal,cutoff_mask);
1126
1127             /* Update vectorial force */
1128             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1129             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1130             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1131
1132             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1133             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1134             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1135
1136             }
1137
1138             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1139
1140             /* Inner loop uses 213 flops */
1141         }
1142
1143         if(jidx<j_index_end)
1144         {
1145
1146             jnrA             = jjnr[jidx];
1147             j_coord_offsetA  = DIM*jnrA;
1148
1149             /* load j atom coordinates */
1150             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1151                                               &jx0,&jy0,&jz0);
1152
1153             /* Calculate displacement vector */
1154             dx00             = _mm_sub_pd(ix0,jx0);
1155             dy00             = _mm_sub_pd(iy0,jy0);
1156             dz00             = _mm_sub_pd(iz0,jz0);
1157             dx10             = _mm_sub_pd(ix1,jx0);
1158             dy10             = _mm_sub_pd(iy1,jy0);
1159             dz10             = _mm_sub_pd(iz1,jz0);
1160             dx20             = _mm_sub_pd(ix2,jx0);
1161             dy20             = _mm_sub_pd(iy2,jy0);
1162             dz20             = _mm_sub_pd(iz2,jz0);
1163
1164             /* Calculate squared distance and things based on it */
1165             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1166             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1167             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1168
1169             rinv00           = avx128fma_invsqrt_d(rsq00);
1170             rinv10           = avx128fma_invsqrt_d(rsq10);
1171             rinv20           = avx128fma_invsqrt_d(rsq20);
1172
1173             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1174             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1175             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1176
1177             /* Load parameters for j particles */
1178             jq0              = _mm_load_sd(charge+jnrA+0);
1179             vdwjidx0A        = 2*vdwtype[jnrA+0];
1180
1181             fjx0             = _mm_setzero_pd();
1182             fjy0             = _mm_setzero_pd();
1183             fjz0             = _mm_setzero_pd();
1184
1185             /**************************
1186              * CALCULATE INTERACTIONS *
1187              **************************/
1188
1189             if (gmx_mm_any_lt(rsq00,rcutoff2))
1190             {
1191
1192             r00              = _mm_mul_pd(rsq00,rinv00);
1193
1194             /* Compute parameters for interactions between i and j atoms */
1195             qq00             = _mm_mul_pd(iq0,jq0);
1196             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1197
1198             /* EWALD ELECTROSTATICS */
1199
1200             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1201             ewrt             = _mm_mul_pd(r00,ewtabscale);
1202             ewitab           = _mm_cvttpd_epi32(ewrt);
1203 #ifdef __XOP__
1204             eweps            = _mm_frcz_pd(ewrt);
1205 #else
1206             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1207 #endif
1208             twoeweps         = _mm_add_pd(eweps,eweps);
1209             ewitab           = _mm_slli_epi32(ewitab,2);
1210             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1211             ewtabD           = _mm_setzero_pd();
1212             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1213             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1214             ewtabFn          = _mm_setzero_pd();
1215             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1216             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1217             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1218             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
1219             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1220
1221             /* LENNARD-JONES DISPERSION/REPULSION */
1222
1223             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1224             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1225             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1226             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1227             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1228
1229             d                = _mm_sub_pd(r00,rswitch);
1230             d                = _mm_max_pd(d,_mm_setzero_pd());
1231             d2               = _mm_mul_pd(d,d);
1232             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1233
1234             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1235
1236             /* Evaluate switch function */
1237             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1238             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
1239             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1240             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1241
1242             fscal            = _mm_add_pd(felec,fvdw);
1243
1244             fscal            = _mm_and_pd(fscal,cutoff_mask);
1245
1246             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1247
1248             /* Update vectorial force */
1249             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1250             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1251             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1252
1253             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1254             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1255             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1256
1257             }
1258
1259             /**************************
1260              * CALCULATE INTERACTIONS *
1261              **************************/
1262
1263             if (gmx_mm_any_lt(rsq10,rcutoff2))
1264             {
1265
1266             r10              = _mm_mul_pd(rsq10,rinv10);
1267
1268             /* Compute parameters for interactions between i and j atoms */
1269             qq10             = _mm_mul_pd(iq1,jq0);
1270
1271             /* EWALD ELECTROSTATICS */
1272
1273             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1274             ewrt             = _mm_mul_pd(r10,ewtabscale);
1275             ewitab           = _mm_cvttpd_epi32(ewrt);
1276 #ifdef __XOP__
1277             eweps            = _mm_frcz_pd(ewrt);
1278 #else
1279             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1280 #endif
1281             twoeweps         = _mm_add_pd(eweps,eweps);
1282             ewitab           = _mm_slli_epi32(ewitab,2);
1283             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1284             ewtabD           = _mm_setzero_pd();
1285             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1286             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1287             ewtabFn          = _mm_setzero_pd();
1288             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1289             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1290             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1291             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1292             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1293
1294             d                = _mm_sub_pd(r10,rswitch);
1295             d                = _mm_max_pd(d,_mm_setzero_pd());
1296             d2               = _mm_mul_pd(d,d);
1297             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1298
1299             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1300
1301             /* Evaluate switch function */
1302             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1303             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1304             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1305
1306             fscal            = felec;
1307
1308             fscal            = _mm_and_pd(fscal,cutoff_mask);
1309
1310             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1311
1312             /* Update vectorial force */
1313             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1314             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1315             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1316
1317             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1318             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1319             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1320
1321             }
1322
1323             /**************************
1324              * CALCULATE INTERACTIONS *
1325              **************************/
1326
1327             if (gmx_mm_any_lt(rsq20,rcutoff2))
1328             {
1329
1330             r20              = _mm_mul_pd(rsq20,rinv20);
1331
1332             /* Compute parameters for interactions between i and j atoms */
1333             qq20             = _mm_mul_pd(iq2,jq0);
1334
1335             /* EWALD ELECTROSTATICS */
1336
1337             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1338             ewrt             = _mm_mul_pd(r20,ewtabscale);
1339             ewitab           = _mm_cvttpd_epi32(ewrt);
1340 #ifdef __XOP__
1341             eweps            = _mm_frcz_pd(ewrt);
1342 #else
1343             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1344 #endif
1345             twoeweps         = _mm_add_pd(eweps,eweps);
1346             ewitab           = _mm_slli_epi32(ewitab,2);
1347             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1348             ewtabD           = _mm_setzero_pd();
1349             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1350             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1351             ewtabFn          = _mm_setzero_pd();
1352             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1353             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1354             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1355             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1356             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1357
1358             d                = _mm_sub_pd(r20,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             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1368             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1369
1370             fscal            = felec;
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             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1378             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1379             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1380
1381             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1382             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1383             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1384
1385             }
1386
1387             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1388
1389             /* Inner loop uses 213 flops */
1390         }
1391
1392         /* End of innermost loop */
1393
1394         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1395                                               f+i_coord_offset,fshift+i_shift_offset);
1396
1397         /* Increment number of inner iterations */
1398         inneriter                  += j_index_end - j_index_start;
1399
1400         /* Outer loop uses 18 flops */
1401     }
1402
1403     /* Increment number of outer iterations */
1404     outeriter        += nri;
1405
1406     /* Update outer/inner flops */
1407
1408     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*213);
1409 }