src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  92     real             rswitch_scalar,d_scalar;
  93     __m128d          dummy_mask,cutoff_mask;
  94     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  95     __m128d          one     = _mm_set1_pd(1.0);
  96     __m128d          two     = _mm_set1_pd(2.0);
  97     x                = xx[0];
  98     f                = ff[0];
  99
 100     nri              = nlist->nri;
 101     iinr             = nlist->iinr;
 102     jindex           = nlist->jindex;
 103     jjnr             = nlist->jjnr;
 104     shiftidx         = nlist->shift;
 105     gid              = nlist->gid;
 106     shiftvec         = fr->shift_vec[0];
 107     fshift           = fr->fshift[0];
 108     facel            = _mm_set1_pd(fr->epsfac);
 109     charge           = mdatoms->chargeA;
 110     nvdwtype         = fr->ntype;
 111     vdwparam         = fr->nbfp;
 112     vdwtype          = mdatoms->typeA;
 113
 114     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 115     ewtab            = fr->ic->tabq_coul_FDV0;
 116     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 117     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 122     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 123     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 127     rcutoff_scalar   = fr->rcoulomb;
 128     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 129     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 130
 131     rswitch_scalar   = fr->rcoulomb_switch;
 132     rswitch          = _mm_set1_pd(rswitch_scalar);
 133     /* Setup switch parameters */
 134     d_scalar         = rcutoff_scalar-rswitch_scalar;
 135     d                = _mm_set1_pd(d_scalar);
 136     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 137     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 138     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 139     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 140     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 141     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 142
 143     /* Avoid stupid compiler warnings */
 144     jnrA = jnrB = 0;
 145     j_coord_offsetA = 0;
 146     j_coord_offsetB = 0;
 147
 148     outeriter        = 0;
 149     inneriter        = 0;
 150
 151     /* Start outer loop over neighborlists */
 152     for(iidx=0; iidx<nri; iidx++)
 153     {
 154         /* Load shift vector for this list */
 155         i_shift_offset   = DIM*shiftidx[iidx];
 156
 157         /* Load limits for loop over neighbors */
 158         j_index_start    = jindex[iidx];
 159         j_index_end      = jindex[iidx+1];
 160
 161         /* Get outer coordinate index */
 162         inr              = iinr[iidx];
 163         i_coord_offset   = DIM*inr;
 164
 165         /* Load i particle coords and add shift vector */
 166         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 167                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 168
 169         fix0             = _mm_setzero_pd();
 170         fiy0             = _mm_setzero_pd();
 171         fiz0             = _mm_setzero_pd();
 172         fix1             = _mm_setzero_pd();
 173         fiy1             = _mm_setzero_pd();
 174         fiz1             = _mm_setzero_pd();
 175         fix2             = _mm_setzero_pd();
 176         fiy2             = _mm_setzero_pd();
 177         fiz2             = _mm_setzero_pd();
 178         fix3             = _mm_setzero_pd();
 179         fiy3             = _mm_setzero_pd();
 180         fiz3             = _mm_setzero_pd();
 181
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_pd();
 184         vvdwsum          = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_pd(ix0,jx0);
 202             dy00             = _mm_sub_pd(iy0,jy0);
 203             dz00             = _mm_sub_pd(iz0,jz0);
 204             dx10             = _mm_sub_pd(ix1,jx0);
 205             dy10             = _mm_sub_pd(iy1,jy0);
 206             dz10             = _mm_sub_pd(iz1,jz0);
 207             dx20             = _mm_sub_pd(ix2,jx0);
 208             dy20             = _mm_sub_pd(iy2,jy0);
 209             dz20             = _mm_sub_pd(iz2,jz0);
 210             dx30             = _mm_sub_pd(ix3,jx0);
 211             dy30             = _mm_sub_pd(iy3,jy0);
 212             dz30             = _mm_sub_pd(iz3,jz0);
 213
 214             /* Calculate squared distance and things based on it */
 215             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 216             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 217             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 218             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 219
 220             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 221             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 222             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 223             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 224
 225             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 226             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 227             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 228             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 229
 230             /* Load parameters for j particles */
 231             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 232             vdwjidx0A        = 2*vdwtype[jnrA+0];
 233             vdwjidx0B        = 2*vdwtype[jnrB+0];
 234
 235             fjx0             = _mm_setzero_pd();
 236             fjy0             = _mm_setzero_pd();
 237             fjz0             = _mm_setzero_pd();
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             if (gmx_mm_any_lt(rsq00,rcutoff2))
 244             {
 245
 246             r00              = _mm_mul_pd(rsq00,rinv00);
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 251
 252             /* LENNARD-JONES DISPERSION/REPULSION */
 253
 254             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 255             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 256             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 257             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 258             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 259
 260             d                = _mm_sub_pd(r00,rswitch);
 261             d                = _mm_max_pd(d,_mm_setzero_pd());
 262             d2               = _mm_mul_pd(d,d);
 263             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 264
 265             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 266
 267             /* Evaluate switch function */
 268             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 269             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 270             vvdw             = _mm_mul_pd(vvdw,sw);
 271             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 272
 273             /* Update potential sum for this i atom from the interaction with this j atom. */
 274             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 275             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 276
 277             fscal            = fvdw;
 278
 279             fscal            = _mm_and_pd(fscal,cutoff_mask);
 280
 281             /* Update vectorial force */
 282             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 283             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 284             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 285
 286             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 287             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 288             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 289
 290             }
 291
 292             /**************************
 293              * CALCULATE INTERACTIONS *
 294              **************************/
 295
 296             if (gmx_mm_any_lt(rsq10,rcutoff2))
 297             {
 298
 299             r10              = _mm_mul_pd(rsq10,rinv10);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq10             = _mm_mul_pd(iq1,jq0);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _mm_mul_pd(r10,ewtabscale);
 308             ewitab           = _mm_cvttpd_epi32(ewrt);
 309 #ifdef __XOP__
 310             eweps            = _mm_frcz_pd(ewrt);
 311 #else
 312             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 313 #endif
 314             twoeweps         = _mm_add_pd(eweps,eweps);
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 323             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 324             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 326
 327             d                = _mm_sub_pd(r10,rswitch);
 328             d                = _mm_max_pd(d,_mm_setzero_pd());
 329             d2               = _mm_mul_pd(d,d);
 330             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 331
 332             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 333
 334             /* Evaluate switch function */
 335             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 336             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 337             velec            = _mm_mul_pd(velec,sw);
 338             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 339
 340             /* Update potential sum for this i atom from the interaction with this j atom. */
 341             velec            = _mm_and_pd(velec,cutoff_mask);
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             fscal            = _mm_and_pd(fscal,cutoff_mask);
 347
 348             /* Update vectorial force */
 349             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 350             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 351             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 352
 353             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 354             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 355             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 356
 357             }
 358
 359             /**************************
 360              * CALCULATE INTERACTIONS *
 361              **************************/
 362
 363             if (gmx_mm_any_lt(rsq20,rcutoff2))
 364             {
 365
 366             r20              = _mm_mul_pd(rsq20,rinv20);
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq20             = _mm_mul_pd(iq2,jq0);
 370
 371             /* EWALD ELECTROSTATICS */
 372
 373             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 374             ewrt             = _mm_mul_pd(r20,ewtabscale);
 375             ewitab           = _mm_cvttpd_epi32(ewrt);
 376 #ifdef __XOP__
 377             eweps            = _mm_frcz_pd(ewrt);
 378 #else
 379             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 380 #endif
 381             twoeweps         = _mm_add_pd(eweps,eweps);
 382             ewitab           = _mm_slli_epi32(ewitab,2);
 383             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 384             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 385             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 386             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 387             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 388             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 389             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 390             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 391             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 392             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 393
 394             d                = _mm_sub_pd(r20,rswitch);
 395             d                = _mm_max_pd(d,_mm_setzero_pd());
 396             d2               = _mm_mul_pd(d,d);
 397             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 398
 399             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 400
 401             /* Evaluate switch function */
 402             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 403             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 404             velec            = _mm_mul_pd(velec,sw);
 405             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 406
 407             /* Update potential sum for this i atom from the interaction with this j atom. */
 408             velec            = _mm_and_pd(velec,cutoff_mask);
 409             velecsum         = _mm_add_pd(velecsum,velec);
 410
 411             fscal            = felec;
 412
 413             fscal            = _mm_and_pd(fscal,cutoff_mask);
 414
 415             /* Update vectorial force */
 416             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 417             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 418             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 419
 420             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 421             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 422             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 423
 424             }
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             if (gmx_mm_any_lt(rsq30,rcutoff2))
 431             {
 432
 433             r30              = _mm_mul_pd(rsq30,rinv30);
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             qq30             = _mm_mul_pd(iq3,jq0);
 437
 438             /* EWALD ELECTROSTATICS */
 439
 440             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 441             ewrt             = _mm_mul_pd(r30,ewtabscale);
 442             ewitab           = _mm_cvttpd_epi32(ewrt);
 443 #ifdef __XOP__
 444             eweps            = _mm_frcz_pd(ewrt);
 445 #else
 446             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 447 #endif
 448             twoeweps         = _mm_add_pd(eweps,eweps);
 449             ewitab           = _mm_slli_epi32(ewitab,2);
 450             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 451             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 452             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 453             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 454             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 455             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 456             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 457             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 458             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 459             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 460
 461             d                = _mm_sub_pd(r30,rswitch);
 462             d                = _mm_max_pd(d,_mm_setzero_pd());
 463             d2               = _mm_mul_pd(d,d);
 464             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 465
 466             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 467
 468             /* Evaluate switch function */
 469             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 470             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 471             velec            = _mm_mul_pd(velec,sw);
 472             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 473
 474             /* Update potential sum for this i atom from the interaction with this j atom. */
 475             velec            = _mm_and_pd(velec,cutoff_mask);
 476             velecsum         = _mm_add_pd(velecsum,velec);
 477
 478             fscal            = felec;
 479
 480             fscal            = _mm_and_pd(fscal,cutoff_mask);
 481
 482             /* Update vectorial force */
 483             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 484             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 485             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 486
 487             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 488             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 489             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 490
 491             }
 492
 493             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 494
 495             /* Inner loop uses 269 flops */
 496         }
 497
 498         if(jidx<j_index_end)
 499         {
 500
 501             jnrA             = jjnr[jidx];
 502             j_coord_offsetA  = DIM*jnrA;
 503
 504             /* load j atom coordinates */
 505             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 506                                               &jx0,&jy0,&jz0);
 507
 508             /* Calculate displacement vector */
 509             dx00             = _mm_sub_pd(ix0,jx0);
 510             dy00             = _mm_sub_pd(iy0,jy0);
 511             dz00             = _mm_sub_pd(iz0,jz0);
 512             dx10             = _mm_sub_pd(ix1,jx0);
 513             dy10             = _mm_sub_pd(iy1,jy0);
 514             dz10             = _mm_sub_pd(iz1,jz0);
 515             dx20             = _mm_sub_pd(ix2,jx0);
 516             dy20             = _mm_sub_pd(iy2,jy0);
 517             dz20             = _mm_sub_pd(iz2,jz0);
 518             dx30             = _mm_sub_pd(ix3,jx0);
 519             dy30             = _mm_sub_pd(iy3,jy0);
 520             dz30             = _mm_sub_pd(iz3,jz0);
 521
 522             /* Calculate squared distance and things based on it */
 523             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 524             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 525             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 526             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 527
 528             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 529             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 530             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 531             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 532
 533             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 534             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 535             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 536             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 537
 538             /* Load parameters for j particles */
 539             jq0              = _mm_load_sd(charge+jnrA+0);
 540             vdwjidx0A        = 2*vdwtype[jnrA+0];
 541
 542             fjx0             = _mm_setzero_pd();
 543             fjy0             = _mm_setzero_pd();
 544             fjz0             = _mm_setzero_pd();
 545
 546             /**************************
 547              * CALCULATE INTERACTIONS *
 548              **************************/
 549
 550             if (gmx_mm_any_lt(rsq00,rcutoff2))
 551             {
 552
 553             r00              = _mm_mul_pd(rsq00,rinv00);
 554
 555             /* Compute parameters for interactions between i and j atoms */
 556             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 557
 558             /* LENNARD-JONES DISPERSION/REPULSION */
 559
 560             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 561             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 562             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 563             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 564             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 565
 566             d                = _mm_sub_pd(r00,rswitch);
 567             d                = _mm_max_pd(d,_mm_setzero_pd());
 568             d2               = _mm_mul_pd(d,d);
 569             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 570
 571             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 572
 573             /* Evaluate switch function */
 574             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 575             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 576             vvdw             = _mm_mul_pd(vvdw,sw);
 577             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 578
 579             /* Update potential sum for this i atom from the interaction with this j atom. */
 580             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 581             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 582             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 583
 584             fscal            = fvdw;
 585
 586             fscal            = _mm_and_pd(fscal,cutoff_mask);
 587
 588             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 589
 590             /* Update vectorial force */
 591             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 592             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 593             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 594
 595             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 596             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 597             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 598
 599             }
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             if (gmx_mm_any_lt(rsq10,rcutoff2))
 606             {
 607
 608             r10              = _mm_mul_pd(rsq10,rinv10);
 609
 610             /* Compute parameters for interactions between i and j atoms */
 611             qq10             = _mm_mul_pd(iq1,jq0);
 612
 613             /* EWALD ELECTROSTATICS */
 614
 615             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 616             ewrt             = _mm_mul_pd(r10,ewtabscale);
 617             ewitab           = _mm_cvttpd_epi32(ewrt);
 618 #ifdef __XOP__
 619             eweps            = _mm_frcz_pd(ewrt);
 620 #else
 621             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 622 #endif
 623             twoeweps         = _mm_add_pd(eweps,eweps);
 624             ewitab           = _mm_slli_epi32(ewitab,2);
 625             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 626             ewtabD           = _mm_setzero_pd();
 627             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 628             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 629             ewtabFn          = _mm_setzero_pd();
 630             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 631             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 632             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 633             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 634             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 635
 636             d                = _mm_sub_pd(r10,rswitch);
 637             d                = _mm_max_pd(d,_mm_setzero_pd());
 638             d2               = _mm_mul_pd(d,d);
 639             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 640
 641             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 642
 643             /* Evaluate switch function */
 644             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 645             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
 646             velec            = _mm_mul_pd(velec,sw);
 647             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 648
 649             /* Update potential sum for this i atom from the interaction with this j atom. */
 650             velec            = _mm_and_pd(velec,cutoff_mask);
 651             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 652             velecsum         = _mm_add_pd(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm_and_pd(fscal,cutoff_mask);
 657
 658             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 659
 660             /* Update vectorial force */
 661             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 662             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 663             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 664
 665             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 666             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 667             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 668
 669             }
 670
 671             /**************************
 672              * CALCULATE INTERACTIONS *
 673              **************************/
 674
 675             if (gmx_mm_any_lt(rsq20,rcutoff2))
 676             {
 677
 678             r20              = _mm_mul_pd(rsq20,rinv20);
 679
 680             /* Compute parameters for interactions between i and j atoms */
 681             qq20             = _mm_mul_pd(iq2,jq0);
 682
 683             /* EWALD ELECTROSTATICS */
 684
 685             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 686             ewrt             = _mm_mul_pd(r20,ewtabscale);
 687             ewitab           = _mm_cvttpd_epi32(ewrt);
 688 #ifdef __XOP__
 689             eweps            = _mm_frcz_pd(ewrt);
 690 #else
 691             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 692 #endif
 693             twoeweps         = _mm_add_pd(eweps,eweps);
 694             ewitab           = _mm_slli_epi32(ewitab,2);
 695             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 696             ewtabD           = _mm_setzero_pd();
 697             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 698             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 699             ewtabFn          = _mm_setzero_pd();
 700             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 701             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 702             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 703             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 704             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 705
 706             d                = _mm_sub_pd(r20,rswitch);
 707             d                = _mm_max_pd(d,_mm_setzero_pd());
 708             d2               = _mm_mul_pd(d,d);
 709             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 710
 711             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 712
 713             /* Evaluate switch function */
 714             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 715             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
 716             velec            = _mm_mul_pd(velec,sw);
 717             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 718
 719             /* Update potential sum for this i atom from the interaction with this j atom. */
 720             velec            = _mm_and_pd(velec,cutoff_mask);
 721             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 722             velecsum         = _mm_add_pd(velecsum,velec);
 723
 724             fscal            = felec;
 725
 726             fscal            = _mm_and_pd(fscal,cutoff_mask);
 727
 728             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 729
 730             /* Update vectorial force */
 731             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 732             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 733             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 734
 735             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 736             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 737             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 738
 739             }
 740
 741             /**************************
 742              * CALCULATE INTERACTIONS *
 743              **************************/
 744
 745             if (gmx_mm_any_lt(rsq30,rcutoff2))
 746             {
 747
 748             r30              = _mm_mul_pd(rsq30,rinv30);
 749
 750             /* Compute parameters for interactions between i and j atoms */
 751             qq30             = _mm_mul_pd(iq3,jq0);
 752
 753             /* EWALD ELECTROSTATICS */
 754
 755             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 756             ewrt             = _mm_mul_pd(r30,ewtabscale);
 757             ewitab           = _mm_cvttpd_epi32(ewrt);
 758 #ifdef __XOP__
 759             eweps            = _mm_frcz_pd(ewrt);
 760 #else
 761             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 762 #endif
 763             twoeweps         = _mm_add_pd(eweps,eweps);
 764             ewitab           = _mm_slli_epi32(ewitab,2);
 765             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 766             ewtabD           = _mm_setzero_pd();
 767             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 768             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 769             ewtabFn          = _mm_setzero_pd();
 770             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 771             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 772             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 773             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 774             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 775
 776             d                = _mm_sub_pd(r30,rswitch);
 777             d                = _mm_max_pd(d,_mm_setzero_pd());
 778             d2               = _mm_mul_pd(d,d);
 779             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
 780
 781             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
 782
 783             /* Evaluate switch function */
 784             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 785             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
 786             velec            = _mm_mul_pd(velec,sw);
 787             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 788
 789             /* Update potential sum for this i atom from the interaction with this j atom. */
 790             velec            = _mm_and_pd(velec,cutoff_mask);
 791             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 792             velecsum         = _mm_add_pd(velecsum,velec);
 793
 794             fscal            = felec;
 795
 796             fscal            = _mm_and_pd(fscal,cutoff_mask);
 797
 798             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 799
 800             /* Update vectorial force */
 801             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 802             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 803             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 804
 805             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 806             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 807             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 808
 809             }
 810
 811             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 812
 813             /* Inner loop uses 269 flops */
 814         }
 815
 816         /* End of innermost loop */
 817
 818         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 819                                               f+i_coord_offset,fshift+i_shift_offset);
 820
 821         ggid                        = gid[iidx];
 822         /* Update potential energies */
 823         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 824         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 825
 826         /* Increment number of inner iterations */
 827         inneriter                  += j_index_end - j_index_start;
 828
 829         /* Outer loop uses 26 flops */
 830     }
 831
 832     /* Increment number of outer iterations */
 833     outeriter        += nri;
 834
 835     /* Update outer/inner flops */
 836
 837     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*269);
 838 }
 839 /*
 840  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 841  * Electrostatics interaction: Ewald
 842  * VdW interaction:            LennardJones
 843  * Geometry:                   Water4-Particle
 844  * Calculate force/pot:        Force
 845  */
 846 void
 847 nb_kernel_ElecEwSw_VdwLJSw_GeomW4P1_F_avx_128_fma_double
 848                     (t_nblist * gmx_restrict                nlist,
 849                      rvec * gmx_restrict                    xx,
 850                      rvec * gmx_restrict                    ff,
 851                      t_forcerec * gmx_restrict              fr,
 852                      t_mdatoms * gmx_restrict               mdatoms,
 853                      nb_kernel_data_t * gmx_restrict        kernel_data,
 854                      t_nrnb * gmx_restrict                  nrnb)
 855 {
 856     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 857      * just 0 for non-waters.
 858      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 859      * jnr indices corresponding to data put in the four positions in the SIMD register.
 860      */
 861     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 862     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 863     int              jnrA,jnrB;
 864     int              j_coord_offsetA,j_coord_offsetB;
 865     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 866     real             rcutoff_scalar;
 867     real             *shiftvec,*fshift,*x,*f;
 868     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 869     int              vdwioffset0;
 870     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 871     int              vdwioffset1;
 872     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 873     int              vdwioffset2;
 874     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 875     int              vdwioffset3;
 876     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 877     int              vdwjidx0A,vdwjidx0B;
 878     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 879     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 880     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 881     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 882     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 883     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 884     real             *charge;
 885     int              nvdwtype;
 886     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 887     int              *vdwtype;
 888     real             *vdwparam;
 889     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 890     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 891     __m128i          ewitab;
 892     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 893     real             *ewtab;
 894     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 895     real             rswitch_scalar,d_scalar;
 896     __m128d          dummy_mask,cutoff_mask;
 897     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 898     __m128d          one     = _mm_set1_pd(1.0);
 899     __m128d          two     = _mm_set1_pd(2.0);
 900     x                = xx[0];
 901     f                = ff[0];
 902
 903     nri              = nlist->nri;
 904     iinr             = nlist->iinr;
 905     jindex           = nlist->jindex;
 906     jjnr             = nlist->jjnr;
 907     shiftidx         = nlist->shift;
 908     gid              = nlist->gid;
 909     shiftvec         = fr->shift_vec[0];
 910     fshift           = fr->fshift[0];
 911     facel            = _mm_set1_pd(fr->epsfac);
 912     charge           = mdatoms->chargeA;
 913     nvdwtype         = fr->ntype;
 914     vdwparam         = fr->nbfp;
 915     vdwtype          = mdatoms->typeA;
 916
 917     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 918     ewtab            = fr->ic->tabq_coul_FDV0;
 919     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 920     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 921
 922     /* Setup water-specific parameters */
 923     inr              = nlist->iinr[0];
 924     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 925     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 926     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 927     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 928
 929     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 930     rcutoff_scalar   = fr->rcoulomb;
 931     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 932     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 933
 934     rswitch_scalar   = fr->rcoulomb_switch;
 935     rswitch          = _mm_set1_pd(rswitch_scalar);
 936     /* Setup switch parameters */
 937     d_scalar         = rcutoff_scalar-rswitch_scalar;
 938     d                = _mm_set1_pd(d_scalar);
 939     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 940     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 941     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 942     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 943     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 944     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 945
 946     /* Avoid stupid compiler warnings */
 947     jnrA = jnrB = 0;
 948     j_coord_offsetA = 0;
 949     j_coord_offsetB = 0;
 950
 951     outeriter        = 0;
 952     inneriter        = 0;
 953
 954     /* Start outer loop over neighborlists */
 955     for(iidx=0; iidx<nri; iidx++)
 956     {
 957         /* Load shift vector for this list */
 958         i_shift_offset   = DIM*shiftidx[iidx];
 959
 960         /* Load limits for loop over neighbors */
 961         j_index_start    = jindex[iidx];
 962         j_index_end      = jindex[iidx+1];
 963
 964         /* Get outer coordinate index */
 965         inr              = iinr[iidx];
 966         i_coord_offset   = DIM*inr;
 967
 968         /* Load i particle coords and add shift vector */
 969         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 970                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 971
 972         fix0             = _mm_setzero_pd();
 973         fiy0             = _mm_setzero_pd();
 974         fiz0             = _mm_setzero_pd();
 975         fix1             = _mm_setzero_pd();
 976         fiy1             = _mm_setzero_pd();
 977         fiz1             = _mm_setzero_pd();
 978         fix2             = _mm_setzero_pd();
 979         fiy2             = _mm_setzero_pd();
 980         fiz2             = _mm_setzero_pd();
 981         fix3             = _mm_setzero_pd();
 982         fiy3             = _mm_setzero_pd();
 983         fiz3             = _mm_setzero_pd();
 984
 985         /* Start inner kernel loop */
 986         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 987         {
 988
 989             /* Get j neighbor index, and coordinate index */
 990             jnrA             = jjnr[jidx];
 991             jnrB             = jjnr[jidx+1];
 992             j_coord_offsetA  = DIM*jnrA;
 993             j_coord_offsetB  = DIM*jnrB;
 994
 995             /* load j atom coordinates */
 996             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 997                                               &jx0,&jy0,&jz0);
 998
 999             /* Calculate displacement vector */
1000             dx00             = _mm_sub_pd(ix0,jx0);
1001             dy00             = _mm_sub_pd(iy0,jy0);
1002             dz00             = _mm_sub_pd(iz0,jz0);
1003             dx10             = _mm_sub_pd(ix1,jx0);
1004             dy10             = _mm_sub_pd(iy1,jy0);
1005             dz10             = _mm_sub_pd(iz1,jz0);
1006             dx20             = _mm_sub_pd(ix2,jx0);
1007             dy20             = _mm_sub_pd(iy2,jy0);
1008             dz20             = _mm_sub_pd(iz2,jz0);
1009             dx30             = _mm_sub_pd(ix3,jx0);
1010             dy30             = _mm_sub_pd(iy3,jy0);
1011             dz30             = _mm_sub_pd(iz3,jz0);
1012
1013             /* Calculate squared distance and things based on it */
1014             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1015             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1016             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1017             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1018
1019             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1020             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1021             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1022             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1023
1024             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1025             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1026             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1027             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1028
1029             /* Load parameters for j particles */
1030             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
1031             vdwjidx0A        = 2*vdwtype[jnrA+0];
1032             vdwjidx0B        = 2*vdwtype[jnrB+0];
1033
1034             fjx0             = _mm_setzero_pd();
1035             fjy0             = _mm_setzero_pd();
1036             fjz0             = _mm_setzero_pd();
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             if (gmx_mm_any_lt(rsq00,rcutoff2))
1043             {
1044
1045             r00              = _mm_mul_pd(rsq00,rinv00);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
1049                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
1050
1051             /* LENNARD-JONES DISPERSION/REPULSION */
1052
1053             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1054             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1055             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1056             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1057             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1058
1059             d                = _mm_sub_pd(r00,rswitch);
1060             d                = _mm_max_pd(d,_mm_setzero_pd());
1061             d2               = _mm_mul_pd(d,d);
1062             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1063
1064             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1065
1066             /* Evaluate switch function */
1067             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1068             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1069             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1070
1071             fscal            = fvdw;
1072
1073             fscal            = _mm_and_pd(fscal,cutoff_mask);
1074
1075             /* Update vectorial force */
1076             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1077             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1078             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1079
1080             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1081             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1082             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1083
1084             }
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             if (gmx_mm_any_lt(rsq10,rcutoff2))
1091             {
1092
1093             r10              = _mm_mul_pd(rsq10,rinv10);
1094
1095             /* Compute parameters for interactions between i and j atoms */
1096             qq10             = _mm_mul_pd(iq1,jq0);
1097
1098             /* EWALD ELECTROSTATICS */
1099
1100             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1101             ewrt             = _mm_mul_pd(r10,ewtabscale);
1102             ewitab           = _mm_cvttpd_epi32(ewrt);
1103 #ifdef __XOP__
1104             eweps            = _mm_frcz_pd(ewrt);
1105 #else
1106             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1107 #endif
1108             twoeweps         = _mm_add_pd(eweps,eweps);
1109             ewitab           = _mm_slli_epi32(ewitab,2);
1110             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1111             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1112             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1113             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1114             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1115             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1116             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1117             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1118             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1119             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1120
1121             d                = _mm_sub_pd(r10,rswitch);
1122             d                = _mm_max_pd(d,_mm_setzero_pd());
1123             d2               = _mm_mul_pd(d,d);
1124             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1125
1126             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1127
1128             /* Evaluate switch function */
1129             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1130             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1131             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1132
1133             fscal            = felec;
1134
1135             fscal            = _mm_and_pd(fscal,cutoff_mask);
1136
1137             /* Update vectorial force */
1138             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1139             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1140             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1141
1142             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1143             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1144             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1145
1146             }
1147
1148             /**************************
1149              * CALCULATE INTERACTIONS *
1150              **************************/
1151
1152             if (gmx_mm_any_lt(rsq20,rcutoff2))
1153             {
1154
1155             r20              = _mm_mul_pd(rsq20,rinv20);
1156
1157             /* Compute parameters for interactions between i and j atoms */
1158             qq20             = _mm_mul_pd(iq2,jq0);
1159
1160             /* EWALD ELECTROSTATICS */
1161
1162             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1163             ewrt             = _mm_mul_pd(r20,ewtabscale);
1164             ewitab           = _mm_cvttpd_epi32(ewrt);
1165 #ifdef __XOP__
1166             eweps            = _mm_frcz_pd(ewrt);
1167 #else
1168             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1169 #endif
1170             twoeweps         = _mm_add_pd(eweps,eweps);
1171             ewitab           = _mm_slli_epi32(ewitab,2);
1172             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1173             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1174             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1175             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1176             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1177             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1178             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1179             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1180             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1181             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1182
1183             d                = _mm_sub_pd(r20,rswitch);
1184             d                = _mm_max_pd(d,_mm_setzero_pd());
1185             d2               = _mm_mul_pd(d,d);
1186             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1187
1188             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1189
1190             /* Evaluate switch function */
1191             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1192             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1193             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1194
1195             fscal            = felec;
1196
1197             fscal            = _mm_and_pd(fscal,cutoff_mask);
1198
1199             /* Update vectorial force */
1200             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1201             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1202             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1203
1204             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1205             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1206             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1207
1208             }
1209
1210             /**************************
1211              * CALCULATE INTERACTIONS *
1212              **************************/
1213
1214             if (gmx_mm_any_lt(rsq30,rcutoff2))
1215             {
1216
1217             r30              = _mm_mul_pd(rsq30,rinv30);
1218
1219             /* Compute parameters for interactions between i and j atoms */
1220             qq30             = _mm_mul_pd(iq3,jq0);
1221
1222             /* EWALD ELECTROSTATICS */
1223
1224             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1225             ewrt             = _mm_mul_pd(r30,ewtabscale);
1226             ewitab           = _mm_cvttpd_epi32(ewrt);
1227 #ifdef __XOP__
1228             eweps            = _mm_frcz_pd(ewrt);
1229 #else
1230             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1231 #endif
1232             twoeweps         = _mm_add_pd(eweps,eweps);
1233             ewitab           = _mm_slli_epi32(ewitab,2);
1234             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1235             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
1236             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1237             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1238             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
1239             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1240             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1241             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1242             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1243             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1244
1245             d                = _mm_sub_pd(r30,rswitch);
1246             d                = _mm_max_pd(d,_mm_setzero_pd());
1247             d2               = _mm_mul_pd(d,d);
1248             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1249
1250             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1251
1252             /* Evaluate switch function */
1253             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1254             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1255             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1256
1257             fscal            = felec;
1258
1259             fscal            = _mm_and_pd(fscal,cutoff_mask);
1260
1261             /* Update vectorial force */
1262             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1263             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1264             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1265
1266             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1267             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1268             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1269
1270             }
1271
1272             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1273
1274             /* Inner loop uses 257 flops */
1275         }
1276
1277         if(jidx<j_index_end)
1278         {
1279
1280             jnrA             = jjnr[jidx];
1281             j_coord_offsetA  = DIM*jnrA;
1282
1283             /* load j atom coordinates */
1284             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1285                                               &jx0,&jy0,&jz0);
1286
1287             /* Calculate displacement vector */
1288             dx00             = _mm_sub_pd(ix0,jx0);
1289             dy00             = _mm_sub_pd(iy0,jy0);
1290             dz00             = _mm_sub_pd(iz0,jz0);
1291             dx10             = _mm_sub_pd(ix1,jx0);
1292             dy10             = _mm_sub_pd(iy1,jy0);
1293             dz10             = _mm_sub_pd(iz1,jz0);
1294             dx20             = _mm_sub_pd(ix2,jx0);
1295             dy20             = _mm_sub_pd(iy2,jy0);
1296             dz20             = _mm_sub_pd(iz2,jz0);
1297             dx30             = _mm_sub_pd(ix3,jx0);
1298             dy30             = _mm_sub_pd(iy3,jy0);
1299             dz30             = _mm_sub_pd(iz3,jz0);
1300
1301             /* Calculate squared distance and things based on it */
1302             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1303             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1304             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1305             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1306
1307             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1308             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1309             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1310             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1311
1312             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1313             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1314             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1315             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1316
1317             /* Load parameters for j particles */
1318             jq0              = _mm_load_sd(charge+jnrA+0);
1319             vdwjidx0A        = 2*vdwtype[jnrA+0];
1320
1321             fjx0             = _mm_setzero_pd();
1322             fjy0             = _mm_setzero_pd();
1323             fjz0             = _mm_setzero_pd();
1324
1325             /**************************
1326              * CALCULATE INTERACTIONS *
1327              **************************/
1328
1329             if (gmx_mm_any_lt(rsq00,rcutoff2))
1330             {
1331
1332             r00              = _mm_mul_pd(rsq00,rinv00);
1333
1334             /* Compute parameters for interactions between i and j atoms */
1335             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1336
1337             /* LENNARD-JONES DISPERSION/REPULSION */
1338
1339             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1340             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
1341             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
1342             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
1343             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
1344
1345             d                = _mm_sub_pd(r00,rswitch);
1346             d                = _mm_max_pd(d,_mm_setzero_pd());
1347             d2               = _mm_mul_pd(d,d);
1348             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1349
1350             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1351
1352             /* Evaluate switch function */
1353             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1354             fvdw             = _mm_msub_pd( fvdw,sw , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
1355             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1356
1357             fscal            = fvdw;
1358
1359             fscal            = _mm_and_pd(fscal,cutoff_mask);
1360
1361             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1362
1363             /* Update vectorial force */
1364             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1365             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1366             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1367
1368             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1369             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1370             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1371
1372             }
1373
1374             /**************************
1375              * CALCULATE INTERACTIONS *
1376              **************************/
1377
1378             if (gmx_mm_any_lt(rsq10,rcutoff2))
1379             {
1380
1381             r10              = _mm_mul_pd(rsq10,rinv10);
1382
1383             /* Compute parameters for interactions between i and j atoms */
1384             qq10             = _mm_mul_pd(iq1,jq0);
1385
1386             /* EWALD ELECTROSTATICS */
1387
1388             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1389             ewrt             = _mm_mul_pd(r10,ewtabscale);
1390             ewitab           = _mm_cvttpd_epi32(ewrt);
1391 #ifdef __XOP__
1392             eweps            = _mm_frcz_pd(ewrt);
1393 #else
1394             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1395 #endif
1396             twoeweps         = _mm_add_pd(eweps,eweps);
1397             ewitab           = _mm_slli_epi32(ewitab,2);
1398             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1399             ewtabD           = _mm_setzero_pd();
1400             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1401             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1402             ewtabFn          = _mm_setzero_pd();
1403             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1404             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1405             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1406             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
1407             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1408
1409             d                = _mm_sub_pd(r10,rswitch);
1410             d                = _mm_max_pd(d,_mm_setzero_pd());
1411             d2               = _mm_mul_pd(d,d);
1412             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1413
1414             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1415
1416             /* Evaluate switch function */
1417             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1418             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv10,_mm_mul_pd(velec,dsw)) );
1419             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1420
1421             fscal            = felec;
1422
1423             fscal            = _mm_and_pd(fscal,cutoff_mask);
1424
1425             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1426
1427             /* Update vectorial force */
1428             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1429             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1430             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1431
1432             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1433             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1434             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1435
1436             }
1437
1438             /**************************
1439              * CALCULATE INTERACTIONS *
1440              **************************/
1441
1442             if (gmx_mm_any_lt(rsq20,rcutoff2))
1443             {
1444
1445             r20              = _mm_mul_pd(rsq20,rinv20);
1446
1447             /* Compute parameters for interactions between i and j atoms */
1448             qq20             = _mm_mul_pd(iq2,jq0);
1449
1450             /* EWALD ELECTROSTATICS */
1451
1452             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1453             ewrt             = _mm_mul_pd(r20,ewtabscale);
1454             ewitab           = _mm_cvttpd_epi32(ewrt);
1455 #ifdef __XOP__
1456             eweps            = _mm_frcz_pd(ewrt);
1457 #else
1458             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1459 #endif
1460             twoeweps         = _mm_add_pd(eweps,eweps);
1461             ewitab           = _mm_slli_epi32(ewitab,2);
1462             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1463             ewtabD           = _mm_setzero_pd();
1464             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1465             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1466             ewtabFn          = _mm_setzero_pd();
1467             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1468             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1469             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1470             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
1471             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1472
1473             d                = _mm_sub_pd(r20,rswitch);
1474             d                = _mm_max_pd(d,_mm_setzero_pd());
1475             d2               = _mm_mul_pd(d,d);
1476             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1477
1478             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1479
1480             /* Evaluate switch function */
1481             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1482             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv20,_mm_mul_pd(velec,dsw)) );
1483             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1484
1485             fscal            = felec;
1486
1487             fscal            = _mm_and_pd(fscal,cutoff_mask);
1488
1489             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1490
1491             /* Update vectorial force */
1492             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1493             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1494             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1495
1496             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1497             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1498             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1499
1500             }
1501
1502             /**************************
1503              * CALCULATE INTERACTIONS *
1504              **************************/
1505
1506             if (gmx_mm_any_lt(rsq30,rcutoff2))
1507             {
1508
1509             r30              = _mm_mul_pd(rsq30,rinv30);
1510
1511             /* Compute parameters for interactions between i and j atoms */
1512             qq30             = _mm_mul_pd(iq3,jq0);
1513
1514             /* EWALD ELECTROSTATICS */
1515
1516             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1517             ewrt             = _mm_mul_pd(r30,ewtabscale);
1518             ewitab           = _mm_cvttpd_epi32(ewrt);
1519 #ifdef __XOP__
1520             eweps            = _mm_frcz_pd(ewrt);
1521 #else
1522             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1523 #endif
1524             twoeweps         = _mm_add_pd(eweps,eweps);
1525             ewitab           = _mm_slli_epi32(ewitab,2);
1526             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
1527             ewtabD           = _mm_setzero_pd();
1528             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
1529             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
1530             ewtabFn          = _mm_setzero_pd();
1531             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
1532             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
1533             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
1534             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
1535             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1536
1537             d                = _mm_sub_pd(r30,rswitch);
1538             d                = _mm_max_pd(d,_mm_setzero_pd());
1539             d2               = _mm_mul_pd(d,d);
1540             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_macc_pd(d,_mm_macc_pd(d,swV5,swV4),swV3))));
1541
1542             dsw              = _mm_mul_pd(d2,_mm_macc_pd(d,_mm_macc_pd(d,swF4,swF3),swF2));
1543
1544             /* Evaluate switch function */
1545             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
1546             felec            = _mm_msub_pd( felec,sw , _mm_mul_pd(rinv30,_mm_mul_pd(velec,dsw)) );
1547             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1548
1549             fscal            = felec;
1550
1551             fscal            = _mm_and_pd(fscal,cutoff_mask);
1552
1553             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1554
1555             /* Update vectorial force */
1556             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1557             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1558             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1559
1560             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1561             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1562             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1563
1564             }
1565
1566             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1567
1568             /* Inner loop uses 257 flops */
1569         }
1570
1571         /* End of innermost loop */
1572
1573         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1574                                               f+i_coord_offset,fshift+i_shift_offset);
1575
1576         /* Increment number of inner iterations */
1577         inneriter                  += j_index_end - j_index_start;
1578
1579         /* Outer loop uses 24 flops */
1580     }
1581
1582     /* Increment number of outer iterations */
1583     outeriter        += nri;
1584
1585     /* Update outer/inner flops */
1586
1587     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*257);
1588 }