src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_sse4_1_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 sse4_1_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_sse4_1_double.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_sse4_1_double
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_sse4_1_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              vdwioffset1;
  80     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  81     int              vdwioffset2;
  82     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  83     int              vdwioffset3;
  84     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  85     int              vdwjidx0A,vdwjidx0B;
  86     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  87     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  88     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  89     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  90     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  91     real             *charge;
  92     __m128i          ewitab;
  93     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     real             *ewtab;
  95     __m128d          dummy_mask,cutoff_mask;
  96     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  97     __m128d          one     = _mm_set1_pd(1.0);
  98     __m128d          two     = _mm_set1_pd(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_pd(fr->ic->epsfac);
 111     charge           = mdatoms->chargeA;
 112
 113     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 114     ewtab            = fr->ic->tabq_coul_FDV0;
 115     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 116     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 117
 118     /* Setup water-specific parameters */
 119     inr              = nlist->iinr[0];
 120     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 121     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 122     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 123
 124     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 125     rcutoff_scalar   = fr->ic->rcoulomb;
 126     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 127     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 128
 129     /* Avoid stupid compiler warnings */
 130     jnrA = jnrB = 0;
 131     j_coord_offsetA = 0;
 132     j_coord_offsetB = 0;
 133
 134     outeriter        = 0;
 135     inneriter        = 0;
 136
 137     /* Start outer loop over neighborlists */
 138     for(iidx=0; iidx<nri; iidx++)
 139     {
 140         /* Load shift vector for this list */
 141         i_shift_offset   = DIM*shiftidx[iidx];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 153                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 154
 155         fix1             = _mm_setzero_pd();
 156         fiy1             = _mm_setzero_pd();
 157         fiz1             = _mm_setzero_pd();
 158         fix2             = _mm_setzero_pd();
 159         fiy2             = _mm_setzero_pd();
 160         fiz2             = _mm_setzero_pd();
 161         fix3             = _mm_setzero_pd();
 162         fiy3             = _mm_setzero_pd();
 163         fiz3             = _mm_setzero_pd();
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx10             = _mm_sub_pd(ix1,jx0);
 184             dy10             = _mm_sub_pd(iy1,jy0);
 185             dz10             = _mm_sub_pd(iz1,jz0);
 186             dx20             = _mm_sub_pd(ix2,jx0);
 187             dy20             = _mm_sub_pd(iy2,jy0);
 188             dz20             = _mm_sub_pd(iz2,jz0);
 189             dx30             = _mm_sub_pd(ix3,jx0);
 190             dy30             = _mm_sub_pd(iy3,jy0);
 191             dz30             = _mm_sub_pd(iz3,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 195             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 196             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 197
 198             rinv10           = sse41_invsqrt_d(rsq10);
 199             rinv20           = sse41_invsqrt_d(rsq20);
 200             rinv30           = sse41_invsqrt_d(rsq30);
 201
 202             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 203             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 204             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 208
 209             fjx0             = _mm_setzero_pd();
 210             fjy0             = _mm_setzero_pd();
 211             fjz0             = _mm_setzero_pd();
 212
 213             /**************************
 214              * CALCULATE INTERACTIONS *
 215              **************************/
 216
 217             if (gmx_mm_any_lt(rsq10,rcutoff2))
 218             {
 219
 220             r10              = _mm_mul_pd(rsq10,rinv10);
 221
 222             /* Compute parameters for interactions between i and j atoms */
 223             qq10             = _mm_mul_pd(iq1,jq0);
 224
 225             /* EWALD ELECTROSTATICS */
 226
 227             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 228             ewrt             = _mm_mul_pd(r10,ewtabscale);
 229             ewitab           = _mm_cvttpd_epi32(ewrt);
 230             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 231             ewitab           = _mm_slli_epi32(ewitab,2);
 232             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 233             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 234             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 235             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 236             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 237             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 238             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 239             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 240             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 241             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 242
 243             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velec            = _mm_and_pd(velec,cutoff_mask);
 247             velecsum         = _mm_add_pd(velecsum,velec);
 248
 249             fscal            = felec;
 250
 251             fscal            = _mm_and_pd(fscal,cutoff_mask);
 252
 253             /* Calculate temporary vectorial force */
 254             tx               = _mm_mul_pd(fscal,dx10);
 255             ty               = _mm_mul_pd(fscal,dy10);
 256             tz               = _mm_mul_pd(fscal,dz10);
 257
 258             /* Update vectorial force */
 259             fix1             = _mm_add_pd(fix1,tx);
 260             fiy1             = _mm_add_pd(fiy1,ty);
 261             fiz1             = _mm_add_pd(fiz1,tz);
 262
 263             fjx0             = _mm_add_pd(fjx0,tx);
 264             fjy0             = _mm_add_pd(fjy0,ty);
 265             fjz0             = _mm_add_pd(fjz0,tz);
 266
 267             }
 268
 269             /**************************
 270              * CALCULATE INTERACTIONS *
 271              **************************/
 272
 273             if (gmx_mm_any_lt(rsq20,rcutoff2))
 274             {
 275
 276             r20              = _mm_mul_pd(rsq20,rinv20);
 277
 278             /* Compute parameters for interactions between i and j atoms */
 279             qq20             = _mm_mul_pd(iq2,jq0);
 280
 281             /* EWALD ELECTROSTATICS */
 282
 283             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 284             ewrt             = _mm_mul_pd(r20,ewtabscale);
 285             ewitab           = _mm_cvttpd_epi32(ewrt);
 286             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 287             ewitab           = _mm_slli_epi32(ewitab,2);
 288             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 289             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 290             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 291             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 292             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 293             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 294             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 295             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 296             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 297             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 298
 299             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velec            = _mm_and_pd(velec,cutoff_mask);
 303             velecsum         = _mm_add_pd(velecsum,velec);
 304
 305             fscal            = felec;
 306
 307             fscal            = _mm_and_pd(fscal,cutoff_mask);
 308
 309             /* Calculate temporary vectorial force */
 310             tx               = _mm_mul_pd(fscal,dx20);
 311             ty               = _mm_mul_pd(fscal,dy20);
 312             tz               = _mm_mul_pd(fscal,dz20);
 313
 314             /* Update vectorial force */
 315             fix2             = _mm_add_pd(fix2,tx);
 316             fiy2             = _mm_add_pd(fiy2,ty);
 317             fiz2             = _mm_add_pd(fiz2,tz);
 318
 319             fjx0             = _mm_add_pd(fjx0,tx);
 320             fjy0             = _mm_add_pd(fjy0,ty);
 321             fjz0             = _mm_add_pd(fjz0,tz);
 322
 323             }
 324
 325             /**************************
 326              * CALCULATE INTERACTIONS *
 327              **************************/
 328
 329             if (gmx_mm_any_lt(rsq30,rcutoff2))
 330             {
 331
 332             r30              = _mm_mul_pd(rsq30,rinv30);
 333
 334             /* Compute parameters for interactions between i and j atoms */
 335             qq30             = _mm_mul_pd(iq3,jq0);
 336
 337             /* EWALD ELECTROSTATICS */
 338
 339             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 340             ewrt             = _mm_mul_pd(r30,ewtabscale);
 341             ewitab           = _mm_cvttpd_epi32(ewrt);
 342             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 343             ewitab           = _mm_slli_epi32(ewitab,2);
 344             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 345             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 346             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 347             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 348             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 349             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 350             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 351             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 352             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 353             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 354
 355             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm_and_pd(velec,cutoff_mask);
 359             velecsum         = _mm_add_pd(velecsum,velec);
 360
 361             fscal            = felec;
 362
 363             fscal            = _mm_and_pd(fscal,cutoff_mask);
 364
 365             /* Calculate temporary vectorial force */
 366             tx               = _mm_mul_pd(fscal,dx30);
 367             ty               = _mm_mul_pd(fscal,dy30);
 368             tz               = _mm_mul_pd(fscal,dz30);
 369
 370             /* Update vectorial force */
 371             fix3             = _mm_add_pd(fix3,tx);
 372             fiy3             = _mm_add_pd(fiy3,ty);
 373             fiz3             = _mm_add_pd(fiz3,tz);
 374
 375             fjx0             = _mm_add_pd(fjx0,tx);
 376             fjy0             = _mm_add_pd(fjy0,ty);
 377             fjz0             = _mm_add_pd(fjz0,tz);
 378
 379             }
 380
 381             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 382
 383             /* Inner loop uses 141 flops */
 384         }
 385
 386         if(jidx<j_index_end)
 387         {
 388
 389             jnrA             = jjnr[jidx];
 390             j_coord_offsetA  = DIM*jnrA;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 394                                               &jx0,&jy0,&jz0);
 395
 396             /* Calculate displacement vector */
 397             dx10             = _mm_sub_pd(ix1,jx0);
 398             dy10             = _mm_sub_pd(iy1,jy0);
 399             dz10             = _mm_sub_pd(iz1,jz0);
 400             dx20             = _mm_sub_pd(ix2,jx0);
 401             dy20             = _mm_sub_pd(iy2,jy0);
 402             dz20             = _mm_sub_pd(iz2,jz0);
 403             dx30             = _mm_sub_pd(ix3,jx0);
 404             dy30             = _mm_sub_pd(iy3,jy0);
 405             dz30             = _mm_sub_pd(iz3,jz0);
 406
 407             /* Calculate squared distance and things based on it */
 408             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 409             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 410             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 411
 412             rinv10           = sse41_invsqrt_d(rsq10);
 413             rinv20           = sse41_invsqrt_d(rsq20);
 414             rinv30           = sse41_invsqrt_d(rsq30);
 415
 416             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 417             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 418             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 419
 420             /* Load parameters for j particles */
 421             jq0              = _mm_load_sd(charge+jnrA+0);
 422
 423             fjx0             = _mm_setzero_pd();
 424             fjy0             = _mm_setzero_pd();
 425             fjz0             = _mm_setzero_pd();
 426
 427             /**************************
 428              * CALCULATE INTERACTIONS *
 429              **************************/
 430
 431             if (gmx_mm_any_lt(rsq10,rcutoff2))
 432             {
 433
 434             r10              = _mm_mul_pd(rsq10,rinv10);
 435
 436             /* Compute parameters for interactions between i and j atoms */
 437             qq10             = _mm_mul_pd(iq1,jq0);
 438
 439             /* EWALD ELECTROSTATICS */
 440
 441             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 442             ewrt             = _mm_mul_pd(r10,ewtabscale);
 443             ewitab           = _mm_cvttpd_epi32(ewrt);
 444             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 445             ewitab           = _mm_slli_epi32(ewitab,2);
 446             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 447             ewtabD           = _mm_setzero_pd();
 448             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 449             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 450             ewtabFn          = _mm_setzero_pd();
 451             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 452             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 453             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 454             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 455             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 456
 457             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 458
 459             /* Update potential sum for this i atom from the interaction with this j atom. */
 460             velec            = _mm_and_pd(velec,cutoff_mask);
 461             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 462             velecsum         = _mm_add_pd(velecsum,velec);
 463
 464             fscal            = felec;
 465
 466             fscal            = _mm_and_pd(fscal,cutoff_mask);
 467
 468             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 469
 470             /* Calculate temporary vectorial force */
 471             tx               = _mm_mul_pd(fscal,dx10);
 472             ty               = _mm_mul_pd(fscal,dy10);
 473             tz               = _mm_mul_pd(fscal,dz10);
 474
 475             /* Update vectorial force */
 476             fix1             = _mm_add_pd(fix1,tx);
 477             fiy1             = _mm_add_pd(fiy1,ty);
 478             fiz1             = _mm_add_pd(fiz1,tz);
 479
 480             fjx0             = _mm_add_pd(fjx0,tx);
 481             fjy0             = _mm_add_pd(fjy0,ty);
 482             fjz0             = _mm_add_pd(fjz0,tz);
 483
 484             }
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             if (gmx_mm_any_lt(rsq20,rcutoff2))
 491             {
 492
 493             r20              = _mm_mul_pd(rsq20,rinv20);
 494
 495             /* Compute parameters for interactions between i and j atoms */
 496             qq20             = _mm_mul_pd(iq2,jq0);
 497
 498             /* EWALD ELECTROSTATICS */
 499
 500             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 501             ewrt             = _mm_mul_pd(r20,ewtabscale);
 502             ewitab           = _mm_cvttpd_epi32(ewrt);
 503             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 504             ewitab           = _mm_slli_epi32(ewitab,2);
 505             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 506             ewtabD           = _mm_setzero_pd();
 507             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 508             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 509             ewtabFn          = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 511             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 512             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 513             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 514             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 515
 516             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 517
 518             /* Update potential sum for this i atom from the interaction with this j atom. */
 519             velec            = _mm_and_pd(velec,cutoff_mask);
 520             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 521             velecsum         = _mm_add_pd(velecsum,velec);
 522
 523             fscal            = felec;
 524
 525             fscal            = _mm_and_pd(fscal,cutoff_mask);
 526
 527             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 528
 529             /* Calculate temporary vectorial force */
 530             tx               = _mm_mul_pd(fscal,dx20);
 531             ty               = _mm_mul_pd(fscal,dy20);
 532             tz               = _mm_mul_pd(fscal,dz20);
 533
 534             /* Update vectorial force */
 535             fix2             = _mm_add_pd(fix2,tx);
 536             fiy2             = _mm_add_pd(fiy2,ty);
 537             fiz2             = _mm_add_pd(fiz2,tz);
 538
 539             fjx0             = _mm_add_pd(fjx0,tx);
 540             fjy0             = _mm_add_pd(fjy0,ty);
 541             fjz0             = _mm_add_pd(fjz0,tz);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq30,rcutoff2))
 550             {
 551
 552             r30              = _mm_mul_pd(rsq30,rinv30);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq30             = _mm_mul_pd(iq3,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 560             ewrt             = _mm_mul_pd(r30,ewtabscale);
 561             ewitab           = _mm_cvttpd_epi32(ewrt);
 562             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 563             ewitab           = _mm_slli_epi32(ewitab,2);
 564             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 565             ewtabD           = _mm_setzero_pd();
 566             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 567             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 568             ewtabFn          = _mm_setzero_pd();
 569             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 570             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 571             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 572             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 573             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 574
 575             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 576
 577             /* Update potential sum for this i atom from the interaction with this j atom. */
 578             velec            = _mm_and_pd(velec,cutoff_mask);
 579             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 580             velecsum         = _mm_add_pd(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm_and_pd(fscal,cutoff_mask);
 585
 586             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 587
 588             /* Calculate temporary vectorial force */
 589             tx               = _mm_mul_pd(fscal,dx30);
 590             ty               = _mm_mul_pd(fscal,dy30);
 591             tz               = _mm_mul_pd(fscal,dz30);
 592
 593             /* Update vectorial force */
 594             fix3             = _mm_add_pd(fix3,tx);
 595             fiy3             = _mm_add_pd(fiy3,ty);
 596             fiz3             = _mm_add_pd(fiz3,tz);
 597
 598             fjx0             = _mm_add_pd(fjx0,tx);
 599             fjy0             = _mm_add_pd(fjy0,ty);
 600             fjz0             = _mm_add_pd(fjz0,tz);
 601
 602             }
 603
 604             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 605
 606             /* Inner loop uses 141 flops */
 607         }
 608
 609         /* End of innermost loop */
 610
 611         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 612                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 613
 614         ggid                        = gid[iidx];
 615         /* Update potential energies */
 616         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 617
 618         /* Increment number of inner iterations */
 619         inneriter                  += j_index_end - j_index_start;
 620
 621         /* Outer loop uses 19 flops */
 622     }
 623
 624     /* Increment number of outer iterations */
 625     outeriter        += nri;
 626
 627     /* Update outer/inner flops */
 628
 629     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
 630 }
 631 /*
 632  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse4_1_double
 633  * Electrostatics interaction: Ewald
 634  * VdW interaction:            None
 635  * Geometry:                   Water4-Particle
 636  * Calculate force/pot:        Force
 637  */
 638 void
 639 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse4_1_double
 640                     (t_nblist                    * gmx_restrict       nlist,
 641                      rvec                        * gmx_restrict          xx,
 642                      rvec                        * gmx_restrict          ff,
 643                      struct t_forcerec           * gmx_restrict          fr,
 644                      t_mdatoms                   * gmx_restrict     mdatoms,
 645                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 646                      t_nrnb                      * gmx_restrict        nrnb)
 647 {
 648     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 649      * just 0 for non-waters.
 650      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 651      * jnr indices corresponding to data put in the four positions in the SIMD register.
 652      */
 653     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 654     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 655     int              jnrA,jnrB;
 656     int              j_coord_offsetA,j_coord_offsetB;
 657     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 658     real             rcutoff_scalar;
 659     real             *shiftvec,*fshift,*x,*f;
 660     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 661     int              vdwioffset1;
 662     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 663     int              vdwioffset2;
 664     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 665     int              vdwioffset3;
 666     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 667     int              vdwjidx0A,vdwjidx0B;
 668     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 669     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 670     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 671     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 672     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 673     real             *charge;
 674     __m128i          ewitab;
 675     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 676     real             *ewtab;
 677     __m128d          dummy_mask,cutoff_mask;
 678     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 679     __m128d          one     = _mm_set1_pd(1.0);
 680     __m128d          two     = _mm_set1_pd(2.0);
 681     x                = xx[0];
 682     f                = ff[0];
 683
 684     nri              = nlist->nri;
 685     iinr             = nlist->iinr;
 686     jindex           = nlist->jindex;
 687     jjnr             = nlist->jjnr;
 688     shiftidx         = nlist->shift;
 689     gid              = nlist->gid;
 690     shiftvec         = fr->shift_vec[0];
 691     fshift           = fr->fshift[0];
 692     facel            = _mm_set1_pd(fr->ic->epsfac);
 693     charge           = mdatoms->chargeA;
 694
 695     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 696     ewtab            = fr->ic->tabq_coul_F;
 697     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 698     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 699
 700     /* Setup water-specific parameters */
 701     inr              = nlist->iinr[0];
 702     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 703     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 704     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 705
 706     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 707     rcutoff_scalar   = fr->ic->rcoulomb;
 708     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 709     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 710
 711     /* Avoid stupid compiler warnings */
 712     jnrA = jnrB = 0;
 713     j_coord_offsetA = 0;
 714     j_coord_offsetB = 0;
 715
 716     outeriter        = 0;
 717     inneriter        = 0;
 718
 719     /* Start outer loop over neighborlists */
 720     for(iidx=0; iidx<nri; iidx++)
 721     {
 722         /* Load shift vector for this list */
 723         i_shift_offset   = DIM*shiftidx[iidx];
 724
 725         /* Load limits for loop over neighbors */
 726         j_index_start    = jindex[iidx];
 727         j_index_end      = jindex[iidx+1];
 728
 729         /* Get outer coordinate index */
 730         inr              = iinr[iidx];
 731         i_coord_offset   = DIM*inr;
 732
 733         /* Load i particle coords and add shift vector */
 734         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 735                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 736
 737         fix1             = _mm_setzero_pd();
 738         fiy1             = _mm_setzero_pd();
 739         fiz1             = _mm_setzero_pd();
 740         fix2             = _mm_setzero_pd();
 741         fiy2             = _mm_setzero_pd();
 742         fiz2             = _mm_setzero_pd();
 743         fix3             = _mm_setzero_pd();
 744         fiy3             = _mm_setzero_pd();
 745         fiz3             = _mm_setzero_pd();
 746
 747         /* Start inner kernel loop */
 748         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrA             = jjnr[jidx];
 753             jnrB             = jjnr[jidx+1];
 754             j_coord_offsetA  = DIM*jnrA;
 755             j_coord_offsetB  = DIM*jnrB;
 756
 757             /* load j atom coordinates */
 758             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 759                                               &jx0,&jy0,&jz0);
 760
 761             /* Calculate displacement vector */
 762             dx10             = _mm_sub_pd(ix1,jx0);
 763             dy10             = _mm_sub_pd(iy1,jy0);
 764             dz10             = _mm_sub_pd(iz1,jz0);
 765             dx20             = _mm_sub_pd(ix2,jx0);
 766             dy20             = _mm_sub_pd(iy2,jy0);
 767             dz20             = _mm_sub_pd(iz2,jz0);
 768             dx30             = _mm_sub_pd(ix3,jx0);
 769             dy30             = _mm_sub_pd(iy3,jy0);
 770             dz30             = _mm_sub_pd(iz3,jz0);
 771
 772             /* Calculate squared distance and things based on it */
 773             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 774             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 775             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 776
 777             rinv10           = sse41_invsqrt_d(rsq10);
 778             rinv20           = sse41_invsqrt_d(rsq20);
 779             rinv30           = sse41_invsqrt_d(rsq30);
 780
 781             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 782             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 783             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 784
 785             /* Load parameters for j particles */
 786             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 787
 788             fjx0             = _mm_setzero_pd();
 789             fjy0             = _mm_setzero_pd();
 790             fjz0             = _mm_setzero_pd();
 791
 792             /**************************
 793              * CALCULATE INTERACTIONS *
 794              **************************/
 795
 796             if (gmx_mm_any_lt(rsq10,rcutoff2))
 797             {
 798
 799             r10              = _mm_mul_pd(rsq10,rinv10);
 800
 801             /* Compute parameters for interactions between i and j atoms */
 802             qq10             = _mm_mul_pd(iq1,jq0);
 803
 804             /* EWALD ELECTROSTATICS */
 805
 806             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 807             ewrt             = _mm_mul_pd(r10,ewtabscale);
 808             ewitab           = _mm_cvttpd_epi32(ewrt);
 809             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 810             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 811                                          &ewtabF,&ewtabFn);
 812             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 813             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 814
 815             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 816
 817             fscal            = felec;
 818
 819             fscal            = _mm_and_pd(fscal,cutoff_mask);
 820
 821             /* Calculate temporary vectorial force */
 822             tx               = _mm_mul_pd(fscal,dx10);
 823             ty               = _mm_mul_pd(fscal,dy10);
 824             tz               = _mm_mul_pd(fscal,dz10);
 825
 826             /* Update vectorial force */
 827             fix1             = _mm_add_pd(fix1,tx);
 828             fiy1             = _mm_add_pd(fiy1,ty);
 829             fiz1             = _mm_add_pd(fiz1,tz);
 830
 831             fjx0             = _mm_add_pd(fjx0,tx);
 832             fjy0             = _mm_add_pd(fjy0,ty);
 833             fjz0             = _mm_add_pd(fjz0,tz);
 834
 835             }
 836
 837             /**************************
 838              * CALCULATE INTERACTIONS *
 839              **************************/
 840
 841             if (gmx_mm_any_lt(rsq20,rcutoff2))
 842             {
 843
 844             r20              = _mm_mul_pd(rsq20,rinv20);
 845
 846             /* Compute parameters for interactions between i and j atoms */
 847             qq20             = _mm_mul_pd(iq2,jq0);
 848
 849             /* EWALD ELECTROSTATICS */
 850
 851             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 852             ewrt             = _mm_mul_pd(r20,ewtabscale);
 853             ewitab           = _mm_cvttpd_epi32(ewrt);
 854             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 855             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 856                                          &ewtabF,&ewtabFn);
 857             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 858             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 859
 860             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 861
 862             fscal            = felec;
 863
 864             fscal            = _mm_and_pd(fscal,cutoff_mask);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm_mul_pd(fscal,dx20);
 868             ty               = _mm_mul_pd(fscal,dy20);
 869             tz               = _mm_mul_pd(fscal,dz20);
 870
 871             /* Update vectorial force */
 872             fix2             = _mm_add_pd(fix2,tx);
 873             fiy2             = _mm_add_pd(fiy2,ty);
 874             fiz2             = _mm_add_pd(fiz2,tz);
 875
 876             fjx0             = _mm_add_pd(fjx0,tx);
 877             fjy0             = _mm_add_pd(fjy0,ty);
 878             fjz0             = _mm_add_pd(fjz0,tz);
 879
 880             }
 881
 882             /**************************
 883              * CALCULATE INTERACTIONS *
 884              **************************/
 885
 886             if (gmx_mm_any_lt(rsq30,rcutoff2))
 887             {
 888
 889             r30              = _mm_mul_pd(rsq30,rinv30);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq30             = _mm_mul_pd(iq3,jq0);
 893
 894             /* EWALD ELECTROSTATICS */
 895
 896             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 897             ewrt             = _mm_mul_pd(r30,ewtabscale);
 898             ewitab           = _mm_cvttpd_epi32(ewrt);
 899             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 900             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 901                                          &ewtabF,&ewtabFn);
 902             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 903             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 904
 905             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 906
 907             fscal            = felec;
 908
 909             fscal            = _mm_and_pd(fscal,cutoff_mask);
 910
 911             /* Calculate temporary vectorial force */
 912             tx               = _mm_mul_pd(fscal,dx30);
 913             ty               = _mm_mul_pd(fscal,dy30);
 914             tz               = _mm_mul_pd(fscal,dz30);
 915
 916             /* Update vectorial force */
 917             fix3             = _mm_add_pd(fix3,tx);
 918             fiy3             = _mm_add_pd(fiy3,ty);
 919             fiz3             = _mm_add_pd(fiz3,tz);
 920
 921             fjx0             = _mm_add_pd(fjx0,tx);
 922             fjy0             = _mm_add_pd(fjy0,ty);
 923             fjz0             = _mm_add_pd(fjz0,tz);
 924
 925             }
 926
 927             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 928
 929             /* Inner loop uses 120 flops */
 930         }
 931
 932         if(jidx<j_index_end)
 933         {
 934
 935             jnrA             = jjnr[jidx];
 936             j_coord_offsetA  = DIM*jnrA;
 937
 938             /* load j atom coordinates */
 939             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 940                                               &jx0,&jy0,&jz0);
 941
 942             /* Calculate displacement vector */
 943             dx10             = _mm_sub_pd(ix1,jx0);
 944             dy10             = _mm_sub_pd(iy1,jy0);
 945             dz10             = _mm_sub_pd(iz1,jz0);
 946             dx20             = _mm_sub_pd(ix2,jx0);
 947             dy20             = _mm_sub_pd(iy2,jy0);
 948             dz20             = _mm_sub_pd(iz2,jz0);
 949             dx30             = _mm_sub_pd(ix3,jx0);
 950             dy30             = _mm_sub_pd(iy3,jy0);
 951             dz30             = _mm_sub_pd(iz3,jz0);
 952
 953             /* Calculate squared distance and things based on it */
 954             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 955             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 956             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 957
 958             rinv10           = sse41_invsqrt_d(rsq10);
 959             rinv20           = sse41_invsqrt_d(rsq20);
 960             rinv30           = sse41_invsqrt_d(rsq30);
 961
 962             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 963             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 964             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 965
 966             /* Load parameters for j particles */
 967             jq0              = _mm_load_sd(charge+jnrA+0);
 968
 969             fjx0             = _mm_setzero_pd();
 970             fjy0             = _mm_setzero_pd();
 971             fjz0             = _mm_setzero_pd();
 972
 973             /**************************
 974              * CALCULATE INTERACTIONS *
 975              **************************/
 976
 977             if (gmx_mm_any_lt(rsq10,rcutoff2))
 978             {
 979
 980             r10              = _mm_mul_pd(rsq10,rinv10);
 981
 982             /* Compute parameters for interactions between i and j atoms */
 983             qq10             = _mm_mul_pd(iq1,jq0);
 984
 985             /* EWALD ELECTROSTATICS */
 986
 987             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 988             ewrt             = _mm_mul_pd(r10,ewtabscale);
 989             ewitab           = _mm_cvttpd_epi32(ewrt);
 990             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 991             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 992             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 993             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 994
 995             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 996
 997             fscal            = felec;
 998
 999             fscal            = _mm_and_pd(fscal,cutoff_mask);
1000
1001             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1002
1003             /* Calculate temporary vectorial force */
1004             tx               = _mm_mul_pd(fscal,dx10);
1005             ty               = _mm_mul_pd(fscal,dy10);
1006             tz               = _mm_mul_pd(fscal,dz10);
1007
1008             /* Update vectorial force */
1009             fix1             = _mm_add_pd(fix1,tx);
1010             fiy1             = _mm_add_pd(fiy1,ty);
1011             fiz1             = _mm_add_pd(fiz1,tz);
1012
1013             fjx0             = _mm_add_pd(fjx0,tx);
1014             fjy0             = _mm_add_pd(fjy0,ty);
1015             fjz0             = _mm_add_pd(fjz0,tz);
1016
1017             }
1018
1019             /**************************
1020              * CALCULATE INTERACTIONS *
1021              **************************/
1022
1023             if (gmx_mm_any_lt(rsq20,rcutoff2))
1024             {
1025
1026             r20              = _mm_mul_pd(rsq20,rinv20);
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq20             = _mm_mul_pd(iq2,jq0);
1030
1031             /* EWALD ELECTROSTATICS */
1032
1033             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1034             ewrt             = _mm_mul_pd(r20,ewtabscale);
1035             ewitab           = _mm_cvttpd_epi32(ewrt);
1036             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1037             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1038             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1039             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1040
1041             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1042
1043             fscal            = felec;
1044
1045             fscal            = _mm_and_pd(fscal,cutoff_mask);
1046
1047             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1048
1049             /* Calculate temporary vectorial force */
1050             tx               = _mm_mul_pd(fscal,dx20);
1051             ty               = _mm_mul_pd(fscal,dy20);
1052             tz               = _mm_mul_pd(fscal,dz20);
1053
1054             /* Update vectorial force */
1055             fix2             = _mm_add_pd(fix2,tx);
1056             fiy2             = _mm_add_pd(fiy2,ty);
1057             fiz2             = _mm_add_pd(fiz2,tz);
1058
1059             fjx0             = _mm_add_pd(fjx0,tx);
1060             fjy0             = _mm_add_pd(fjy0,ty);
1061             fjz0             = _mm_add_pd(fjz0,tz);
1062
1063             }
1064
1065             /**************************
1066              * CALCULATE INTERACTIONS *
1067              **************************/
1068
1069             if (gmx_mm_any_lt(rsq30,rcutoff2))
1070             {
1071
1072             r30              = _mm_mul_pd(rsq30,rinv30);
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq30             = _mm_mul_pd(iq3,jq0);
1076
1077             /* EWALD ELECTROSTATICS */
1078
1079             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1080             ewrt             = _mm_mul_pd(r30,ewtabscale);
1081             ewitab           = _mm_cvttpd_epi32(ewrt);
1082             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1083             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1084             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1085             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1086
1087             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1088
1089             fscal            = felec;
1090
1091             fscal            = _mm_and_pd(fscal,cutoff_mask);
1092
1093             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1094
1095             /* Calculate temporary vectorial force */
1096             tx               = _mm_mul_pd(fscal,dx30);
1097             ty               = _mm_mul_pd(fscal,dy30);
1098             tz               = _mm_mul_pd(fscal,dz30);
1099
1100             /* Update vectorial force */
1101             fix3             = _mm_add_pd(fix3,tx);
1102             fiy3             = _mm_add_pd(fiy3,ty);
1103             fiz3             = _mm_add_pd(fiz3,tz);
1104
1105             fjx0             = _mm_add_pd(fjx0,tx);
1106             fjy0             = _mm_add_pd(fjy0,ty);
1107             fjz0             = _mm_add_pd(fjz0,tz);
1108
1109             }
1110
1111             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1112
1113             /* Inner loop uses 120 flops */
1114         }
1115
1116         /* End of innermost loop */
1117
1118         gmx_mm_update_iforce_3atom_swizzle_pd(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1119                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1120
1121         /* Increment number of inner iterations */
1122         inneriter                  += j_index_end - j_index_start;
1123
1124         /* Outer loop uses 18 flops */
1125     }
1126
1127     /* Increment number of outer iterations */
1128     outeriter        += nri;
1129
1130     /* Update outer/inner flops */
1131
1132     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1133 }