src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_sse4_1_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, 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 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse4_1_double.h"
  50 #include "kernelutil_x86_sse4_1_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_sse4_1_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_sse4_1_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwjidx0A,vdwjidx0B;
  89     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  92     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  93     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          ewitab;
  96     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128d          dummy_mask,cutoff_mask;
  99     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 100     __m128d          one     = _mm_set1_pd(1.0);
 101     __m128d          two     = _mm_set1_pd(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_pd(fr->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 124     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 125     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->rcoulomb;
 129     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136
 137     outeriter        = 0;
 138     inneriter        = 0;
 139
 140     /* Start outer loop over neighborlists */
 141     for(iidx=0; iidx<nri; iidx++)
 142     {
 143         /* Load shift vector for this list */
 144         i_shift_offset   = DIM*shiftidx[iidx];
 145
 146         /* Load limits for loop over neighbors */
 147         j_index_start    = jindex[iidx];
 148         j_index_end      = jindex[iidx+1];
 149
 150         /* Get outer coordinate index */
 151         inr              = iinr[iidx];
 152         i_coord_offset   = DIM*inr;
 153
 154         /* Load i particle coords and add shift vector */
 155         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 156                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 157
 158         fix0             = _mm_setzero_pd();
 159         fiy0             = _mm_setzero_pd();
 160         fiz0             = _mm_setzero_pd();
 161         fix1             = _mm_setzero_pd();
 162         fiy1             = _mm_setzero_pd();
 163         fiz1             = _mm_setzero_pd();
 164         fix2             = _mm_setzero_pd();
 165         fiy2             = _mm_setzero_pd();
 166         fiz2             = _mm_setzero_pd();
 167
 168         /* Reset potential sums */
 169         velecsum         = _mm_setzero_pd();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180
 181             /* load j atom coordinates */
 182             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 183                                               &jx0,&jy0,&jz0);
 184
 185             /* Calculate displacement vector */
 186             dx00             = _mm_sub_pd(ix0,jx0);
 187             dy00             = _mm_sub_pd(iy0,jy0);
 188             dz00             = _mm_sub_pd(iz0,jz0);
 189             dx10             = _mm_sub_pd(ix1,jx0);
 190             dy10             = _mm_sub_pd(iy1,jy0);
 191             dz10             = _mm_sub_pd(iz1,jz0);
 192             dx20             = _mm_sub_pd(ix2,jx0);
 193             dy20             = _mm_sub_pd(iy2,jy0);
 194             dz20             = _mm_sub_pd(iz2,jz0);
 195
 196             /* Calculate squared distance and things based on it */
 197             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 198             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 199             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 200
 201             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 202             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 203             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 204
 205             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 206             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 207             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 211
 212             fjx0             = _mm_setzero_pd();
 213             fjy0             = _mm_setzero_pd();
 214             fjz0             = _mm_setzero_pd();
 215
 216             /**************************
 217              * CALCULATE INTERACTIONS *
 218              **************************/
 219
 220             if (gmx_mm_any_lt(rsq00,rcutoff2))
 221             {
 222
 223             r00              = _mm_mul_pd(rsq00,rinv00);
 224
 225             /* Compute parameters for interactions between i and j atoms */
 226             qq00             = _mm_mul_pd(iq0,jq0);
 227
 228             /* EWALD ELECTROSTATICS */
 229
 230             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 231             ewrt             = _mm_mul_pd(r00,ewtabscale);
 232             ewitab           = _mm_cvttpd_epi32(ewrt);
 233             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 234             ewitab           = _mm_slli_epi32(ewitab,2);
 235             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 236             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 237             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 238             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 239             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 240             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 241             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 242             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 243             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 244             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 245
 246             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velec            = _mm_and_pd(velec,cutoff_mask);
 250             velecsum         = _mm_add_pd(velecsum,velec);
 251
 252             fscal            = felec;
 253
 254             fscal            = _mm_and_pd(fscal,cutoff_mask);
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm_mul_pd(fscal,dx00);
 258             ty               = _mm_mul_pd(fscal,dy00);
 259             tz               = _mm_mul_pd(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm_add_pd(fix0,tx);
 263             fiy0             = _mm_add_pd(fiy0,ty);
 264             fiz0             = _mm_add_pd(fiz0,tz);
 265
 266             fjx0             = _mm_add_pd(fjx0,tx);
 267             fjy0             = _mm_add_pd(fjy0,ty);
 268             fjz0             = _mm_add_pd(fjz0,tz);
 269
 270             }
 271
 272             /**************************
 273              * CALCULATE INTERACTIONS *
 274              **************************/
 275
 276             if (gmx_mm_any_lt(rsq10,rcutoff2))
 277             {
 278
 279             r10              = _mm_mul_pd(rsq10,rinv10);
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq10             = _mm_mul_pd(iq1,jq0);
 283
 284             /* EWALD ELECTROSTATICS */
 285
 286             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 287             ewrt             = _mm_mul_pd(r10,ewtabscale);
 288             ewitab           = _mm_cvttpd_epi32(ewrt);
 289             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 290             ewitab           = _mm_slli_epi32(ewitab,2);
 291             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 292             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 293             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 294             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 295             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 296             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 297             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 298             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 299             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 300             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 301
 302             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_and_pd(velec,cutoff_mask);
 306             velecsum         = _mm_add_pd(velecsum,velec);
 307
 308             fscal            = felec;
 309
 310             fscal            = _mm_and_pd(fscal,cutoff_mask);
 311
 312             /* Calculate temporary vectorial force */
 313             tx               = _mm_mul_pd(fscal,dx10);
 314             ty               = _mm_mul_pd(fscal,dy10);
 315             tz               = _mm_mul_pd(fscal,dz10);
 316
 317             /* Update vectorial force */
 318             fix1             = _mm_add_pd(fix1,tx);
 319             fiy1             = _mm_add_pd(fiy1,ty);
 320             fiz1             = _mm_add_pd(fiz1,tz);
 321
 322             fjx0             = _mm_add_pd(fjx0,tx);
 323             fjy0             = _mm_add_pd(fjy0,ty);
 324             fjz0             = _mm_add_pd(fjz0,tz);
 325
 326             }
 327
 328             /**************************
 329              * CALCULATE INTERACTIONS *
 330              **************************/
 331
 332             if (gmx_mm_any_lt(rsq20,rcutoff2))
 333             {
 334
 335             r20              = _mm_mul_pd(rsq20,rinv20);
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq20             = _mm_mul_pd(iq2,jq0);
 339
 340             /* EWALD ELECTROSTATICS */
 341
 342             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 343             ewrt             = _mm_mul_pd(r20,ewtabscale);
 344             ewitab           = _mm_cvttpd_epi32(ewrt);
 345             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 346             ewitab           = _mm_slli_epi32(ewitab,2);
 347             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 348             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 349             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 350             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 351             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 352             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 353             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 354             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 355             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 356             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 357
 358             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 359
 360             /* Update potential sum for this i atom from the interaction with this j atom. */
 361             velec            = _mm_and_pd(velec,cutoff_mask);
 362             velecsum         = _mm_add_pd(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             fscal            = _mm_and_pd(fscal,cutoff_mask);
 367
 368             /* Calculate temporary vectorial force */
 369             tx               = _mm_mul_pd(fscal,dx20);
 370             ty               = _mm_mul_pd(fscal,dy20);
 371             tz               = _mm_mul_pd(fscal,dz20);
 372
 373             /* Update vectorial force */
 374             fix2             = _mm_add_pd(fix2,tx);
 375             fiy2             = _mm_add_pd(fiy2,ty);
 376             fiz2             = _mm_add_pd(fiz2,tz);
 377
 378             fjx0             = _mm_add_pd(fjx0,tx);
 379             fjy0             = _mm_add_pd(fjy0,ty);
 380             fjz0             = _mm_add_pd(fjz0,tz);
 381
 382             }
 383
 384             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 385
 386             /* Inner loop uses 141 flops */
 387         }
 388
 389         if(jidx<j_index_end)
 390         {
 391
 392             jnrA             = jjnr[jidx];
 393             j_coord_offsetA  = DIM*jnrA;
 394
 395             /* load j atom coordinates */
 396             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 397                                               &jx0,&jy0,&jz0);
 398
 399             /* Calculate displacement vector */
 400             dx00             = _mm_sub_pd(ix0,jx0);
 401             dy00             = _mm_sub_pd(iy0,jy0);
 402             dz00             = _mm_sub_pd(iz0,jz0);
 403             dx10             = _mm_sub_pd(ix1,jx0);
 404             dy10             = _mm_sub_pd(iy1,jy0);
 405             dz10             = _mm_sub_pd(iz1,jz0);
 406             dx20             = _mm_sub_pd(ix2,jx0);
 407             dy20             = _mm_sub_pd(iy2,jy0);
 408             dz20             = _mm_sub_pd(iz2,jz0);
 409
 410             /* Calculate squared distance and things based on it */
 411             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 412             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 413             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 414
 415             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 416             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 417             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 418
 419             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 420             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 421             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 422
 423             /* Load parameters for j particles */
 424             jq0              = _mm_load_sd(charge+jnrA+0);
 425
 426             fjx0             = _mm_setzero_pd();
 427             fjy0             = _mm_setzero_pd();
 428             fjz0             = _mm_setzero_pd();
 429
 430             /**************************
 431              * CALCULATE INTERACTIONS *
 432              **************************/
 433
 434             if (gmx_mm_any_lt(rsq00,rcutoff2))
 435             {
 436
 437             r00              = _mm_mul_pd(rsq00,rinv00);
 438
 439             /* Compute parameters for interactions between i and j atoms */
 440             qq00             = _mm_mul_pd(iq0,jq0);
 441
 442             /* EWALD ELECTROSTATICS */
 443
 444             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 445             ewrt             = _mm_mul_pd(r00,ewtabscale);
 446             ewitab           = _mm_cvttpd_epi32(ewrt);
 447             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 448             ewitab           = _mm_slli_epi32(ewitab,2);
 449             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 450             ewtabD           = _mm_setzero_pd();
 451             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 452             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 453             ewtabFn          = _mm_setzero_pd();
 454             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 455             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 456             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 457             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_sub_pd(rinv00,sh_ewald),velec));
 458             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 459
 460             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 461
 462             /* Update potential sum for this i atom from the interaction with this j atom. */
 463             velec            = _mm_and_pd(velec,cutoff_mask);
 464             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 465             velecsum         = _mm_add_pd(velecsum,velec);
 466
 467             fscal            = felec;
 468
 469             fscal            = _mm_and_pd(fscal,cutoff_mask);
 470
 471             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 472
 473             /* Calculate temporary vectorial force */
 474             tx               = _mm_mul_pd(fscal,dx00);
 475             ty               = _mm_mul_pd(fscal,dy00);
 476             tz               = _mm_mul_pd(fscal,dz00);
 477
 478             /* Update vectorial force */
 479             fix0             = _mm_add_pd(fix0,tx);
 480             fiy0             = _mm_add_pd(fiy0,ty);
 481             fiz0             = _mm_add_pd(fiz0,tz);
 482
 483             fjx0             = _mm_add_pd(fjx0,tx);
 484             fjy0             = _mm_add_pd(fjy0,ty);
 485             fjz0             = _mm_add_pd(fjz0,tz);
 486
 487             }
 488
 489             /**************************
 490              * CALCULATE INTERACTIONS *
 491              **************************/
 492
 493             if (gmx_mm_any_lt(rsq10,rcutoff2))
 494             {
 495
 496             r10              = _mm_mul_pd(rsq10,rinv10);
 497
 498             /* Compute parameters for interactions between i and j atoms */
 499             qq10             = _mm_mul_pd(iq1,jq0);
 500
 501             /* EWALD ELECTROSTATICS */
 502
 503             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 504             ewrt             = _mm_mul_pd(r10,ewtabscale);
 505             ewitab           = _mm_cvttpd_epi32(ewrt);
 506             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 507             ewitab           = _mm_slli_epi32(ewitab,2);
 508             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 509             ewtabD           = _mm_setzero_pd();
 510             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 511             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 512             ewtabFn          = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 514             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 515             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 516             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 517             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 518
 519             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 520
 521             /* Update potential sum for this i atom from the interaction with this j atom. */
 522             velec            = _mm_and_pd(velec,cutoff_mask);
 523             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 524             velecsum         = _mm_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_and_pd(fscal,cutoff_mask);
 529
 530             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 531
 532             /* Calculate temporary vectorial force */
 533             tx               = _mm_mul_pd(fscal,dx10);
 534             ty               = _mm_mul_pd(fscal,dy10);
 535             tz               = _mm_mul_pd(fscal,dz10);
 536
 537             /* Update vectorial force */
 538             fix1             = _mm_add_pd(fix1,tx);
 539             fiy1             = _mm_add_pd(fiy1,ty);
 540             fiz1             = _mm_add_pd(fiz1,tz);
 541
 542             fjx0             = _mm_add_pd(fjx0,tx);
 543             fjy0             = _mm_add_pd(fjy0,ty);
 544             fjz0             = _mm_add_pd(fjz0,tz);
 545
 546             }
 547
 548             /**************************
 549              * CALCULATE INTERACTIONS *
 550              **************************/
 551
 552             if (gmx_mm_any_lt(rsq20,rcutoff2))
 553             {
 554
 555             r20              = _mm_mul_pd(rsq20,rinv20);
 556
 557             /* Compute parameters for interactions between i and j atoms */
 558             qq20             = _mm_mul_pd(iq2,jq0);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 563             ewrt             = _mm_mul_pd(r20,ewtabscale);
 564             ewitab           = _mm_cvttpd_epi32(ewrt);
 565             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 566             ewitab           = _mm_slli_epi32(ewitab,2);
 567             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 568             ewtabD           = _mm_setzero_pd();
 569             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 570             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 571             ewtabFn          = _mm_setzero_pd();
 572             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 573             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 574             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 575             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 576             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 577
 578             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 579
 580             /* Update potential sum for this i atom from the interaction with this j atom. */
 581             velec            = _mm_and_pd(velec,cutoff_mask);
 582             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 583             velecsum         = _mm_add_pd(velecsum,velec);
 584
 585             fscal            = felec;
 586
 587             fscal            = _mm_and_pd(fscal,cutoff_mask);
 588
 589             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 590
 591             /* Calculate temporary vectorial force */
 592             tx               = _mm_mul_pd(fscal,dx20);
 593             ty               = _mm_mul_pd(fscal,dy20);
 594             tz               = _mm_mul_pd(fscal,dz20);
 595
 596             /* Update vectorial force */
 597             fix2             = _mm_add_pd(fix2,tx);
 598             fiy2             = _mm_add_pd(fiy2,ty);
 599             fiz2             = _mm_add_pd(fiz2,tz);
 600
 601             fjx0             = _mm_add_pd(fjx0,tx);
 602             fjy0             = _mm_add_pd(fjy0,ty);
 603             fjz0             = _mm_add_pd(fjz0,tz);
 604
 605             }
 606
 607             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 608
 609             /* Inner loop uses 141 flops */
 610         }
 611
 612         /* End of innermost loop */
 613
 614         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 615                                               f+i_coord_offset,fshift+i_shift_offset);
 616
 617         ggid                        = gid[iidx];
 618         /* Update potential energies */
 619         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 620
 621         /* Increment number of inner iterations */
 622         inneriter                  += j_index_end - j_index_start;
 623
 624         /* Outer loop uses 19 flops */
 625     }
 626
 627     /* Increment number of outer iterations */
 628     outeriter        += nri;
 629
 630     /* Update outer/inner flops */
 631
 632     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*141);
 633 }
 634 /*
 635  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse4_1_double
 636  * Electrostatics interaction: Ewald
 637  * VdW interaction:            None
 638  * Geometry:                   Water3-Particle
 639  * Calculate force/pot:        Force
 640  */
 641 void
 642 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse4_1_double
 643                     (t_nblist                    * gmx_restrict       nlist,
 644                      rvec                        * gmx_restrict          xx,
 645                      rvec                        * gmx_restrict          ff,
 646                      t_forcerec                  * gmx_restrict          fr,
 647                      t_mdatoms                   * gmx_restrict     mdatoms,
 648                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 649                      t_nrnb                      * gmx_restrict        nrnb)
 650 {
 651     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 652      * just 0 for non-waters.
 653      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 654      * jnr indices corresponding to data put in the four positions in the SIMD register.
 655      */
 656     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 657     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 658     int              jnrA,jnrB;
 659     int              j_coord_offsetA,j_coord_offsetB;
 660     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 661     real             rcutoff_scalar;
 662     real             *shiftvec,*fshift,*x,*f;
 663     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 664     int              vdwioffset0;
 665     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 666     int              vdwioffset1;
 667     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 668     int              vdwioffset2;
 669     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 670     int              vdwjidx0A,vdwjidx0B;
 671     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 672     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 673     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 674     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 675     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 676     real             *charge;
 677     __m128i          ewitab;
 678     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 679     real             *ewtab;
 680     __m128d          dummy_mask,cutoff_mask;
 681     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 682     __m128d          one     = _mm_set1_pd(1.0);
 683     __m128d          two     = _mm_set1_pd(2.0);
 684     x                = xx[0];
 685     f                = ff[0];
 686
 687     nri              = nlist->nri;
 688     iinr             = nlist->iinr;
 689     jindex           = nlist->jindex;
 690     jjnr             = nlist->jjnr;
 691     shiftidx         = nlist->shift;
 692     gid              = nlist->gid;
 693     shiftvec         = fr->shift_vec[0];
 694     fshift           = fr->fshift[0];
 695     facel            = _mm_set1_pd(fr->epsfac);
 696     charge           = mdatoms->chargeA;
 697
 698     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 699     ewtab            = fr->ic->tabq_coul_F;
 700     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 701     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 702
 703     /* Setup water-specific parameters */
 704     inr              = nlist->iinr[0];
 705     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 706     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 707     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 708
 709     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 710     rcutoff_scalar   = fr->rcoulomb;
 711     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 712     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 713
 714     /* Avoid stupid compiler warnings */
 715     jnrA = jnrB = 0;
 716     j_coord_offsetA = 0;
 717     j_coord_offsetB = 0;
 718
 719     outeriter        = 0;
 720     inneriter        = 0;
 721
 722     /* Start outer loop over neighborlists */
 723     for(iidx=0; iidx<nri; iidx++)
 724     {
 725         /* Load shift vector for this list */
 726         i_shift_offset   = DIM*shiftidx[iidx];
 727
 728         /* Load limits for loop over neighbors */
 729         j_index_start    = jindex[iidx];
 730         j_index_end      = jindex[iidx+1];
 731
 732         /* Get outer coordinate index */
 733         inr              = iinr[iidx];
 734         i_coord_offset   = DIM*inr;
 735
 736         /* Load i particle coords and add shift vector */
 737         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 738                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 739
 740         fix0             = _mm_setzero_pd();
 741         fiy0             = _mm_setzero_pd();
 742         fiz0             = _mm_setzero_pd();
 743         fix1             = _mm_setzero_pd();
 744         fiy1             = _mm_setzero_pd();
 745         fiz1             = _mm_setzero_pd();
 746         fix2             = _mm_setzero_pd();
 747         fiy2             = _mm_setzero_pd();
 748         fiz2             = _mm_setzero_pd();
 749
 750         /* Start inner kernel loop */
 751         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 752         {
 753
 754             /* Get j neighbor index, and coordinate index */
 755             jnrA             = jjnr[jidx];
 756             jnrB             = jjnr[jidx+1];
 757             j_coord_offsetA  = DIM*jnrA;
 758             j_coord_offsetB  = DIM*jnrB;
 759
 760             /* load j atom coordinates */
 761             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 762                                               &jx0,&jy0,&jz0);
 763
 764             /* Calculate displacement vector */
 765             dx00             = _mm_sub_pd(ix0,jx0);
 766             dy00             = _mm_sub_pd(iy0,jy0);
 767             dz00             = _mm_sub_pd(iz0,jz0);
 768             dx10             = _mm_sub_pd(ix1,jx0);
 769             dy10             = _mm_sub_pd(iy1,jy0);
 770             dz10             = _mm_sub_pd(iz1,jz0);
 771             dx20             = _mm_sub_pd(ix2,jx0);
 772             dy20             = _mm_sub_pd(iy2,jy0);
 773             dz20             = _mm_sub_pd(iz2,jz0);
 774
 775             /* Calculate squared distance and things based on it */
 776             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 777             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 778             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 779
 780             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 781             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 782             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 783
 784             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 785             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 786             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 787
 788             /* Load parameters for j particles */
 789             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 790
 791             fjx0             = _mm_setzero_pd();
 792             fjy0             = _mm_setzero_pd();
 793             fjz0             = _mm_setzero_pd();
 794
 795             /**************************
 796              * CALCULATE INTERACTIONS *
 797              **************************/
 798
 799             if (gmx_mm_any_lt(rsq00,rcutoff2))
 800             {
 801
 802             r00              = _mm_mul_pd(rsq00,rinv00);
 803
 804             /* Compute parameters for interactions between i and j atoms */
 805             qq00             = _mm_mul_pd(iq0,jq0);
 806
 807             /* EWALD ELECTROSTATICS */
 808
 809             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 810             ewrt             = _mm_mul_pd(r00,ewtabscale);
 811             ewitab           = _mm_cvttpd_epi32(ewrt);
 812             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 813             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 814                                          &ewtabF,&ewtabFn);
 815             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 816             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 817
 818             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 819
 820             fscal            = felec;
 821
 822             fscal            = _mm_and_pd(fscal,cutoff_mask);
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm_mul_pd(fscal,dx00);
 826             ty               = _mm_mul_pd(fscal,dy00);
 827             tz               = _mm_mul_pd(fscal,dz00);
 828
 829             /* Update vectorial force */
 830             fix0             = _mm_add_pd(fix0,tx);
 831             fiy0             = _mm_add_pd(fiy0,ty);
 832             fiz0             = _mm_add_pd(fiz0,tz);
 833
 834             fjx0             = _mm_add_pd(fjx0,tx);
 835             fjy0             = _mm_add_pd(fjy0,ty);
 836             fjz0             = _mm_add_pd(fjz0,tz);
 837
 838             }
 839
 840             /**************************
 841              * CALCULATE INTERACTIONS *
 842              **************************/
 843
 844             if (gmx_mm_any_lt(rsq10,rcutoff2))
 845             {
 846
 847             r10              = _mm_mul_pd(rsq10,rinv10);
 848
 849             /* Compute parameters for interactions between i and j atoms */
 850             qq10             = _mm_mul_pd(iq1,jq0);
 851
 852             /* EWALD ELECTROSTATICS */
 853
 854             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 855             ewrt             = _mm_mul_pd(r10,ewtabscale);
 856             ewitab           = _mm_cvttpd_epi32(ewrt);
 857             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 858             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 859                                          &ewtabF,&ewtabFn);
 860             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 861             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 862
 863             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 864
 865             fscal            = felec;
 866
 867             fscal            = _mm_and_pd(fscal,cutoff_mask);
 868
 869             /* Calculate temporary vectorial force */
 870             tx               = _mm_mul_pd(fscal,dx10);
 871             ty               = _mm_mul_pd(fscal,dy10);
 872             tz               = _mm_mul_pd(fscal,dz10);
 873
 874             /* Update vectorial force */
 875             fix1             = _mm_add_pd(fix1,tx);
 876             fiy1             = _mm_add_pd(fiy1,ty);
 877             fiz1             = _mm_add_pd(fiz1,tz);
 878
 879             fjx0             = _mm_add_pd(fjx0,tx);
 880             fjy0             = _mm_add_pd(fjy0,ty);
 881             fjz0             = _mm_add_pd(fjz0,tz);
 882
 883             }
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             if (gmx_mm_any_lt(rsq20,rcutoff2))
 890             {
 891
 892             r20              = _mm_mul_pd(rsq20,rinv20);
 893
 894             /* Compute parameters for interactions between i and j atoms */
 895             qq20             = _mm_mul_pd(iq2,jq0);
 896
 897             /* EWALD ELECTROSTATICS */
 898
 899             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 900             ewrt             = _mm_mul_pd(r20,ewtabscale);
 901             ewitab           = _mm_cvttpd_epi32(ewrt);
 902             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 903             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 904                                          &ewtabF,&ewtabFn);
 905             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 906             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 907
 908             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 909
 910             fscal            = felec;
 911
 912             fscal            = _mm_and_pd(fscal,cutoff_mask);
 913
 914             /* Calculate temporary vectorial force */
 915             tx               = _mm_mul_pd(fscal,dx20);
 916             ty               = _mm_mul_pd(fscal,dy20);
 917             tz               = _mm_mul_pd(fscal,dz20);
 918
 919             /* Update vectorial force */
 920             fix2             = _mm_add_pd(fix2,tx);
 921             fiy2             = _mm_add_pd(fiy2,ty);
 922             fiz2             = _mm_add_pd(fiz2,tz);
 923
 924             fjx0             = _mm_add_pd(fjx0,tx);
 925             fjy0             = _mm_add_pd(fjy0,ty);
 926             fjz0             = _mm_add_pd(fjz0,tz);
 927
 928             }
 929
 930             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 931
 932             /* Inner loop uses 120 flops */
 933         }
 934
 935         if(jidx<j_index_end)
 936         {
 937
 938             jnrA             = jjnr[jidx];
 939             j_coord_offsetA  = DIM*jnrA;
 940
 941             /* load j atom coordinates */
 942             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 943                                               &jx0,&jy0,&jz0);
 944
 945             /* Calculate displacement vector */
 946             dx00             = _mm_sub_pd(ix0,jx0);
 947             dy00             = _mm_sub_pd(iy0,jy0);
 948             dz00             = _mm_sub_pd(iz0,jz0);
 949             dx10             = _mm_sub_pd(ix1,jx0);
 950             dy10             = _mm_sub_pd(iy1,jy0);
 951             dz10             = _mm_sub_pd(iz1,jz0);
 952             dx20             = _mm_sub_pd(ix2,jx0);
 953             dy20             = _mm_sub_pd(iy2,jy0);
 954             dz20             = _mm_sub_pd(iz2,jz0);
 955
 956             /* Calculate squared distance and things based on it */
 957             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 958             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 959             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 960
 961             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 962             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 963             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 964
 965             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 966             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 967             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 968
 969             /* Load parameters for j particles */
 970             jq0              = _mm_load_sd(charge+jnrA+0);
 971
 972             fjx0             = _mm_setzero_pd();
 973             fjy0             = _mm_setzero_pd();
 974             fjz0             = _mm_setzero_pd();
 975
 976             /**************************
 977              * CALCULATE INTERACTIONS *
 978              **************************/
 979
 980             if (gmx_mm_any_lt(rsq00,rcutoff2))
 981             {
 982
 983             r00              = _mm_mul_pd(rsq00,rinv00);
 984
 985             /* Compute parameters for interactions between i and j atoms */
 986             qq00             = _mm_mul_pd(iq0,jq0);
 987
 988             /* EWALD ELECTROSTATICS */
 989
 990             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 991             ewrt             = _mm_mul_pd(r00,ewtabscale);
 992             ewitab           = _mm_cvttpd_epi32(ewrt);
 993             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 994             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 995             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 996             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 997
 998             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_and_pd(fscal,cutoff_mask);
1003
1004             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1005
1006             /* Calculate temporary vectorial force */
1007             tx               = _mm_mul_pd(fscal,dx00);
1008             ty               = _mm_mul_pd(fscal,dy00);
1009             tz               = _mm_mul_pd(fscal,dz00);
1010
1011             /* Update vectorial force */
1012             fix0             = _mm_add_pd(fix0,tx);
1013             fiy0             = _mm_add_pd(fiy0,ty);
1014             fiz0             = _mm_add_pd(fiz0,tz);
1015
1016             fjx0             = _mm_add_pd(fjx0,tx);
1017             fjy0             = _mm_add_pd(fjy0,ty);
1018             fjz0             = _mm_add_pd(fjz0,tz);
1019
1020             }
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             if (gmx_mm_any_lt(rsq10,rcutoff2))
1027             {
1028
1029             r10              = _mm_mul_pd(rsq10,rinv10);
1030
1031             /* Compute parameters for interactions between i and j atoms */
1032             qq10             = _mm_mul_pd(iq1,jq0);
1033
1034             /* EWALD ELECTROSTATICS */
1035
1036             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1037             ewrt             = _mm_mul_pd(r10,ewtabscale);
1038             ewitab           = _mm_cvttpd_epi32(ewrt);
1039             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1040             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1041             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1042             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1043
1044             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1045
1046             fscal            = felec;
1047
1048             fscal            = _mm_and_pd(fscal,cutoff_mask);
1049
1050             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1051
1052             /* Calculate temporary vectorial force */
1053             tx               = _mm_mul_pd(fscal,dx10);
1054             ty               = _mm_mul_pd(fscal,dy10);
1055             tz               = _mm_mul_pd(fscal,dz10);
1056
1057             /* Update vectorial force */
1058             fix1             = _mm_add_pd(fix1,tx);
1059             fiy1             = _mm_add_pd(fiy1,ty);
1060             fiz1             = _mm_add_pd(fiz1,tz);
1061
1062             fjx0             = _mm_add_pd(fjx0,tx);
1063             fjy0             = _mm_add_pd(fjy0,ty);
1064             fjz0             = _mm_add_pd(fjz0,tz);
1065
1066             }
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             if (gmx_mm_any_lt(rsq20,rcutoff2))
1073             {
1074
1075             r20              = _mm_mul_pd(rsq20,rinv20);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq20             = _mm_mul_pd(iq2,jq0);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1083             ewrt             = _mm_mul_pd(r20,ewtabscale);
1084             ewitab           = _mm_cvttpd_epi32(ewrt);
1085             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1086             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1087             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1088             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1089
1090             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1091
1092             fscal            = felec;
1093
1094             fscal            = _mm_and_pd(fscal,cutoff_mask);
1095
1096             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1097
1098             /* Calculate temporary vectorial force */
1099             tx               = _mm_mul_pd(fscal,dx20);
1100             ty               = _mm_mul_pd(fscal,dy20);
1101             tz               = _mm_mul_pd(fscal,dz20);
1102
1103             /* Update vectorial force */
1104             fix2             = _mm_add_pd(fix2,tx);
1105             fiy2             = _mm_add_pd(fiy2,ty);
1106             fiz2             = _mm_add_pd(fiz2,tz);
1107
1108             fjx0             = _mm_add_pd(fjx0,tx);
1109             fjy0             = _mm_add_pd(fjy0,ty);
1110             fjz0             = _mm_add_pd(fjz0,tz);
1111
1112             }
1113
1114             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1115
1116             /* Inner loop uses 120 flops */
1117         }
1118
1119         /* End of innermost loop */
1120
1121         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1122                                               f+i_coord_offset,fshift+i_shift_offset);
1123
1124         /* Increment number of inner iterations */
1125         inneriter                  += j_index_end - j_index_start;
1126
1127         /* Outer loop uses 18 flops */
1128     }
1129
1130     /* Increment number of outer iterations */
1131     outeriter        += nri;
1132
1133     /* Update outer/inner flops */
1134
1135     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1136 }