src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_double/nb_kernel_ElecEw_VdwLJ_GeomW4P1_sse2_double.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, 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 sse2_double kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_double.h"
  48 #include "kernelutil_x86_sse2_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_sse2_double
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB;
  75     int              j_coord_offsetA,j_coord_offsetB;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  80     int              vdwioffset0;
  81     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  82     int              vdwioffset1;
  83     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  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          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  94     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     int              nvdwtype;
  97     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  98     int              *vdwtype;
  99     real             *vdwparam;
 100     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 101     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 102     __m128i          ewitab;
 103     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 104     real             *ewtab;
 105     __m128d          dummy_mask,cutoff_mask;
 106     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 107     __m128d          one     = _mm_set1_pd(1.0);
 108     __m128d          two     = _mm_set1_pd(2.0);
 109     x                = xx[0];
 110     f                = ff[0];
 111
 112     nri              = nlist->nri;
 113     iinr             = nlist->iinr;
 114     jindex           = nlist->jindex;
 115     jjnr             = nlist->jjnr;
 116     shiftidx         = nlist->shift;
 117     gid              = nlist->gid;
 118     shiftvec         = fr->shift_vec[0];
 119     fshift           = fr->fshift[0];
 120     facel            = _mm_set1_pd(fr->epsfac);
 121     charge           = mdatoms->chargeA;
 122     nvdwtype         = fr->ntype;
 123     vdwparam         = fr->nbfp;
 124     vdwtype          = mdatoms->typeA;
 125
 126     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 127     ewtab            = fr->ic->tabq_coul_FDV0;
 128     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 129     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 130
 131     /* Setup water-specific parameters */
 132     inr              = nlist->iinr[0];
 133     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 134     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 135     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 136     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 137
 138     /* Avoid stupid compiler warnings */
 139     jnrA = jnrB = 0;
 140     j_coord_offsetA = 0;
 141     j_coord_offsetB = 0;
 142
 143     outeriter        = 0;
 144     inneriter        = 0;
 145
 146     /* Start outer loop over neighborlists */
 147     for(iidx=0; iidx<nri; iidx++)
 148     {
 149         /* Load shift vector for this list */
 150         i_shift_offset   = DIM*shiftidx[iidx];
 151
 152         /* Load limits for loop over neighbors */
 153         j_index_start    = jindex[iidx];
 154         j_index_end      = jindex[iidx+1];
 155
 156         /* Get outer coordinate index */
 157         inr              = iinr[iidx];
 158         i_coord_offset   = DIM*inr;
 159
 160         /* Load i particle coords and add shift vector */
 161         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 162                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 163
 164         fix0             = _mm_setzero_pd();
 165         fiy0             = _mm_setzero_pd();
 166         fiz0             = _mm_setzero_pd();
 167         fix1             = _mm_setzero_pd();
 168         fiy1             = _mm_setzero_pd();
 169         fiz1             = _mm_setzero_pd();
 170         fix2             = _mm_setzero_pd();
 171         fiy2             = _mm_setzero_pd();
 172         fiz2             = _mm_setzero_pd();
 173         fix3             = _mm_setzero_pd();
 174         fiy3             = _mm_setzero_pd();
 175         fiz3             = _mm_setzero_pd();
 176
 177         /* Reset potential sums */
 178         velecsum         = _mm_setzero_pd();
 179         vvdwsum          = _mm_setzero_pd();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             j_coord_offsetA  = DIM*jnrA;
 189             j_coord_offsetB  = DIM*jnrB;
 190
 191             /* load j atom coordinates */
 192             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_pd(ix0,jx0);
 197             dy00             = _mm_sub_pd(iy0,jy0);
 198             dz00             = _mm_sub_pd(iz0,jz0);
 199             dx10             = _mm_sub_pd(ix1,jx0);
 200             dy10             = _mm_sub_pd(iy1,jy0);
 201             dz10             = _mm_sub_pd(iz1,jz0);
 202             dx20             = _mm_sub_pd(ix2,jx0);
 203             dy20             = _mm_sub_pd(iy2,jy0);
 204             dz20             = _mm_sub_pd(iz2,jz0);
 205             dx30             = _mm_sub_pd(ix3,jx0);
 206             dy30             = _mm_sub_pd(iy3,jy0);
 207             dz30             = _mm_sub_pd(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 214
 215             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 216             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 217             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 218
 219             rinvsq00         = gmx_mm_inv_pd(rsq00);
 220             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 221             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 222             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228
 229             fjx0             = _mm_setzero_pd();
 230             fjy0             = _mm_setzero_pd();
 231             fjz0             = _mm_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 240
 241             /* LENNARD-JONES DISPERSION/REPULSION */
 242
 243             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 244             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 245             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 246             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 247             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 251
 252             fscal            = fvdw;
 253
 254             /* Calculate temporary vectorial force */
 255             tx               = _mm_mul_pd(fscal,dx00);
 256             ty               = _mm_mul_pd(fscal,dy00);
 257             tz               = _mm_mul_pd(fscal,dz00);
 258
 259             /* Update vectorial force */
 260             fix0             = _mm_add_pd(fix0,tx);
 261             fiy0             = _mm_add_pd(fiy0,ty);
 262             fiz0             = _mm_add_pd(fiz0,tz);
 263
 264             fjx0             = _mm_add_pd(fjx0,tx);
 265             fjy0             = _mm_add_pd(fjy0,ty);
 266             fjz0             = _mm_add_pd(fjz0,tz);
 267
 268             /**************************
 269              * CALCULATE INTERACTIONS *
 270              **************************/
 271
 272             r10              = _mm_mul_pd(rsq10,rinv10);
 273
 274             /* Compute parameters for interactions between i and j atoms */
 275             qq10             = _mm_mul_pd(iq1,jq0);
 276
 277             /* EWALD ELECTROSTATICS */
 278
 279             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 280             ewrt             = _mm_mul_pd(r10,ewtabscale);
 281             ewitab           = _mm_cvttpd_epi32(ewrt);
 282             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 283             ewitab           = _mm_slli_epi32(ewitab,2);
 284             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 285             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 286             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 287             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 288             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 289             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 290             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 291             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 292             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 293             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 294
 295             /* Update potential sum for this i atom from the interaction with this j atom. */
 296             velecsum         = _mm_add_pd(velecsum,velec);
 297
 298             fscal            = felec;
 299
 300             /* Calculate temporary vectorial force */
 301             tx               = _mm_mul_pd(fscal,dx10);
 302             ty               = _mm_mul_pd(fscal,dy10);
 303             tz               = _mm_mul_pd(fscal,dz10);
 304
 305             /* Update vectorial force */
 306             fix1             = _mm_add_pd(fix1,tx);
 307             fiy1             = _mm_add_pd(fiy1,ty);
 308             fiz1             = _mm_add_pd(fiz1,tz);
 309
 310             fjx0             = _mm_add_pd(fjx0,tx);
 311             fjy0             = _mm_add_pd(fjy0,ty);
 312             fjz0             = _mm_add_pd(fjz0,tz);
 313
 314             /**************************
 315              * CALCULATE INTERACTIONS *
 316              **************************/
 317
 318             r20              = _mm_mul_pd(rsq20,rinv20);
 319
 320             /* Compute parameters for interactions between i and j atoms */
 321             qq20             = _mm_mul_pd(iq2,jq0);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm_mul_pd(r20,ewtabscale);
 327             ewitab           = _mm_cvttpd_epi32(ewrt);
 328             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 329             ewitab           = _mm_slli_epi32(ewitab,2);
 330             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 331             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 332             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 333             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 334             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 335             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 336             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 337             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 338             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 339             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_pd(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_pd(fscal,dx20);
 348             ty               = _mm_mul_pd(fscal,dy20);
 349             tz               = _mm_mul_pd(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_pd(fix2,tx);
 353             fiy2             = _mm_add_pd(fiy2,ty);
 354             fiz2             = _mm_add_pd(fiz2,tz);
 355
 356             fjx0             = _mm_add_pd(fjx0,tx);
 357             fjy0             = _mm_add_pd(fjy0,ty);
 358             fjz0             = _mm_add_pd(fjz0,tz);
 359
 360             /**************************
 361              * CALCULATE INTERACTIONS *
 362              **************************/
 363
 364             r30              = _mm_mul_pd(rsq30,rinv30);
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq30             = _mm_mul_pd(iq3,jq0);
 368
 369             /* EWALD ELECTROSTATICS */
 370
 371             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 372             ewrt             = _mm_mul_pd(r30,ewtabscale);
 373             ewitab           = _mm_cvttpd_epi32(ewrt);
 374             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 375             ewitab           = _mm_slli_epi32(ewitab,2);
 376             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 377             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 378             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 379             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 380             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 381             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 382             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 383             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 384             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 385             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 386
 387             /* Update potential sum for this i atom from the interaction with this j atom. */
 388             velecsum         = _mm_add_pd(velecsum,velec);
 389
 390             fscal            = felec;
 391
 392             /* Calculate temporary vectorial force */
 393             tx               = _mm_mul_pd(fscal,dx30);
 394             ty               = _mm_mul_pd(fscal,dy30);
 395             tz               = _mm_mul_pd(fscal,dz30);
 396
 397             /* Update vectorial force */
 398             fix3             = _mm_add_pd(fix3,tx);
 399             fiy3             = _mm_add_pd(fiy3,ty);
 400             fiz3             = _mm_add_pd(fiz3,tz);
 401
 402             fjx0             = _mm_add_pd(fjx0,tx);
 403             fjy0             = _mm_add_pd(fjy0,ty);
 404             fjz0             = _mm_add_pd(fjz0,tz);
 405
 406             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 407
 408             /* Inner loop uses 158 flops */
 409         }
 410
 411         if(jidx<j_index_end)
 412         {
 413
 414             jnrA             = jjnr[jidx];
 415             j_coord_offsetA  = DIM*jnrA;
 416
 417             /* load j atom coordinates */
 418             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 419                                               &jx0,&jy0,&jz0);
 420
 421             /* Calculate displacement vector */
 422             dx00             = _mm_sub_pd(ix0,jx0);
 423             dy00             = _mm_sub_pd(iy0,jy0);
 424             dz00             = _mm_sub_pd(iz0,jz0);
 425             dx10             = _mm_sub_pd(ix1,jx0);
 426             dy10             = _mm_sub_pd(iy1,jy0);
 427             dz10             = _mm_sub_pd(iz1,jz0);
 428             dx20             = _mm_sub_pd(ix2,jx0);
 429             dy20             = _mm_sub_pd(iy2,jy0);
 430             dz20             = _mm_sub_pd(iz2,jz0);
 431             dx30             = _mm_sub_pd(ix3,jx0);
 432             dy30             = _mm_sub_pd(iy3,jy0);
 433             dz30             = _mm_sub_pd(iz3,jz0);
 434
 435             /* Calculate squared distance and things based on it */
 436             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 437             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 438             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 439             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 440
 441             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 442             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 443             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 444
 445             rinvsq00         = gmx_mm_inv_pd(rsq00);
 446             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 447             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 448             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 449
 450             /* Load parameters for j particles */
 451             jq0              = _mm_load_sd(charge+jnrA+0);
 452             vdwjidx0A        = 2*vdwtype[jnrA+0];
 453
 454             fjx0             = _mm_setzero_pd();
 455             fjy0             = _mm_setzero_pd();
 456             fjz0             = _mm_setzero_pd();
 457
 458             /**************************
 459              * CALCULATE INTERACTIONS *
 460              **************************/
 461
 462             /* Compute parameters for interactions between i and j atoms */
 463             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 464
 465             /* LENNARD-JONES DISPERSION/REPULSION */
 466
 467             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 468             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 469             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 470             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 471             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 472
 473             /* Update potential sum for this i atom from the interaction with this j atom. */
 474             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 475             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 476
 477             fscal            = fvdw;
 478
 479             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 480
 481             /* Calculate temporary vectorial force */
 482             tx               = _mm_mul_pd(fscal,dx00);
 483             ty               = _mm_mul_pd(fscal,dy00);
 484             tz               = _mm_mul_pd(fscal,dz00);
 485
 486             /* Update vectorial force */
 487             fix0             = _mm_add_pd(fix0,tx);
 488             fiy0             = _mm_add_pd(fiy0,ty);
 489             fiz0             = _mm_add_pd(fiz0,tz);
 490
 491             fjx0             = _mm_add_pd(fjx0,tx);
 492             fjy0             = _mm_add_pd(fjy0,ty);
 493             fjz0             = _mm_add_pd(fjz0,tz);
 494
 495             /**************************
 496              * CALCULATE INTERACTIONS *
 497              **************************/
 498
 499             r10              = _mm_mul_pd(rsq10,rinv10);
 500
 501             /* Compute parameters for interactions between i and j atoms */
 502             qq10             = _mm_mul_pd(iq1,jq0);
 503
 504             /* EWALD ELECTROSTATICS */
 505
 506             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 507             ewrt             = _mm_mul_pd(r10,ewtabscale);
 508             ewitab           = _mm_cvttpd_epi32(ewrt);
 509             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 510             ewitab           = _mm_slli_epi32(ewitab,2);
 511             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 512             ewtabD           = _mm_setzero_pd();
 513             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 514             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 515             ewtabFn          = _mm_setzero_pd();
 516             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 517             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 518             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 519             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 520             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 521
 522             /* Update potential sum for this i atom from the interaction with this j atom. */
 523             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 524             velecsum         = _mm_add_pd(velecsum,velec);
 525
 526             fscal            = felec;
 527
 528             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 529
 530             /* Calculate temporary vectorial force */
 531             tx               = _mm_mul_pd(fscal,dx10);
 532             ty               = _mm_mul_pd(fscal,dy10);
 533             tz               = _mm_mul_pd(fscal,dz10);
 534
 535             /* Update vectorial force */
 536             fix1             = _mm_add_pd(fix1,tx);
 537             fiy1             = _mm_add_pd(fiy1,ty);
 538             fiz1             = _mm_add_pd(fiz1,tz);
 539
 540             fjx0             = _mm_add_pd(fjx0,tx);
 541             fjy0             = _mm_add_pd(fjy0,ty);
 542             fjz0             = _mm_add_pd(fjz0,tz);
 543
 544             /**************************
 545              * CALCULATE INTERACTIONS *
 546              **************************/
 547
 548             r20              = _mm_mul_pd(rsq20,rinv20);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq20             = _mm_mul_pd(iq2,jq0);
 552
 553             /* EWALD ELECTROSTATICS */
 554
 555             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 556             ewrt             = _mm_mul_pd(r20,ewtabscale);
 557             ewitab           = _mm_cvttpd_epi32(ewrt);
 558             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 559             ewitab           = _mm_slli_epi32(ewitab,2);
 560             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 561             ewtabD           = _mm_setzero_pd();
 562             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 563             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 564             ewtabFn          = _mm_setzero_pd();
 565             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 566             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 567             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 568             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 569             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 570
 571             /* Update potential sum for this i atom from the interaction with this j atom. */
 572             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 573             velecsum         = _mm_add_pd(velecsum,velec);
 574
 575             fscal            = felec;
 576
 577             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_pd(fscal,dx20);
 581             ty               = _mm_mul_pd(fscal,dy20);
 582             tz               = _mm_mul_pd(fscal,dz20);
 583
 584             /* Update vectorial force */
 585             fix2             = _mm_add_pd(fix2,tx);
 586             fiy2             = _mm_add_pd(fiy2,ty);
 587             fiz2             = _mm_add_pd(fiz2,tz);
 588
 589             fjx0             = _mm_add_pd(fjx0,tx);
 590             fjy0             = _mm_add_pd(fjy0,ty);
 591             fjz0             = _mm_add_pd(fjz0,tz);
 592
 593             /**************************
 594              * CALCULATE INTERACTIONS *
 595              **************************/
 596
 597             r30              = _mm_mul_pd(rsq30,rinv30);
 598
 599             /* Compute parameters for interactions between i and j atoms */
 600             qq30             = _mm_mul_pd(iq3,jq0);
 601
 602             /* EWALD ELECTROSTATICS */
 603
 604             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 605             ewrt             = _mm_mul_pd(r30,ewtabscale);
 606             ewitab           = _mm_cvttpd_epi32(ewrt);
 607             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 608             ewitab           = _mm_slli_epi32(ewitab,2);
 609             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 610             ewtabD           = _mm_setzero_pd();
 611             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 612             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 613             ewtabFn          = _mm_setzero_pd();
 614             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 615             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 616             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 617             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 618             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 619
 620             /* Update potential sum for this i atom from the interaction with this j atom. */
 621             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 622             velecsum         = _mm_add_pd(velecsum,velec);
 623
 624             fscal            = felec;
 625
 626             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 627
 628             /* Calculate temporary vectorial force */
 629             tx               = _mm_mul_pd(fscal,dx30);
 630             ty               = _mm_mul_pd(fscal,dy30);
 631             tz               = _mm_mul_pd(fscal,dz30);
 632
 633             /* Update vectorial force */
 634             fix3             = _mm_add_pd(fix3,tx);
 635             fiy3             = _mm_add_pd(fiy3,ty);
 636             fiz3             = _mm_add_pd(fiz3,tz);
 637
 638             fjx0             = _mm_add_pd(fjx0,tx);
 639             fjy0             = _mm_add_pd(fjy0,ty);
 640             fjz0             = _mm_add_pd(fjz0,tz);
 641
 642             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 643
 644             /* Inner loop uses 158 flops */
 645         }
 646
 647         /* End of innermost loop */
 648
 649         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 650                                               f+i_coord_offset,fshift+i_shift_offset);
 651
 652         ggid                        = gid[iidx];
 653         /* Update potential energies */
 654         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 655         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 656
 657         /* Increment number of inner iterations */
 658         inneriter                  += j_index_end - j_index_start;
 659
 660         /* Outer loop uses 26 flops */
 661     }
 662
 663     /* Increment number of outer iterations */
 664     outeriter        += nri;
 665
 666     /* Update outer/inner flops */
 667
 668     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*158);
 669 }
 670 /*
 671  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_double
 672  * Electrostatics interaction: Ewald
 673  * VdW interaction:            LennardJones
 674  * Geometry:                   Water4-Particle
 675  * Calculate force/pot:        Force
 676  */
 677 void
 678 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_double
 679                     (t_nblist                    * gmx_restrict       nlist,
 680                      rvec                        * gmx_restrict          xx,
 681                      rvec                        * gmx_restrict          ff,
 682                      t_forcerec                  * gmx_restrict          fr,
 683                      t_mdatoms                   * gmx_restrict     mdatoms,
 684                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 685                      t_nrnb                      * gmx_restrict        nrnb)
 686 {
 687     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 688      * just 0 for non-waters.
 689      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 690      * jnr indices corresponding to data put in the four positions in the SIMD register.
 691      */
 692     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 693     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 694     int              jnrA,jnrB;
 695     int              j_coord_offsetA,j_coord_offsetB;
 696     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 697     real             rcutoff_scalar;
 698     real             *shiftvec,*fshift,*x,*f;
 699     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 700     int              vdwioffset0;
 701     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 702     int              vdwioffset1;
 703     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 704     int              vdwioffset2;
 705     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 706     int              vdwioffset3;
 707     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 708     int              vdwjidx0A,vdwjidx0B;
 709     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 710     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 711     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 712     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 713     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 714     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 715     real             *charge;
 716     int              nvdwtype;
 717     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 718     int              *vdwtype;
 719     real             *vdwparam;
 720     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 721     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 722     __m128i          ewitab;
 723     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 724     real             *ewtab;
 725     __m128d          dummy_mask,cutoff_mask;
 726     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 727     __m128d          one     = _mm_set1_pd(1.0);
 728     __m128d          two     = _mm_set1_pd(2.0);
 729     x                = xx[0];
 730     f                = ff[0];
 731
 732     nri              = nlist->nri;
 733     iinr             = nlist->iinr;
 734     jindex           = nlist->jindex;
 735     jjnr             = nlist->jjnr;
 736     shiftidx         = nlist->shift;
 737     gid              = nlist->gid;
 738     shiftvec         = fr->shift_vec[0];
 739     fshift           = fr->fshift[0];
 740     facel            = _mm_set1_pd(fr->epsfac);
 741     charge           = mdatoms->chargeA;
 742     nvdwtype         = fr->ntype;
 743     vdwparam         = fr->nbfp;
 744     vdwtype          = mdatoms->typeA;
 745
 746     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 747     ewtab            = fr->ic->tabq_coul_F;
 748     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 749     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 750
 751     /* Setup water-specific parameters */
 752     inr              = nlist->iinr[0];
 753     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 754     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 755     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 756     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 757
 758     /* Avoid stupid compiler warnings */
 759     jnrA = jnrB = 0;
 760     j_coord_offsetA = 0;
 761     j_coord_offsetB = 0;
 762
 763     outeriter        = 0;
 764     inneriter        = 0;
 765
 766     /* Start outer loop over neighborlists */
 767     for(iidx=0; iidx<nri; iidx++)
 768     {
 769         /* Load shift vector for this list */
 770         i_shift_offset   = DIM*shiftidx[iidx];
 771
 772         /* Load limits for loop over neighbors */
 773         j_index_start    = jindex[iidx];
 774         j_index_end      = jindex[iidx+1];
 775
 776         /* Get outer coordinate index */
 777         inr              = iinr[iidx];
 778         i_coord_offset   = DIM*inr;
 779
 780         /* Load i particle coords and add shift vector */
 781         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 782                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 783
 784         fix0             = _mm_setzero_pd();
 785         fiy0             = _mm_setzero_pd();
 786         fiz0             = _mm_setzero_pd();
 787         fix1             = _mm_setzero_pd();
 788         fiy1             = _mm_setzero_pd();
 789         fiz1             = _mm_setzero_pd();
 790         fix2             = _mm_setzero_pd();
 791         fiy2             = _mm_setzero_pd();
 792         fiz2             = _mm_setzero_pd();
 793         fix3             = _mm_setzero_pd();
 794         fiy3             = _mm_setzero_pd();
 795         fiz3             = _mm_setzero_pd();
 796
 797         /* Start inner kernel loop */
 798         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 799         {
 800
 801             /* Get j neighbor index, and coordinate index */
 802             jnrA             = jjnr[jidx];
 803             jnrB             = jjnr[jidx+1];
 804             j_coord_offsetA  = DIM*jnrA;
 805             j_coord_offsetB  = DIM*jnrB;
 806
 807             /* load j atom coordinates */
 808             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 809                                               &jx0,&jy0,&jz0);
 810
 811             /* Calculate displacement vector */
 812             dx00             = _mm_sub_pd(ix0,jx0);
 813             dy00             = _mm_sub_pd(iy0,jy0);
 814             dz00             = _mm_sub_pd(iz0,jz0);
 815             dx10             = _mm_sub_pd(ix1,jx0);
 816             dy10             = _mm_sub_pd(iy1,jy0);
 817             dz10             = _mm_sub_pd(iz1,jz0);
 818             dx20             = _mm_sub_pd(ix2,jx0);
 819             dy20             = _mm_sub_pd(iy2,jy0);
 820             dz20             = _mm_sub_pd(iz2,jz0);
 821             dx30             = _mm_sub_pd(ix3,jx0);
 822             dy30             = _mm_sub_pd(iy3,jy0);
 823             dz30             = _mm_sub_pd(iz3,jz0);
 824
 825             /* Calculate squared distance and things based on it */
 826             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 827             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 828             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 829             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 830
 831             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 832             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 833             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 834
 835             rinvsq00         = gmx_mm_inv_pd(rsq00);
 836             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 837             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 838             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 839
 840             /* Load parameters for j particles */
 841             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 842             vdwjidx0A        = 2*vdwtype[jnrA+0];
 843             vdwjidx0B        = 2*vdwtype[jnrB+0];
 844
 845             fjx0             = _mm_setzero_pd();
 846             fjy0             = _mm_setzero_pd();
 847             fjz0             = _mm_setzero_pd();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             /* Compute parameters for interactions between i and j atoms */
 854             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 855                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 856
 857             /* LENNARD-JONES DISPERSION/REPULSION */
 858
 859             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 860             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 861
 862             fscal            = fvdw;
 863
 864             /* Calculate temporary vectorial force */
 865             tx               = _mm_mul_pd(fscal,dx00);
 866             ty               = _mm_mul_pd(fscal,dy00);
 867             tz               = _mm_mul_pd(fscal,dz00);
 868
 869             /* Update vectorial force */
 870             fix0             = _mm_add_pd(fix0,tx);
 871             fiy0             = _mm_add_pd(fiy0,ty);
 872             fiz0             = _mm_add_pd(fiz0,tz);
 873
 874             fjx0             = _mm_add_pd(fjx0,tx);
 875             fjy0             = _mm_add_pd(fjy0,ty);
 876             fjz0             = _mm_add_pd(fjz0,tz);
 877
 878             /**************************
 879              * CALCULATE INTERACTIONS *
 880              **************************/
 881
 882             r10              = _mm_mul_pd(rsq10,rinv10);
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq10             = _mm_mul_pd(iq1,jq0);
 886
 887             /* EWALD ELECTROSTATICS */
 888
 889             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 890             ewrt             = _mm_mul_pd(r10,ewtabscale);
 891             ewitab           = _mm_cvttpd_epi32(ewrt);
 892             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 893             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 894                                          &ewtabF,&ewtabFn);
 895             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 896             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 897
 898             fscal            = felec;
 899
 900             /* Calculate temporary vectorial force */
 901             tx               = _mm_mul_pd(fscal,dx10);
 902             ty               = _mm_mul_pd(fscal,dy10);
 903             tz               = _mm_mul_pd(fscal,dz10);
 904
 905             /* Update vectorial force */
 906             fix1             = _mm_add_pd(fix1,tx);
 907             fiy1             = _mm_add_pd(fiy1,ty);
 908             fiz1             = _mm_add_pd(fiz1,tz);
 909
 910             fjx0             = _mm_add_pd(fjx0,tx);
 911             fjy0             = _mm_add_pd(fjy0,ty);
 912             fjz0             = _mm_add_pd(fjz0,tz);
 913
 914             /**************************
 915              * CALCULATE INTERACTIONS *
 916              **************************/
 917
 918             r20              = _mm_mul_pd(rsq20,rinv20);
 919
 920             /* Compute parameters for interactions between i and j atoms */
 921             qq20             = _mm_mul_pd(iq2,jq0);
 922
 923             /* EWALD ELECTROSTATICS */
 924
 925             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 926             ewrt             = _mm_mul_pd(r20,ewtabscale);
 927             ewitab           = _mm_cvttpd_epi32(ewrt);
 928             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 929             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 930                                          &ewtabF,&ewtabFn);
 931             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 932             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 933
 934             fscal            = felec;
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm_mul_pd(fscal,dx20);
 938             ty               = _mm_mul_pd(fscal,dy20);
 939             tz               = _mm_mul_pd(fscal,dz20);
 940
 941             /* Update vectorial force */
 942             fix2             = _mm_add_pd(fix2,tx);
 943             fiy2             = _mm_add_pd(fiy2,ty);
 944             fiz2             = _mm_add_pd(fiz2,tz);
 945
 946             fjx0             = _mm_add_pd(fjx0,tx);
 947             fjy0             = _mm_add_pd(fjy0,ty);
 948             fjz0             = _mm_add_pd(fjz0,tz);
 949
 950             /**************************
 951              * CALCULATE INTERACTIONS *
 952              **************************/
 953
 954             r30              = _mm_mul_pd(rsq30,rinv30);
 955
 956             /* Compute parameters for interactions between i and j atoms */
 957             qq30             = _mm_mul_pd(iq3,jq0);
 958
 959             /* EWALD ELECTROSTATICS */
 960
 961             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 962             ewrt             = _mm_mul_pd(r30,ewtabscale);
 963             ewitab           = _mm_cvttpd_epi32(ewrt);
 964             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
 965             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 966                                          &ewtabF,&ewtabFn);
 967             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 968             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 969
 970             fscal            = felec;
 971
 972             /* Calculate temporary vectorial force */
 973             tx               = _mm_mul_pd(fscal,dx30);
 974             ty               = _mm_mul_pd(fscal,dy30);
 975             tz               = _mm_mul_pd(fscal,dz30);
 976
 977             /* Update vectorial force */
 978             fix3             = _mm_add_pd(fix3,tx);
 979             fiy3             = _mm_add_pd(fiy3,ty);
 980             fiz3             = _mm_add_pd(fiz3,tz);
 981
 982             fjx0             = _mm_add_pd(fjx0,tx);
 983             fjy0             = _mm_add_pd(fjy0,ty);
 984             fjz0             = _mm_add_pd(fjz0,tz);
 985
 986             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 987
 988             /* Inner loop uses 138 flops */
 989         }
 990
 991         if(jidx<j_index_end)
 992         {
 993
 994             jnrA             = jjnr[jidx];
 995             j_coord_offsetA  = DIM*jnrA;
 996
 997             /* load j atom coordinates */
 998             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 999                                               &jx0,&jy0,&jz0);
1000
1001             /* Calculate displacement vector */
1002             dx00             = _mm_sub_pd(ix0,jx0);
1003             dy00             = _mm_sub_pd(iy0,jy0);
1004             dz00             = _mm_sub_pd(iz0,jz0);
1005             dx10             = _mm_sub_pd(ix1,jx0);
1006             dy10             = _mm_sub_pd(iy1,jy0);
1007             dz10             = _mm_sub_pd(iz1,jz0);
1008             dx20             = _mm_sub_pd(ix2,jx0);
1009             dy20             = _mm_sub_pd(iy2,jy0);
1010             dz20             = _mm_sub_pd(iz2,jz0);
1011             dx30             = _mm_sub_pd(ix3,jx0);
1012             dy30             = _mm_sub_pd(iy3,jy0);
1013             dz30             = _mm_sub_pd(iz3,jz0);
1014
1015             /* Calculate squared distance and things based on it */
1016             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1017             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1018             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1019             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1020
1021             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1022             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1023             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1024
1025             rinvsq00         = gmx_mm_inv_pd(rsq00);
1026             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1027             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1028             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1029
1030             /* Load parameters for j particles */
1031             jq0              = _mm_load_sd(charge+jnrA+0);
1032             vdwjidx0A        = 2*vdwtype[jnrA+0];
1033
1034             fjx0             = _mm_setzero_pd();
1035             fjy0             = _mm_setzero_pd();
1036             fjz0             = _mm_setzero_pd();
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1044
1045             /* LENNARD-JONES DISPERSION/REPULSION */
1046
1047             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1048             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1049
1050             fscal            = fvdw;
1051
1052             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1053
1054             /* Calculate temporary vectorial force */
1055             tx               = _mm_mul_pd(fscal,dx00);
1056             ty               = _mm_mul_pd(fscal,dy00);
1057             tz               = _mm_mul_pd(fscal,dz00);
1058
1059             /* Update vectorial force */
1060             fix0             = _mm_add_pd(fix0,tx);
1061             fiy0             = _mm_add_pd(fiy0,ty);
1062             fiz0             = _mm_add_pd(fiz0,tz);
1063
1064             fjx0             = _mm_add_pd(fjx0,tx);
1065             fjy0             = _mm_add_pd(fjy0,ty);
1066             fjz0             = _mm_add_pd(fjz0,tz);
1067
1068             /**************************
1069              * CALCULATE INTERACTIONS *
1070              **************************/
1071
1072             r10              = _mm_mul_pd(rsq10,rinv10);
1073
1074             /* Compute parameters for interactions between i and j atoms */
1075             qq10             = _mm_mul_pd(iq1,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(r10,ewtabscale);
1081             ewitab           = _mm_cvttpd_epi32(ewrt);
1082             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
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(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1086
1087             fscal            = felec;
1088
1089             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1090
1091             /* Calculate temporary vectorial force */
1092             tx               = _mm_mul_pd(fscal,dx10);
1093             ty               = _mm_mul_pd(fscal,dy10);
1094             tz               = _mm_mul_pd(fscal,dz10);
1095
1096             /* Update vectorial force */
1097             fix1             = _mm_add_pd(fix1,tx);
1098             fiy1             = _mm_add_pd(fiy1,ty);
1099             fiz1             = _mm_add_pd(fiz1,tz);
1100
1101             fjx0             = _mm_add_pd(fjx0,tx);
1102             fjy0             = _mm_add_pd(fjy0,ty);
1103             fjz0             = _mm_add_pd(fjz0,tz);
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             r20              = _mm_mul_pd(rsq20,rinv20);
1110
1111             /* Compute parameters for interactions between i and j atoms */
1112             qq20             = _mm_mul_pd(iq2,jq0);
1113
1114             /* EWALD ELECTROSTATICS */
1115
1116             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1117             ewrt             = _mm_mul_pd(r20,ewtabscale);
1118             ewitab           = _mm_cvttpd_epi32(ewrt);
1119             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1120             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1121             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1122             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1123
1124             fscal            = felec;
1125
1126             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1127
1128             /* Calculate temporary vectorial force */
1129             tx               = _mm_mul_pd(fscal,dx20);
1130             ty               = _mm_mul_pd(fscal,dy20);
1131             tz               = _mm_mul_pd(fscal,dz20);
1132
1133             /* Update vectorial force */
1134             fix2             = _mm_add_pd(fix2,tx);
1135             fiy2             = _mm_add_pd(fiy2,ty);
1136             fiz2             = _mm_add_pd(fiz2,tz);
1137
1138             fjx0             = _mm_add_pd(fjx0,tx);
1139             fjy0             = _mm_add_pd(fjy0,ty);
1140             fjz0             = _mm_add_pd(fjz0,tz);
1141
1142             /**************************
1143              * CALCULATE INTERACTIONS *
1144              **************************/
1145
1146             r30              = _mm_mul_pd(rsq30,rinv30);
1147
1148             /* Compute parameters for interactions between i and j atoms */
1149             qq30             = _mm_mul_pd(iq3,jq0);
1150
1151             /* EWALD ELECTROSTATICS */
1152
1153             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1154             ewrt             = _mm_mul_pd(r30,ewtabscale);
1155             ewitab           = _mm_cvttpd_epi32(ewrt);
1156             eweps            = _mm_sub_pd(ewrt,_mm_cvtepi32_pd(ewitab));
1157             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1158             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
1159             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1160
1161             fscal            = felec;
1162
1163             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1164
1165             /* Calculate temporary vectorial force */
1166             tx               = _mm_mul_pd(fscal,dx30);
1167             ty               = _mm_mul_pd(fscal,dy30);
1168             tz               = _mm_mul_pd(fscal,dz30);
1169
1170             /* Update vectorial force */
1171             fix3             = _mm_add_pd(fix3,tx);
1172             fiy3             = _mm_add_pd(fiy3,ty);
1173             fiz3             = _mm_add_pd(fiz3,tz);
1174
1175             fjx0             = _mm_add_pd(fjx0,tx);
1176             fjy0             = _mm_add_pd(fjy0,ty);
1177             fjz0             = _mm_add_pd(fjz0,tz);
1178
1179             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1180
1181             /* Inner loop uses 138 flops */
1182         }
1183
1184         /* End of innermost loop */
1185
1186         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1187                                               f+i_coord_offset,fshift+i_shift_offset);
1188
1189         /* Increment number of inner iterations */
1190         inneriter                  += j_index_end - j_index_start;
1191
1192         /* Outer loop uses 24 flops */
1193     }
1194
1195     /* Increment number of outer iterations */
1196     outeriter        += nri;
1197
1198     /* Update outer/inner flops */
1199
1200     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*138);
1201 }