src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_avx_128_fma_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 avx_128_fma_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_avx_128_fma_double.h"
  48 #include "kernelutil_x86_avx_128_fma_double.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Water4-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_VF_avx_128_fma_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,twoeweps,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     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 139     rcutoff_scalar   = fr->rcoulomb;
 140     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 141     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 142
 143     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 144     rvdw             = _mm_set1_pd(fr->rvdw);
 145
 146     /* Avoid stupid compiler warnings */
 147     jnrA = jnrB = 0;
 148     j_coord_offsetA = 0;
 149     j_coord_offsetB = 0;
 150
 151     outeriter        = 0;
 152     inneriter        = 0;
 153
 154     /* Start outer loop over neighborlists */
 155     for(iidx=0; iidx<nri; iidx++)
 156     {
 157         /* Load shift vector for this list */
 158         i_shift_offset   = DIM*shiftidx[iidx];
 159
 160         /* Load limits for loop over neighbors */
 161         j_index_start    = jindex[iidx];
 162         j_index_end      = jindex[iidx+1];
 163
 164         /* Get outer coordinate index */
 165         inr              = iinr[iidx];
 166         i_coord_offset   = DIM*inr;
 167
 168         /* Load i particle coords and add shift vector */
 169         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 171
 172         fix0             = _mm_setzero_pd();
 173         fiy0             = _mm_setzero_pd();
 174         fiz0             = _mm_setzero_pd();
 175         fix1             = _mm_setzero_pd();
 176         fiy1             = _mm_setzero_pd();
 177         fiz1             = _mm_setzero_pd();
 178         fix2             = _mm_setzero_pd();
 179         fiy2             = _mm_setzero_pd();
 180         fiz2             = _mm_setzero_pd();
 181         fix3             = _mm_setzero_pd();
 182         fiy3             = _mm_setzero_pd();
 183         fiz3             = _mm_setzero_pd();
 184
 185         /* Reset potential sums */
 186         velecsum         = _mm_setzero_pd();
 187         vvdwsum          = _mm_setzero_pd();
 188
 189         /* Start inner kernel loop */
 190         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 191         {
 192
 193             /* Get j neighbor index, and coordinate index */
 194             jnrA             = jjnr[jidx];
 195             jnrB             = jjnr[jidx+1];
 196             j_coord_offsetA  = DIM*jnrA;
 197             j_coord_offsetB  = DIM*jnrB;
 198
 199             /* load j atom coordinates */
 200             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 201                                               &jx0,&jy0,&jz0);
 202
 203             /* Calculate displacement vector */
 204             dx00             = _mm_sub_pd(ix0,jx0);
 205             dy00             = _mm_sub_pd(iy0,jy0);
 206             dz00             = _mm_sub_pd(iz0,jz0);
 207             dx10             = _mm_sub_pd(ix1,jx0);
 208             dy10             = _mm_sub_pd(iy1,jy0);
 209             dz10             = _mm_sub_pd(iz1,jz0);
 210             dx20             = _mm_sub_pd(ix2,jx0);
 211             dy20             = _mm_sub_pd(iy2,jy0);
 212             dz20             = _mm_sub_pd(iz2,jz0);
 213             dx30             = _mm_sub_pd(ix3,jx0);
 214             dy30             = _mm_sub_pd(iy3,jy0);
 215             dz30             = _mm_sub_pd(iz3,jz0);
 216
 217             /* Calculate squared distance and things based on it */
 218             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 219             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 220             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 221             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 222
 223             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 224             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 225             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 226
 227             rinvsq00         = gmx_mm_inv_pd(rsq00);
 228             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 229             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 230             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 231
 232             /* Load parameters for j particles */
 233             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236
 237             fjx0             = _mm_setzero_pd();
 238             fjy0             = _mm_setzero_pd();
 239             fjz0             = _mm_setzero_pd();
 240
 241             /**************************
 242              * CALCULATE INTERACTIONS *
 243              **************************/
 244
 245             if (gmx_mm_any_lt(rsq00,rcutoff2))
 246             {
 247
 248             /* Compute parameters for interactions between i and j atoms */
 249             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 250                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 251
 252             /* LENNARD-JONES DISPERSION/REPULSION */
 253
 254             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 255             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 256             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 257             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 258                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 259             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 260
 261             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 265             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 266
 267             fscal            = fvdw;
 268
 269             fscal            = _mm_and_pd(fscal,cutoff_mask);
 270
 271             /* Update vectorial force */
 272             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 273             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 274             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 275
 276             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 277             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 278             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 279
 280             }
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             if (gmx_mm_any_lt(rsq10,rcutoff2))
 287             {
 288
 289             r10              = _mm_mul_pd(rsq10,rinv10);
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_pd(iq1,jq0);
 293
 294             /* EWALD ELECTROSTATICS */
 295
 296             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 297             ewrt             = _mm_mul_pd(r10,ewtabscale);
 298             ewitab           = _mm_cvttpd_epi32(ewrt);
 299 #ifdef __XOP__
 300             eweps            = _mm_frcz_pd(ewrt);
 301 #else
 302             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 303 #endif
 304             twoeweps         = _mm_add_pd(eweps,eweps);
 305             ewitab           = _mm_slli_epi32(ewitab,2);
 306             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 307             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 308             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 309             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 310             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 311             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 312             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 313             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 314             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 315             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 316
 317             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_and_pd(velec,cutoff_mask);
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_pd(fscal,cutoff_mask);
 326
 327             /* Update vectorial force */
 328             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 329             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 330             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 331
 332             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 333             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 334             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 335
 336             }
 337
 338             /**************************
 339              * CALCULATE INTERACTIONS *
 340              **************************/
 341
 342             if (gmx_mm_any_lt(rsq20,rcutoff2))
 343             {
 344
 345             r20              = _mm_mul_pd(rsq20,rinv20);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq20             = _mm_mul_pd(iq2,jq0);
 349
 350             /* EWALD ELECTROSTATICS */
 351
 352             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 353             ewrt             = _mm_mul_pd(r20,ewtabscale);
 354             ewitab           = _mm_cvttpd_epi32(ewrt);
 355 #ifdef __XOP__
 356             eweps            = _mm_frcz_pd(ewrt);
 357 #else
 358             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 359 #endif
 360             twoeweps         = _mm_add_pd(eweps,eweps);
 361             ewitab           = _mm_slli_epi32(ewitab,2);
 362             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 363             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 364             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 365             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 366             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 367             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 368             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 369             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 370             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 371             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 372
 373             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_and_pd(velec,cutoff_mask);
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm_and_pd(fscal,cutoff_mask);
 382
 383             /* Update vectorial force */
 384             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 385             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 386             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 387
 388             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 389             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 390             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 391
 392             }
 393
 394             /**************************
 395              * CALCULATE INTERACTIONS *
 396              **************************/
 397
 398             if (gmx_mm_any_lt(rsq30,rcutoff2))
 399             {
 400
 401             r30              = _mm_mul_pd(rsq30,rinv30);
 402
 403             /* Compute parameters for interactions between i and j atoms */
 404             qq30             = _mm_mul_pd(iq3,jq0);
 405
 406             /* EWALD ELECTROSTATICS */
 407
 408             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 409             ewrt             = _mm_mul_pd(r30,ewtabscale);
 410             ewitab           = _mm_cvttpd_epi32(ewrt);
 411 #ifdef __XOP__
 412             eweps            = _mm_frcz_pd(ewrt);
 413 #else
 414             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 415 #endif
 416             twoeweps         = _mm_add_pd(eweps,eweps);
 417             ewitab           = _mm_slli_epi32(ewitab,2);
 418             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 419             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 420             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 421             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 422             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 423             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 424             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 425             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 426             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 427             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 428
 429             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 430
 431             /* Update potential sum for this i atom from the interaction with this j atom. */
 432             velec            = _mm_and_pd(velec,cutoff_mask);
 433             velecsum         = _mm_add_pd(velecsum,velec);
 434
 435             fscal            = felec;
 436
 437             fscal            = _mm_and_pd(fscal,cutoff_mask);
 438
 439             /* Update vectorial force */
 440             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 441             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 442             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 443
 444             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 445             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 446             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 447
 448             }
 449
 450             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 451
 452             /* Inner loop uses 194 flops */
 453         }
 454
 455         if(jidx<j_index_end)
 456         {
 457
 458             jnrA             = jjnr[jidx];
 459             j_coord_offsetA  = DIM*jnrA;
 460
 461             /* load j atom coordinates */
 462             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 463                                               &jx0,&jy0,&jz0);
 464
 465             /* Calculate displacement vector */
 466             dx00             = _mm_sub_pd(ix0,jx0);
 467             dy00             = _mm_sub_pd(iy0,jy0);
 468             dz00             = _mm_sub_pd(iz0,jz0);
 469             dx10             = _mm_sub_pd(ix1,jx0);
 470             dy10             = _mm_sub_pd(iy1,jy0);
 471             dz10             = _mm_sub_pd(iz1,jz0);
 472             dx20             = _mm_sub_pd(ix2,jx0);
 473             dy20             = _mm_sub_pd(iy2,jy0);
 474             dz20             = _mm_sub_pd(iz2,jz0);
 475             dx30             = _mm_sub_pd(ix3,jx0);
 476             dy30             = _mm_sub_pd(iy3,jy0);
 477             dz30             = _mm_sub_pd(iz3,jz0);
 478
 479             /* Calculate squared distance and things based on it */
 480             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 481             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 482             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 483             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 484
 485             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 486             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 487             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 488
 489             rinvsq00         = gmx_mm_inv_pd(rsq00);
 490             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 491             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 492             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 493
 494             /* Load parameters for j particles */
 495             jq0              = _mm_load_sd(charge+jnrA+0);
 496             vdwjidx0A        = 2*vdwtype[jnrA+0];
 497
 498             fjx0             = _mm_setzero_pd();
 499             fjy0             = _mm_setzero_pd();
 500             fjz0             = _mm_setzero_pd();
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             if (gmx_mm_any_lt(rsq00,rcutoff2))
 507             {
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 511
 512             /* LENNARD-JONES DISPERSION/REPULSION */
 513
 514             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 515             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 516             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 517             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 518                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 519             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 520
 521             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 522
 523             /* Update potential sum for this i atom from the interaction with this j atom. */
 524             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 525             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 526             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 527
 528             fscal            = fvdw;
 529
 530             fscal            = _mm_and_pd(fscal,cutoff_mask);
 531
 532             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 533
 534             /* Update vectorial force */
 535             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 536             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 537             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 538
 539             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 540             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 541             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 542
 543             }
 544
 545             /**************************
 546              * CALCULATE INTERACTIONS *
 547              **************************/
 548
 549             if (gmx_mm_any_lt(rsq10,rcutoff2))
 550             {
 551
 552             r10              = _mm_mul_pd(rsq10,rinv10);
 553
 554             /* Compute parameters for interactions between i and j atoms */
 555             qq10             = _mm_mul_pd(iq1,jq0);
 556
 557             /* EWALD ELECTROSTATICS */
 558
 559             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 560             ewrt             = _mm_mul_pd(r10,ewtabscale);
 561             ewitab           = _mm_cvttpd_epi32(ewrt);
 562 #ifdef __XOP__
 563             eweps            = _mm_frcz_pd(ewrt);
 564 #else
 565             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 566 #endif
 567             twoeweps         = _mm_add_pd(eweps,eweps);
 568             ewitab           = _mm_slli_epi32(ewitab,2);
 569             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 570             ewtabD           = _mm_setzero_pd();
 571             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 572             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 573             ewtabFn          = _mm_setzero_pd();
 574             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 575             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 576             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 577             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_sub_pd(rinv10,sh_ewald),velec));
 578             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 579
 580             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 581
 582             /* Update potential sum for this i atom from the interaction with this j atom. */
 583             velec            = _mm_and_pd(velec,cutoff_mask);
 584             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 585             velecsum         = _mm_add_pd(velecsum,velec);
 586
 587             fscal            = felec;
 588
 589             fscal            = _mm_and_pd(fscal,cutoff_mask);
 590
 591             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 592
 593             /* Update vectorial force */
 594             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 595             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 596             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 597
 598             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 599             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 600             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 601
 602             }
 603
 604             /**************************
 605              * CALCULATE INTERACTIONS *
 606              **************************/
 607
 608             if (gmx_mm_any_lt(rsq20,rcutoff2))
 609             {
 610
 611             r20              = _mm_mul_pd(rsq20,rinv20);
 612
 613             /* Compute parameters for interactions between i and j atoms */
 614             qq20             = _mm_mul_pd(iq2,jq0);
 615
 616             /* EWALD ELECTROSTATICS */
 617
 618             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 619             ewrt             = _mm_mul_pd(r20,ewtabscale);
 620             ewitab           = _mm_cvttpd_epi32(ewrt);
 621 #ifdef __XOP__
 622             eweps            = _mm_frcz_pd(ewrt);
 623 #else
 624             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 625 #endif
 626             twoeweps         = _mm_add_pd(eweps,eweps);
 627             ewitab           = _mm_slli_epi32(ewitab,2);
 628             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 629             ewtabD           = _mm_setzero_pd();
 630             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 631             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 632             ewtabFn          = _mm_setzero_pd();
 633             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 634             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 635             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 636             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_sub_pd(rinv20,sh_ewald),velec));
 637             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 638
 639             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 640
 641             /* Update potential sum for this i atom from the interaction with this j atom. */
 642             velec            = _mm_and_pd(velec,cutoff_mask);
 643             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 644             velecsum         = _mm_add_pd(velecsum,velec);
 645
 646             fscal            = felec;
 647
 648             fscal            = _mm_and_pd(fscal,cutoff_mask);
 649
 650             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 651
 652             /* Update vectorial force */
 653             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 654             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 655             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 656
 657             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 658             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 659             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 660
 661             }
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             if (gmx_mm_any_lt(rsq30,rcutoff2))
 668             {
 669
 670             r30              = _mm_mul_pd(rsq30,rinv30);
 671
 672             /* Compute parameters for interactions between i and j atoms */
 673             qq30             = _mm_mul_pd(iq3,jq0);
 674
 675             /* EWALD ELECTROSTATICS */
 676
 677             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 678             ewrt             = _mm_mul_pd(r30,ewtabscale);
 679             ewitab           = _mm_cvttpd_epi32(ewrt);
 680 #ifdef __XOP__
 681             eweps            = _mm_frcz_pd(ewrt);
 682 #else
 683             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 684 #endif
 685             twoeweps         = _mm_add_pd(eweps,eweps);
 686             ewitab           = _mm_slli_epi32(ewitab,2);
 687             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 688             ewtabD           = _mm_setzero_pd();
 689             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 690             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 691             ewtabFn          = _mm_setzero_pd();
 692             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 693             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 694             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 695             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_sub_pd(rinv30,sh_ewald),velec));
 696             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 697
 698             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
 699
 700             /* Update potential sum for this i atom from the interaction with this j atom. */
 701             velec            = _mm_and_pd(velec,cutoff_mask);
 702             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 703             velecsum         = _mm_add_pd(velecsum,velec);
 704
 705             fscal            = felec;
 706
 707             fscal            = _mm_and_pd(fscal,cutoff_mask);
 708
 709             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 710
 711             /* Update vectorial force */
 712             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 713             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 714             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 715
 716             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 717             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 718             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 719
 720             }
 721
 722             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 723
 724             /* Inner loop uses 194 flops */
 725         }
 726
 727         /* End of innermost loop */
 728
 729         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 730                                               f+i_coord_offset,fshift+i_shift_offset);
 731
 732         ggid                        = gid[iidx];
 733         /* Update potential energies */
 734         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 735         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 736
 737         /* Increment number of inner iterations */
 738         inneriter                  += j_index_end - j_index_start;
 739
 740         /* Outer loop uses 26 flops */
 741     }
 742
 743     /* Increment number of outer iterations */
 744     outeriter        += nri;
 745
 746     /* Update outer/inner flops */
 747
 748     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*194);
 749 }
 750 /*
 751  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_double
 752  * Electrostatics interaction: Ewald
 753  * VdW interaction:            LennardJones
 754  * Geometry:                   Water4-Particle
 755  * Calculate force/pot:        Force
 756  */
 757 void
 758 nb_kernel_ElecEwSh_VdwLJSh_GeomW4P1_F_avx_128_fma_double
 759                     (t_nblist                    * gmx_restrict       nlist,
 760                      rvec                        * gmx_restrict          xx,
 761                      rvec                        * gmx_restrict          ff,
 762                      t_forcerec                  * gmx_restrict          fr,
 763                      t_mdatoms                   * gmx_restrict     mdatoms,
 764                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 765                      t_nrnb                      * gmx_restrict        nrnb)
 766 {
 767     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 768      * just 0 for non-waters.
 769      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 770      * jnr indices corresponding to data put in the four positions in the SIMD register.
 771      */
 772     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 773     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 774     int              jnrA,jnrB;
 775     int              j_coord_offsetA,j_coord_offsetB;
 776     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 777     real             rcutoff_scalar;
 778     real             *shiftvec,*fshift,*x,*f;
 779     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 780     int              vdwioffset0;
 781     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 782     int              vdwioffset1;
 783     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 784     int              vdwioffset2;
 785     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 786     int              vdwioffset3;
 787     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 788     int              vdwjidx0A,vdwjidx0B;
 789     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 790     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 791     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 792     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 793     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 794     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 795     real             *charge;
 796     int              nvdwtype;
 797     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 798     int              *vdwtype;
 799     real             *vdwparam;
 800     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 801     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 802     __m128i          ewitab;
 803     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 804     real             *ewtab;
 805     __m128d          dummy_mask,cutoff_mask;
 806     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 807     __m128d          one     = _mm_set1_pd(1.0);
 808     __m128d          two     = _mm_set1_pd(2.0);
 809     x                = xx[0];
 810     f                = ff[0];
 811
 812     nri              = nlist->nri;
 813     iinr             = nlist->iinr;
 814     jindex           = nlist->jindex;
 815     jjnr             = nlist->jjnr;
 816     shiftidx         = nlist->shift;
 817     gid              = nlist->gid;
 818     shiftvec         = fr->shift_vec[0];
 819     fshift           = fr->fshift[0];
 820     facel            = _mm_set1_pd(fr->epsfac);
 821     charge           = mdatoms->chargeA;
 822     nvdwtype         = fr->ntype;
 823     vdwparam         = fr->nbfp;
 824     vdwtype          = mdatoms->typeA;
 825
 826     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 827     ewtab            = fr->ic->tabq_coul_F;
 828     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 829     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 830
 831     /* Setup water-specific parameters */
 832     inr              = nlist->iinr[0];
 833     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 834     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 835     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 836     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 837
 838     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 839     rcutoff_scalar   = fr->rcoulomb;
 840     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 841     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 842
 843     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 844     rvdw             = _mm_set1_pd(fr->rvdw);
 845
 846     /* Avoid stupid compiler warnings */
 847     jnrA = jnrB = 0;
 848     j_coord_offsetA = 0;
 849     j_coord_offsetB = 0;
 850
 851     outeriter        = 0;
 852     inneriter        = 0;
 853
 854     /* Start outer loop over neighborlists */
 855     for(iidx=0; iidx<nri; iidx++)
 856     {
 857         /* Load shift vector for this list */
 858         i_shift_offset   = DIM*shiftidx[iidx];
 859
 860         /* Load limits for loop over neighbors */
 861         j_index_start    = jindex[iidx];
 862         j_index_end      = jindex[iidx+1];
 863
 864         /* Get outer coordinate index */
 865         inr              = iinr[iidx];
 866         i_coord_offset   = DIM*inr;
 867
 868         /* Load i particle coords and add shift vector */
 869         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 870                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 871
 872         fix0             = _mm_setzero_pd();
 873         fiy0             = _mm_setzero_pd();
 874         fiz0             = _mm_setzero_pd();
 875         fix1             = _mm_setzero_pd();
 876         fiy1             = _mm_setzero_pd();
 877         fiz1             = _mm_setzero_pd();
 878         fix2             = _mm_setzero_pd();
 879         fiy2             = _mm_setzero_pd();
 880         fiz2             = _mm_setzero_pd();
 881         fix3             = _mm_setzero_pd();
 882         fiy3             = _mm_setzero_pd();
 883         fiz3             = _mm_setzero_pd();
 884
 885         /* Start inner kernel loop */
 886         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 887         {
 888
 889             /* Get j neighbor index, and coordinate index */
 890             jnrA             = jjnr[jidx];
 891             jnrB             = jjnr[jidx+1];
 892             j_coord_offsetA  = DIM*jnrA;
 893             j_coord_offsetB  = DIM*jnrB;
 894
 895             /* load j atom coordinates */
 896             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 897                                               &jx0,&jy0,&jz0);
 898
 899             /* Calculate displacement vector */
 900             dx00             = _mm_sub_pd(ix0,jx0);
 901             dy00             = _mm_sub_pd(iy0,jy0);
 902             dz00             = _mm_sub_pd(iz0,jz0);
 903             dx10             = _mm_sub_pd(ix1,jx0);
 904             dy10             = _mm_sub_pd(iy1,jy0);
 905             dz10             = _mm_sub_pd(iz1,jz0);
 906             dx20             = _mm_sub_pd(ix2,jx0);
 907             dy20             = _mm_sub_pd(iy2,jy0);
 908             dz20             = _mm_sub_pd(iz2,jz0);
 909             dx30             = _mm_sub_pd(ix3,jx0);
 910             dy30             = _mm_sub_pd(iy3,jy0);
 911             dz30             = _mm_sub_pd(iz3,jz0);
 912
 913             /* Calculate squared distance and things based on it */
 914             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 915             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 916             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 917             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 918
 919             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 920             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 921             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 922
 923             rinvsq00         = gmx_mm_inv_pd(rsq00);
 924             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 925             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 926             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 927
 928             /* Load parameters for j particles */
 929             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 930             vdwjidx0A        = 2*vdwtype[jnrA+0];
 931             vdwjidx0B        = 2*vdwtype[jnrB+0];
 932
 933             fjx0             = _mm_setzero_pd();
 934             fjy0             = _mm_setzero_pd();
 935             fjz0             = _mm_setzero_pd();
 936
 937             /**************************
 938              * CALCULATE INTERACTIONS *
 939              **************************/
 940
 941             if (gmx_mm_any_lt(rsq00,rcutoff2))
 942             {
 943
 944             /* Compute parameters for interactions between i and j atoms */
 945             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 946                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 947
 948             /* LENNARD-JONES DISPERSION/REPULSION */
 949
 950             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 951             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 952
 953             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 954
 955             fscal            = fvdw;
 956
 957             fscal            = _mm_and_pd(fscal,cutoff_mask);
 958
 959             /* Update vectorial force */
 960             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 961             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 962             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 963
 964             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 965             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 966             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 967
 968             }
 969
 970             /**************************
 971              * CALCULATE INTERACTIONS *
 972              **************************/
 973
 974             if (gmx_mm_any_lt(rsq10,rcutoff2))
 975             {
 976
 977             r10              = _mm_mul_pd(rsq10,rinv10);
 978
 979             /* Compute parameters for interactions between i and j atoms */
 980             qq10             = _mm_mul_pd(iq1,jq0);
 981
 982             /* EWALD ELECTROSTATICS */
 983
 984             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 985             ewrt             = _mm_mul_pd(r10,ewtabscale);
 986             ewitab           = _mm_cvttpd_epi32(ewrt);
 987 #ifdef __XOP__
 988             eweps            = _mm_frcz_pd(ewrt);
 989 #else
 990             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 991 #endif
 992             twoeweps         = _mm_add_pd(eweps,eweps);
 993             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 994                                          &ewtabF,&ewtabFn);
 995             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 996             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 997
 998             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 999
1000             fscal            = felec;
1001
1002             fscal            = _mm_and_pd(fscal,cutoff_mask);
1003
1004             /* Update vectorial force */
1005             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1006             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1007             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1008
1009             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1010             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1011             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1012
1013             }
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             if (gmx_mm_any_lt(rsq20,rcutoff2))
1020             {
1021
1022             r20              = _mm_mul_pd(rsq20,rinv20);
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq20             = _mm_mul_pd(iq2,jq0);
1026
1027             /* EWALD ELECTROSTATICS */
1028
1029             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1030             ewrt             = _mm_mul_pd(r20,ewtabscale);
1031             ewitab           = _mm_cvttpd_epi32(ewrt);
1032 #ifdef __XOP__
1033             eweps            = _mm_frcz_pd(ewrt);
1034 #else
1035             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1036 #endif
1037             twoeweps         = _mm_add_pd(eweps,eweps);
1038             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1039                                          &ewtabF,&ewtabFn);
1040             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1041             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1042
1043             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1044
1045             fscal            = felec;
1046
1047             fscal            = _mm_and_pd(fscal,cutoff_mask);
1048
1049             /* Update vectorial force */
1050             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1051             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1052             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1053
1054             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1055             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1056             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1057
1058             }
1059
1060             /**************************
1061              * CALCULATE INTERACTIONS *
1062              **************************/
1063
1064             if (gmx_mm_any_lt(rsq30,rcutoff2))
1065             {
1066
1067             r30              = _mm_mul_pd(rsq30,rinv30);
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq30             = _mm_mul_pd(iq3,jq0);
1071
1072             /* EWALD ELECTROSTATICS */
1073
1074             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1075             ewrt             = _mm_mul_pd(r30,ewtabscale);
1076             ewitab           = _mm_cvttpd_epi32(ewrt);
1077 #ifdef __XOP__
1078             eweps            = _mm_frcz_pd(ewrt);
1079 #else
1080             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1081 #endif
1082             twoeweps         = _mm_add_pd(eweps,eweps);
1083             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1084                                          &ewtabF,&ewtabFn);
1085             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1086             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1087
1088             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1089
1090             fscal            = felec;
1091
1092             fscal            = _mm_and_pd(fscal,cutoff_mask);
1093
1094             /* Update vectorial force */
1095             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1096             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1097             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1098
1099             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1100             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1101             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1102
1103             }
1104
1105             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1106
1107             /* Inner loop uses 162 flops */
1108         }
1109
1110         if(jidx<j_index_end)
1111         {
1112
1113             jnrA             = jjnr[jidx];
1114             j_coord_offsetA  = DIM*jnrA;
1115
1116             /* load j atom coordinates */
1117             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1118                                               &jx0,&jy0,&jz0);
1119
1120             /* Calculate displacement vector */
1121             dx00             = _mm_sub_pd(ix0,jx0);
1122             dy00             = _mm_sub_pd(iy0,jy0);
1123             dz00             = _mm_sub_pd(iz0,jz0);
1124             dx10             = _mm_sub_pd(ix1,jx0);
1125             dy10             = _mm_sub_pd(iy1,jy0);
1126             dz10             = _mm_sub_pd(iz1,jz0);
1127             dx20             = _mm_sub_pd(ix2,jx0);
1128             dy20             = _mm_sub_pd(iy2,jy0);
1129             dz20             = _mm_sub_pd(iz2,jz0);
1130             dx30             = _mm_sub_pd(ix3,jx0);
1131             dy30             = _mm_sub_pd(iy3,jy0);
1132             dz30             = _mm_sub_pd(iz3,jz0);
1133
1134             /* Calculate squared distance and things based on it */
1135             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1136             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1137             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1138             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1139
1140             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1141             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1142             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1143
1144             rinvsq00         = gmx_mm_inv_pd(rsq00);
1145             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1146             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1147             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1148
1149             /* Load parameters for j particles */
1150             jq0              = _mm_load_sd(charge+jnrA+0);
1151             vdwjidx0A        = 2*vdwtype[jnrA+0];
1152
1153             fjx0             = _mm_setzero_pd();
1154             fjy0             = _mm_setzero_pd();
1155             fjz0             = _mm_setzero_pd();
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             if (gmx_mm_any_lt(rsq00,rcutoff2))
1162             {
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1166
1167             /* LENNARD-JONES DISPERSION/REPULSION */
1168
1169             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1170             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1171
1172             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1173
1174             fscal            = fvdw;
1175
1176             fscal            = _mm_and_pd(fscal,cutoff_mask);
1177
1178             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1179
1180             /* Update vectorial force */
1181             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1182             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1183             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1184
1185             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1186             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1187             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1188
1189             }
1190
1191             /**************************
1192              * CALCULATE INTERACTIONS *
1193              **************************/
1194
1195             if (gmx_mm_any_lt(rsq10,rcutoff2))
1196             {
1197
1198             r10              = _mm_mul_pd(rsq10,rinv10);
1199
1200             /* Compute parameters for interactions between i and j atoms */
1201             qq10             = _mm_mul_pd(iq1,jq0);
1202
1203             /* EWALD ELECTROSTATICS */
1204
1205             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1206             ewrt             = _mm_mul_pd(r10,ewtabscale);
1207             ewitab           = _mm_cvttpd_epi32(ewrt);
1208 #ifdef __XOP__
1209             eweps            = _mm_frcz_pd(ewrt);
1210 #else
1211             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1212 #endif
1213             twoeweps         = _mm_add_pd(eweps,eweps);
1214             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1215             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1216             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1217
1218             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_and_pd(fscal,cutoff_mask);
1223
1224             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1225
1226             /* Update vectorial force */
1227             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1228             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1229             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1230
1231             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1232             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1233             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1234
1235             }
1236
1237             /**************************
1238              * CALCULATE INTERACTIONS *
1239              **************************/
1240
1241             if (gmx_mm_any_lt(rsq20,rcutoff2))
1242             {
1243
1244             r20              = _mm_mul_pd(rsq20,rinv20);
1245
1246             /* Compute parameters for interactions between i and j atoms */
1247             qq20             = _mm_mul_pd(iq2,jq0);
1248
1249             /* EWALD ELECTROSTATICS */
1250
1251             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1252             ewrt             = _mm_mul_pd(r20,ewtabscale);
1253             ewitab           = _mm_cvttpd_epi32(ewrt);
1254 #ifdef __XOP__
1255             eweps            = _mm_frcz_pd(ewrt);
1256 #else
1257             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1258 #endif
1259             twoeweps         = _mm_add_pd(eweps,eweps);
1260             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1261             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1262             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1263
1264             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1265
1266             fscal            = felec;
1267
1268             fscal            = _mm_and_pd(fscal,cutoff_mask);
1269
1270             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1271
1272             /* Update vectorial force */
1273             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1274             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1275             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1276
1277             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1278             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1279             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1280
1281             }
1282
1283             /**************************
1284              * CALCULATE INTERACTIONS *
1285              **************************/
1286
1287             if (gmx_mm_any_lt(rsq30,rcutoff2))
1288             {
1289
1290             r30              = _mm_mul_pd(rsq30,rinv30);
1291
1292             /* Compute parameters for interactions between i and j atoms */
1293             qq30             = _mm_mul_pd(iq3,jq0);
1294
1295             /* EWALD ELECTROSTATICS */
1296
1297             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1298             ewrt             = _mm_mul_pd(r30,ewtabscale);
1299             ewitab           = _mm_cvttpd_epi32(ewrt);
1300 #ifdef __XOP__
1301             eweps            = _mm_frcz_pd(ewrt);
1302 #else
1303             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1304 #endif
1305             twoeweps         = _mm_add_pd(eweps,eweps);
1306             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1307             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1308             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1309
1310             cutoff_mask      = _mm_cmplt_pd(rsq30,rcutoff2);
1311
1312             fscal            = felec;
1313
1314             fscal            = _mm_and_pd(fscal,cutoff_mask);
1315
1316             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1317
1318             /* Update vectorial force */
1319             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1320             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1321             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1322
1323             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1324             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1325             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1326
1327             }
1328
1329             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1330
1331             /* Inner loop uses 162 flops */
1332         }
1333
1334         /* End of innermost loop */
1335
1336         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1337                                               f+i_coord_offset,fshift+i_shift_offset);
1338
1339         /* Increment number of inner iterations */
1340         inneriter                  += j_index_end - j_index_start;
1341
1342         /* Outer loop uses 24 flops */
1343     }
1344
1345     /* Increment number of outer iterations */
1346     outeriter        += nri;
1347
1348     /* Update outer/inner flops */
1349
1350     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*162);
1351 }