src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_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 "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/legacyheaders/types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "gromacs/legacyheaders/nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_avx_128_fma_double.h"
  50 #include "kernelutil_x86_avx_128_fma_double.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_double
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            LJEwald
  56  * Geometry:                   Water4-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_VF_avx_128_fma_double
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB;
  77     int              j_coord_offsetA,j_coord_offsetB;
  78     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  79     real             rcutoff_scalar;
  80     real             *shiftvec,*fshift,*x,*f;
  81     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset0;
  83     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  84     int              vdwioffset1;
  85     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  86     int              vdwioffset2;
  87     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  88     int              vdwioffset3;
  89     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  90     int              vdwjidx0A,vdwjidx0B;
  91     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  92     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  93     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  94     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  95     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  96     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 103     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 104     __m128d           c6grid_00;
 105     __m128d           c6grid_10;
 106     __m128d           c6grid_20;
 107     __m128d           c6grid_30;
 108     real             *vdwgridparam;
 109     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 110     __m128d           one_half  = _mm_set1_pd(0.5);
 111     __m128d           minus_one = _mm_set1_pd(-1.0);
 112     __m128i          ewitab;
 113     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 114     real             *ewtab;
 115     __m128d          dummy_mask,cutoff_mask;
 116     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 117     __m128d          one     = _mm_set1_pd(1.0);
 118     __m128d          two     = _mm_set1_pd(2.0);
 119     x                = xx[0];
 120     f                = ff[0];
 121
 122     nri              = nlist->nri;
 123     iinr             = nlist->iinr;
 124     jindex           = nlist->jindex;
 125     jjnr             = nlist->jjnr;
 126     shiftidx         = nlist->shift;
 127     gid              = nlist->gid;
 128     shiftvec         = fr->shift_vec[0];
 129     fshift           = fr->fshift[0];
 130     facel            = _mm_set1_pd(fr->epsfac);
 131     charge           = mdatoms->chargeA;
 132     nvdwtype         = fr->ntype;
 133     vdwparam         = fr->nbfp;
 134     vdwtype          = mdatoms->typeA;
 135     vdwgridparam     = fr->ljpme_c6grid;
 136     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 137     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 138     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 139
 140     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 141     ewtab            = fr->ic->tabq_coul_FDV0;
 142     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 143     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 144
 145     /* Setup water-specific parameters */
 146     inr              = nlist->iinr[0];
 147     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 148     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 149     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 150     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 151
 152     /* Avoid stupid compiler warnings */
 153     jnrA = jnrB = 0;
 154     j_coord_offsetA = 0;
 155     j_coord_offsetB = 0;
 156
 157     outeriter        = 0;
 158     inneriter        = 0;
 159
 160     /* Start outer loop over neighborlists */
 161     for(iidx=0; iidx<nri; iidx++)
 162     {
 163         /* Load shift vector for this list */
 164         i_shift_offset   = DIM*shiftidx[iidx];
 165
 166         /* Load limits for loop over neighbors */
 167         j_index_start    = jindex[iidx];
 168         j_index_end      = jindex[iidx+1];
 169
 170         /* Get outer coordinate index */
 171         inr              = iinr[iidx];
 172         i_coord_offset   = DIM*inr;
 173
 174         /* Load i particle coords and add shift vector */
 175         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 176                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 177
 178         fix0             = _mm_setzero_pd();
 179         fiy0             = _mm_setzero_pd();
 180         fiz0             = _mm_setzero_pd();
 181         fix1             = _mm_setzero_pd();
 182         fiy1             = _mm_setzero_pd();
 183         fiz1             = _mm_setzero_pd();
 184         fix2             = _mm_setzero_pd();
 185         fiy2             = _mm_setzero_pd();
 186         fiz2             = _mm_setzero_pd();
 187         fix3             = _mm_setzero_pd();
 188         fiy3             = _mm_setzero_pd();
 189         fiz3             = _mm_setzero_pd();
 190
 191         /* Reset potential sums */
 192         velecsum         = _mm_setzero_pd();
 193         vvdwsum          = _mm_setzero_pd();
 194
 195         /* Start inner kernel loop */
 196         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 197         {
 198
 199             /* Get j neighbor index, and coordinate index */
 200             jnrA             = jjnr[jidx];
 201             jnrB             = jjnr[jidx+1];
 202             j_coord_offsetA  = DIM*jnrA;
 203             j_coord_offsetB  = DIM*jnrB;
 204
 205             /* load j atom coordinates */
 206             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 207                                               &jx0,&jy0,&jz0);
 208
 209             /* Calculate displacement vector */
 210             dx00             = _mm_sub_pd(ix0,jx0);
 211             dy00             = _mm_sub_pd(iy0,jy0);
 212             dz00             = _mm_sub_pd(iz0,jz0);
 213             dx10             = _mm_sub_pd(ix1,jx0);
 214             dy10             = _mm_sub_pd(iy1,jy0);
 215             dz10             = _mm_sub_pd(iz1,jz0);
 216             dx20             = _mm_sub_pd(ix2,jx0);
 217             dy20             = _mm_sub_pd(iy2,jy0);
 218             dz20             = _mm_sub_pd(iz2,jz0);
 219             dx30             = _mm_sub_pd(ix3,jx0);
 220             dy30             = _mm_sub_pd(iy3,jy0);
 221             dz30             = _mm_sub_pd(iz3,jz0);
 222
 223             /* Calculate squared distance and things based on it */
 224             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 225             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 226             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 227             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 228
 229             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 230             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 231             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 232             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 233
 234             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 235             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 236             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 237             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 238
 239             /* Load parameters for j particles */
 240             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 241             vdwjidx0A        = 2*vdwtype[jnrA+0];
 242             vdwjidx0B        = 2*vdwtype[jnrB+0];
 243
 244             fjx0             = _mm_setzero_pd();
 245             fjy0             = _mm_setzero_pd();
 246             fjz0             = _mm_setzero_pd();
 247
 248             /**************************
 249              * CALCULATE INTERACTIONS *
 250              **************************/
 251
 252             r00              = _mm_mul_pd(rsq00,rinv00);
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 256                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 257             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 258                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 259
 260             /* Analytical LJ-PME */
 261             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 262             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 263             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 264             exponent         = gmx_simd_exp_d(ewcljrsq);
 265             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 266             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 267             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 268             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 269             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 270             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 271             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 272             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 276
 277             fscal            = fvdw;
 278
 279             /* Update vectorial force */
 280             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 281             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 282             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 283
 284             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 285             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 286             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 287
 288             /**************************
 289              * CALCULATE INTERACTIONS *
 290              **************************/
 291
 292             r10              = _mm_mul_pd(rsq10,rinv10);
 293
 294             /* Compute parameters for interactions between i and j atoms */
 295             qq10             = _mm_mul_pd(iq1,jq0);
 296
 297             /* EWALD ELECTROSTATICS */
 298
 299             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 300             ewrt             = _mm_mul_pd(r10,ewtabscale);
 301             ewitab           = _mm_cvttpd_epi32(ewrt);
 302 #ifdef __XOP__
 303             eweps            = _mm_frcz_pd(ewrt);
 304 #else
 305             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 306 #endif
 307             twoeweps         = _mm_add_pd(eweps,eweps);
 308             ewitab           = _mm_slli_epi32(ewitab,2);
 309             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 310             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 311             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 312             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 313             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 314             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 315             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 316             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 317             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 318             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 319
 320             /* Update potential sum for this i atom from the interaction with this j atom. */
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             /* Update vectorial force */
 326             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 327             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 328             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 329
 330             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 331             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 332             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 333
 334             /**************************
 335              * CALCULATE INTERACTIONS *
 336              **************************/
 337
 338             r20              = _mm_mul_pd(rsq20,rinv20);
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq20             = _mm_mul_pd(iq2,jq0);
 342
 343             /* EWALD ELECTROSTATICS */
 344
 345             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 346             ewrt             = _mm_mul_pd(r20,ewtabscale);
 347             ewitab           = _mm_cvttpd_epi32(ewrt);
 348 #ifdef __XOP__
 349             eweps            = _mm_frcz_pd(ewrt);
 350 #else
 351             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 352 #endif
 353             twoeweps         = _mm_add_pd(eweps,eweps);
 354             ewitab           = _mm_slli_epi32(ewitab,2);
 355             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 356             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 357             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 358             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 359             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 360             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 361             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 362             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 363             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 364             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 365
 366             /* Update potential sum for this i atom from the interaction with this j atom. */
 367             velecsum         = _mm_add_pd(velecsum,velec);
 368
 369             fscal            = felec;
 370
 371             /* Update vectorial force */
 372             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 373             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 374             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 375
 376             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 377             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 378             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             r30              = _mm_mul_pd(rsq30,rinv30);
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq30             = _mm_mul_pd(iq3,jq0);
 388
 389             /* EWALD ELECTROSTATICS */
 390
 391             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 392             ewrt             = _mm_mul_pd(r30,ewtabscale);
 393             ewitab           = _mm_cvttpd_epi32(ewrt);
 394 #ifdef __XOP__
 395             eweps            = _mm_frcz_pd(ewrt);
 396 #else
 397             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 398 #endif
 399             twoeweps         = _mm_add_pd(eweps,eweps);
 400             ewitab           = _mm_slli_epi32(ewitab,2);
 401             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 402             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 403             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 404             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 405             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 406             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 407             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 408             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 409             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 410             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 411
 412             /* Update potential sum for this i atom from the interaction with this j atom. */
 413             velecsum         = _mm_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             /* Update vectorial force */
 418             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 419             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 420             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 421
 422             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 423             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 424             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 425
 426             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 427
 428             /* Inner loop uses 185 flops */
 429         }
 430
 431         if(jidx<j_index_end)
 432         {
 433
 434             jnrA             = jjnr[jidx];
 435             j_coord_offsetA  = DIM*jnrA;
 436
 437             /* load j atom coordinates */
 438             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 439                                               &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm_sub_pd(ix0,jx0);
 443             dy00             = _mm_sub_pd(iy0,jy0);
 444             dz00             = _mm_sub_pd(iz0,jz0);
 445             dx10             = _mm_sub_pd(ix1,jx0);
 446             dy10             = _mm_sub_pd(iy1,jy0);
 447             dz10             = _mm_sub_pd(iz1,jz0);
 448             dx20             = _mm_sub_pd(ix2,jx0);
 449             dy20             = _mm_sub_pd(iy2,jy0);
 450             dz20             = _mm_sub_pd(iz2,jz0);
 451             dx30             = _mm_sub_pd(ix3,jx0);
 452             dy30             = _mm_sub_pd(iy3,jy0);
 453             dz30             = _mm_sub_pd(iz3,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 457             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 458             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 459             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 460
 461             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 462             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 463             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 464             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 465
 466             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 467             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 468             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 469             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 470
 471             /* Load parameters for j particles */
 472             jq0              = _mm_load_sd(charge+jnrA+0);
 473             vdwjidx0A        = 2*vdwtype[jnrA+0];
 474
 475             fjx0             = _mm_setzero_pd();
 476             fjy0             = _mm_setzero_pd();
 477             fjz0             = _mm_setzero_pd();
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             r00              = _mm_mul_pd(rsq00,rinv00);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 487             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 488
 489             /* Analytical LJ-PME */
 490             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 491             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 492             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 493             exponent         = gmx_simd_exp_d(ewcljrsq);
 494             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 495             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 496             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 497             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 498             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 499             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 500             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 501             fvdw             = _mm_mul_pd(_mm_add_pd(vvdw12,_mm_msub_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6),vvdw6)),rinvsq00);
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 505             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 506
 507             fscal            = fvdw;
 508
 509             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 510
 511             /* Update vectorial force */
 512             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 513             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 514             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 515
 516             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 517             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 518             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             r10              = _mm_mul_pd(rsq10,rinv10);
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq10             = _mm_mul_pd(iq1,jq0);
 528
 529             /* EWALD ELECTROSTATICS */
 530
 531             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 532             ewrt             = _mm_mul_pd(r10,ewtabscale);
 533             ewitab           = _mm_cvttpd_epi32(ewrt);
 534 #ifdef __XOP__
 535             eweps            = _mm_frcz_pd(ewrt);
 536 #else
 537             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 538 #endif
 539             twoeweps         = _mm_add_pd(eweps,eweps);
 540             ewitab           = _mm_slli_epi32(ewitab,2);
 541             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 542             ewtabD           = _mm_setzero_pd();
 543             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 544             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 545             ewtabFn          = _mm_setzero_pd();
 546             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 547             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 548             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 549             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 550             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 551
 552             /* Update potential sum for this i atom from the interaction with this j atom. */
 553             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 554             velecsum         = _mm_add_pd(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 559
 560             /* Update vectorial force */
 561             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 562             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 563             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 564
 565             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 566             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 567             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r20              = _mm_mul_pd(rsq20,rinv20);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq20             = _mm_mul_pd(iq2,jq0);
 577
 578             /* EWALD ELECTROSTATICS */
 579
 580             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 581             ewrt             = _mm_mul_pd(r20,ewtabscale);
 582             ewitab           = _mm_cvttpd_epi32(ewrt);
 583 #ifdef __XOP__
 584             eweps            = _mm_frcz_pd(ewrt);
 585 #else
 586             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 587 #endif
 588             twoeweps         = _mm_add_pd(eweps,eweps);
 589             ewitab           = _mm_slli_epi32(ewitab,2);
 590             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 591             ewtabD           = _mm_setzero_pd();
 592             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 593             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 594             ewtabFn          = _mm_setzero_pd();
 595             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 596             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 597             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 598             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 599             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 600
 601             /* Update potential sum for this i atom from the interaction with this j atom. */
 602             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 603             velecsum         = _mm_add_pd(velecsum,velec);
 604
 605             fscal            = felec;
 606
 607             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 608
 609             /* Update vectorial force */
 610             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 611             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 612             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 613
 614             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 615             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 616             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 617
 618             /**************************
 619              * CALCULATE INTERACTIONS *
 620              **************************/
 621
 622             r30              = _mm_mul_pd(rsq30,rinv30);
 623
 624             /* Compute parameters for interactions between i and j atoms */
 625             qq30             = _mm_mul_pd(iq3,jq0);
 626
 627             /* EWALD ELECTROSTATICS */
 628
 629             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 630             ewrt             = _mm_mul_pd(r30,ewtabscale);
 631             ewitab           = _mm_cvttpd_epi32(ewrt);
 632 #ifdef __XOP__
 633             eweps            = _mm_frcz_pd(ewrt);
 634 #else
 635             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 636 #endif
 637             twoeweps         = _mm_add_pd(eweps,eweps);
 638             ewitab           = _mm_slli_epi32(ewitab,2);
 639             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 640             ewtabD           = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 642             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 643             ewtabFn          = _mm_setzero_pd();
 644             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 645             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 646             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 647             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 648             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 649
 650             /* Update potential sum for this i atom from the interaction with this j atom. */
 651             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 652             velecsum         = _mm_add_pd(velecsum,velec);
 653
 654             fscal            = felec;
 655
 656             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 657
 658             /* Update vectorial force */
 659             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 660             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 661             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 662
 663             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 664             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 665             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 666
 667             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 668
 669             /* Inner loop uses 185 flops */
 670         }
 671
 672         /* End of innermost loop */
 673
 674         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 675                                               f+i_coord_offset,fshift+i_shift_offset);
 676
 677         ggid                        = gid[iidx];
 678         /* Update potential energies */
 679         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 680         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 681
 682         /* Increment number of inner iterations */
 683         inneriter                  += j_index_end - j_index_start;
 684
 685         /* Outer loop uses 26 flops */
 686     }
 687
 688     /* Increment number of outer iterations */
 689     outeriter        += nri;
 690
 691     /* Update outer/inner flops */
 692
 693     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*185);
 694 }
 695 /*
 696  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
 697  * Electrostatics interaction: Ewald
 698  * VdW interaction:            LJEwald
 699  * Geometry:                   Water4-Particle
 700  * Calculate force/pot:        Force
 701  */
 702 void
 703 nb_kernel_ElecEw_VdwLJEw_GeomW4P1_F_avx_128_fma_double
 704                     (t_nblist                    * gmx_restrict       nlist,
 705                      rvec                        * gmx_restrict          xx,
 706                      rvec                        * gmx_restrict          ff,
 707                      t_forcerec                  * gmx_restrict          fr,
 708                      t_mdatoms                   * gmx_restrict     mdatoms,
 709                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 710                      t_nrnb                      * gmx_restrict        nrnb)
 711 {
 712     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 713      * just 0 for non-waters.
 714      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 715      * jnr indices corresponding to data put in the four positions in the SIMD register.
 716      */
 717     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 718     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 719     int              jnrA,jnrB;
 720     int              j_coord_offsetA,j_coord_offsetB;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 725     int              vdwioffset0;
 726     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 727     int              vdwioffset1;
 728     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 729     int              vdwioffset2;
 730     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 731     int              vdwioffset3;
 732     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 733     int              vdwjidx0A,vdwjidx0B;
 734     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 736     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 737     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 738     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 739     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 740     real             *charge;
 741     int              nvdwtype;
 742     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 743     int              *vdwtype;
 744     real             *vdwparam;
 745     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 746     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 747     __m128d           c6grid_00;
 748     __m128d           c6grid_10;
 749     __m128d           c6grid_20;
 750     __m128d           c6grid_30;
 751     real             *vdwgridparam;
 752     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 753     __m128d           one_half  = _mm_set1_pd(0.5);
 754     __m128d           minus_one = _mm_set1_pd(-1.0);
 755     __m128i          ewitab;
 756     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 757     real             *ewtab;
 758     __m128d          dummy_mask,cutoff_mask;
 759     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 760     __m128d          one     = _mm_set1_pd(1.0);
 761     __m128d          two     = _mm_set1_pd(2.0);
 762     x                = xx[0];
 763     f                = ff[0];
 764
 765     nri              = nlist->nri;
 766     iinr             = nlist->iinr;
 767     jindex           = nlist->jindex;
 768     jjnr             = nlist->jjnr;
 769     shiftidx         = nlist->shift;
 770     gid              = nlist->gid;
 771     shiftvec         = fr->shift_vec[0];
 772     fshift           = fr->fshift[0];
 773     facel            = _mm_set1_pd(fr->epsfac);
 774     charge           = mdatoms->chargeA;
 775     nvdwtype         = fr->ntype;
 776     vdwparam         = fr->nbfp;
 777     vdwtype          = mdatoms->typeA;
 778     vdwgridparam     = fr->ljpme_c6grid;
 779     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 780     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 781     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 782
 783     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 784     ewtab            = fr->ic->tabq_coul_F;
 785     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 786     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 787
 788     /* Setup water-specific parameters */
 789     inr              = nlist->iinr[0];
 790     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 791     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 792     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 793     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 794
 795     /* Avoid stupid compiler warnings */
 796     jnrA = jnrB = 0;
 797     j_coord_offsetA = 0;
 798     j_coord_offsetB = 0;
 799
 800     outeriter        = 0;
 801     inneriter        = 0;
 802
 803     /* Start outer loop over neighborlists */
 804     for(iidx=0; iidx<nri; iidx++)
 805     {
 806         /* Load shift vector for this list */
 807         i_shift_offset   = DIM*shiftidx[iidx];
 808
 809         /* Load limits for loop over neighbors */
 810         j_index_start    = jindex[iidx];
 811         j_index_end      = jindex[iidx+1];
 812
 813         /* Get outer coordinate index */
 814         inr              = iinr[iidx];
 815         i_coord_offset   = DIM*inr;
 816
 817         /* Load i particle coords and add shift vector */
 818         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 819                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 820
 821         fix0             = _mm_setzero_pd();
 822         fiy0             = _mm_setzero_pd();
 823         fiz0             = _mm_setzero_pd();
 824         fix1             = _mm_setzero_pd();
 825         fiy1             = _mm_setzero_pd();
 826         fiz1             = _mm_setzero_pd();
 827         fix2             = _mm_setzero_pd();
 828         fiy2             = _mm_setzero_pd();
 829         fiz2             = _mm_setzero_pd();
 830         fix3             = _mm_setzero_pd();
 831         fiy3             = _mm_setzero_pd();
 832         fiz3             = _mm_setzero_pd();
 833
 834         /* Start inner kernel loop */
 835         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 836         {
 837
 838             /* Get j neighbor index, and coordinate index */
 839             jnrA             = jjnr[jidx];
 840             jnrB             = jjnr[jidx+1];
 841             j_coord_offsetA  = DIM*jnrA;
 842             j_coord_offsetB  = DIM*jnrB;
 843
 844             /* load j atom coordinates */
 845             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 846                                               &jx0,&jy0,&jz0);
 847
 848             /* Calculate displacement vector */
 849             dx00             = _mm_sub_pd(ix0,jx0);
 850             dy00             = _mm_sub_pd(iy0,jy0);
 851             dz00             = _mm_sub_pd(iz0,jz0);
 852             dx10             = _mm_sub_pd(ix1,jx0);
 853             dy10             = _mm_sub_pd(iy1,jy0);
 854             dz10             = _mm_sub_pd(iz1,jz0);
 855             dx20             = _mm_sub_pd(ix2,jx0);
 856             dy20             = _mm_sub_pd(iy2,jy0);
 857             dz20             = _mm_sub_pd(iz2,jz0);
 858             dx30             = _mm_sub_pd(ix3,jx0);
 859             dy30             = _mm_sub_pd(iy3,jy0);
 860             dz30             = _mm_sub_pd(iz3,jz0);
 861
 862             /* Calculate squared distance and things based on it */
 863             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 864             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 865             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 866             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 867
 868             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 869             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 870             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 871             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 872
 873             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 874             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 875             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 876             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 877
 878             /* Load parameters for j particles */
 879             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 880             vdwjidx0A        = 2*vdwtype[jnrA+0];
 881             vdwjidx0B        = 2*vdwtype[jnrB+0];
 882
 883             fjx0             = _mm_setzero_pd();
 884             fjy0             = _mm_setzero_pd();
 885             fjz0             = _mm_setzero_pd();
 886
 887             /**************************
 888              * CALCULATE INTERACTIONS *
 889              **************************/
 890
 891             r00              = _mm_mul_pd(rsq00,rinv00);
 892
 893             /* Compute parameters for interactions between i and j atoms */
 894             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 895                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 896             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 897                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 898
 899             /* Analytical LJ-PME */
 900             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 901             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 902             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 903             exponent         = gmx_simd_exp_d(ewcljrsq);
 904             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 905             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 906             /* f6A = 6 * C6grid * (1 - poly) */
 907             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 908             /* f6B = C6grid * exponent * beta^6 */
 909             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 910             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 911             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 912
 913             fscal            = fvdw;
 914
 915             /* Update vectorial force */
 916             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 917             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 918             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 919
 920             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 921             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 922             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 923
 924             /**************************
 925              * CALCULATE INTERACTIONS *
 926              **************************/
 927
 928             r10              = _mm_mul_pd(rsq10,rinv10);
 929
 930             /* Compute parameters for interactions between i and j atoms */
 931             qq10             = _mm_mul_pd(iq1,jq0);
 932
 933             /* EWALD ELECTROSTATICS */
 934
 935             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 936             ewrt             = _mm_mul_pd(r10,ewtabscale);
 937             ewitab           = _mm_cvttpd_epi32(ewrt);
 938 #ifdef __XOP__
 939             eweps            = _mm_frcz_pd(ewrt);
 940 #else
 941             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 942 #endif
 943             twoeweps         = _mm_add_pd(eweps,eweps);
 944             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 945                                          &ewtabF,&ewtabFn);
 946             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 947             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 948
 949             fscal            = felec;
 950
 951             /* Update vectorial force */
 952             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 953             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 954             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 955
 956             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 957             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 958             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r20              = _mm_mul_pd(rsq20,rinv20);
 965
 966             /* Compute parameters for interactions between i and j atoms */
 967             qq20             = _mm_mul_pd(iq2,jq0);
 968
 969             /* EWALD ELECTROSTATICS */
 970
 971             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 972             ewrt             = _mm_mul_pd(r20,ewtabscale);
 973             ewitab           = _mm_cvttpd_epi32(ewrt);
 974 #ifdef __XOP__
 975             eweps            = _mm_frcz_pd(ewrt);
 976 #else
 977             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 978 #endif
 979             twoeweps         = _mm_add_pd(eweps,eweps);
 980             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 981                                          &ewtabF,&ewtabFn);
 982             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 983             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 984
 985             fscal            = felec;
 986
 987             /* Update vectorial force */
 988             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 989             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 990             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 991
 992             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 993             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 994             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             r30              = _mm_mul_pd(rsq30,rinv30);
1001
1002             /* Compute parameters for interactions between i and j atoms */
1003             qq30             = _mm_mul_pd(iq3,jq0);
1004
1005             /* EWALD ELECTROSTATICS */
1006
1007             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1008             ewrt             = _mm_mul_pd(r30,ewtabscale);
1009             ewitab           = _mm_cvttpd_epi32(ewrt);
1010 #ifdef __XOP__
1011             eweps            = _mm_frcz_pd(ewrt);
1012 #else
1013             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1014 #endif
1015             twoeweps         = _mm_add_pd(eweps,eweps);
1016             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
1017                                          &ewtabF,&ewtabFn);
1018             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1019             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1020
1021             fscal            = felec;
1022
1023             /* Update vectorial force */
1024             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1025             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1026             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1027
1028             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1029             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1030             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1031
1032             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
1033
1034             /* Inner loop uses 167 flops */
1035         }
1036
1037         if(jidx<j_index_end)
1038         {
1039
1040             jnrA             = jjnr[jidx];
1041             j_coord_offsetA  = DIM*jnrA;
1042
1043             /* load j atom coordinates */
1044             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
1045                                               &jx0,&jy0,&jz0);
1046
1047             /* Calculate displacement vector */
1048             dx00             = _mm_sub_pd(ix0,jx0);
1049             dy00             = _mm_sub_pd(iy0,jy0);
1050             dz00             = _mm_sub_pd(iz0,jz0);
1051             dx10             = _mm_sub_pd(ix1,jx0);
1052             dy10             = _mm_sub_pd(iy1,jy0);
1053             dz10             = _mm_sub_pd(iz1,jz0);
1054             dx20             = _mm_sub_pd(ix2,jx0);
1055             dy20             = _mm_sub_pd(iy2,jy0);
1056             dz20             = _mm_sub_pd(iz2,jz0);
1057             dx30             = _mm_sub_pd(ix3,jx0);
1058             dy30             = _mm_sub_pd(iy3,jy0);
1059             dz30             = _mm_sub_pd(iz3,jz0);
1060
1061             /* Calculate squared distance and things based on it */
1062             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1063             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1064             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1065             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
1066
1067             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1068             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1069             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1070             rinv30           = gmx_mm_invsqrt_pd(rsq30);
1071
1072             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1073             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1074             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1075             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1076
1077             /* Load parameters for j particles */
1078             jq0              = _mm_load_sd(charge+jnrA+0);
1079             vdwjidx0A        = 2*vdwtype[jnrA+0];
1080
1081             fjx0             = _mm_setzero_pd();
1082             fjy0             = _mm_setzero_pd();
1083             fjz0             = _mm_setzero_pd();
1084
1085             /**************************
1086              * CALCULATE INTERACTIONS *
1087              **************************/
1088
1089             r00              = _mm_mul_pd(rsq00,rinv00);
1090
1091             /* Compute parameters for interactions between i and j atoms */
1092             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1093             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
1094
1095             /* Analytical LJ-PME */
1096             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1097             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1098             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1099             exponent         = gmx_simd_exp_d(ewcljrsq);
1100             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1101             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1102             /* f6A = 6 * C6grid * (1 - poly) */
1103             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1104             /* f6B = C6grid * exponent * beta^6 */
1105             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1106             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1107             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1108
1109             fscal            = fvdw;
1110
1111             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1112
1113             /* Update vectorial force */
1114             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1115             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1116             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1117
1118             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1119             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1120             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1121
1122             /**************************
1123              * CALCULATE INTERACTIONS *
1124              **************************/
1125
1126             r10              = _mm_mul_pd(rsq10,rinv10);
1127
1128             /* Compute parameters for interactions between i and j atoms */
1129             qq10             = _mm_mul_pd(iq1,jq0);
1130
1131             /* EWALD ELECTROSTATICS */
1132
1133             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1134             ewrt             = _mm_mul_pd(r10,ewtabscale);
1135             ewitab           = _mm_cvttpd_epi32(ewrt);
1136 #ifdef __XOP__
1137             eweps            = _mm_frcz_pd(ewrt);
1138 #else
1139             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1140 #endif
1141             twoeweps         = _mm_add_pd(eweps,eweps);
1142             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1143             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1144             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1145
1146             fscal            = felec;
1147
1148             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1149
1150             /* Update vectorial force */
1151             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1152             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1153             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1154
1155             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1156             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1157             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1158
1159             /**************************
1160              * CALCULATE INTERACTIONS *
1161              **************************/
1162
1163             r20              = _mm_mul_pd(rsq20,rinv20);
1164
1165             /* Compute parameters for interactions between i and j atoms */
1166             qq20             = _mm_mul_pd(iq2,jq0);
1167
1168             /* EWALD ELECTROSTATICS */
1169
1170             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1171             ewrt             = _mm_mul_pd(r20,ewtabscale);
1172             ewitab           = _mm_cvttpd_epi32(ewrt);
1173 #ifdef __XOP__
1174             eweps            = _mm_frcz_pd(ewrt);
1175 #else
1176             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1177 #endif
1178             twoeweps         = _mm_add_pd(eweps,eweps);
1179             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1180             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1181             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1182
1183             fscal            = felec;
1184
1185             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1186
1187             /* Update vectorial force */
1188             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1189             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1190             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1191
1192             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1193             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1194             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1195
1196             /**************************
1197              * CALCULATE INTERACTIONS *
1198              **************************/
1199
1200             r30              = _mm_mul_pd(rsq30,rinv30);
1201
1202             /* Compute parameters for interactions between i and j atoms */
1203             qq30             = _mm_mul_pd(iq3,jq0);
1204
1205             /* EWALD ELECTROSTATICS */
1206
1207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1208             ewrt             = _mm_mul_pd(r30,ewtabscale);
1209             ewitab           = _mm_cvttpd_epi32(ewrt);
1210 #ifdef __XOP__
1211             eweps            = _mm_frcz_pd(ewrt);
1212 #else
1213             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1214 #endif
1215             twoeweps         = _mm_add_pd(eweps,eweps);
1216             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1217             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1218             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1223
1224             /* Update vectorial force */
1225             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1226             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1227             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1228
1229             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1230             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1231             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1232
1233             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1234
1235             /* Inner loop uses 167 flops */
1236         }
1237
1238         /* End of innermost loop */
1239
1240         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1241                                               f+i_coord_offset,fshift+i_shift_offset);
1242
1243         /* Increment number of inner iterations */
1244         inneriter                  += j_index_end - j_index_start;
1245
1246         /* Outer loop uses 24 flops */
1247     }
1248
1249     /* Increment number of outer iterations */
1250     outeriter        += nri;
1251
1252     /* Update outer/inner flops */
1253
1254     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*167);
1255 }