src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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_ElecEw_VdwLJEw_GeomW3P1_VF_avx_128_fma_double
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LJEwald
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_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              vdwjidx0A,vdwjidx0B;
  87     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  88     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  89     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  90     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  91     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  98     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  99     __m128d           c6grid_00;
 100     __m128d           c6grid_10;
 101     __m128d           c6grid_20;
 102     real             *vdwgridparam;
 103     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 104     __m128d           one_half  = _mm_set1_pd(0.5);
 105     __m128d           minus_one = _mm_set1_pd(-1.0);
 106     __m128i          ewitab;
 107     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 108     real             *ewtab;
 109     __m128d          dummy_mask,cutoff_mask;
 110     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 111     __m128d          one     = _mm_set1_pd(1.0);
 112     __m128d          two     = _mm_set1_pd(2.0);
 113     x                = xx[0];
 114     f                = ff[0];
 115
 116     nri              = nlist->nri;
 117     iinr             = nlist->iinr;
 118     jindex           = nlist->jindex;
 119     jjnr             = nlist->jjnr;
 120     shiftidx         = nlist->shift;
 121     gid              = nlist->gid;
 122     shiftvec         = fr->shift_vec[0];
 123     fshift           = fr->fshift[0];
 124     facel            = _mm_set1_pd(fr->epsfac);
 125     charge           = mdatoms->chargeA;
 126     nvdwtype         = fr->ntype;
 127     vdwparam         = fr->nbfp;
 128     vdwtype          = mdatoms->typeA;
 129     vdwgridparam     = fr->ljpme_c6grid;
 130     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 131     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 132     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 133
 134     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 135     ewtab            = fr->ic->tabq_coul_FDV0;
 136     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 137     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 138
 139     /* Setup water-specific parameters */
 140     inr              = nlist->iinr[0];
 141     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 142     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 143     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 144     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 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_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 170                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 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
 182         /* Reset potential sums */
 183         velecsum         = _mm_setzero_pd();
 184         vvdwsum          = _mm_setzero_pd();
 185
 186         /* Start inner kernel loop */
 187         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 188         {
 189
 190             /* Get j neighbor index, and coordinate index */
 191             jnrA             = jjnr[jidx];
 192             jnrB             = jjnr[jidx+1];
 193             j_coord_offsetA  = DIM*jnrA;
 194             j_coord_offsetB  = DIM*jnrB;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_pd(ix0,jx0);
 202             dy00             = _mm_sub_pd(iy0,jy0);
 203             dz00             = _mm_sub_pd(iz0,jz0);
 204             dx10             = _mm_sub_pd(ix1,jx0);
 205             dy10             = _mm_sub_pd(iy1,jy0);
 206             dz10             = _mm_sub_pd(iz1,jz0);
 207             dx20             = _mm_sub_pd(ix2,jx0);
 208             dy20             = _mm_sub_pd(iy2,jy0);
 209             dz20             = _mm_sub_pd(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 217             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 218             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 219
 220             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 226             vdwjidx0A        = 2*vdwtype[jnrA+0];
 227             vdwjidx0B        = 2*vdwtype[jnrB+0];
 228
 229             fjx0             = _mm_setzero_pd();
 230             fjy0             = _mm_setzero_pd();
 231             fjz0             = _mm_setzero_pd();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             r00              = _mm_mul_pd(rsq00,rinv00);
 238
 239             /* Compute parameters for interactions between i and j atoms */
 240             qq00             = _mm_mul_pd(iq0,jq0);
 241             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 242                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 243             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 244                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 245
 246             /* EWALD ELECTROSTATICS */
 247
 248             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 249             ewrt             = _mm_mul_pd(r00,ewtabscale);
 250             ewitab           = _mm_cvttpd_epi32(ewrt);
 251 #ifdef __XOP__
 252             eweps            = _mm_frcz_pd(ewrt);
 253 #else
 254             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 255 #endif
 256             twoeweps         = _mm_add_pd(eweps,eweps);
 257             ewitab           = _mm_slli_epi32(ewitab,2);
 258             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 259             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 260             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 261             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 262             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 263             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 264             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 265             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 266             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 267             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 268
 269             /* Analytical LJ-PME */
 270             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 271             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 272             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 273             exponent         = gmx_simd_exp_d(ewcljrsq);
 274             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 275             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 276             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 277             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 278             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 279             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 280             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 281             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);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velecsum         = _mm_add_pd(velecsum,velec);
 285             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 286
 287             fscal            = _mm_add_pd(felec,fvdw);
 288
 289             /* Update vectorial force */
 290             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 291             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 292             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 293
 294             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 295             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 296             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm_mul_pd(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_pd(iq1,jq0);
 306
 307             /* EWALD ELECTROSTATICS */
 308
 309             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 310             ewrt             = _mm_mul_pd(r10,ewtabscale);
 311             ewitab           = _mm_cvttpd_epi32(ewrt);
 312 #ifdef __XOP__
 313             eweps            = _mm_frcz_pd(ewrt);
 314 #else
 315             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 316 #endif
 317             twoeweps         = _mm_add_pd(eweps,eweps);
 318             ewitab           = _mm_slli_epi32(ewitab,2);
 319             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 320             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 321             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 322             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 323             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 324             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 325             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 326             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 327             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 328             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm_add_pd(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335             /* Update vectorial force */
 336             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 337             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 338             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 339
 340             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 341             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 342             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r20              = _mm_mul_pd(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_pd(iq2,jq0);
 352
 353             /* EWALD ELECTROSTATICS */
 354
 355             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 356             ewrt             = _mm_mul_pd(r20,ewtabscale);
 357             ewitab           = _mm_cvttpd_epi32(ewrt);
 358 #ifdef __XOP__
 359             eweps            = _mm_frcz_pd(ewrt);
 360 #else
 361             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 362 #endif
 363             twoeweps         = _mm_add_pd(eweps,eweps);
 364             ewitab           = _mm_slli_epi32(ewitab,2);
 365             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 366             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 367             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 368             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 369             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 370             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 371             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 372             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 373             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 374             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             /* Update vectorial force */
 382             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 383             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 384             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 385
 386             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 387             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 388             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 389
 390             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 391
 392             /* Inner loop uses 159 flops */
 393         }
 394
 395         if(jidx<j_index_end)
 396         {
 397
 398             jnrA             = jjnr[jidx];
 399             j_coord_offsetA  = DIM*jnrA;
 400
 401             /* load j atom coordinates */
 402             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 403                                               &jx0,&jy0,&jz0);
 404
 405             /* Calculate displacement vector */
 406             dx00             = _mm_sub_pd(ix0,jx0);
 407             dy00             = _mm_sub_pd(iy0,jy0);
 408             dz00             = _mm_sub_pd(iz0,jz0);
 409             dx10             = _mm_sub_pd(ix1,jx0);
 410             dy10             = _mm_sub_pd(iy1,jy0);
 411             dz10             = _mm_sub_pd(iz1,jz0);
 412             dx20             = _mm_sub_pd(ix2,jx0);
 413             dy20             = _mm_sub_pd(iy2,jy0);
 414             dz20             = _mm_sub_pd(iz2,jz0);
 415
 416             /* Calculate squared distance and things based on it */
 417             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 418             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 419             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 420
 421             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 422             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 423             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 424
 425             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 426             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 427             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 428
 429             /* Load parameters for j particles */
 430             jq0              = _mm_load_sd(charge+jnrA+0);
 431             vdwjidx0A        = 2*vdwtype[jnrA+0];
 432
 433             fjx0             = _mm_setzero_pd();
 434             fjy0             = _mm_setzero_pd();
 435             fjz0             = _mm_setzero_pd();
 436
 437             /**************************
 438              * CALCULATE INTERACTIONS *
 439              **************************/
 440
 441             r00              = _mm_mul_pd(rsq00,rinv00);
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             qq00             = _mm_mul_pd(iq0,jq0);
 445             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 446             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 447
 448             /* EWALD ELECTROSTATICS */
 449
 450             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 451             ewrt             = _mm_mul_pd(r00,ewtabscale);
 452             ewitab           = _mm_cvttpd_epi32(ewrt);
 453 #ifdef __XOP__
 454             eweps            = _mm_frcz_pd(ewrt);
 455 #else
 456             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 457 #endif
 458             twoeweps         = _mm_add_pd(eweps,eweps);
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm_setzero_pd();
 462             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 463             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 464             ewtabFn          = _mm_setzero_pd();
 465             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 466             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 467             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 468             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 469             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 470
 471             /* Analytical LJ-PME */
 472             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 473             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 474             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 475             exponent         = gmx_simd_exp_d(ewcljrsq);
 476             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 477             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 478             /* vvdw6 = [C6 - C6grid * (1-poly)]/r6 */
 479             vvdw6            = _mm_mul_pd(_mm_macc_pd(-c6grid_00,_mm_sub_pd(one,poly),c6_00),rinvsix);
 480             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 481             vvdw             = _mm_msub_pd(vvdw12,one_twelfth,_mm_mul_pd(vvdw6,one_sixth));
 482             /* fvdw = vvdw12/r - (vvdw6/r + (C6grid * exponent * beta^6)/r) */
 483             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);
 484
 485             /* Update potential sum for this i atom from the interaction with this j atom. */
 486             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 487             velecsum         = _mm_add_pd(velecsum,velec);
 488             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 489             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 490
 491             fscal            = _mm_add_pd(felec,fvdw);
 492
 493             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 494
 495             /* Update vectorial force */
 496             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 497             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 498             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 499
 500             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 501             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 502             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 503
 504             /**************************
 505              * CALCULATE INTERACTIONS *
 506              **************************/
 507
 508             r10              = _mm_mul_pd(rsq10,rinv10);
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq10             = _mm_mul_pd(iq1,jq0);
 512
 513             /* EWALD ELECTROSTATICS */
 514
 515             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 516             ewrt             = _mm_mul_pd(r10,ewtabscale);
 517             ewitab           = _mm_cvttpd_epi32(ewrt);
 518 #ifdef __XOP__
 519             eweps            = _mm_frcz_pd(ewrt);
 520 #else
 521             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 522 #endif
 523             twoeweps         = _mm_add_pd(eweps,eweps);
 524             ewitab           = _mm_slli_epi32(ewitab,2);
 525             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 526             ewtabD           = _mm_setzero_pd();
 527             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 528             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 529             ewtabFn          = _mm_setzero_pd();
 530             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 531             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 532             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 533             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 534             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 535
 536             /* Update potential sum for this i atom from the interaction with this j atom. */
 537             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 538             velecsum         = _mm_add_pd(velecsum,velec);
 539
 540             fscal            = felec;
 541
 542             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 543
 544             /* Update vectorial force */
 545             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 546             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 547             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 548
 549             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 550             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 551             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 552
 553             /**************************
 554              * CALCULATE INTERACTIONS *
 555              **************************/
 556
 557             r20              = _mm_mul_pd(rsq20,rinv20);
 558
 559             /* Compute parameters for interactions between i and j atoms */
 560             qq20             = _mm_mul_pd(iq2,jq0);
 561
 562             /* EWALD ELECTROSTATICS */
 563
 564             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 565             ewrt             = _mm_mul_pd(r20,ewtabscale);
 566             ewitab           = _mm_cvttpd_epi32(ewrt);
 567 #ifdef __XOP__
 568             eweps            = _mm_frcz_pd(ewrt);
 569 #else
 570             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 571 #endif
 572             twoeweps         = _mm_add_pd(eweps,eweps);
 573             ewitab           = _mm_slli_epi32(ewitab,2);
 574             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 575             ewtabD           = _mm_setzero_pd();
 576             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 577             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 578             ewtabFn          = _mm_setzero_pd();
 579             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 580             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 581             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 582             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 583             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 584
 585             /* Update potential sum for this i atom from the interaction with this j atom. */
 586             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 587             velecsum         = _mm_add_pd(velecsum,velec);
 588
 589             fscal            = felec;
 590
 591             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 592
 593             /* Update vectorial force */
 594             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 595             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 596             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 597
 598             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 599             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 600             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 601
 602             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 603
 604             /* Inner loop uses 159 flops */
 605         }
 606
 607         /* End of innermost loop */
 608
 609         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 610                                               f+i_coord_offset,fshift+i_shift_offset);
 611
 612         ggid                        = gid[iidx];
 613         /* Update potential energies */
 614         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 615         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 616
 617         /* Increment number of inner iterations */
 618         inneriter                  += j_index_end - j_index_start;
 619
 620         /* Outer loop uses 20 flops */
 621     }
 622
 623     /* Increment number of outer iterations */
 624     outeriter        += nri;
 625
 626     /* Update outer/inner flops */
 627
 628     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*159);
 629 }
 630 /*
 631  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_double
 632  * Electrostatics interaction: Ewald
 633  * VdW interaction:            LJEwald
 634  * Geometry:                   Water3-Particle
 635  * Calculate force/pot:        Force
 636  */
 637 void
 638 nb_kernel_ElecEw_VdwLJEw_GeomW3P1_F_avx_128_fma_double
 639                     (t_nblist                    * gmx_restrict       nlist,
 640                      rvec                        * gmx_restrict          xx,
 641                      rvec                        * gmx_restrict          ff,
 642                      t_forcerec                  * gmx_restrict          fr,
 643                      t_mdatoms                   * gmx_restrict     mdatoms,
 644                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 645                      t_nrnb                      * gmx_restrict        nrnb)
 646 {
 647     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 648      * just 0 for non-waters.
 649      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 650      * jnr indices corresponding to data put in the four positions in the SIMD register.
 651      */
 652     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 653     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 654     int              jnrA,jnrB;
 655     int              j_coord_offsetA,j_coord_offsetB;
 656     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 657     real             rcutoff_scalar;
 658     real             *shiftvec,*fshift,*x,*f;
 659     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 660     int              vdwioffset0;
 661     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 662     int              vdwioffset1;
 663     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 664     int              vdwioffset2;
 665     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 666     int              vdwjidx0A,vdwjidx0B;
 667     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 668     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 669     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 670     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 671     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 672     real             *charge;
 673     int              nvdwtype;
 674     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 675     int              *vdwtype;
 676     real             *vdwparam;
 677     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 678     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 679     __m128d           c6grid_00;
 680     __m128d           c6grid_10;
 681     __m128d           c6grid_20;
 682     real             *vdwgridparam;
 683     __m128d           ewclj,ewclj2,ewclj6,ewcljrsq,poly,exponent,f6A,f6B,sh_lj_ewald;
 684     __m128d           one_half  = _mm_set1_pd(0.5);
 685     __m128d           minus_one = _mm_set1_pd(-1.0);
 686     __m128i          ewitab;
 687     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 688     real             *ewtab;
 689     __m128d          dummy_mask,cutoff_mask;
 690     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 691     __m128d          one     = _mm_set1_pd(1.0);
 692     __m128d          two     = _mm_set1_pd(2.0);
 693     x                = xx[0];
 694     f                = ff[0];
 695
 696     nri              = nlist->nri;
 697     iinr             = nlist->iinr;
 698     jindex           = nlist->jindex;
 699     jjnr             = nlist->jjnr;
 700     shiftidx         = nlist->shift;
 701     gid              = nlist->gid;
 702     shiftvec         = fr->shift_vec[0];
 703     fshift           = fr->fshift[0];
 704     facel            = _mm_set1_pd(fr->epsfac);
 705     charge           = mdatoms->chargeA;
 706     nvdwtype         = fr->ntype;
 707     vdwparam         = fr->nbfp;
 708     vdwtype          = mdatoms->typeA;
 709     vdwgridparam     = fr->ljpme_c6grid;
 710     sh_lj_ewald      = _mm_set1_pd(fr->ic->sh_lj_ewald);
 711     ewclj            = _mm_set1_pd(fr->ewaldcoeff_lj);
 712     ewclj2           = _mm_mul_pd(minus_one,_mm_mul_pd(ewclj,ewclj));
 713
 714     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 715     ewtab            = fr->ic->tabq_coul_F;
 716     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 717     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 718
 719     /* Setup water-specific parameters */
 720     inr              = nlist->iinr[0];
 721     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 722     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 723     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 724     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 725
 726     /* Avoid stupid compiler warnings */
 727     jnrA = jnrB = 0;
 728     j_coord_offsetA = 0;
 729     j_coord_offsetB = 0;
 730
 731     outeriter        = 0;
 732     inneriter        = 0;
 733
 734     /* Start outer loop over neighborlists */
 735     for(iidx=0; iidx<nri; iidx++)
 736     {
 737         /* Load shift vector for this list */
 738         i_shift_offset   = DIM*shiftidx[iidx];
 739
 740         /* Load limits for loop over neighbors */
 741         j_index_start    = jindex[iidx];
 742         j_index_end      = jindex[iidx+1];
 743
 744         /* Get outer coordinate index */
 745         inr              = iinr[iidx];
 746         i_coord_offset   = DIM*inr;
 747
 748         /* Load i particle coords and add shift vector */
 749         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 750                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 751
 752         fix0             = _mm_setzero_pd();
 753         fiy0             = _mm_setzero_pd();
 754         fiz0             = _mm_setzero_pd();
 755         fix1             = _mm_setzero_pd();
 756         fiy1             = _mm_setzero_pd();
 757         fiz1             = _mm_setzero_pd();
 758         fix2             = _mm_setzero_pd();
 759         fiy2             = _mm_setzero_pd();
 760         fiz2             = _mm_setzero_pd();
 761
 762         /* Start inner kernel loop */
 763         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 764         {
 765
 766             /* Get j neighbor index, and coordinate index */
 767             jnrA             = jjnr[jidx];
 768             jnrB             = jjnr[jidx+1];
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771
 772             /* load j atom coordinates */
 773             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 774                                               &jx0,&jy0,&jz0);
 775
 776             /* Calculate displacement vector */
 777             dx00             = _mm_sub_pd(ix0,jx0);
 778             dy00             = _mm_sub_pd(iy0,jy0);
 779             dz00             = _mm_sub_pd(iz0,jz0);
 780             dx10             = _mm_sub_pd(ix1,jx0);
 781             dy10             = _mm_sub_pd(iy1,jy0);
 782             dz10             = _mm_sub_pd(iz1,jz0);
 783             dx20             = _mm_sub_pd(ix2,jx0);
 784             dy20             = _mm_sub_pd(iy2,jy0);
 785             dz20             = _mm_sub_pd(iz2,jz0);
 786
 787             /* Calculate squared distance and things based on it */
 788             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 789             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 790             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 791
 792             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 793             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 794             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 795
 796             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 797             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 798             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 799
 800             /* Load parameters for j particles */
 801             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 802             vdwjidx0A        = 2*vdwtype[jnrA+0];
 803             vdwjidx0B        = 2*vdwtype[jnrB+0];
 804
 805             fjx0             = _mm_setzero_pd();
 806             fjy0             = _mm_setzero_pd();
 807             fjz0             = _mm_setzero_pd();
 808
 809             /**************************
 810              * CALCULATE INTERACTIONS *
 811              **************************/
 812
 813             r00              = _mm_mul_pd(rsq00,rinv00);
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq00             = _mm_mul_pd(iq0,jq0);
 817             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 818                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 819             c6grid_00       = gmx_mm_load_2real_swizzle_pd(vdwgridparam+vdwioffset0+vdwjidx0A,
 820                                                                vdwgridparam+vdwioffset0+vdwjidx0B);
 821
 822             /* EWALD ELECTROSTATICS */
 823
 824             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 825             ewrt             = _mm_mul_pd(r00,ewtabscale);
 826             ewitab           = _mm_cvttpd_epi32(ewrt);
 827 #ifdef __XOP__
 828             eweps            = _mm_frcz_pd(ewrt);
 829 #else
 830             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 831 #endif
 832             twoeweps         = _mm_add_pd(eweps,eweps);
 833             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 834                                          &ewtabF,&ewtabFn);
 835             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 836             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 837
 838             /* Analytical LJ-PME */
 839             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 840             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
 841             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
 842             exponent         = gmx_simd_exp_d(ewcljrsq);
 843             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
 844             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
 845             /* f6A = 6 * C6grid * (1 - poly) */
 846             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
 847             /* f6B = C6grid * exponent * beta^6 */
 848             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
 849             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
 850             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
 851
 852             fscal            = _mm_add_pd(felec,fvdw);
 853
 854             /* Update vectorial force */
 855             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 856             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 857             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 858
 859             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 860             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 861             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 862
 863             /**************************
 864              * CALCULATE INTERACTIONS *
 865              **************************/
 866
 867             r10              = _mm_mul_pd(rsq10,rinv10);
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq10             = _mm_mul_pd(iq1,jq0);
 871
 872             /* EWALD ELECTROSTATICS */
 873
 874             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 875             ewrt             = _mm_mul_pd(r10,ewtabscale);
 876             ewitab           = _mm_cvttpd_epi32(ewrt);
 877 #ifdef __XOP__
 878             eweps            = _mm_frcz_pd(ewrt);
 879 #else
 880             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 881 #endif
 882             twoeweps         = _mm_add_pd(eweps,eweps);
 883             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 884                                          &ewtabF,&ewtabFn);
 885             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 886             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 887
 888             fscal            = felec;
 889
 890             /* Update vectorial force */
 891             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 892             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 893             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 894
 895             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 896             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 897             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 898
 899             /**************************
 900              * CALCULATE INTERACTIONS *
 901              **************************/
 902
 903             r20              = _mm_mul_pd(rsq20,rinv20);
 904
 905             /* Compute parameters for interactions between i and j atoms */
 906             qq20             = _mm_mul_pd(iq2,jq0);
 907
 908             /* EWALD ELECTROSTATICS */
 909
 910             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 911             ewrt             = _mm_mul_pd(r20,ewtabscale);
 912             ewitab           = _mm_cvttpd_epi32(ewrt);
 913 #ifdef __XOP__
 914             eweps            = _mm_frcz_pd(ewrt);
 915 #else
 916             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 917 #endif
 918             twoeweps         = _mm_add_pd(eweps,eweps);
 919             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 920                                          &ewtabF,&ewtabFn);
 921             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 922             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 923
 924             fscal            = felec;
 925
 926             /* Update vectorial force */
 927             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 928             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 929             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 930
 931             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 932             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 933             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 934
 935             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 936
 937             /* Inner loop uses 141 flops */
 938         }
 939
 940         if(jidx<j_index_end)
 941         {
 942
 943             jnrA             = jjnr[jidx];
 944             j_coord_offsetA  = DIM*jnrA;
 945
 946             /* load j atom coordinates */
 947             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 948                                               &jx0,&jy0,&jz0);
 949
 950             /* Calculate displacement vector */
 951             dx00             = _mm_sub_pd(ix0,jx0);
 952             dy00             = _mm_sub_pd(iy0,jy0);
 953             dz00             = _mm_sub_pd(iz0,jz0);
 954             dx10             = _mm_sub_pd(ix1,jx0);
 955             dy10             = _mm_sub_pd(iy1,jy0);
 956             dz10             = _mm_sub_pd(iz1,jz0);
 957             dx20             = _mm_sub_pd(ix2,jx0);
 958             dy20             = _mm_sub_pd(iy2,jy0);
 959             dz20             = _mm_sub_pd(iz2,jz0);
 960
 961             /* Calculate squared distance and things based on it */
 962             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 963             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 964             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 965
 966             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 967             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 968             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 969
 970             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 971             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 972             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 973
 974             /* Load parameters for j particles */
 975             jq0              = _mm_load_sd(charge+jnrA+0);
 976             vdwjidx0A        = 2*vdwtype[jnrA+0];
 977
 978             fjx0             = _mm_setzero_pd();
 979             fjy0             = _mm_setzero_pd();
 980             fjz0             = _mm_setzero_pd();
 981
 982             /**************************
 983              * CALCULATE INTERACTIONS *
 984              **************************/
 985
 986             r00              = _mm_mul_pd(rsq00,rinv00);
 987
 988             /* Compute parameters for interactions between i and j atoms */
 989             qq00             = _mm_mul_pd(iq0,jq0);
 990             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 991             c6grid_00       = gmx_mm_load_1real_pd(vdwgridparam+vdwioffset0+vdwjidx0A);
 992
 993             /* EWALD ELECTROSTATICS */
 994
 995             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 996             ewrt             = _mm_mul_pd(r00,ewtabscale);
 997             ewitab           = _mm_cvttpd_epi32(ewrt);
 998 #ifdef __XOP__
 999             eweps            = _mm_frcz_pd(ewrt);
1000 #else
1001             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1002 #endif
1003             twoeweps         = _mm_add_pd(eweps,eweps);
1004             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1005             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1006             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1007
1008             /* Analytical LJ-PME */
1009             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1010             ewcljrsq         = _mm_mul_pd(ewclj2,rsq00);
1011             ewclj6           = _mm_mul_pd(ewclj2,_mm_mul_pd(ewclj2,ewclj2));
1012             exponent         = gmx_simd_exp_d(ewcljrsq);
1013             /* poly = exp(-(beta*r)^2) * (1 + (beta*r)^2 + (beta*r)^4 /2) */
1014             poly             = _mm_mul_pd(exponent,_mm_macc_pd(_mm_mul_pd(ewcljrsq,ewcljrsq),one_half,_mm_sub_pd(one,ewcljrsq)));
1015             /* f6A = 6 * C6grid * (1 - poly) */
1016             f6A              = _mm_mul_pd(c6grid_00,_mm_sub_pd(one,poly));
1017             /* f6B = C6grid * exponent * beta^6 */
1018             f6B              = _mm_mul_pd(_mm_mul_pd(c6grid_00,one_sixth),_mm_mul_pd(exponent,ewclj6));
1019             /* fvdw = 12*C12/r13 - ((6*C6 - f6A)/r6 + f6B)/r */
1020             fvdw              = _mm_mul_pd(_mm_macc_pd(_mm_msub_pd(c12_00,rinvsix,_mm_sub_pd(c6_00,f6A)),rinvsix,f6B),rinvsq00);
1021
1022             fscal            = _mm_add_pd(felec,fvdw);
1023
1024             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1025
1026             /* Update vectorial force */
1027             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1028             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1029             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1030
1031             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1032             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1033             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1034
1035             /**************************
1036              * CALCULATE INTERACTIONS *
1037              **************************/
1038
1039             r10              = _mm_mul_pd(rsq10,rinv10);
1040
1041             /* Compute parameters for interactions between i and j atoms */
1042             qq10             = _mm_mul_pd(iq1,jq0);
1043
1044             /* EWALD ELECTROSTATICS */
1045
1046             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1047             ewrt             = _mm_mul_pd(r10,ewtabscale);
1048             ewitab           = _mm_cvttpd_epi32(ewrt);
1049 #ifdef __XOP__
1050             eweps            = _mm_frcz_pd(ewrt);
1051 #else
1052             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1053 #endif
1054             twoeweps         = _mm_add_pd(eweps,eweps);
1055             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1056             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1057             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1058
1059             fscal            = felec;
1060
1061             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1062
1063             /* Update vectorial force */
1064             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1065             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1066             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1067
1068             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1069             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1070             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1071
1072             /**************************
1073              * CALCULATE INTERACTIONS *
1074              **************************/
1075
1076             r20              = _mm_mul_pd(rsq20,rinv20);
1077
1078             /* Compute parameters for interactions between i and j atoms */
1079             qq20             = _mm_mul_pd(iq2,jq0);
1080
1081             /* EWALD ELECTROSTATICS */
1082
1083             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1084             ewrt             = _mm_mul_pd(r20,ewtabscale);
1085             ewitab           = _mm_cvttpd_epi32(ewrt);
1086 #ifdef __XOP__
1087             eweps            = _mm_frcz_pd(ewrt);
1088 #else
1089             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1090 #endif
1091             twoeweps         = _mm_add_pd(eweps,eweps);
1092             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1093             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1094             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1095
1096             fscal            = felec;
1097
1098             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1099
1100             /* Update vectorial force */
1101             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1102             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1103             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1104
1105             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1106             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1107             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1108
1109             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1110
1111             /* Inner loop uses 141 flops */
1112         }
1113
1114         /* End of innermost loop */
1115
1116         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1117                                               f+i_coord_offset,fshift+i_shift_offset);
1118
1119         /* Increment number of inner iterations */
1120         inneriter                  += j_index_end - j_index_start;
1121
1122         /* Outer loop uses 18 flops */
1123     }
1124
1125     /* Increment number of outer iterations */
1126     outeriter        += nri;
1127
1128     /* Update outer/inner flops */
1129
1130     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*141);
1131 }