src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_GeomW4P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomW4P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwioffset3;
  73     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B;
  75     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  87     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128d          dummy_mask,cutoff_mask;
  92     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  93     __m128d          one     = _mm_set1_pd(1.0);
  94     __m128d          two     = _mm_set1_pd(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_pd(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 120     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 121     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 122     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128
 129     outeriter        = 0;
 130     inneriter        = 0;
 131
 132     /* Start outer loop over neighborlists */
 133     for(iidx=0; iidx<nri; iidx++)
 134     {
 135         /* Load shift vector for this list */
 136         i_shift_offset   = DIM*shiftidx[iidx];
 137
 138         /* Load limits for loop over neighbors */
 139         j_index_start    = jindex[iidx];
 140         j_index_end      = jindex[iidx+1];
 141
 142         /* Get outer coordinate index */
 143         inr              = iinr[iidx];
 144         i_coord_offset   = DIM*inr;
 145
 146         /* Load i particle coords and add shift vector */
 147         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 148                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 149
 150         fix0             = _mm_setzero_pd();
 151         fiy0             = _mm_setzero_pd();
 152         fiz0             = _mm_setzero_pd();
 153         fix1             = _mm_setzero_pd();
 154         fiy1             = _mm_setzero_pd();
 155         fiz1             = _mm_setzero_pd();
 156         fix2             = _mm_setzero_pd();
 157         fiy2             = _mm_setzero_pd();
 158         fiz2             = _mm_setzero_pd();
 159         fix3             = _mm_setzero_pd();
 160         fiy3             = _mm_setzero_pd();
 161         fiz3             = _mm_setzero_pd();
 162
 163         /* Reset potential sums */
 164         velecsum         = _mm_setzero_pd();
 165         vvdwsum          = _mm_setzero_pd();
 166
 167         /* Start inner kernel loop */
 168         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 169         {
 170
 171             /* Get j neighbor index, and coordinate index */
 172             jnrA             = jjnr[jidx];
 173             jnrB             = jjnr[jidx+1];
 174             j_coord_offsetA  = DIM*jnrA;
 175             j_coord_offsetB  = DIM*jnrB;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               &jx0,&jy0,&jz0);
 180
 181             /* Calculate displacement vector */
 182             dx00             = _mm_sub_pd(ix0,jx0);
 183             dy00             = _mm_sub_pd(iy0,jy0);
 184             dz00             = _mm_sub_pd(iz0,jz0);
 185             dx10             = _mm_sub_pd(ix1,jx0);
 186             dy10             = _mm_sub_pd(iy1,jy0);
 187             dz10             = _mm_sub_pd(iz1,jz0);
 188             dx20             = _mm_sub_pd(ix2,jx0);
 189             dy20             = _mm_sub_pd(iy2,jy0);
 190             dz20             = _mm_sub_pd(iz2,jz0);
 191             dx30             = _mm_sub_pd(ix3,jx0);
 192             dy30             = _mm_sub_pd(iy3,jy0);
 193             dz30             = _mm_sub_pd(iz3,jz0);
 194
 195             /* Calculate squared distance and things based on it */
 196             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 197             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 198             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 199             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 200
 201             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 202             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 203             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 204
 205             rinvsq00         = gmx_mm_inv_pd(rsq00);
 206             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 207             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 208             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 209
 210             /* Load parameters for j particles */
 211             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 212             vdwjidx0A        = 2*vdwtype[jnrA+0];
 213             vdwjidx0B        = 2*vdwtype[jnrB+0];
 214
 215             fjx0             = _mm_setzero_pd();
 216             fjy0             = _mm_setzero_pd();
 217             fjz0             = _mm_setzero_pd();
 218
 219             /**************************
 220              * CALCULATE INTERACTIONS *
 221              **************************/
 222
 223             /* Compute parameters for interactions between i and j atoms */
 224             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 225                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 226
 227             /* LENNARD-JONES DISPERSION/REPULSION */
 228
 229             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 230             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 231             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 232             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 233             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 234
 235             /* Update potential sum for this i atom from the interaction with this j atom. */
 236             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 237
 238             fscal            = fvdw;
 239
 240             /* Update vectorial force */
 241             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 242             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 243             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 244
 245             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 246             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 247             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             r10              = _mm_mul_pd(rsq10,rinv10);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq10             = _mm_mul_pd(iq1,jq0);
 257
 258             /* EWALD ELECTROSTATICS */
 259
 260             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 261             ewrt             = _mm_mul_pd(r10,ewtabscale);
 262             ewitab           = _mm_cvttpd_epi32(ewrt);
 263 #ifdef __XOP__
 264             eweps            = _mm_frcz_pd(ewrt);
 265 #else
 266             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 267 #endif
 268             twoeweps         = _mm_add_pd(eweps,eweps);
 269             ewitab           = _mm_slli_epi32(ewitab,2);
 270             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 271             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 272             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 273             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 274             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 275             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 276             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 277             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 278             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 279             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_pd(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286             /* Update vectorial force */
 287             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 288             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 289             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 290
 291             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 292             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 293             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 294
 295             /**************************
 296              * CALCULATE INTERACTIONS *
 297              **************************/
 298
 299             r20              = _mm_mul_pd(rsq20,rinv20);
 300
 301             /* Compute parameters for interactions between i and j atoms */
 302             qq20             = _mm_mul_pd(iq2,jq0);
 303
 304             /* EWALD ELECTROSTATICS */
 305
 306             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 307             ewrt             = _mm_mul_pd(r20,ewtabscale);
 308             ewitab           = _mm_cvttpd_epi32(ewrt);
 309 #ifdef __XOP__
 310             eweps            = _mm_frcz_pd(ewrt);
 311 #else
 312             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 313 #endif
 314             twoeweps         = _mm_add_pd(eweps,eweps);
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 318             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 319             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 320             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 321             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 322             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 323             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 324             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 325             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 326
 327             /* Update potential sum for this i atom from the interaction with this j atom. */
 328             velecsum         = _mm_add_pd(velecsum,velec);
 329
 330             fscal            = felec;
 331
 332             /* Update vectorial force */
 333             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 334             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 335             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 336
 337             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 338             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 339             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 340
 341             /**************************
 342              * CALCULATE INTERACTIONS *
 343              **************************/
 344
 345             r30              = _mm_mul_pd(rsq30,rinv30);
 346
 347             /* Compute parameters for interactions between i and j atoms */
 348             qq30             = _mm_mul_pd(iq3,jq0);
 349
 350             /* EWALD ELECTROSTATICS */
 351
 352             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 353             ewrt             = _mm_mul_pd(r30,ewtabscale);
 354             ewitab           = _mm_cvttpd_epi32(ewrt);
 355 #ifdef __XOP__
 356             eweps            = _mm_frcz_pd(ewrt);
 357 #else
 358             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 359 #endif
 360             twoeweps         = _mm_add_pd(eweps,eweps);
 361             ewitab           = _mm_slli_epi32(ewitab,2);
 362             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 363             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 364             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 365             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 366             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 367             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 368             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 369             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 370             velec            = _mm_mul_pd(qq30,_mm_sub_pd(rinv30,velec));
 371             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velecsum         = _mm_add_pd(velecsum,velec);
 375
 376             fscal            = felec;
 377
 378             /* Update vectorial force */
 379             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 380             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 381             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 382
 383             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 384             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 385             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 386
 387             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 388
 389             /* Inner loop uses 170 flops */
 390         }
 391
 392         if(jidx<j_index_end)
 393         {
 394
 395             jnrA             = jjnr[jidx];
 396             j_coord_offsetA  = DIM*jnrA;
 397
 398             /* load j atom coordinates */
 399             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 400                                               &jx0,&jy0,&jz0);
 401
 402             /* Calculate displacement vector */
 403             dx00             = _mm_sub_pd(ix0,jx0);
 404             dy00             = _mm_sub_pd(iy0,jy0);
 405             dz00             = _mm_sub_pd(iz0,jz0);
 406             dx10             = _mm_sub_pd(ix1,jx0);
 407             dy10             = _mm_sub_pd(iy1,jy0);
 408             dz10             = _mm_sub_pd(iz1,jz0);
 409             dx20             = _mm_sub_pd(ix2,jx0);
 410             dy20             = _mm_sub_pd(iy2,jy0);
 411             dz20             = _mm_sub_pd(iz2,jz0);
 412             dx30             = _mm_sub_pd(ix3,jx0);
 413             dy30             = _mm_sub_pd(iy3,jy0);
 414             dz30             = _mm_sub_pd(iz3,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             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 421
 422             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 423             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 424             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 425
 426             rinvsq00         = gmx_mm_inv_pd(rsq00);
 427             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 428             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 429             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 430
 431             /* Load parameters for j particles */
 432             jq0              = _mm_load_sd(charge+jnrA+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434
 435             fjx0             = _mm_setzero_pd();
 436             fjy0             = _mm_setzero_pd();
 437             fjz0             = _mm_setzero_pd();
 438
 439             /**************************
 440              * CALCULATE INTERACTIONS *
 441              **************************/
 442
 443             /* Compute parameters for interactions between i and j atoms */
 444             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 445
 446             /* LENNARD-JONES DISPERSION/REPULSION */
 447
 448             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 449             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 450             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 451             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 452             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 453
 454             /* Update potential sum for this i atom from the interaction with this j atom. */
 455             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 456             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 457
 458             fscal            = fvdw;
 459
 460             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 461
 462             /* Update vectorial force */
 463             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 464             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 465             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 466
 467             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 468             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 469             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 470
 471             /**************************
 472              * CALCULATE INTERACTIONS *
 473              **************************/
 474
 475             r10              = _mm_mul_pd(rsq10,rinv10);
 476
 477             /* Compute parameters for interactions between i and j atoms */
 478             qq10             = _mm_mul_pd(iq1,jq0);
 479
 480             /* EWALD ELECTROSTATICS */
 481
 482             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 483             ewrt             = _mm_mul_pd(r10,ewtabscale);
 484             ewitab           = _mm_cvttpd_epi32(ewrt);
 485 #ifdef __XOP__
 486             eweps            = _mm_frcz_pd(ewrt);
 487 #else
 488             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 489 #endif
 490             twoeweps         = _mm_add_pd(eweps,eweps);
 491             ewitab           = _mm_slli_epi32(ewitab,2);
 492             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 493             ewtabD           = _mm_setzero_pd();
 494             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 495             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 496             ewtabFn          = _mm_setzero_pd();
 497             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 498             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 499             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 500             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 501             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 502
 503             /* Update potential sum for this i atom from the interaction with this j atom. */
 504             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 505             velecsum         = _mm_add_pd(velecsum,velec);
 506
 507             fscal            = felec;
 508
 509             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 510
 511             /* Update vectorial force */
 512             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 513             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 514             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 515
 516             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 517             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 518             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 519
 520             /**************************
 521              * CALCULATE INTERACTIONS *
 522              **************************/
 523
 524             r20              = _mm_mul_pd(rsq20,rinv20);
 525
 526             /* Compute parameters for interactions between i and j atoms */
 527             qq20             = _mm_mul_pd(iq2,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(r20,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(qq20,_mm_sub_pd(rinv20,velec));
 550             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,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             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 562             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 563             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 564
 565             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 566             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 567             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 568
 569             /**************************
 570              * CALCULATE INTERACTIONS *
 571              **************************/
 572
 573             r30              = _mm_mul_pd(rsq30,rinv30);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq30             = _mm_mul_pd(iq3,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(r30,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(qq30,_mm_sub_pd(rinv30,velec));
 599             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,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             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 611             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 612             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 613
 614             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 615             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 616             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 617
 618             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 619
 620             /* Inner loop uses 170 flops */
 621         }
 622
 623         /* End of innermost loop */
 624
 625         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 626                                               f+i_coord_offset,fshift+i_shift_offset);
 627
 628         ggid                        = gid[iidx];
 629         /* Update potential energies */
 630         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 631         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 632
 633         /* Increment number of inner iterations */
 634         inneriter                  += j_index_end - j_index_start;
 635
 636         /* Outer loop uses 26 flops */
 637     }
 638
 639     /* Increment number of outer iterations */
 640     outeriter        += nri;
 641
 642     /* Update outer/inner flops */
 643
 644     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*170);
 645 }
 646 /*
 647  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_double
 648  * Electrostatics interaction: Ewald
 649  * VdW interaction:            LennardJones
 650  * Geometry:                   Water4-Particle
 651  * Calculate force/pot:        Force
 652  */
 653 void
 654 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_avx_128_fma_double
 655                     (t_nblist * gmx_restrict                nlist,
 656                      rvec * gmx_restrict                    xx,
 657                      rvec * gmx_restrict                    ff,
 658                      t_forcerec * gmx_restrict              fr,
 659                      t_mdatoms * gmx_restrict               mdatoms,
 660                      nb_kernel_data_t * gmx_restrict        kernel_data,
 661                      t_nrnb * gmx_restrict                  nrnb)
 662 {
 663     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 664      * just 0 for non-waters.
 665      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 666      * jnr indices corresponding to data put in the four positions in the SIMD register.
 667      */
 668     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 669     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 670     int              jnrA,jnrB;
 671     int              j_coord_offsetA,j_coord_offsetB;
 672     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 673     real             rcutoff_scalar;
 674     real             *shiftvec,*fshift,*x,*f;
 675     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 676     int              vdwioffset0;
 677     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 678     int              vdwioffset1;
 679     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 680     int              vdwioffset2;
 681     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 682     int              vdwioffset3;
 683     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 684     int              vdwjidx0A,vdwjidx0B;
 685     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 686     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 687     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 688     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 689     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 690     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 691     real             *charge;
 692     int              nvdwtype;
 693     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 694     int              *vdwtype;
 695     real             *vdwparam;
 696     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 697     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 698     __m128i          ewitab;
 699     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 700     real             *ewtab;
 701     __m128d          dummy_mask,cutoff_mask;
 702     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 703     __m128d          one     = _mm_set1_pd(1.0);
 704     __m128d          two     = _mm_set1_pd(2.0);
 705     x                = xx[0];
 706     f                = ff[0];
 707
 708     nri              = nlist->nri;
 709     iinr             = nlist->iinr;
 710     jindex           = nlist->jindex;
 711     jjnr             = nlist->jjnr;
 712     shiftidx         = nlist->shift;
 713     gid              = nlist->gid;
 714     shiftvec         = fr->shift_vec[0];
 715     fshift           = fr->fshift[0];
 716     facel            = _mm_set1_pd(fr->epsfac);
 717     charge           = mdatoms->chargeA;
 718     nvdwtype         = fr->ntype;
 719     vdwparam         = fr->nbfp;
 720     vdwtype          = mdatoms->typeA;
 721
 722     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 723     ewtab            = fr->ic->tabq_coul_F;
 724     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 725     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 726
 727     /* Setup water-specific parameters */
 728     inr              = nlist->iinr[0];
 729     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 730     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 731     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 732     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 733
 734     /* Avoid stupid compiler warnings */
 735     jnrA = jnrB = 0;
 736     j_coord_offsetA = 0;
 737     j_coord_offsetB = 0;
 738
 739     outeriter        = 0;
 740     inneriter        = 0;
 741
 742     /* Start outer loop over neighborlists */
 743     for(iidx=0; iidx<nri; iidx++)
 744     {
 745         /* Load shift vector for this list */
 746         i_shift_offset   = DIM*shiftidx[iidx];
 747
 748         /* Load limits for loop over neighbors */
 749         j_index_start    = jindex[iidx];
 750         j_index_end      = jindex[iidx+1];
 751
 752         /* Get outer coordinate index */
 753         inr              = iinr[iidx];
 754         i_coord_offset   = DIM*inr;
 755
 756         /* Load i particle coords and add shift vector */
 757         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 758                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 759
 760         fix0             = _mm_setzero_pd();
 761         fiy0             = _mm_setzero_pd();
 762         fiz0             = _mm_setzero_pd();
 763         fix1             = _mm_setzero_pd();
 764         fiy1             = _mm_setzero_pd();
 765         fiz1             = _mm_setzero_pd();
 766         fix2             = _mm_setzero_pd();
 767         fiy2             = _mm_setzero_pd();
 768         fiz2             = _mm_setzero_pd();
 769         fix3             = _mm_setzero_pd();
 770         fiy3             = _mm_setzero_pd();
 771         fiz3             = _mm_setzero_pd();
 772
 773         /* Start inner kernel loop */
 774         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 775         {
 776
 777             /* Get j neighbor index, and coordinate index */
 778             jnrA             = jjnr[jidx];
 779             jnrB             = jjnr[jidx+1];
 780             j_coord_offsetA  = DIM*jnrA;
 781             j_coord_offsetB  = DIM*jnrB;
 782
 783             /* load j atom coordinates */
 784             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 785                                               &jx0,&jy0,&jz0);
 786
 787             /* Calculate displacement vector */
 788             dx00             = _mm_sub_pd(ix0,jx0);
 789             dy00             = _mm_sub_pd(iy0,jy0);
 790             dz00             = _mm_sub_pd(iz0,jz0);
 791             dx10             = _mm_sub_pd(ix1,jx0);
 792             dy10             = _mm_sub_pd(iy1,jy0);
 793             dz10             = _mm_sub_pd(iz1,jz0);
 794             dx20             = _mm_sub_pd(ix2,jx0);
 795             dy20             = _mm_sub_pd(iy2,jy0);
 796             dz20             = _mm_sub_pd(iz2,jz0);
 797             dx30             = _mm_sub_pd(ix3,jx0);
 798             dy30             = _mm_sub_pd(iy3,jy0);
 799             dz30             = _mm_sub_pd(iz3,jz0);
 800
 801             /* Calculate squared distance and things based on it */
 802             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 803             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 804             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 805             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 806
 807             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 808             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 809             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 810
 811             rinvsq00         = gmx_mm_inv_pd(rsq00);
 812             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 813             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 814             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 815
 816             /* Load parameters for j particles */
 817             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 818             vdwjidx0A        = 2*vdwtype[jnrA+0];
 819             vdwjidx0B        = 2*vdwtype[jnrB+0];
 820
 821             fjx0             = _mm_setzero_pd();
 822             fjy0             = _mm_setzero_pd();
 823             fjz0             = _mm_setzero_pd();
 824
 825             /**************************
 826              * CALCULATE INTERACTIONS *
 827              **************************/
 828
 829             /* Compute parameters for interactions between i and j atoms */
 830             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 831                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 832
 833             /* LENNARD-JONES DISPERSION/REPULSION */
 834
 835             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 836             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 837
 838             fscal            = fvdw;
 839
 840             /* Update vectorial force */
 841             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 842             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 843             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 844
 845             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 846             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 847             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             r10              = _mm_mul_pd(rsq10,rinv10);
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq10             = _mm_mul_pd(iq1,jq0);
 857
 858             /* EWALD ELECTROSTATICS */
 859
 860             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 861             ewrt             = _mm_mul_pd(r10,ewtabscale);
 862             ewitab           = _mm_cvttpd_epi32(ewrt);
 863 #ifdef __XOP__
 864             eweps            = _mm_frcz_pd(ewrt);
 865 #else
 866             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 867 #endif
 868             twoeweps         = _mm_add_pd(eweps,eweps);
 869             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 870                                          &ewtabF,&ewtabFn);
 871             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 872             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 873
 874             fscal            = felec;
 875
 876             /* Update vectorial force */
 877             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 878             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 879             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 880
 881             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 882             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 883             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 884
 885             /**************************
 886              * CALCULATE INTERACTIONS *
 887              **************************/
 888
 889             r20              = _mm_mul_pd(rsq20,rinv20);
 890
 891             /* Compute parameters for interactions between i and j atoms */
 892             qq20             = _mm_mul_pd(iq2,jq0);
 893
 894             /* EWALD ELECTROSTATICS */
 895
 896             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 897             ewrt             = _mm_mul_pd(r20,ewtabscale);
 898             ewitab           = _mm_cvttpd_epi32(ewrt);
 899 #ifdef __XOP__
 900             eweps            = _mm_frcz_pd(ewrt);
 901 #else
 902             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 903 #endif
 904             twoeweps         = _mm_add_pd(eweps,eweps);
 905             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 906                                          &ewtabF,&ewtabFn);
 907             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 908             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 909
 910             fscal            = felec;
 911
 912             /* Update vectorial force */
 913             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 914             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 915             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 916
 917             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 918             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 919             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 920
 921             /**************************
 922              * CALCULATE INTERACTIONS *
 923              **************************/
 924
 925             r30              = _mm_mul_pd(rsq30,rinv30);
 926
 927             /* Compute parameters for interactions between i and j atoms */
 928             qq30             = _mm_mul_pd(iq3,jq0);
 929
 930             /* EWALD ELECTROSTATICS */
 931
 932             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 933             ewrt             = _mm_mul_pd(r30,ewtabscale);
 934             ewitab           = _mm_cvttpd_epi32(ewrt);
 935 #ifdef __XOP__
 936             eweps            = _mm_frcz_pd(ewrt);
 937 #else
 938             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 939 #endif
 940             twoeweps         = _mm_add_pd(eweps,eweps);
 941             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 942                                          &ewtabF,&ewtabFn);
 943             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 944             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
 945
 946             fscal            = felec;
 947
 948             /* Update vectorial force */
 949             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 950             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 951             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 952
 953             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 954             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 955             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 956
 957             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 958
 959             /* Inner loop uses 150 flops */
 960         }
 961
 962         if(jidx<j_index_end)
 963         {
 964
 965             jnrA             = jjnr[jidx];
 966             j_coord_offsetA  = DIM*jnrA;
 967
 968             /* load j atom coordinates */
 969             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 970                                               &jx0,&jy0,&jz0);
 971
 972             /* Calculate displacement vector */
 973             dx00             = _mm_sub_pd(ix0,jx0);
 974             dy00             = _mm_sub_pd(iy0,jy0);
 975             dz00             = _mm_sub_pd(iz0,jz0);
 976             dx10             = _mm_sub_pd(ix1,jx0);
 977             dy10             = _mm_sub_pd(iy1,jy0);
 978             dz10             = _mm_sub_pd(iz1,jz0);
 979             dx20             = _mm_sub_pd(ix2,jx0);
 980             dy20             = _mm_sub_pd(iy2,jy0);
 981             dz20             = _mm_sub_pd(iz2,jz0);
 982             dx30             = _mm_sub_pd(ix3,jx0);
 983             dy30             = _mm_sub_pd(iy3,jy0);
 984             dz30             = _mm_sub_pd(iz3,jz0);
 985
 986             /* Calculate squared distance and things based on it */
 987             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 988             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 989             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 990             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 991
 992             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 993             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 994             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 995
 996             rinvsq00         = gmx_mm_inv_pd(rsq00);
 997             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 998             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 999             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
1000
1001             /* Load parameters for j particles */
1002             jq0              = _mm_load_sd(charge+jnrA+0);
1003             vdwjidx0A        = 2*vdwtype[jnrA+0];
1004
1005             fjx0             = _mm_setzero_pd();
1006             fjy0             = _mm_setzero_pd();
1007             fjz0             = _mm_setzero_pd();
1008
1009             /**************************
1010              * CALCULATE INTERACTIONS *
1011              **************************/
1012
1013             /* Compute parameters for interactions between i and j atoms */
1014             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1015
1016             /* LENNARD-JONES DISPERSION/REPULSION */
1017
1018             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
1019             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
1020
1021             fscal            = fvdw;
1022
1023             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1024
1025             /* Update vectorial force */
1026             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1027             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1028             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1029
1030             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1031             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1032             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1033
1034             /**************************
1035              * CALCULATE INTERACTIONS *
1036              **************************/
1037
1038             r10              = _mm_mul_pd(rsq10,rinv10);
1039
1040             /* Compute parameters for interactions between i and j atoms */
1041             qq10             = _mm_mul_pd(iq1,jq0);
1042
1043             /* EWALD ELECTROSTATICS */
1044
1045             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1046             ewrt             = _mm_mul_pd(r10,ewtabscale);
1047             ewitab           = _mm_cvttpd_epi32(ewrt);
1048 #ifdef __XOP__
1049             eweps            = _mm_frcz_pd(ewrt);
1050 #else
1051             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1052 #endif
1053             twoeweps         = _mm_add_pd(eweps,eweps);
1054             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1055             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1056             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1057
1058             fscal            = felec;
1059
1060             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1061
1062             /* Update vectorial force */
1063             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1064             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1065             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1066
1067             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1068             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1069             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1070
1071             /**************************
1072              * CALCULATE INTERACTIONS *
1073              **************************/
1074
1075             r20              = _mm_mul_pd(rsq20,rinv20);
1076
1077             /* Compute parameters for interactions between i and j atoms */
1078             qq20             = _mm_mul_pd(iq2,jq0);
1079
1080             /* EWALD ELECTROSTATICS */
1081
1082             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1083             ewrt             = _mm_mul_pd(r20,ewtabscale);
1084             ewitab           = _mm_cvttpd_epi32(ewrt);
1085 #ifdef __XOP__
1086             eweps            = _mm_frcz_pd(ewrt);
1087 #else
1088             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1089 #endif
1090             twoeweps         = _mm_add_pd(eweps,eweps);
1091             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1092             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1093             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1098
1099             /* Update vectorial force */
1100             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1101             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1102             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1103
1104             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1105             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1106             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             r30              = _mm_mul_pd(rsq30,rinv30);
1113
1114             /* Compute parameters for interactions between i and j atoms */
1115             qq30             = _mm_mul_pd(iq3,jq0);
1116
1117             /* EWALD ELECTROSTATICS */
1118
1119             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1120             ewrt             = _mm_mul_pd(r30,ewtabscale);
1121             ewitab           = _mm_cvttpd_epi32(ewrt);
1122 #ifdef __XOP__
1123             eweps            = _mm_frcz_pd(ewrt);
1124 #else
1125             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1126 #endif
1127             twoeweps         = _mm_add_pd(eweps,eweps);
1128             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1129             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1130             felec            = _mm_mul_pd(_mm_mul_pd(qq30,rinv30),_mm_sub_pd(rinvsq30,felec));
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1135
1136             /* Update vectorial force */
1137             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1138             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1139             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1140
1141             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1142             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1143             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1144
1145             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1146
1147             /* Inner loop uses 150 flops */
1148         }
1149
1150         /* End of innermost loop */
1151
1152         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1153                                               f+i_coord_offset,fshift+i_shift_offset);
1154
1155         /* Increment number of inner iterations */
1156         inneriter                  += j_index_end - j_index_start;
1157
1158         /* Outer loop uses 24 flops */
1159     }
1160
1161     /* Increment number of outer iterations */
1162     outeriter        += nri;
1163
1164     /* Update outer/inner flops */
1165
1166     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*150);
1167 }