src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_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              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128i          vfitab;
  86     __m128i          ifour       = _mm_set1_epi32(4);
  87     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  88     real             *vftab;
  89     __m128i          ewitab;
  90     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  91     real             *ewtab;
  92     __m128d          dummy_mask,cutoff_mask;
  93     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  94     __m128d          one     = _mm_set1_pd(1.0);
  95     __m128d          two     = _mm_set1_pd(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_pd(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     nvdwtype         = fr->ntype;
 110     vdwparam         = fr->nbfp;
 111     vdwtype          = mdatoms->typeA;
 112
 113     vftab            = kernel_data->table_vdw->data;
 114     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 115
 116     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 124     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 125     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 126     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     /* Start outer loop over neighborlists */
 137     for(iidx=0; iidx<nri; iidx++)
 138     {
 139         /* Load shift vector for this list */
 140         i_shift_offset   = DIM*shiftidx[iidx];
 141
 142         /* Load limits for loop over neighbors */
 143         j_index_start    = jindex[iidx];
 144         j_index_end      = jindex[iidx+1];
 145
 146         /* Get outer coordinate index */
 147         inr              = iinr[iidx];
 148         i_coord_offset   = DIM*inr;
 149
 150         /* Load i particle coords and add shift vector */
 151         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 152                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 153
 154         fix0             = _mm_setzero_pd();
 155         fiy0             = _mm_setzero_pd();
 156         fiz0             = _mm_setzero_pd();
 157         fix1             = _mm_setzero_pd();
 158         fiy1             = _mm_setzero_pd();
 159         fiz1             = _mm_setzero_pd();
 160         fix2             = _mm_setzero_pd();
 161         fiy2             = _mm_setzero_pd();
 162         fiz2             = _mm_setzero_pd();
 163
 164         /* Reset potential sums */
 165         velecsum         = _mm_setzero_pd();
 166         vvdwsum          = _mm_setzero_pd();
 167
 168         /* Start inner kernel loop */
 169         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 170         {
 171
 172             /* Get j neighbor index, and coordinate index */
 173             jnrA             = jjnr[jidx];
 174             jnrB             = jjnr[jidx+1];
 175             j_coord_offsetA  = DIM*jnrA;
 176             j_coord_offsetB  = DIM*jnrB;
 177
 178             /* load j atom coordinates */
 179             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx00             = _mm_sub_pd(ix0,jx0);
 184             dy00             = _mm_sub_pd(iy0,jy0);
 185             dz00             = _mm_sub_pd(iz0,jz0);
 186             dx10             = _mm_sub_pd(ix1,jx0);
 187             dy10             = _mm_sub_pd(iy1,jy0);
 188             dz10             = _mm_sub_pd(iz1,jz0);
 189             dx20             = _mm_sub_pd(ix2,jx0);
 190             dy20             = _mm_sub_pd(iy2,jy0);
 191             dz20             = _mm_sub_pd(iz2,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 195             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 196             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 197
 198             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 199             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 200             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 201
 202             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 203             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 204             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 208             vdwjidx0A        = 2*vdwtype[jnrA+0];
 209             vdwjidx0B        = 2*vdwtype[jnrB+0];
 210
 211             fjx0             = _mm_setzero_pd();
 212             fjy0             = _mm_setzero_pd();
 213             fjz0             = _mm_setzero_pd();
 214
 215             /**************************
 216              * CALCULATE INTERACTIONS *
 217              **************************/
 218
 219             r00              = _mm_mul_pd(rsq00,rinv00);
 220
 221             /* Compute parameters for interactions between i and j atoms */
 222             qq00             = _mm_mul_pd(iq0,jq0);
 223             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 224                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 225
 226             /* Calculate table index by multiplying r with table scale and truncate to integer */
 227             rt               = _mm_mul_pd(r00,vftabscale);
 228             vfitab           = _mm_cvttpd_epi32(rt);
 229 #ifdef __XOP__
 230             vfeps            = _mm_frcz_pd(rt);
 231 #else
 232             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 233 #endif
 234             twovfeps         = _mm_add_pd(vfeps,vfeps);
 235             vfitab           = _mm_slli_epi32(vfitab,3);
 236
 237             /* EWALD ELECTROSTATICS */
 238
 239             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 240             ewrt             = _mm_mul_pd(r00,ewtabscale);
 241             ewitab           = _mm_cvttpd_epi32(ewrt);
 242 #ifdef __XOP__
 243             eweps            = _mm_frcz_pd(ewrt);
 244 #else
 245             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 246 #endif
 247             twoeweps         = _mm_add_pd(eweps,eweps);
 248             ewitab           = _mm_slli_epi32(ewitab,2);
 249             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 250             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 251             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 252             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 253             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 254             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 255             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 256             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 257             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 258             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 259
 260             /* CUBIC SPLINE TABLE DISPERSION */
 261             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 262             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 263             GMX_MM_TRANSPOSE2_PD(Y,F);
 264             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 265             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 266             GMX_MM_TRANSPOSE2_PD(G,H);
 267             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 268             VV               = _mm_macc_pd(vfeps,Fp,Y);
 269             vvdw6            = _mm_mul_pd(c6_00,VV);
 270             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 271             fvdw6            = _mm_mul_pd(c6_00,FF);
 272
 273             /* CUBIC SPLINE TABLE REPULSION */
 274             vfitab           = _mm_add_epi32(vfitab,ifour);
 275             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 277             GMX_MM_TRANSPOSE2_PD(Y,F);
 278             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 279             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 280             GMX_MM_TRANSPOSE2_PD(G,H);
 281             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 282             VV               = _mm_macc_pd(vfeps,Fp,Y);
 283             vvdw12           = _mm_mul_pd(c12_00,VV);
 284             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 285             fvdw12           = _mm_mul_pd(c12_00,FF);
 286             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 287             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 288
 289             /* Update potential sum for this i atom from the interaction with this j atom. */
 290             velecsum         = _mm_add_pd(velecsum,velec);
 291             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 292
 293             fscal            = _mm_add_pd(felec,fvdw);
 294
 295             /* Update vectorial force */
 296             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 297             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 298             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 299
 300             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 301             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 302             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 303
 304             /**************************
 305              * CALCULATE INTERACTIONS *
 306              **************************/
 307
 308             r10              = _mm_mul_pd(rsq10,rinv10);
 309
 310             /* Compute parameters for interactions between i and j atoms */
 311             qq10             = _mm_mul_pd(iq1,jq0);
 312
 313             /* EWALD ELECTROSTATICS */
 314
 315             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 316             ewrt             = _mm_mul_pd(r10,ewtabscale);
 317             ewitab           = _mm_cvttpd_epi32(ewrt);
 318 #ifdef __XOP__
 319             eweps            = _mm_frcz_pd(ewrt);
 320 #else
 321             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 322 #endif
 323             twoeweps         = _mm_add_pd(eweps,eweps);
 324             ewitab           = _mm_slli_epi32(ewitab,2);
 325             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 326             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 327             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 328             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 329             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 330             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 331             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 332             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 333             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 334             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 335
 336             /* Update potential sum for this i atom from the interaction with this j atom. */
 337             velecsum         = _mm_add_pd(velecsum,velec);
 338
 339             fscal            = felec;
 340
 341             /* Update vectorial force */
 342             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 343             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 344             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 345
 346             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 347             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 348             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             r20              = _mm_mul_pd(rsq20,rinv20);
 355
 356             /* Compute parameters for interactions between i and j atoms */
 357             qq20             = _mm_mul_pd(iq2,jq0);
 358
 359             /* EWALD ELECTROSTATICS */
 360
 361             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 362             ewrt             = _mm_mul_pd(r20,ewtabscale);
 363             ewitab           = _mm_cvttpd_epi32(ewrt);
 364 #ifdef __XOP__
 365             eweps            = _mm_frcz_pd(ewrt);
 366 #else
 367             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 368 #endif
 369             twoeweps         = _mm_add_pd(eweps,eweps);
 370             ewitab           = _mm_slli_epi32(ewitab,2);
 371             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 372             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 373             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 374             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 375             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 376             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 377             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 378             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 379             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 380             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 381
 382             /* Update potential sum for this i atom from the interaction with this j atom. */
 383             velecsum         = _mm_add_pd(velecsum,velec);
 384
 385             fscal            = felec;
 386
 387             /* Update vectorial force */
 388             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 389             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 390             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 391
 392             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 393             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 394             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 395
 396             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 397
 398             /* Inner loop uses 169 flops */
 399         }
 400
 401         if(jidx<j_index_end)
 402         {
 403
 404             jnrA             = jjnr[jidx];
 405             j_coord_offsetA  = DIM*jnrA;
 406
 407             /* load j atom coordinates */
 408             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 409                                               &jx0,&jy0,&jz0);
 410
 411             /* Calculate displacement vector */
 412             dx00             = _mm_sub_pd(ix0,jx0);
 413             dy00             = _mm_sub_pd(iy0,jy0);
 414             dz00             = _mm_sub_pd(iz0,jz0);
 415             dx10             = _mm_sub_pd(ix1,jx0);
 416             dy10             = _mm_sub_pd(iy1,jy0);
 417             dz10             = _mm_sub_pd(iz1,jz0);
 418             dx20             = _mm_sub_pd(ix2,jx0);
 419             dy20             = _mm_sub_pd(iy2,jy0);
 420             dz20             = _mm_sub_pd(iz2,jz0);
 421
 422             /* Calculate squared distance and things based on it */
 423             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 424             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 425             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 426
 427             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 428             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 429             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 430
 431             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 432             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 433             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 434
 435             /* Load parameters for j particles */
 436             jq0              = _mm_load_sd(charge+jnrA+0);
 437             vdwjidx0A        = 2*vdwtype[jnrA+0];
 438
 439             fjx0             = _mm_setzero_pd();
 440             fjy0             = _mm_setzero_pd();
 441             fjz0             = _mm_setzero_pd();
 442
 443             /**************************
 444              * CALCULATE INTERACTIONS *
 445              **************************/
 446
 447             r00              = _mm_mul_pd(rsq00,rinv00);
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             qq00             = _mm_mul_pd(iq0,jq0);
 451             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 452
 453             /* Calculate table index by multiplying r with table scale and truncate to integer */
 454             rt               = _mm_mul_pd(r00,vftabscale);
 455             vfitab           = _mm_cvttpd_epi32(rt);
 456 #ifdef __XOP__
 457             vfeps            = _mm_frcz_pd(rt);
 458 #else
 459             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 460 #endif
 461             twovfeps         = _mm_add_pd(vfeps,vfeps);
 462             vfitab           = _mm_slli_epi32(vfitab,3);
 463
 464             /* EWALD ELECTROSTATICS */
 465
 466             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 467             ewrt             = _mm_mul_pd(r00,ewtabscale);
 468             ewitab           = _mm_cvttpd_epi32(ewrt);
 469 #ifdef __XOP__
 470             eweps            = _mm_frcz_pd(ewrt);
 471 #else
 472             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 473 #endif
 474             twoeweps         = _mm_add_pd(eweps,eweps);
 475             ewitab           = _mm_slli_epi32(ewitab,2);
 476             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 477             ewtabD           = _mm_setzero_pd();
 478             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 479             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 480             ewtabFn          = _mm_setzero_pd();
 481             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 482             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 483             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 484             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 485             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 486
 487             /* CUBIC SPLINE TABLE DISPERSION */
 488             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 489             F                = _mm_setzero_pd();
 490             GMX_MM_TRANSPOSE2_PD(Y,F);
 491             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 492             H                = _mm_setzero_pd();
 493             GMX_MM_TRANSPOSE2_PD(G,H);
 494             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 495             VV               = _mm_macc_pd(vfeps,Fp,Y);
 496             vvdw6            = _mm_mul_pd(c6_00,VV);
 497             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 498             fvdw6            = _mm_mul_pd(c6_00,FF);
 499
 500             /* CUBIC SPLINE TABLE REPULSION */
 501             vfitab           = _mm_add_epi32(vfitab,ifour);
 502             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 503             F                = _mm_setzero_pd();
 504             GMX_MM_TRANSPOSE2_PD(Y,F);
 505             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 506             H                = _mm_setzero_pd();
 507             GMX_MM_TRANSPOSE2_PD(G,H);
 508             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 509             VV               = _mm_macc_pd(vfeps,Fp,Y);
 510             vvdw12           = _mm_mul_pd(c12_00,VV);
 511             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 512             fvdw12           = _mm_mul_pd(c12_00,FF);
 513             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 514             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 515
 516             /* Update potential sum for this i atom from the interaction with this j atom. */
 517             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 518             velecsum         = _mm_add_pd(velecsum,velec);
 519             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 520             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 521
 522             fscal            = _mm_add_pd(felec,fvdw);
 523
 524             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 528             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 529             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 530
 531             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 532             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 533             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             r10              = _mm_mul_pd(rsq10,rinv10);
 540
 541             /* Compute parameters for interactions between i and j atoms */
 542             qq10             = _mm_mul_pd(iq1,jq0);
 543
 544             /* EWALD ELECTROSTATICS */
 545
 546             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 547             ewrt             = _mm_mul_pd(r10,ewtabscale);
 548             ewitab           = _mm_cvttpd_epi32(ewrt);
 549 #ifdef __XOP__
 550             eweps            = _mm_frcz_pd(ewrt);
 551 #else
 552             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 553 #endif
 554             twoeweps         = _mm_add_pd(eweps,eweps);
 555             ewitab           = _mm_slli_epi32(ewitab,2);
 556             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 557             ewtabD           = _mm_setzero_pd();
 558             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 559             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 560             ewtabFn          = _mm_setzero_pd();
 561             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 562             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 563             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 564             velec            = _mm_mul_pd(qq10,_mm_sub_pd(rinv10,velec));
 565             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 566
 567             /* Update potential sum for this i atom from the interaction with this j atom. */
 568             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 569             velecsum         = _mm_add_pd(velecsum,velec);
 570
 571             fscal            = felec;
 572
 573             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 574
 575             /* Update vectorial force */
 576             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 577             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 578             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 579
 580             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 581             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 582             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             r20              = _mm_mul_pd(rsq20,rinv20);
 589
 590             /* Compute parameters for interactions between i and j atoms */
 591             qq20             = _mm_mul_pd(iq2,jq0);
 592
 593             /* EWALD ELECTROSTATICS */
 594
 595             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 596             ewrt             = _mm_mul_pd(r20,ewtabscale);
 597             ewitab           = _mm_cvttpd_epi32(ewrt);
 598 #ifdef __XOP__
 599             eweps            = _mm_frcz_pd(ewrt);
 600 #else
 601             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 602 #endif
 603             twoeweps         = _mm_add_pd(eweps,eweps);
 604             ewitab           = _mm_slli_epi32(ewitab,2);
 605             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 606             ewtabD           = _mm_setzero_pd();
 607             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 608             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 609             ewtabFn          = _mm_setzero_pd();
 610             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 611             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 612             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 613             velec            = _mm_mul_pd(qq20,_mm_sub_pd(rinv20,velec));
 614             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 618             velecsum         = _mm_add_pd(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 623
 624             /* Update vectorial force */
 625             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 626             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 627             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 628
 629             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 630             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 631             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 632
 633             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 634
 635             /* Inner loop uses 169 flops */
 636         }
 637
 638         /* End of innermost loop */
 639
 640         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 641                                               f+i_coord_offset,fshift+i_shift_offset);
 642
 643         ggid                        = gid[iidx];
 644         /* Update potential energies */
 645         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 646         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 647
 648         /* Increment number of inner iterations */
 649         inneriter                  += j_index_end - j_index_start;
 650
 651         /* Outer loop uses 20 flops */
 652     }
 653
 654     /* Increment number of outer iterations */
 655     outeriter        += nri;
 656
 657     /* Update outer/inner flops */
 658
 659     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*169);
 660 }
 661 /*
 662  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 663  * Electrostatics interaction: Ewald
 664  * VdW interaction:            CubicSplineTable
 665  * Geometry:                   Water3-Particle
 666  * Calculate force/pot:        Force
 667  */
 668 void
 669 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 670                     (t_nblist * gmx_restrict                nlist,
 671                      rvec * gmx_restrict                    xx,
 672                      rvec * gmx_restrict                    ff,
 673                      t_forcerec * gmx_restrict              fr,
 674                      t_mdatoms * gmx_restrict               mdatoms,
 675                      nb_kernel_data_t * gmx_restrict        kernel_data,
 676                      t_nrnb * gmx_restrict                  nrnb)
 677 {
 678     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 679      * just 0 for non-waters.
 680      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 681      * jnr indices corresponding to data put in the four positions in the SIMD register.
 682      */
 683     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 684     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 685     int              jnrA,jnrB;
 686     int              j_coord_offsetA,j_coord_offsetB;
 687     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 688     real             rcutoff_scalar;
 689     real             *shiftvec,*fshift,*x,*f;
 690     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 691     int              vdwioffset0;
 692     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 693     int              vdwioffset1;
 694     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 695     int              vdwioffset2;
 696     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 697     int              vdwjidx0A,vdwjidx0B;
 698     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 699     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 700     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 701     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 702     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 703     real             *charge;
 704     int              nvdwtype;
 705     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 706     int              *vdwtype;
 707     real             *vdwparam;
 708     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 709     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 710     __m128i          vfitab;
 711     __m128i          ifour       = _mm_set1_epi32(4);
 712     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 713     real             *vftab;
 714     __m128i          ewitab;
 715     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 716     real             *ewtab;
 717     __m128d          dummy_mask,cutoff_mask;
 718     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 719     __m128d          one     = _mm_set1_pd(1.0);
 720     __m128d          two     = _mm_set1_pd(2.0);
 721     x                = xx[0];
 722     f                = ff[0];
 723
 724     nri              = nlist->nri;
 725     iinr             = nlist->iinr;
 726     jindex           = nlist->jindex;
 727     jjnr             = nlist->jjnr;
 728     shiftidx         = nlist->shift;
 729     gid              = nlist->gid;
 730     shiftvec         = fr->shift_vec[0];
 731     fshift           = fr->fshift[0];
 732     facel            = _mm_set1_pd(fr->epsfac);
 733     charge           = mdatoms->chargeA;
 734     nvdwtype         = fr->ntype;
 735     vdwparam         = fr->nbfp;
 736     vdwtype          = mdatoms->typeA;
 737
 738     vftab            = kernel_data->table_vdw->data;
 739     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 740
 741     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 742     ewtab            = fr->ic->tabq_coul_F;
 743     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 744     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 745
 746     /* Setup water-specific parameters */
 747     inr              = nlist->iinr[0];
 748     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 749     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 750     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 751     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 752
 753     /* Avoid stupid compiler warnings */
 754     jnrA = jnrB = 0;
 755     j_coord_offsetA = 0;
 756     j_coord_offsetB = 0;
 757
 758     outeriter        = 0;
 759     inneriter        = 0;
 760
 761     /* Start outer loop over neighborlists */
 762     for(iidx=0; iidx<nri; iidx++)
 763     {
 764         /* Load shift vector for this list */
 765         i_shift_offset   = DIM*shiftidx[iidx];
 766
 767         /* Load limits for loop over neighbors */
 768         j_index_start    = jindex[iidx];
 769         j_index_end      = jindex[iidx+1];
 770
 771         /* Get outer coordinate index */
 772         inr              = iinr[iidx];
 773         i_coord_offset   = DIM*inr;
 774
 775         /* Load i particle coords and add shift vector */
 776         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 777                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 778
 779         fix0             = _mm_setzero_pd();
 780         fiy0             = _mm_setzero_pd();
 781         fiz0             = _mm_setzero_pd();
 782         fix1             = _mm_setzero_pd();
 783         fiy1             = _mm_setzero_pd();
 784         fiz1             = _mm_setzero_pd();
 785         fix2             = _mm_setzero_pd();
 786         fiy2             = _mm_setzero_pd();
 787         fiz2             = _mm_setzero_pd();
 788
 789         /* Start inner kernel loop */
 790         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 791         {
 792
 793             /* Get j neighbor index, and coordinate index */
 794             jnrA             = jjnr[jidx];
 795             jnrB             = jjnr[jidx+1];
 796             j_coord_offsetA  = DIM*jnrA;
 797             j_coord_offsetB  = DIM*jnrB;
 798
 799             /* load j atom coordinates */
 800             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 801                                               &jx0,&jy0,&jz0);
 802
 803             /* Calculate displacement vector */
 804             dx00             = _mm_sub_pd(ix0,jx0);
 805             dy00             = _mm_sub_pd(iy0,jy0);
 806             dz00             = _mm_sub_pd(iz0,jz0);
 807             dx10             = _mm_sub_pd(ix1,jx0);
 808             dy10             = _mm_sub_pd(iy1,jy0);
 809             dz10             = _mm_sub_pd(iz1,jz0);
 810             dx20             = _mm_sub_pd(ix2,jx0);
 811             dy20             = _mm_sub_pd(iy2,jy0);
 812             dz20             = _mm_sub_pd(iz2,jz0);
 813
 814             /* Calculate squared distance and things based on it */
 815             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 816             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 817             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 818
 819             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 820             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 821             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 822
 823             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 824             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 825             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 826
 827             /* Load parameters for j particles */
 828             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 829             vdwjidx0A        = 2*vdwtype[jnrA+0];
 830             vdwjidx0B        = 2*vdwtype[jnrB+0];
 831
 832             fjx0             = _mm_setzero_pd();
 833             fjy0             = _mm_setzero_pd();
 834             fjz0             = _mm_setzero_pd();
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             r00              = _mm_mul_pd(rsq00,rinv00);
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             qq00             = _mm_mul_pd(iq0,jq0);
 844             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 845                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 846
 847             /* Calculate table index by multiplying r with table scale and truncate to integer */
 848             rt               = _mm_mul_pd(r00,vftabscale);
 849             vfitab           = _mm_cvttpd_epi32(rt);
 850 #ifdef __XOP__
 851             vfeps            = _mm_frcz_pd(rt);
 852 #else
 853             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 854 #endif
 855             twovfeps         = _mm_add_pd(vfeps,vfeps);
 856             vfitab           = _mm_slli_epi32(vfitab,3);
 857
 858             /* EWALD ELECTROSTATICS */
 859
 860             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 861             ewrt             = _mm_mul_pd(r00,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(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 873
 874             /* CUBIC SPLINE TABLE DISPERSION */
 875             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 876             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 877             GMX_MM_TRANSPOSE2_PD(Y,F);
 878             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 879             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 880             GMX_MM_TRANSPOSE2_PD(G,H);
 881             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 882             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 883             fvdw6            = _mm_mul_pd(c6_00,FF);
 884
 885             /* CUBIC SPLINE TABLE REPULSION */
 886             vfitab           = _mm_add_epi32(vfitab,ifour);
 887             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 888             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 889             GMX_MM_TRANSPOSE2_PD(Y,F);
 890             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 891             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 892             GMX_MM_TRANSPOSE2_PD(G,H);
 893             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 894             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 895             fvdw12           = _mm_mul_pd(c12_00,FF);
 896             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 897
 898             fscal            = _mm_add_pd(felec,fvdw);
 899
 900             /* Update vectorial force */
 901             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 902             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 903             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 904
 905             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 906             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 907             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 908
 909             /**************************
 910              * CALCULATE INTERACTIONS *
 911              **************************/
 912
 913             r10              = _mm_mul_pd(rsq10,rinv10);
 914
 915             /* Compute parameters for interactions between i and j atoms */
 916             qq10             = _mm_mul_pd(iq1,jq0);
 917
 918             /* EWALD ELECTROSTATICS */
 919
 920             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 921             ewrt             = _mm_mul_pd(r10,ewtabscale);
 922             ewitab           = _mm_cvttpd_epi32(ewrt);
 923 #ifdef __XOP__
 924             eweps            = _mm_frcz_pd(ewrt);
 925 #else
 926             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 927 #endif
 928             twoeweps         = _mm_add_pd(eweps,eweps);
 929             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 930                                          &ewtabF,&ewtabFn);
 931             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 932             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
 933
 934             fscal            = felec;
 935
 936             /* Update vectorial force */
 937             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 938             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 939             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 940
 941             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 942             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 943             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 944
 945             /**************************
 946              * CALCULATE INTERACTIONS *
 947              **************************/
 948
 949             r20              = _mm_mul_pd(rsq20,rinv20);
 950
 951             /* Compute parameters for interactions between i and j atoms */
 952             qq20             = _mm_mul_pd(iq2,jq0);
 953
 954             /* EWALD ELECTROSTATICS */
 955
 956             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 957             ewrt             = _mm_mul_pd(r20,ewtabscale);
 958             ewitab           = _mm_cvttpd_epi32(ewrt);
 959 #ifdef __XOP__
 960             eweps            = _mm_frcz_pd(ewrt);
 961 #else
 962             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 963 #endif
 964             twoeweps         = _mm_add_pd(eweps,eweps);
 965             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 966                                          &ewtabF,&ewtabFn);
 967             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 968             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
 969
 970             fscal            = felec;
 971
 972             /* Update vectorial force */
 973             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 974             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 975             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 976
 977             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 978             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 979             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 980
 981             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 982
 983             /* Inner loop uses 146 flops */
 984         }
 985
 986         if(jidx<j_index_end)
 987         {
 988
 989             jnrA             = jjnr[jidx];
 990             j_coord_offsetA  = DIM*jnrA;
 991
 992             /* load j atom coordinates */
 993             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 994                                               &jx0,&jy0,&jz0);
 995
 996             /* Calculate displacement vector */
 997             dx00             = _mm_sub_pd(ix0,jx0);
 998             dy00             = _mm_sub_pd(iy0,jy0);
 999             dz00             = _mm_sub_pd(iz0,jz0);
1000             dx10             = _mm_sub_pd(ix1,jx0);
1001             dy10             = _mm_sub_pd(iy1,jy0);
1002             dz10             = _mm_sub_pd(iz1,jz0);
1003             dx20             = _mm_sub_pd(ix2,jx0);
1004             dy20             = _mm_sub_pd(iy2,jy0);
1005             dz20             = _mm_sub_pd(iz2,jz0);
1006
1007             /* Calculate squared distance and things based on it */
1008             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
1009             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
1010             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
1011
1012             rinv00           = gmx_mm_invsqrt_pd(rsq00);
1013             rinv10           = gmx_mm_invsqrt_pd(rsq10);
1014             rinv20           = gmx_mm_invsqrt_pd(rsq20);
1015
1016             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
1017             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
1018             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
1019
1020             /* Load parameters for j particles */
1021             jq0              = _mm_load_sd(charge+jnrA+0);
1022             vdwjidx0A        = 2*vdwtype[jnrA+0];
1023
1024             fjx0             = _mm_setzero_pd();
1025             fjy0             = _mm_setzero_pd();
1026             fjz0             = _mm_setzero_pd();
1027
1028             /**************************
1029              * CALCULATE INTERACTIONS *
1030              **************************/
1031
1032             r00              = _mm_mul_pd(rsq00,rinv00);
1033
1034             /* Compute parameters for interactions between i and j atoms */
1035             qq00             = _mm_mul_pd(iq0,jq0);
1036             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
1037
1038             /* Calculate table index by multiplying r with table scale and truncate to integer */
1039             rt               = _mm_mul_pd(r00,vftabscale);
1040             vfitab           = _mm_cvttpd_epi32(rt);
1041 #ifdef __XOP__
1042             vfeps            = _mm_frcz_pd(rt);
1043 #else
1044             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
1045 #endif
1046             twovfeps         = _mm_add_pd(vfeps,vfeps);
1047             vfitab           = _mm_slli_epi32(vfitab,3);
1048
1049             /* EWALD ELECTROSTATICS */
1050
1051             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1052             ewrt             = _mm_mul_pd(r00,ewtabscale);
1053             ewitab           = _mm_cvttpd_epi32(ewrt);
1054 #ifdef __XOP__
1055             eweps            = _mm_frcz_pd(ewrt);
1056 #else
1057             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1058 #endif
1059             twoeweps         = _mm_add_pd(eweps,eweps);
1060             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1061             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1062             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
1063
1064             /* CUBIC SPLINE TABLE DISPERSION */
1065             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1066             F                = _mm_setzero_pd();
1067             GMX_MM_TRANSPOSE2_PD(Y,F);
1068             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1069             H                = _mm_setzero_pd();
1070             GMX_MM_TRANSPOSE2_PD(G,H);
1071             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1072             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1073             fvdw6            = _mm_mul_pd(c6_00,FF);
1074
1075             /* CUBIC SPLINE TABLE REPULSION */
1076             vfitab           = _mm_add_epi32(vfitab,ifour);
1077             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
1078             F                = _mm_setzero_pd();
1079             GMX_MM_TRANSPOSE2_PD(Y,F);
1080             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
1081             H                = _mm_setzero_pd();
1082             GMX_MM_TRANSPOSE2_PD(G,H);
1083             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
1084             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1085             fvdw12           = _mm_mul_pd(c12_00,FF);
1086             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1087
1088             fscal            = _mm_add_pd(felec,fvdw);
1089
1090             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1091
1092             /* Update vectorial force */
1093             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1094             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1095             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1096
1097             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1098             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1099             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1100
1101             /**************************
1102              * CALCULATE INTERACTIONS *
1103              **************************/
1104
1105             r10              = _mm_mul_pd(rsq10,rinv10);
1106
1107             /* Compute parameters for interactions between i and j atoms */
1108             qq10             = _mm_mul_pd(iq1,jq0);
1109
1110             /* EWALD ELECTROSTATICS */
1111
1112             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1113             ewrt             = _mm_mul_pd(r10,ewtabscale);
1114             ewitab           = _mm_cvttpd_epi32(ewrt);
1115 #ifdef __XOP__
1116             eweps            = _mm_frcz_pd(ewrt);
1117 #else
1118             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1119 #endif
1120             twoeweps         = _mm_add_pd(eweps,eweps);
1121             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1122             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1123             felec            = _mm_mul_pd(_mm_mul_pd(qq10,rinv10),_mm_sub_pd(rinvsq10,felec));
1124
1125             fscal            = felec;
1126
1127             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1128
1129             /* Update vectorial force */
1130             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1131             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1132             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1133
1134             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1135             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1136             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1137
1138             /**************************
1139              * CALCULATE INTERACTIONS *
1140              **************************/
1141
1142             r20              = _mm_mul_pd(rsq20,rinv20);
1143
1144             /* Compute parameters for interactions between i and j atoms */
1145             qq20             = _mm_mul_pd(iq2,jq0);
1146
1147             /* EWALD ELECTROSTATICS */
1148
1149             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1150             ewrt             = _mm_mul_pd(r20,ewtabscale);
1151             ewitab           = _mm_cvttpd_epi32(ewrt);
1152 #ifdef __XOP__
1153             eweps            = _mm_frcz_pd(ewrt);
1154 #else
1155             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
1156 #endif
1157             twoeweps         = _mm_add_pd(eweps,eweps);
1158             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
1159             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
1160             felec            = _mm_mul_pd(_mm_mul_pd(qq20,rinv20),_mm_sub_pd(rinvsq20,felec));
1161
1162             fscal            = felec;
1163
1164             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1165
1166             /* Update vectorial force */
1167             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1168             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1169             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1170
1171             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1172             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1173             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1174
1175             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1176
1177             /* Inner loop uses 146 flops */
1178         }
1179
1180         /* End of innermost loop */
1181
1182         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1183                                               f+i_coord_offset,fshift+i_shift_offset);
1184
1185         /* Increment number of inner iterations */
1186         inneriter                  += j_index_end - j_index_start;
1187
1188         /* Outer loop uses 18 flops */
1189     }
1190
1191     /* Increment number of outer iterations */
1192     outeriter        += nri;
1193
1194     /* Update outer/inner flops */
1195
1196     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*146);
1197 }