src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwCSTab_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_ElecCoul_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_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          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  91     real             *vftab;
  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     /* Setup water-specific parameters */
 117     inr              = nlist->iinr[0];
 118     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 119     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 120     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136
 137         /* Load limits for loop over neighbors */
 138         j_index_start    = jindex[iidx];
 139         j_index_end      = jindex[iidx+1];
 140
 141         /* Get outer coordinate index */
 142         inr              = iinr[iidx];
 143         i_coord_offset   = DIM*inr;
 144
 145         /* Load i particle coords and add shift vector */
 146         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 147                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 148
 149         fix0             = _mm_setzero_pd();
 150         fiy0             = _mm_setzero_pd();
 151         fiz0             = _mm_setzero_pd();
 152         fix1             = _mm_setzero_pd();
 153         fiy1             = _mm_setzero_pd();
 154         fiz1             = _mm_setzero_pd();
 155         fix2             = _mm_setzero_pd();
 156         fiy2             = _mm_setzero_pd();
 157         fiz2             = _mm_setzero_pd();
 158         fix3             = _mm_setzero_pd();
 159         fiy3             = _mm_setzero_pd();
 160         fiz3             = _mm_setzero_pd();
 161
 162         /* Reset potential sums */
 163         velecsum         = _mm_setzero_pd();
 164         vvdwsum          = _mm_setzero_pd();
 165
 166         /* Start inner kernel loop */
 167         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 168         {
 169
 170             /* Get j neighbor index, and coordinate index */
 171             jnrA             = jjnr[jidx];
 172             jnrB             = jjnr[jidx+1];
 173             j_coord_offsetA  = DIM*jnrA;
 174             j_coord_offsetB  = DIM*jnrB;
 175
 176             /* load j atom coordinates */
 177             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 178                                               &jx0,&jy0,&jz0);
 179
 180             /* Calculate displacement vector */
 181             dx00             = _mm_sub_pd(ix0,jx0);
 182             dy00             = _mm_sub_pd(iy0,jy0);
 183             dz00             = _mm_sub_pd(iz0,jz0);
 184             dx10             = _mm_sub_pd(ix1,jx0);
 185             dy10             = _mm_sub_pd(iy1,jy0);
 186             dz10             = _mm_sub_pd(iz1,jz0);
 187             dx20             = _mm_sub_pd(ix2,jx0);
 188             dy20             = _mm_sub_pd(iy2,jy0);
 189             dz20             = _mm_sub_pd(iz2,jz0);
 190             dx30             = _mm_sub_pd(ix3,jx0);
 191             dy30             = _mm_sub_pd(iy3,jy0);
 192             dz30             = _mm_sub_pd(iz3,jz0);
 193
 194             /* Calculate squared distance and things based on it */
 195             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 196             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 197             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 198             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 199
 200             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 201             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 202             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 203             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 204
 205             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 206             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 207             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 208
 209             /* Load parameters for j particles */
 210             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 211             vdwjidx0A        = 2*vdwtype[jnrA+0];
 212             vdwjidx0B        = 2*vdwtype[jnrB+0];
 213
 214             fjx0             = _mm_setzero_pd();
 215             fjy0             = _mm_setzero_pd();
 216             fjz0             = _mm_setzero_pd();
 217
 218             /**************************
 219              * CALCULATE INTERACTIONS *
 220              **************************/
 221
 222             r00              = _mm_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 226                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 227
 228             /* Calculate table index by multiplying r with table scale and truncate to integer */
 229             rt               = _mm_mul_pd(r00,vftabscale);
 230             vfitab           = _mm_cvttpd_epi32(rt);
 231 #ifdef __XOP__
 232             vfeps            = _mm_frcz_pd(rt);
 233 #else
 234             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 235 #endif
 236             twovfeps         = _mm_add_pd(vfeps,vfeps);
 237             vfitab           = _mm_slli_epi32(vfitab,3);
 238
 239             /* CUBIC SPLINE TABLE DISPERSION */
 240             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 241             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 242             GMX_MM_TRANSPOSE2_PD(Y,F);
 243             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 244             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 245             GMX_MM_TRANSPOSE2_PD(G,H);
 246             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 247             VV               = _mm_macc_pd(vfeps,Fp,Y);
 248             vvdw6            = _mm_mul_pd(c6_00,VV);
 249             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 250             fvdw6            = _mm_mul_pd(c6_00,FF);
 251
 252             /* CUBIC SPLINE TABLE REPULSION */
 253             vfitab           = _mm_add_epi32(vfitab,ifour);
 254             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 255             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 256             GMX_MM_TRANSPOSE2_PD(Y,F);
 257             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 258             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 259             GMX_MM_TRANSPOSE2_PD(G,H);
 260             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 261             VV               = _mm_macc_pd(vfeps,Fp,Y);
 262             vvdw12           = _mm_mul_pd(c12_00,VV);
 263             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 264             fvdw12           = _mm_mul_pd(c12_00,FF);
 265             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 266             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 267
 268             /* Update potential sum for this i atom from the interaction with this j atom. */
 269             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 270
 271             fscal            = fvdw;
 272
 273             /* Update vectorial force */
 274             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 275             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 276             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 277
 278             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 279             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 280             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 281
 282             /**************************
 283              * CALCULATE INTERACTIONS *
 284              **************************/
 285
 286             /* Compute parameters for interactions between i and j atoms */
 287             qq10             = _mm_mul_pd(iq1,jq0);
 288
 289             /* COULOMB ELECTROSTATICS */
 290             velec            = _mm_mul_pd(qq10,rinv10);
 291             felec            = _mm_mul_pd(velec,rinvsq10);
 292
 293             /* Update potential sum for this i atom from the interaction with this j atom. */
 294             velecsum         = _mm_add_pd(velecsum,velec);
 295
 296             fscal            = felec;
 297
 298             /* Update vectorial force */
 299             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 300             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 301             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 302
 303             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 304             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 305             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 306
 307             /**************************
 308              * CALCULATE INTERACTIONS *
 309              **************************/
 310
 311             /* Compute parameters for interactions between i and j atoms */
 312             qq20             = _mm_mul_pd(iq2,jq0);
 313
 314             /* COULOMB ELECTROSTATICS */
 315             velec            = _mm_mul_pd(qq20,rinv20);
 316             felec            = _mm_mul_pd(velec,rinvsq20);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velecsum         = _mm_add_pd(velecsum,velec);
 320
 321             fscal            = felec;
 322
 323             /* Update vectorial force */
 324             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 325             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 326             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 327
 328             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 329             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 330             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 331
 332             /**************************
 333              * CALCULATE INTERACTIONS *
 334              **************************/
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq30             = _mm_mul_pd(iq3,jq0);
 338
 339             /* COULOMB ELECTROSTATICS */
 340             velec            = _mm_mul_pd(qq30,rinv30);
 341             felec            = _mm_mul_pd(velec,rinvsq30);
 342
 343             /* Update potential sum for this i atom from the interaction with this j atom. */
 344             velecsum         = _mm_add_pd(velecsum,velec);
 345
 346             fscal            = felec;
 347
 348             /* Update vectorial force */
 349             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 350             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 351             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 352
 353             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 354             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 355             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 356
 357             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 358
 359             /* Inner loop uses 155 flops */
 360         }
 361
 362         if(jidx<j_index_end)
 363         {
 364
 365             jnrA             = jjnr[jidx];
 366             j_coord_offsetA  = DIM*jnrA;
 367
 368             /* load j atom coordinates */
 369             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 370                                               &jx0,&jy0,&jz0);
 371
 372             /* Calculate displacement vector */
 373             dx00             = _mm_sub_pd(ix0,jx0);
 374             dy00             = _mm_sub_pd(iy0,jy0);
 375             dz00             = _mm_sub_pd(iz0,jz0);
 376             dx10             = _mm_sub_pd(ix1,jx0);
 377             dy10             = _mm_sub_pd(iy1,jy0);
 378             dz10             = _mm_sub_pd(iz1,jz0);
 379             dx20             = _mm_sub_pd(ix2,jx0);
 380             dy20             = _mm_sub_pd(iy2,jy0);
 381             dz20             = _mm_sub_pd(iz2,jz0);
 382             dx30             = _mm_sub_pd(ix3,jx0);
 383             dy30             = _mm_sub_pd(iy3,jy0);
 384             dz30             = _mm_sub_pd(iz3,jz0);
 385
 386             /* Calculate squared distance and things based on it */
 387             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 388             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 389             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 390             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 391
 392             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 393             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 394             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 395             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 396
 397             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 398             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 399             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 400
 401             /* Load parameters for j particles */
 402             jq0              = _mm_load_sd(charge+jnrA+0);
 403             vdwjidx0A        = 2*vdwtype[jnrA+0];
 404
 405             fjx0             = _mm_setzero_pd();
 406             fjy0             = _mm_setzero_pd();
 407             fjz0             = _mm_setzero_pd();
 408
 409             /**************************
 410              * CALCULATE INTERACTIONS *
 411              **************************/
 412
 413             r00              = _mm_mul_pd(rsq00,rinv00);
 414
 415             /* Compute parameters for interactions between i and j atoms */
 416             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 417
 418             /* Calculate table index by multiplying r with table scale and truncate to integer */
 419             rt               = _mm_mul_pd(r00,vftabscale);
 420             vfitab           = _mm_cvttpd_epi32(rt);
 421 #ifdef __XOP__
 422             vfeps            = _mm_frcz_pd(rt);
 423 #else
 424             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 425 #endif
 426             twovfeps         = _mm_add_pd(vfeps,vfeps);
 427             vfitab           = _mm_slli_epi32(vfitab,3);
 428
 429             /* CUBIC SPLINE TABLE DISPERSION */
 430             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 431             F                = _mm_setzero_pd();
 432             GMX_MM_TRANSPOSE2_PD(Y,F);
 433             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 434             H                = _mm_setzero_pd();
 435             GMX_MM_TRANSPOSE2_PD(G,H);
 436             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 437             VV               = _mm_macc_pd(vfeps,Fp,Y);
 438             vvdw6            = _mm_mul_pd(c6_00,VV);
 439             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 440             fvdw6            = _mm_mul_pd(c6_00,FF);
 441
 442             /* CUBIC SPLINE TABLE REPULSION */
 443             vfitab           = _mm_add_epi32(vfitab,ifour);
 444             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 445             F                = _mm_setzero_pd();
 446             GMX_MM_TRANSPOSE2_PD(Y,F);
 447             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 448             H                = _mm_setzero_pd();
 449             GMX_MM_TRANSPOSE2_PD(G,H);
 450             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 451             VV               = _mm_macc_pd(vfeps,Fp,Y);
 452             vvdw12           = _mm_mul_pd(c12_00,VV);
 453             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 454             fvdw12           = _mm_mul_pd(c12_00,FF);
 455             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 456             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 457
 458             /* Update potential sum for this i atom from the interaction with this j atom. */
 459             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 460             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 461
 462             fscal            = fvdw;
 463
 464             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 465
 466             /* Update vectorial force */
 467             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 468             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 469             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 470
 471             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 472             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 473             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 474
 475             /**************************
 476              * CALCULATE INTERACTIONS *
 477              **************************/
 478
 479             /* Compute parameters for interactions between i and j atoms */
 480             qq10             = _mm_mul_pd(iq1,jq0);
 481
 482             /* COULOMB ELECTROSTATICS */
 483             velec            = _mm_mul_pd(qq10,rinv10);
 484             felec            = _mm_mul_pd(velec,rinvsq10);
 485
 486             /* Update potential sum for this i atom from the interaction with this j atom. */
 487             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 488             velecsum         = _mm_add_pd(velecsum,velec);
 489
 490             fscal            = felec;
 491
 492             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 493
 494             /* Update vectorial force */
 495             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 496             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 497             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 498
 499             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 500             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 501             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             /* Compute parameters for interactions between i and j atoms */
 508             qq20             = _mm_mul_pd(iq2,jq0);
 509
 510             /* COULOMB ELECTROSTATICS */
 511             velec            = _mm_mul_pd(qq20,rinv20);
 512             felec            = _mm_mul_pd(velec,rinvsq20);
 513
 514             /* Update potential sum for this i atom from the interaction with this j atom. */
 515             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 516             velecsum         = _mm_add_pd(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 521
 522             /* Update vectorial force */
 523             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 524             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 525             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 526
 527             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 528             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 529             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 530
 531             /**************************
 532              * CALCULATE INTERACTIONS *
 533              **************************/
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq30             = _mm_mul_pd(iq3,jq0);
 537
 538             /* COULOMB ELECTROSTATICS */
 539             velec            = _mm_mul_pd(qq30,rinv30);
 540             felec            = _mm_mul_pd(velec,rinvsq30);
 541
 542             /* Update potential sum for this i atom from the interaction with this j atom. */
 543             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 544             velecsum         = _mm_add_pd(velecsum,velec);
 545
 546             fscal            = felec;
 547
 548             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 549
 550             /* Update vectorial force */
 551             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 552             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 553             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 554
 555             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 556             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 557             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 558
 559             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 560
 561             /* Inner loop uses 155 flops */
 562         }
 563
 564         /* End of innermost loop */
 565
 566         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 567                                               f+i_coord_offset,fshift+i_shift_offset);
 568
 569         ggid                        = gid[iidx];
 570         /* Update potential energies */
 571         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 572         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 573
 574         /* Increment number of inner iterations */
 575         inneriter                  += j_index_end - j_index_start;
 576
 577         /* Outer loop uses 26 flops */
 578     }
 579
 580     /* Increment number of outer iterations */
 581     outeriter        += nri;
 582
 583     /* Update outer/inner flops */
 584
 585     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*155);
 586 }
 587 /*
 588  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 589  * Electrostatics interaction: Coulomb
 590  * VdW interaction:            CubicSplineTable
 591  * Geometry:                   Water4-Particle
 592  * Calculate force/pot:        Force
 593  */
 594 void
 595 nb_kernel_ElecCoul_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 596                     (t_nblist * gmx_restrict                nlist,
 597                      rvec * gmx_restrict                    xx,
 598                      rvec * gmx_restrict                    ff,
 599                      t_forcerec * gmx_restrict              fr,
 600                      t_mdatoms * gmx_restrict               mdatoms,
 601                      nb_kernel_data_t * gmx_restrict        kernel_data,
 602                      t_nrnb * gmx_restrict                  nrnb)
 603 {
 604     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 605      * just 0 for non-waters.
 606      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 607      * jnr indices corresponding to data put in the four positions in the SIMD register.
 608      */
 609     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 610     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 611     int              jnrA,jnrB;
 612     int              j_coord_offsetA,j_coord_offsetB;
 613     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 614     real             rcutoff_scalar;
 615     real             *shiftvec,*fshift,*x,*f;
 616     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 617     int              vdwioffset0;
 618     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 619     int              vdwioffset1;
 620     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 621     int              vdwioffset2;
 622     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 623     int              vdwioffset3;
 624     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 625     int              vdwjidx0A,vdwjidx0B;
 626     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 627     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 628     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 629     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 630     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 631     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 632     real             *charge;
 633     int              nvdwtype;
 634     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 635     int              *vdwtype;
 636     real             *vdwparam;
 637     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 638     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 639     __m128i          vfitab;
 640     __m128i          ifour       = _mm_set1_epi32(4);
 641     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 642     real             *vftab;
 643     __m128d          dummy_mask,cutoff_mask;
 644     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 645     __m128d          one     = _mm_set1_pd(1.0);
 646     __m128d          two     = _mm_set1_pd(2.0);
 647     x                = xx[0];
 648     f                = ff[0];
 649
 650     nri              = nlist->nri;
 651     iinr             = nlist->iinr;
 652     jindex           = nlist->jindex;
 653     jjnr             = nlist->jjnr;
 654     shiftidx         = nlist->shift;
 655     gid              = nlist->gid;
 656     shiftvec         = fr->shift_vec[0];
 657     fshift           = fr->fshift[0];
 658     facel            = _mm_set1_pd(fr->epsfac);
 659     charge           = mdatoms->chargeA;
 660     nvdwtype         = fr->ntype;
 661     vdwparam         = fr->nbfp;
 662     vdwtype          = mdatoms->typeA;
 663
 664     vftab            = kernel_data->table_vdw->data;
 665     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 666
 667     /* Setup water-specific parameters */
 668     inr              = nlist->iinr[0];
 669     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 670     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 671     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 672     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 673
 674     /* Avoid stupid compiler warnings */
 675     jnrA = jnrB = 0;
 676     j_coord_offsetA = 0;
 677     j_coord_offsetB = 0;
 678
 679     outeriter        = 0;
 680     inneriter        = 0;
 681
 682     /* Start outer loop over neighborlists */
 683     for(iidx=0; iidx<nri; iidx++)
 684     {
 685         /* Load shift vector for this list */
 686         i_shift_offset   = DIM*shiftidx[iidx];
 687
 688         /* Load limits for loop over neighbors */
 689         j_index_start    = jindex[iidx];
 690         j_index_end      = jindex[iidx+1];
 691
 692         /* Get outer coordinate index */
 693         inr              = iinr[iidx];
 694         i_coord_offset   = DIM*inr;
 695
 696         /* Load i particle coords and add shift vector */
 697         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 698                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 699
 700         fix0             = _mm_setzero_pd();
 701         fiy0             = _mm_setzero_pd();
 702         fiz0             = _mm_setzero_pd();
 703         fix1             = _mm_setzero_pd();
 704         fiy1             = _mm_setzero_pd();
 705         fiz1             = _mm_setzero_pd();
 706         fix2             = _mm_setzero_pd();
 707         fiy2             = _mm_setzero_pd();
 708         fiz2             = _mm_setzero_pd();
 709         fix3             = _mm_setzero_pd();
 710         fiy3             = _mm_setzero_pd();
 711         fiz3             = _mm_setzero_pd();
 712
 713         /* Start inner kernel loop */
 714         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 715         {
 716
 717             /* Get j neighbor index, and coordinate index */
 718             jnrA             = jjnr[jidx];
 719             jnrB             = jjnr[jidx+1];
 720             j_coord_offsetA  = DIM*jnrA;
 721             j_coord_offsetB  = DIM*jnrB;
 722
 723             /* load j atom coordinates */
 724             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 725                                               &jx0,&jy0,&jz0);
 726
 727             /* Calculate displacement vector */
 728             dx00             = _mm_sub_pd(ix0,jx0);
 729             dy00             = _mm_sub_pd(iy0,jy0);
 730             dz00             = _mm_sub_pd(iz0,jz0);
 731             dx10             = _mm_sub_pd(ix1,jx0);
 732             dy10             = _mm_sub_pd(iy1,jy0);
 733             dz10             = _mm_sub_pd(iz1,jz0);
 734             dx20             = _mm_sub_pd(ix2,jx0);
 735             dy20             = _mm_sub_pd(iy2,jy0);
 736             dz20             = _mm_sub_pd(iz2,jz0);
 737             dx30             = _mm_sub_pd(ix3,jx0);
 738             dy30             = _mm_sub_pd(iy3,jy0);
 739             dz30             = _mm_sub_pd(iz3,jz0);
 740
 741             /* Calculate squared distance and things based on it */
 742             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 743             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 744             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 745             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 746
 747             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 748             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 749             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 750             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 751
 752             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 753             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 754             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 755
 756             /* Load parameters for j particles */
 757             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 758             vdwjidx0A        = 2*vdwtype[jnrA+0];
 759             vdwjidx0B        = 2*vdwtype[jnrB+0];
 760
 761             fjx0             = _mm_setzero_pd();
 762             fjy0             = _mm_setzero_pd();
 763             fjz0             = _mm_setzero_pd();
 764
 765             /**************************
 766              * CALCULATE INTERACTIONS *
 767              **************************/
 768
 769             r00              = _mm_mul_pd(rsq00,rinv00);
 770
 771             /* Compute parameters for interactions between i and j atoms */
 772             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 773                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 774
 775             /* Calculate table index by multiplying r with table scale and truncate to integer */
 776             rt               = _mm_mul_pd(r00,vftabscale);
 777             vfitab           = _mm_cvttpd_epi32(rt);
 778 #ifdef __XOP__
 779             vfeps            = _mm_frcz_pd(rt);
 780 #else
 781             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 782 #endif
 783             twovfeps         = _mm_add_pd(vfeps,vfeps);
 784             vfitab           = _mm_slli_epi32(vfitab,3);
 785
 786             /* CUBIC SPLINE TABLE DISPERSION */
 787             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 788             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 789             GMX_MM_TRANSPOSE2_PD(Y,F);
 790             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 791             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 792             GMX_MM_TRANSPOSE2_PD(G,H);
 793             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 794             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 795             fvdw6            = _mm_mul_pd(c6_00,FF);
 796
 797             /* CUBIC SPLINE TABLE REPULSION */
 798             vfitab           = _mm_add_epi32(vfitab,ifour);
 799             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 800             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 801             GMX_MM_TRANSPOSE2_PD(Y,F);
 802             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 803             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 804             GMX_MM_TRANSPOSE2_PD(G,H);
 805             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 806             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 807             fvdw12           = _mm_mul_pd(c12_00,FF);
 808             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 809
 810             fscal            = fvdw;
 811
 812             /* Update vectorial force */
 813             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 814             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 815             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 816
 817             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 818             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 819             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq10             = _mm_mul_pd(iq1,jq0);
 827
 828             /* COULOMB ELECTROSTATICS */
 829             velec            = _mm_mul_pd(qq10,rinv10);
 830             felec            = _mm_mul_pd(velec,rinvsq10);
 831
 832             fscal            = felec;
 833
 834             /* Update vectorial force */
 835             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 836             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 837             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 838
 839             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 840             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 841             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 842
 843             /**************************
 844              * CALCULATE INTERACTIONS *
 845              **************************/
 846
 847             /* Compute parameters for interactions between i and j atoms */
 848             qq20             = _mm_mul_pd(iq2,jq0);
 849
 850             /* COULOMB ELECTROSTATICS */
 851             velec            = _mm_mul_pd(qq20,rinv20);
 852             felec            = _mm_mul_pd(velec,rinvsq20);
 853
 854             fscal            = felec;
 855
 856             /* Update vectorial force */
 857             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 858             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 859             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 860
 861             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 862             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 863             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq30             = _mm_mul_pd(iq3,jq0);
 871
 872             /* COULOMB ELECTROSTATICS */
 873             velec            = _mm_mul_pd(qq30,rinv30);
 874             felec            = _mm_mul_pd(velec,rinvsq30);
 875
 876             fscal            = felec;
 877
 878             /* Update vectorial force */
 879             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 880             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 881             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 882
 883             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 884             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 885             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 886
 887             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 888
 889             /* Inner loop uses 144 flops */
 890         }
 891
 892         if(jidx<j_index_end)
 893         {
 894
 895             jnrA             = jjnr[jidx];
 896             j_coord_offsetA  = DIM*jnrA;
 897
 898             /* load j atom coordinates */
 899             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 900                                               &jx0,&jy0,&jz0);
 901
 902             /* Calculate displacement vector */
 903             dx00             = _mm_sub_pd(ix0,jx0);
 904             dy00             = _mm_sub_pd(iy0,jy0);
 905             dz00             = _mm_sub_pd(iz0,jz0);
 906             dx10             = _mm_sub_pd(ix1,jx0);
 907             dy10             = _mm_sub_pd(iy1,jy0);
 908             dz10             = _mm_sub_pd(iz1,jz0);
 909             dx20             = _mm_sub_pd(ix2,jx0);
 910             dy20             = _mm_sub_pd(iy2,jy0);
 911             dz20             = _mm_sub_pd(iz2,jz0);
 912             dx30             = _mm_sub_pd(ix3,jx0);
 913             dy30             = _mm_sub_pd(iy3,jy0);
 914             dz30             = _mm_sub_pd(iz3,jz0);
 915
 916             /* Calculate squared distance and things based on it */
 917             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 918             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 919             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 920             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 921
 922             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 923             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 924             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 925             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 926
 927             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 928             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 929             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 930
 931             /* Load parameters for j particles */
 932             jq0              = _mm_load_sd(charge+jnrA+0);
 933             vdwjidx0A        = 2*vdwtype[jnrA+0];
 934
 935             fjx0             = _mm_setzero_pd();
 936             fjy0             = _mm_setzero_pd();
 937             fjz0             = _mm_setzero_pd();
 938
 939             /**************************
 940              * CALCULATE INTERACTIONS *
 941              **************************/
 942
 943             r00              = _mm_mul_pd(rsq00,rinv00);
 944
 945             /* Compute parameters for interactions between i and j atoms */
 946             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 947
 948             /* Calculate table index by multiplying r with table scale and truncate to integer */
 949             rt               = _mm_mul_pd(r00,vftabscale);
 950             vfitab           = _mm_cvttpd_epi32(rt);
 951 #ifdef __XOP__
 952             vfeps            = _mm_frcz_pd(rt);
 953 #else
 954             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 955 #endif
 956             twovfeps         = _mm_add_pd(vfeps,vfeps);
 957             vfitab           = _mm_slli_epi32(vfitab,3);
 958
 959             /* CUBIC SPLINE TABLE DISPERSION */
 960             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 961             F                = _mm_setzero_pd();
 962             GMX_MM_TRANSPOSE2_PD(Y,F);
 963             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 964             H                = _mm_setzero_pd();
 965             GMX_MM_TRANSPOSE2_PD(G,H);
 966             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 967             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 968             fvdw6            = _mm_mul_pd(c6_00,FF);
 969
 970             /* CUBIC SPLINE TABLE REPULSION */
 971             vfitab           = _mm_add_epi32(vfitab,ifour);
 972             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 973             F                = _mm_setzero_pd();
 974             GMX_MM_TRANSPOSE2_PD(Y,F);
 975             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 976             H                = _mm_setzero_pd();
 977             GMX_MM_TRANSPOSE2_PD(G,H);
 978             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 979             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 980             fvdw12           = _mm_mul_pd(c12_00,FF);
 981             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 982
 983             fscal            = fvdw;
 984
 985             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 986
 987             /* Update vectorial force */
 988             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 989             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 990             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 991
 992             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 993             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 994             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             /* Compute parameters for interactions between i and j atoms */
1001             qq10             = _mm_mul_pd(iq1,jq0);
1002
1003             /* COULOMB ELECTROSTATICS */
1004             velec            = _mm_mul_pd(qq10,rinv10);
1005             felec            = _mm_mul_pd(velec,rinvsq10);
1006
1007             fscal            = felec;
1008
1009             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1010
1011             /* Update vectorial force */
1012             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1013             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1014             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1015
1016             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1017             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1018             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1019
1020             /**************************
1021              * CALCULATE INTERACTIONS *
1022              **************************/
1023
1024             /* Compute parameters for interactions between i and j atoms */
1025             qq20             = _mm_mul_pd(iq2,jq0);
1026
1027             /* COULOMB ELECTROSTATICS */
1028             velec            = _mm_mul_pd(qq20,rinv20);
1029             felec            = _mm_mul_pd(velec,rinvsq20);
1030
1031             fscal            = felec;
1032
1033             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1034
1035             /* Update vectorial force */
1036             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1037             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1038             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1039
1040             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1041             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1042             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1043
1044             /**************************
1045              * CALCULATE INTERACTIONS *
1046              **************************/
1047
1048             /* Compute parameters for interactions between i and j atoms */
1049             qq30             = _mm_mul_pd(iq3,jq0);
1050
1051             /* COULOMB ELECTROSTATICS */
1052             velec            = _mm_mul_pd(qq30,rinv30);
1053             felec            = _mm_mul_pd(velec,rinvsq30);
1054
1055             fscal            = felec;
1056
1057             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1058
1059             /* Update vectorial force */
1060             fix3             = _mm_macc_pd(dx30,fscal,fix3);
1061             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
1062             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
1063
1064             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
1065             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
1066             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
1067
1068             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1069
1070             /* Inner loop uses 144 flops */
1071         }
1072
1073         /* End of innermost loop */
1074
1075         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1076                                               f+i_coord_offset,fshift+i_shift_offset);
1077
1078         /* Increment number of inner iterations */
1079         inneriter                  += j_index_end - j_index_start;
1080
1081         /* Outer loop uses 24 flops */
1082     }
1083
1084     /* Increment number of outer iterations */
1085     outeriter        += nri;
1086
1087     /* Update outer/inner flops */
1088
1089     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*144);
1090 }