src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_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_ElecRFCut_VdwCSTab_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_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     __m128d          dummy_mask,cutoff_mask;
  90     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  91     __m128d          one     = _mm_set1_pd(1.0);
  92     __m128d          two     = _mm_set1_pd(2.0);
  93     x                = xx[0];
  94     f                = ff[0];
  95
  96     nri              = nlist->nri;
  97     iinr             = nlist->iinr;
  98     jindex           = nlist->jindex;
  99     jjnr             = nlist->jjnr;
 100     shiftidx         = nlist->shift;
 101     gid              = nlist->gid;
 102     shiftvec         = fr->shift_vec[0];
 103     fshift           = fr->fshift[0];
 104     facel            = _mm_set1_pd(fr->epsfac);
 105     charge           = mdatoms->chargeA;
 106     krf              = _mm_set1_pd(fr->ic->k_rf);
 107     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 108     crf              = _mm_set1_pd(fr->ic->c_rf);
 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     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 119     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 120     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 121     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 122
 123     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 124     rcutoff_scalar   = fr->rcoulomb;
 125     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 126     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 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             if (gmx_mm_any_lt(rsq00,rcutoff2))
 220             {
 221
 222             r00              = _mm_mul_pd(rsq00,rinv00);
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_pd(iq0,jq0);
 226             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 227                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 228
 229             /* Calculate table index by multiplying r with table scale and truncate to integer */
 230             rt               = _mm_mul_pd(r00,vftabscale);
 231             vfitab           = _mm_cvttpd_epi32(rt);
 232 #ifdef __XOP__
 233             vfeps            = _mm_frcz_pd(rt);
 234 #else
 235             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 236 #endif
 237             twovfeps         = _mm_add_pd(vfeps,vfeps);
 238             vfitab           = _mm_slli_epi32(vfitab,3);
 239
 240             /* REACTION-FIELD ELECTROSTATICS */
 241             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 242             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 243
 244             /* CUBIC SPLINE TABLE DISPERSION */
 245             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 246             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 247             GMX_MM_TRANSPOSE2_PD(Y,F);
 248             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 249             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 250             GMX_MM_TRANSPOSE2_PD(G,H);
 251             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 252             VV               = _mm_macc_pd(vfeps,Fp,Y);
 253             vvdw6            = _mm_mul_pd(c6_00,VV);
 254             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 255             fvdw6            = _mm_mul_pd(c6_00,FF);
 256
 257             /* CUBIC SPLINE TABLE REPULSION */
 258             vfitab           = _mm_add_epi32(vfitab,ifour);
 259             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 260             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 261             GMX_MM_TRANSPOSE2_PD(Y,F);
 262             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 263             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 264             GMX_MM_TRANSPOSE2_PD(G,H);
 265             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 266             VV               = _mm_macc_pd(vfeps,Fp,Y);
 267             vvdw12           = _mm_mul_pd(c12_00,VV);
 268             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 269             fvdw12           = _mm_mul_pd(c12_00,FF);
 270             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 271             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 272
 273             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 274
 275             /* Update potential sum for this i atom from the interaction with this j atom. */
 276             velec            = _mm_and_pd(velec,cutoff_mask);
 277             velecsum         = _mm_add_pd(velecsum,velec);
 278             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 279             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 280
 281             fscal            = _mm_add_pd(felec,fvdw);
 282
 283             fscal            = _mm_and_pd(fscal,cutoff_mask);
 284
 285             /* Update vectorial force */
 286             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 287             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 288             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 289
 290             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 291             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 292             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 293
 294             }
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             if (gmx_mm_any_lt(rsq10,rcutoff2))
 301             {
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_pd(iq1,jq0);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 308             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 309
 310             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 311
 312             /* Update potential sum for this i atom from the interaction with this j atom. */
 313             velec            = _mm_and_pd(velec,cutoff_mask);
 314             velecsum         = _mm_add_pd(velecsum,velec);
 315
 316             fscal            = felec;
 317
 318             fscal            = _mm_and_pd(fscal,cutoff_mask);
 319
 320             /* Update vectorial force */
 321             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 322             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 323             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 324
 325             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 326             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 327             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 328
 329             }
 330
 331             /**************************
 332              * CALCULATE INTERACTIONS *
 333              **************************/
 334
 335             if (gmx_mm_any_lt(rsq20,rcutoff2))
 336             {
 337
 338             /* Compute parameters for interactions between i and j atoms */
 339             qq20             = _mm_mul_pd(iq2,jq0);
 340
 341             /* REACTION-FIELD ELECTROSTATICS */
 342             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 343             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 344
 345             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_and_pd(velec,cutoff_mask);
 349             velecsum         = _mm_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm_and_pd(fscal,cutoff_mask);
 354
 355             /* Update vectorial force */
 356             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 357             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 358             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 359
 360             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 361             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 362             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 363
 364             }
 365
 366             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 367
 368             /* Inner loop uses 156 flops */
 369         }
 370
 371         if(jidx<j_index_end)
 372         {
 373
 374             jnrA             = jjnr[jidx];
 375             j_coord_offsetA  = DIM*jnrA;
 376
 377             /* load j atom coordinates */
 378             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 379                                               &jx0,&jy0,&jz0);
 380
 381             /* Calculate displacement vector */
 382             dx00             = _mm_sub_pd(ix0,jx0);
 383             dy00             = _mm_sub_pd(iy0,jy0);
 384             dz00             = _mm_sub_pd(iz0,jz0);
 385             dx10             = _mm_sub_pd(ix1,jx0);
 386             dy10             = _mm_sub_pd(iy1,jy0);
 387             dz10             = _mm_sub_pd(iz1,jz0);
 388             dx20             = _mm_sub_pd(ix2,jx0);
 389             dy20             = _mm_sub_pd(iy2,jy0);
 390             dz20             = _mm_sub_pd(iz2,jz0);
 391
 392             /* Calculate squared distance and things based on it */
 393             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 394             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 395             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 396
 397             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 398             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 399             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 400
 401             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 402             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 403             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 404
 405             /* Load parameters for j particles */
 406             jq0              = _mm_load_sd(charge+jnrA+0);
 407             vdwjidx0A        = 2*vdwtype[jnrA+0];
 408
 409             fjx0             = _mm_setzero_pd();
 410             fjy0             = _mm_setzero_pd();
 411             fjz0             = _mm_setzero_pd();
 412
 413             /**************************
 414              * CALCULATE INTERACTIONS *
 415              **************************/
 416
 417             if (gmx_mm_any_lt(rsq00,rcutoff2))
 418             {
 419
 420             r00              = _mm_mul_pd(rsq00,rinv00);
 421
 422             /* Compute parameters for interactions between i and j atoms */
 423             qq00             = _mm_mul_pd(iq0,jq0);
 424             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 425
 426             /* Calculate table index by multiplying r with table scale and truncate to integer */
 427             rt               = _mm_mul_pd(r00,vftabscale);
 428             vfitab           = _mm_cvttpd_epi32(rt);
 429 #ifdef __XOP__
 430             vfeps            = _mm_frcz_pd(rt);
 431 #else
 432             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 433 #endif
 434             twovfeps         = _mm_add_pd(vfeps,vfeps);
 435             vfitab           = _mm_slli_epi32(vfitab,3);
 436
 437             /* REACTION-FIELD ELECTROSTATICS */
 438             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 439             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 440
 441             /* CUBIC SPLINE TABLE DISPERSION */
 442             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 443             F                = _mm_setzero_pd();
 444             GMX_MM_TRANSPOSE2_PD(Y,F);
 445             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 446             H                = _mm_setzero_pd();
 447             GMX_MM_TRANSPOSE2_PD(G,H);
 448             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 449             VV               = _mm_macc_pd(vfeps,Fp,Y);
 450             vvdw6            = _mm_mul_pd(c6_00,VV);
 451             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 452             fvdw6            = _mm_mul_pd(c6_00,FF);
 453
 454             /* CUBIC SPLINE TABLE REPULSION */
 455             vfitab           = _mm_add_epi32(vfitab,ifour);
 456             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 457             F                = _mm_setzero_pd();
 458             GMX_MM_TRANSPOSE2_PD(Y,F);
 459             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 460             H                = _mm_setzero_pd();
 461             GMX_MM_TRANSPOSE2_PD(G,H);
 462             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 463             VV               = _mm_macc_pd(vfeps,Fp,Y);
 464             vvdw12           = _mm_mul_pd(c12_00,VV);
 465             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 466             fvdw12           = _mm_mul_pd(c12_00,FF);
 467             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 468             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 469
 470             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 471
 472             /* Update potential sum for this i atom from the interaction with this j atom. */
 473             velec            = _mm_and_pd(velec,cutoff_mask);
 474             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 475             velecsum         = _mm_add_pd(velecsum,velec);
 476             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 477             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 478             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 479
 480             fscal            = _mm_add_pd(felec,fvdw);
 481
 482             fscal            = _mm_and_pd(fscal,cutoff_mask);
 483
 484             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 485
 486             /* Update vectorial force */
 487             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 488             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 489             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 490
 491             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 492             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 493             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 494
 495             }
 496
 497             /**************************
 498              * CALCULATE INTERACTIONS *
 499              **************************/
 500
 501             if (gmx_mm_any_lt(rsq10,rcutoff2))
 502             {
 503
 504             /* Compute parameters for interactions between i and j atoms */
 505             qq10             = _mm_mul_pd(iq1,jq0);
 506
 507             /* REACTION-FIELD ELECTROSTATICS */
 508             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 509             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 510
 511             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 512
 513             /* Update potential sum for this i atom from the interaction with this j atom. */
 514             velec            = _mm_and_pd(velec,cutoff_mask);
 515             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 516             velecsum         = _mm_add_pd(velecsum,velec);
 517
 518             fscal            = felec;
 519
 520             fscal            = _mm_and_pd(fscal,cutoff_mask);
 521
 522             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 523
 524             /* Update vectorial force */
 525             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 526             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 527             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 528
 529             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 530             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 531             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 532
 533             }
 534
 535             /**************************
 536              * CALCULATE INTERACTIONS *
 537              **************************/
 538
 539             if (gmx_mm_any_lt(rsq20,rcutoff2))
 540             {
 541
 542             /* Compute parameters for interactions between i and j atoms */
 543             qq20             = _mm_mul_pd(iq2,jq0);
 544
 545             /* REACTION-FIELD ELECTROSTATICS */
 546             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 547             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 548
 549             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 550
 551             /* Update potential sum for this i atom from the interaction with this j atom. */
 552             velec            = _mm_and_pd(velec,cutoff_mask);
 553             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 554             velecsum         = _mm_add_pd(velecsum,velec);
 555
 556             fscal            = felec;
 557
 558             fscal            = _mm_and_pd(fscal,cutoff_mask);
 559
 560             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 561
 562             /* Update vectorial force */
 563             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 564             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 565             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 566
 567             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 568             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 569             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 570
 571             }
 572
 573             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 574
 575             /* Inner loop uses 156 flops */
 576         }
 577
 578         /* End of innermost loop */
 579
 580         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 581                                               f+i_coord_offset,fshift+i_shift_offset);
 582
 583         ggid                        = gid[iidx];
 584         /* Update potential energies */
 585         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 586         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 587
 588         /* Increment number of inner iterations */
 589         inneriter                  += j_index_end - j_index_start;
 590
 591         /* Outer loop uses 20 flops */
 592     }
 593
 594     /* Increment number of outer iterations */
 595     outeriter        += nri;
 596
 597     /* Update outer/inner flops */
 598
 599     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*156);
 600 }
 601 /*
 602  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 603  * Electrostatics interaction: ReactionField
 604  * VdW interaction:            CubicSplineTable
 605  * Geometry:                   Water3-Particle
 606  * Calculate force/pot:        Force
 607  */
 608 void
 609 nb_kernel_ElecRFCut_VdwCSTab_GeomW3P1_F_avx_128_fma_double
 610                     (t_nblist * gmx_restrict                nlist,
 611                      rvec * gmx_restrict                    xx,
 612                      rvec * gmx_restrict                    ff,
 613                      t_forcerec * gmx_restrict              fr,
 614                      t_mdatoms * gmx_restrict               mdatoms,
 615                      nb_kernel_data_t * gmx_restrict        kernel_data,
 616                      t_nrnb * gmx_restrict                  nrnb)
 617 {
 618     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 619      * just 0 for non-waters.
 620      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 621      * jnr indices corresponding to data put in the four positions in the SIMD register.
 622      */
 623     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 624     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 625     int              jnrA,jnrB;
 626     int              j_coord_offsetA,j_coord_offsetB;
 627     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 628     real             rcutoff_scalar;
 629     real             *shiftvec,*fshift,*x,*f;
 630     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 631     int              vdwioffset0;
 632     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 633     int              vdwioffset1;
 634     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 635     int              vdwioffset2;
 636     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 637     int              vdwjidx0A,vdwjidx0B;
 638     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 639     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 640     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 641     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 642     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 643     real             *charge;
 644     int              nvdwtype;
 645     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 646     int              *vdwtype;
 647     real             *vdwparam;
 648     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 649     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 650     __m128i          vfitab;
 651     __m128i          ifour       = _mm_set1_epi32(4);
 652     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 653     real             *vftab;
 654     __m128d          dummy_mask,cutoff_mask;
 655     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 656     __m128d          one     = _mm_set1_pd(1.0);
 657     __m128d          two     = _mm_set1_pd(2.0);
 658     x                = xx[0];
 659     f                = ff[0];
 660
 661     nri              = nlist->nri;
 662     iinr             = nlist->iinr;
 663     jindex           = nlist->jindex;
 664     jjnr             = nlist->jjnr;
 665     shiftidx         = nlist->shift;
 666     gid              = nlist->gid;
 667     shiftvec         = fr->shift_vec[0];
 668     fshift           = fr->fshift[0];
 669     facel            = _mm_set1_pd(fr->epsfac);
 670     charge           = mdatoms->chargeA;
 671     krf              = _mm_set1_pd(fr->ic->k_rf);
 672     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 673     crf              = _mm_set1_pd(fr->ic->c_rf);
 674     nvdwtype         = fr->ntype;
 675     vdwparam         = fr->nbfp;
 676     vdwtype          = mdatoms->typeA;
 677
 678     vftab            = kernel_data->table_vdw->data;
 679     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 680
 681     /* Setup water-specific parameters */
 682     inr              = nlist->iinr[0];
 683     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 684     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 685     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 686     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 687
 688     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 689     rcutoff_scalar   = fr->rcoulomb;
 690     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 691     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 692
 693     /* Avoid stupid compiler warnings */
 694     jnrA = jnrB = 0;
 695     j_coord_offsetA = 0;
 696     j_coord_offsetB = 0;
 697
 698     outeriter        = 0;
 699     inneriter        = 0;
 700
 701     /* Start outer loop over neighborlists */
 702     for(iidx=0; iidx<nri; iidx++)
 703     {
 704         /* Load shift vector for this list */
 705         i_shift_offset   = DIM*shiftidx[iidx];
 706
 707         /* Load limits for loop over neighbors */
 708         j_index_start    = jindex[iidx];
 709         j_index_end      = jindex[iidx+1];
 710
 711         /* Get outer coordinate index */
 712         inr              = iinr[iidx];
 713         i_coord_offset   = DIM*inr;
 714
 715         /* Load i particle coords and add shift vector */
 716         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 717                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 718
 719         fix0             = _mm_setzero_pd();
 720         fiy0             = _mm_setzero_pd();
 721         fiz0             = _mm_setzero_pd();
 722         fix1             = _mm_setzero_pd();
 723         fiy1             = _mm_setzero_pd();
 724         fiz1             = _mm_setzero_pd();
 725         fix2             = _mm_setzero_pd();
 726         fiy2             = _mm_setzero_pd();
 727         fiz2             = _mm_setzero_pd();
 728
 729         /* Start inner kernel loop */
 730         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 731         {
 732
 733             /* Get j neighbor index, and coordinate index */
 734             jnrA             = jjnr[jidx];
 735             jnrB             = jjnr[jidx+1];
 736             j_coord_offsetA  = DIM*jnrA;
 737             j_coord_offsetB  = DIM*jnrB;
 738
 739             /* load j atom coordinates */
 740             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 741                                               &jx0,&jy0,&jz0);
 742
 743             /* Calculate displacement vector */
 744             dx00             = _mm_sub_pd(ix0,jx0);
 745             dy00             = _mm_sub_pd(iy0,jy0);
 746             dz00             = _mm_sub_pd(iz0,jz0);
 747             dx10             = _mm_sub_pd(ix1,jx0);
 748             dy10             = _mm_sub_pd(iy1,jy0);
 749             dz10             = _mm_sub_pd(iz1,jz0);
 750             dx20             = _mm_sub_pd(ix2,jx0);
 751             dy20             = _mm_sub_pd(iy2,jy0);
 752             dz20             = _mm_sub_pd(iz2,jz0);
 753
 754             /* Calculate squared distance and things based on it */
 755             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 756             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 757             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 758
 759             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 760             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 761             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 762
 763             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 764             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 765             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 766
 767             /* Load parameters for j particles */
 768             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 769             vdwjidx0A        = 2*vdwtype[jnrA+0];
 770             vdwjidx0B        = 2*vdwtype[jnrB+0];
 771
 772             fjx0             = _mm_setzero_pd();
 773             fjy0             = _mm_setzero_pd();
 774             fjz0             = _mm_setzero_pd();
 775
 776             /**************************
 777              * CALCULATE INTERACTIONS *
 778              **************************/
 779
 780             if (gmx_mm_any_lt(rsq00,rcutoff2))
 781             {
 782
 783             r00              = _mm_mul_pd(rsq00,rinv00);
 784
 785             /* Compute parameters for interactions between i and j atoms */
 786             qq00             = _mm_mul_pd(iq0,jq0);
 787             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 788                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 789
 790             /* Calculate table index by multiplying r with table scale and truncate to integer */
 791             rt               = _mm_mul_pd(r00,vftabscale);
 792             vfitab           = _mm_cvttpd_epi32(rt);
 793 #ifdef __XOP__
 794             vfeps            = _mm_frcz_pd(rt);
 795 #else
 796             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 797 #endif
 798             twovfeps         = _mm_add_pd(vfeps,vfeps);
 799             vfitab           = _mm_slli_epi32(vfitab,3);
 800
 801             /* REACTION-FIELD ELECTROSTATICS */
 802             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 803
 804             /* CUBIC SPLINE TABLE DISPERSION */
 805             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 806             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 807             GMX_MM_TRANSPOSE2_PD(Y,F);
 808             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 809             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 810             GMX_MM_TRANSPOSE2_PD(G,H);
 811             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 812             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 813             fvdw6            = _mm_mul_pd(c6_00,FF);
 814
 815             /* CUBIC SPLINE TABLE REPULSION */
 816             vfitab           = _mm_add_epi32(vfitab,ifour);
 817             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 818             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 819             GMX_MM_TRANSPOSE2_PD(Y,F);
 820             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 821             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 822             GMX_MM_TRANSPOSE2_PD(G,H);
 823             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 824             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 825             fvdw12           = _mm_mul_pd(c12_00,FF);
 826             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 827
 828             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 829
 830             fscal            = _mm_add_pd(felec,fvdw);
 831
 832             fscal            = _mm_and_pd(fscal,cutoff_mask);
 833
 834             /* Update vectorial force */
 835             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 836             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 837             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 838
 839             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 840             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 841             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 842
 843             }
 844
 845             /**************************
 846              * CALCULATE INTERACTIONS *
 847              **************************/
 848
 849             if (gmx_mm_any_lt(rsq10,rcutoff2))
 850             {
 851
 852             /* Compute parameters for interactions between i and j atoms */
 853             qq10             = _mm_mul_pd(iq1,jq0);
 854
 855             /* REACTION-FIELD ELECTROSTATICS */
 856             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 857
 858             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
 859
 860             fscal            = felec;
 861
 862             fscal            = _mm_and_pd(fscal,cutoff_mask);
 863
 864             /* Update vectorial force */
 865             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 866             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 867             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 868
 869             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 870             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 871             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 872
 873             }
 874
 875             /**************************
 876              * CALCULATE INTERACTIONS *
 877              **************************/
 878
 879             if (gmx_mm_any_lt(rsq20,rcutoff2))
 880             {
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq20             = _mm_mul_pd(iq2,jq0);
 884
 885             /* REACTION-FIELD ELECTROSTATICS */
 886             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 887
 888             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm_and_pd(fscal,cutoff_mask);
 893
 894             /* Update vectorial force */
 895             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 896             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 897             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 898
 899             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 900             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 901             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 902
 903             }
 904
 905             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 906
 907             /* Inner loop uses 129 flops */
 908         }
 909
 910         if(jidx<j_index_end)
 911         {
 912
 913             jnrA             = jjnr[jidx];
 914             j_coord_offsetA  = DIM*jnrA;
 915
 916             /* load j atom coordinates */
 917             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 918                                               &jx0,&jy0,&jz0);
 919
 920             /* Calculate displacement vector */
 921             dx00             = _mm_sub_pd(ix0,jx0);
 922             dy00             = _mm_sub_pd(iy0,jy0);
 923             dz00             = _mm_sub_pd(iz0,jz0);
 924             dx10             = _mm_sub_pd(ix1,jx0);
 925             dy10             = _mm_sub_pd(iy1,jy0);
 926             dz10             = _mm_sub_pd(iz1,jz0);
 927             dx20             = _mm_sub_pd(ix2,jx0);
 928             dy20             = _mm_sub_pd(iy2,jy0);
 929             dz20             = _mm_sub_pd(iz2,jz0);
 930
 931             /* Calculate squared distance and things based on it */
 932             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 933             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 934             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 935
 936             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 937             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 938             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 939
 940             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 941             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 942             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 943
 944             /* Load parameters for j particles */
 945             jq0              = _mm_load_sd(charge+jnrA+0);
 946             vdwjidx0A        = 2*vdwtype[jnrA+0];
 947
 948             fjx0             = _mm_setzero_pd();
 949             fjy0             = _mm_setzero_pd();
 950             fjz0             = _mm_setzero_pd();
 951
 952             /**************************
 953              * CALCULATE INTERACTIONS *
 954              **************************/
 955
 956             if (gmx_mm_any_lt(rsq00,rcutoff2))
 957             {
 958
 959             r00              = _mm_mul_pd(rsq00,rinv00);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             qq00             = _mm_mul_pd(iq0,jq0);
 963             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 964
 965             /* Calculate table index by multiplying r with table scale and truncate to integer */
 966             rt               = _mm_mul_pd(r00,vftabscale);
 967             vfitab           = _mm_cvttpd_epi32(rt);
 968 #ifdef __XOP__
 969             vfeps            = _mm_frcz_pd(rt);
 970 #else
 971             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 972 #endif
 973             twovfeps         = _mm_add_pd(vfeps,vfeps);
 974             vfitab           = _mm_slli_epi32(vfitab,3);
 975
 976             /* REACTION-FIELD ELECTROSTATICS */
 977             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 978
 979             /* CUBIC SPLINE TABLE DISPERSION */
 980             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 981             F                = _mm_setzero_pd();
 982             GMX_MM_TRANSPOSE2_PD(Y,F);
 983             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 984             H                = _mm_setzero_pd();
 985             GMX_MM_TRANSPOSE2_PD(G,H);
 986             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 987             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 988             fvdw6            = _mm_mul_pd(c6_00,FF);
 989
 990             /* CUBIC SPLINE TABLE REPULSION */
 991             vfitab           = _mm_add_epi32(vfitab,ifour);
 992             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 993             F                = _mm_setzero_pd();
 994             GMX_MM_TRANSPOSE2_PD(Y,F);
 995             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 996             H                = _mm_setzero_pd();
 997             GMX_MM_TRANSPOSE2_PD(G,H);
 998             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 999             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
1000             fvdw12           = _mm_mul_pd(c12_00,FF);
1001             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
1002
1003             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
1004
1005             fscal            = _mm_add_pd(felec,fvdw);
1006
1007             fscal            = _mm_and_pd(fscal,cutoff_mask);
1008
1009             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1010
1011             /* Update vectorial force */
1012             fix0             = _mm_macc_pd(dx00,fscal,fix0);
1013             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
1014             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
1015
1016             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
1017             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
1018             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
1019
1020             }
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             if (gmx_mm_any_lt(rsq10,rcutoff2))
1027             {
1028
1029             /* Compute parameters for interactions between i and j atoms */
1030             qq10             = _mm_mul_pd(iq1,jq0);
1031
1032             /* REACTION-FIELD ELECTROSTATICS */
1033             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1034
1035             cutoff_mask      = _mm_cmplt_pd(rsq10,rcutoff2);
1036
1037             fscal            = felec;
1038
1039             fscal            = _mm_and_pd(fscal,cutoff_mask);
1040
1041             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1042
1043             /* Update vectorial force */
1044             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1045             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1046             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1047
1048             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1049             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1050             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1051
1052             }
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             if (gmx_mm_any_lt(rsq20,rcutoff2))
1059             {
1060
1061             /* Compute parameters for interactions between i and j atoms */
1062             qq20             = _mm_mul_pd(iq2,jq0);
1063
1064             /* REACTION-FIELD ELECTROSTATICS */
1065             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1066
1067             cutoff_mask      = _mm_cmplt_pd(rsq20,rcutoff2);
1068
1069             fscal            = felec;
1070
1071             fscal            = _mm_and_pd(fscal,cutoff_mask);
1072
1073             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1074
1075             /* Update vectorial force */
1076             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1077             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1078             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1079
1080             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1081             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1082             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1083
1084             }
1085
1086             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
1087
1088             /* Inner loop uses 129 flops */
1089         }
1090
1091         /* End of innermost loop */
1092
1093         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1094                                               f+i_coord_offset,fshift+i_shift_offset);
1095
1096         /* Increment number of inner iterations */
1097         inneriter                  += j_index_end - j_index_start;
1098
1099         /* Outer loop uses 18 flops */
1100     }
1101
1102     /* Increment number of outer iterations */
1103     outeriter        += nri;
1104
1105     /* Update outer/inner flops */
1106
1107     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*129);
1108 }