src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_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_ElecRF_VdwCSTab_GeomW4P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_pd(fr->ic->k_rf);
 110     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_pd(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 122     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 123     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 150                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155         fix1             = _mm_setzero_pd();
 156         fiy1             = _mm_setzero_pd();
 157         fiz1             = _mm_setzero_pd();
 158         fix2             = _mm_setzero_pd();
 159         fiy2             = _mm_setzero_pd();
 160         fiz2             = _mm_setzero_pd();
 161         fix3             = _mm_setzero_pd();
 162         fiy3             = _mm_setzero_pd();
 163         fiz3             = _mm_setzero_pd();
 164
 165         /* Reset potential sums */
 166         velecsum         = _mm_setzero_pd();
 167         vvdwsum          = _mm_setzero_pd();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             j_coord_offsetA  = DIM*jnrA;
 177             j_coord_offsetB  = DIM*jnrB;
 178
 179             /* load j atom coordinates */
 180             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 181                                               &jx0,&jy0,&jz0);
 182
 183             /* Calculate displacement vector */
 184             dx00             = _mm_sub_pd(ix0,jx0);
 185             dy00             = _mm_sub_pd(iy0,jy0);
 186             dz00             = _mm_sub_pd(iz0,jz0);
 187             dx10             = _mm_sub_pd(ix1,jx0);
 188             dy10             = _mm_sub_pd(iy1,jy0);
 189             dz10             = _mm_sub_pd(iz1,jz0);
 190             dx20             = _mm_sub_pd(ix2,jx0);
 191             dy20             = _mm_sub_pd(iy2,jy0);
 192             dz20             = _mm_sub_pd(iz2,jz0);
 193             dx30             = _mm_sub_pd(ix3,jx0);
 194             dy30             = _mm_sub_pd(iy3,jy0);
 195             dz30             = _mm_sub_pd(iz3,jz0);
 196
 197             /* Calculate squared distance and things based on it */
 198             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 199             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 200             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 201             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 202
 203             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 204             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 205             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 206             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 207
 208             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 209             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 210             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 214             vdwjidx0A        = 2*vdwtype[jnrA+0];
 215             vdwjidx0B        = 2*vdwtype[jnrB+0];
 216
 217             fjx0             = _mm_setzero_pd();
 218             fjy0             = _mm_setzero_pd();
 219             fjz0             = _mm_setzero_pd();
 220
 221             /**************************
 222              * CALCULATE INTERACTIONS *
 223              **************************/
 224
 225             r00              = _mm_mul_pd(rsq00,rinv00);
 226
 227             /* Compute parameters for interactions between i and j atoms */
 228             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 229                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 230
 231             /* Calculate table index by multiplying r with table scale and truncate to integer */
 232             rt               = _mm_mul_pd(r00,vftabscale);
 233             vfitab           = _mm_cvttpd_epi32(rt);
 234 #ifdef __XOP__
 235             vfeps            = _mm_frcz_pd(rt);
 236 #else
 237             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 238 #endif
 239             twovfeps         = _mm_add_pd(vfeps,vfeps);
 240             vfitab           = _mm_slli_epi32(vfitab,3);
 241
 242             /* CUBIC SPLINE TABLE DISPERSION */
 243             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 245             GMX_MM_TRANSPOSE2_PD(Y,F);
 246             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 247             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 248             GMX_MM_TRANSPOSE2_PD(G,H);
 249             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 250             VV               = _mm_macc_pd(vfeps,Fp,Y);
 251             vvdw6            = _mm_mul_pd(c6_00,VV);
 252             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 253             fvdw6            = _mm_mul_pd(c6_00,FF);
 254
 255             /* CUBIC SPLINE TABLE REPULSION */
 256             vfitab           = _mm_add_epi32(vfitab,ifour);
 257             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 258             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 259             GMX_MM_TRANSPOSE2_PD(Y,F);
 260             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 261             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 262             GMX_MM_TRANSPOSE2_PD(G,H);
 263             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 264             VV               = _mm_macc_pd(vfeps,Fp,Y);
 265             vvdw12           = _mm_mul_pd(c12_00,VV);
 266             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 267             fvdw12           = _mm_mul_pd(c12_00,FF);
 268             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 269             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 270
 271             /* Update potential sum for this i atom from the interaction with this j atom. */
 272             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 273
 274             fscal            = fvdw;
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 278             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 279             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 280
 281             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 282             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 283             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             /* Compute parameters for interactions between i and j atoms */
 290             qq10             = _mm_mul_pd(iq1,jq0);
 291
 292             /* REACTION-FIELD ELECTROSTATICS */
 293             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 294             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm_add_pd(velecsum,velec);
 298
 299             fscal            = felec;
 300
 301             /* Update vectorial force */
 302             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 303             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 304             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 305
 306             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 307             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 308             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 309
 310             /**************************
 311              * CALCULATE INTERACTIONS *
 312              **************************/
 313
 314             /* Compute parameters for interactions between i and j atoms */
 315             qq20             = _mm_mul_pd(iq2,jq0);
 316
 317             /* REACTION-FIELD ELECTROSTATICS */
 318             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 319             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 320
 321             /* Update potential sum for this i atom from the interaction with this j atom. */
 322             velecsum         = _mm_add_pd(velecsum,velec);
 323
 324             fscal            = felec;
 325
 326             /* Update vectorial force */
 327             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 328             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 329             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 330
 331             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 332             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 333             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             /* Compute parameters for interactions between i and j atoms */
 340             qq30             = _mm_mul_pd(iq3,jq0);
 341
 342             /* REACTION-FIELD ELECTROSTATICS */
 343             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 344             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 345
 346             /* Update potential sum for this i atom from the interaction with this j atom. */
 347             velecsum         = _mm_add_pd(velecsum,velec);
 348
 349             fscal            = felec;
 350
 351             /* Update vectorial force */
 352             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 353             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 354             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 355
 356             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 357             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 358             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 359
 360             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 361
 362             /* Inner loop uses 167 flops */
 363         }
 364
 365         if(jidx<j_index_end)
 366         {
 367
 368             jnrA             = jjnr[jidx];
 369             j_coord_offsetA  = DIM*jnrA;
 370
 371             /* load j atom coordinates */
 372             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 373                                               &jx0,&jy0,&jz0);
 374
 375             /* Calculate displacement vector */
 376             dx00             = _mm_sub_pd(ix0,jx0);
 377             dy00             = _mm_sub_pd(iy0,jy0);
 378             dz00             = _mm_sub_pd(iz0,jz0);
 379             dx10             = _mm_sub_pd(ix1,jx0);
 380             dy10             = _mm_sub_pd(iy1,jy0);
 381             dz10             = _mm_sub_pd(iz1,jz0);
 382             dx20             = _mm_sub_pd(ix2,jx0);
 383             dy20             = _mm_sub_pd(iy2,jy0);
 384             dz20             = _mm_sub_pd(iz2,jz0);
 385             dx30             = _mm_sub_pd(ix3,jx0);
 386             dy30             = _mm_sub_pd(iy3,jy0);
 387             dz30             = _mm_sub_pd(iz3,jz0);
 388
 389             /* Calculate squared distance and things based on it */
 390             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 391             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 392             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 393             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 394
 395             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 396             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 397             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 398             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 399
 400             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 401             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 402             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 403
 404             /* Load parameters for j particles */
 405             jq0              = _mm_load_sd(charge+jnrA+0);
 406             vdwjidx0A        = 2*vdwtype[jnrA+0];
 407
 408             fjx0             = _mm_setzero_pd();
 409             fjy0             = _mm_setzero_pd();
 410             fjz0             = _mm_setzero_pd();
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             r00              = _mm_mul_pd(rsq00,rinv00);
 417
 418             /* Compute parameters for interactions between i and j atoms */
 419             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 420
 421             /* Calculate table index by multiplying r with table scale and truncate to integer */
 422             rt               = _mm_mul_pd(r00,vftabscale);
 423             vfitab           = _mm_cvttpd_epi32(rt);
 424 #ifdef __XOP__
 425             vfeps            = _mm_frcz_pd(rt);
 426 #else
 427             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 428 #endif
 429             twovfeps         = _mm_add_pd(vfeps,vfeps);
 430             vfitab           = _mm_slli_epi32(vfitab,3);
 431
 432             /* CUBIC SPLINE TABLE DISPERSION */
 433             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 434             F                = _mm_setzero_pd();
 435             GMX_MM_TRANSPOSE2_PD(Y,F);
 436             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 437             H                = _mm_setzero_pd();
 438             GMX_MM_TRANSPOSE2_PD(G,H);
 439             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 440             VV               = _mm_macc_pd(vfeps,Fp,Y);
 441             vvdw6            = _mm_mul_pd(c6_00,VV);
 442             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 443             fvdw6            = _mm_mul_pd(c6_00,FF);
 444
 445             /* CUBIC SPLINE TABLE REPULSION */
 446             vfitab           = _mm_add_epi32(vfitab,ifour);
 447             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 448             F                = _mm_setzero_pd();
 449             GMX_MM_TRANSPOSE2_PD(Y,F);
 450             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 451             H                = _mm_setzero_pd();
 452             GMX_MM_TRANSPOSE2_PD(G,H);
 453             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 454             VV               = _mm_macc_pd(vfeps,Fp,Y);
 455             vvdw12           = _mm_mul_pd(c12_00,VV);
 456             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 457             fvdw12           = _mm_mul_pd(c12_00,FF);
 458             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 459             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 460
 461             /* Update potential sum for this i atom from the interaction with this j atom. */
 462             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 463             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 464
 465             fscal            = fvdw;
 466
 467             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 468
 469             /* Update vectorial force */
 470             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 471             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 472             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 473
 474             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 475             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 476             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             /* Compute parameters for interactions between i and j atoms */
 483             qq10             = _mm_mul_pd(iq1,jq0);
 484
 485             /* REACTION-FIELD ELECTROSTATICS */
 486             velec            = _mm_mul_pd(qq10,_mm_sub_pd(_mm_macc_pd(krf,rsq10,rinv10),crf));
 487             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 488
 489             /* Update potential sum for this i atom from the interaction with this j atom. */
 490             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 491             velecsum         = _mm_add_pd(velecsum,velec);
 492
 493             fscal            = felec;
 494
 495             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 496
 497             /* Update vectorial force */
 498             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 499             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 500             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 501
 502             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 503             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 504             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 505
 506             /**************************
 507              * CALCULATE INTERACTIONS *
 508              **************************/
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq20             = _mm_mul_pd(iq2,jq0);
 512
 513             /* REACTION-FIELD ELECTROSTATICS */
 514             velec            = _mm_mul_pd(qq20,_mm_sub_pd(_mm_macc_pd(krf,rsq20,rinv20),crf));
 515             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 516
 517             /* Update potential sum for this i atom from the interaction with this j atom. */
 518             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 519             velecsum         = _mm_add_pd(velecsum,velec);
 520
 521             fscal            = felec;
 522
 523             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 524
 525             /* Update vectorial force */
 526             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 527             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 528             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 529
 530             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 531             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 532             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 533
 534             /**************************
 535              * CALCULATE INTERACTIONS *
 536              **************************/
 537
 538             /* Compute parameters for interactions between i and j atoms */
 539             qq30             = _mm_mul_pd(iq3,jq0);
 540
 541             /* REACTION-FIELD ELECTROSTATICS */
 542             velec            = _mm_mul_pd(qq30,_mm_sub_pd(_mm_macc_pd(krf,rsq30,rinv30),crf));
 543             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 544
 545             /* Update potential sum for this i atom from the interaction with this j atom. */
 546             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 547             velecsum         = _mm_add_pd(velecsum,velec);
 548
 549             fscal            = felec;
 550
 551             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 552
 553             /* Update vectorial force */
 554             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 555             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 556             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 557
 558             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 559             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 560             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 561
 562             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 563
 564             /* Inner loop uses 167 flops */
 565         }
 566
 567         /* End of innermost loop */
 568
 569         gmx_mm_update_iforce_4atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 570                                               f+i_coord_offset,fshift+i_shift_offset);
 571
 572         ggid                        = gid[iidx];
 573         /* Update potential energies */
 574         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 575         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 576
 577         /* Increment number of inner iterations */
 578         inneriter                  += j_index_end - j_index_start;
 579
 580         /* Outer loop uses 26 flops */
 581     }
 582
 583     /* Increment number of outer iterations */
 584     outeriter        += nri;
 585
 586     /* Update outer/inner flops */
 587
 588     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*167);
 589 }
 590 /*
 591  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 592  * Electrostatics interaction: ReactionField
 593  * VdW interaction:            CubicSplineTable
 594  * Geometry:                   Water4-Particle
 595  * Calculate force/pot:        Force
 596  */
 597 void
 598 nb_kernel_ElecRF_VdwCSTab_GeomW4P1_F_avx_128_fma_double
 599                     (t_nblist * gmx_restrict                nlist,
 600                      rvec * gmx_restrict                    xx,
 601                      rvec * gmx_restrict                    ff,
 602                      t_forcerec * gmx_restrict              fr,
 603                      t_mdatoms * gmx_restrict               mdatoms,
 604                      nb_kernel_data_t * gmx_restrict        kernel_data,
 605                      t_nrnb * gmx_restrict                  nrnb)
 606 {
 607     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 608      * just 0 for non-waters.
 609      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 610      * jnr indices corresponding to data put in the four positions in the SIMD register.
 611      */
 612     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 613     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 614     int              jnrA,jnrB;
 615     int              j_coord_offsetA,j_coord_offsetB;
 616     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 617     real             rcutoff_scalar;
 618     real             *shiftvec,*fshift,*x,*f;
 619     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 620     int              vdwioffset0;
 621     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 622     int              vdwioffset1;
 623     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 624     int              vdwioffset2;
 625     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 626     int              vdwioffset3;
 627     __m128d          ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 628     int              vdwjidx0A,vdwjidx0B;
 629     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 630     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 631     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 632     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 633     __m128d          dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 634     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 635     real             *charge;
 636     int              nvdwtype;
 637     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 638     int              *vdwtype;
 639     real             *vdwparam;
 640     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 641     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 642     __m128i          vfitab;
 643     __m128i          ifour       = _mm_set1_epi32(4);
 644     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 645     real             *vftab;
 646     __m128d          dummy_mask,cutoff_mask;
 647     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 648     __m128d          one     = _mm_set1_pd(1.0);
 649     __m128d          two     = _mm_set1_pd(2.0);
 650     x                = xx[0];
 651     f                = ff[0];
 652
 653     nri              = nlist->nri;
 654     iinr             = nlist->iinr;
 655     jindex           = nlist->jindex;
 656     jjnr             = nlist->jjnr;
 657     shiftidx         = nlist->shift;
 658     gid              = nlist->gid;
 659     shiftvec         = fr->shift_vec[0];
 660     fshift           = fr->fshift[0];
 661     facel            = _mm_set1_pd(fr->epsfac);
 662     charge           = mdatoms->chargeA;
 663     krf              = _mm_set1_pd(fr->ic->k_rf);
 664     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 665     crf              = _mm_set1_pd(fr->ic->c_rf);
 666     nvdwtype         = fr->ntype;
 667     vdwparam         = fr->nbfp;
 668     vdwtype          = mdatoms->typeA;
 669
 670     vftab            = kernel_data->table_vdw->data;
 671     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 672
 673     /* Setup water-specific parameters */
 674     inr              = nlist->iinr[0];
 675     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 676     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 677     iq3              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+3]));
 678     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 679
 680     /* Avoid stupid compiler warnings */
 681     jnrA = jnrB = 0;
 682     j_coord_offsetA = 0;
 683     j_coord_offsetB = 0;
 684
 685     outeriter        = 0;
 686     inneriter        = 0;
 687
 688     /* Start outer loop over neighborlists */
 689     for(iidx=0; iidx<nri; iidx++)
 690     {
 691         /* Load shift vector for this list */
 692         i_shift_offset   = DIM*shiftidx[iidx];
 693
 694         /* Load limits for loop over neighbors */
 695         j_index_start    = jindex[iidx];
 696         j_index_end      = jindex[iidx+1];
 697
 698         /* Get outer coordinate index */
 699         inr              = iinr[iidx];
 700         i_coord_offset   = DIM*inr;
 701
 702         /* Load i particle coords and add shift vector */
 703         gmx_mm_load_shift_and_4rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 704                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 705
 706         fix0             = _mm_setzero_pd();
 707         fiy0             = _mm_setzero_pd();
 708         fiz0             = _mm_setzero_pd();
 709         fix1             = _mm_setzero_pd();
 710         fiy1             = _mm_setzero_pd();
 711         fiz1             = _mm_setzero_pd();
 712         fix2             = _mm_setzero_pd();
 713         fiy2             = _mm_setzero_pd();
 714         fiz2             = _mm_setzero_pd();
 715         fix3             = _mm_setzero_pd();
 716         fiy3             = _mm_setzero_pd();
 717         fiz3             = _mm_setzero_pd();
 718
 719         /* Start inner kernel loop */
 720         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 721         {
 722
 723             /* Get j neighbor index, and coordinate index */
 724             jnrA             = jjnr[jidx];
 725             jnrB             = jjnr[jidx+1];
 726             j_coord_offsetA  = DIM*jnrA;
 727             j_coord_offsetB  = DIM*jnrB;
 728
 729             /* load j atom coordinates */
 730             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 731                                               &jx0,&jy0,&jz0);
 732
 733             /* Calculate displacement vector */
 734             dx00             = _mm_sub_pd(ix0,jx0);
 735             dy00             = _mm_sub_pd(iy0,jy0);
 736             dz00             = _mm_sub_pd(iz0,jz0);
 737             dx10             = _mm_sub_pd(ix1,jx0);
 738             dy10             = _mm_sub_pd(iy1,jy0);
 739             dz10             = _mm_sub_pd(iz1,jz0);
 740             dx20             = _mm_sub_pd(ix2,jx0);
 741             dy20             = _mm_sub_pd(iy2,jy0);
 742             dz20             = _mm_sub_pd(iz2,jz0);
 743             dx30             = _mm_sub_pd(ix3,jx0);
 744             dy30             = _mm_sub_pd(iy3,jy0);
 745             dz30             = _mm_sub_pd(iz3,jz0);
 746
 747             /* Calculate squared distance and things based on it */
 748             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 749             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 750             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 751             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 752
 753             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 754             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 755             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 756             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 757
 758             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 759             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 760             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 761
 762             /* Load parameters for j particles */
 763             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 764             vdwjidx0A        = 2*vdwtype[jnrA+0];
 765             vdwjidx0B        = 2*vdwtype[jnrB+0];
 766
 767             fjx0             = _mm_setzero_pd();
 768             fjy0             = _mm_setzero_pd();
 769             fjz0             = _mm_setzero_pd();
 770
 771             /**************************
 772              * CALCULATE INTERACTIONS *
 773              **************************/
 774
 775             r00              = _mm_mul_pd(rsq00,rinv00);
 776
 777             /* Compute parameters for interactions between i and j atoms */
 778             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 779                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 780
 781             /* Calculate table index by multiplying r with table scale and truncate to integer */
 782             rt               = _mm_mul_pd(r00,vftabscale);
 783             vfitab           = _mm_cvttpd_epi32(rt);
 784 #ifdef __XOP__
 785             vfeps            = _mm_frcz_pd(rt);
 786 #else
 787             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 788 #endif
 789             twovfeps         = _mm_add_pd(vfeps,vfeps);
 790             vfitab           = _mm_slli_epi32(vfitab,3);
 791
 792             /* CUBIC SPLINE TABLE DISPERSION */
 793             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 794             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 795             GMX_MM_TRANSPOSE2_PD(Y,F);
 796             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 797             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 798             GMX_MM_TRANSPOSE2_PD(G,H);
 799             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 800             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 801             fvdw6            = _mm_mul_pd(c6_00,FF);
 802
 803             /* CUBIC SPLINE TABLE REPULSION */
 804             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             fvdw12           = _mm_mul_pd(c12_00,FF);
 814             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 815
 816             fscal            = fvdw;
 817
 818             /* Update vectorial force */
 819             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 820             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 821             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 822
 823             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 824             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 825             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 826
 827             /**************************
 828              * CALCULATE INTERACTIONS *
 829              **************************/
 830
 831             /* Compute parameters for interactions between i and j atoms */
 832             qq10             = _mm_mul_pd(iq1,jq0);
 833
 834             /* REACTION-FIELD ELECTROSTATICS */
 835             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
 836
 837             fscal            = felec;
 838
 839             /* Update vectorial force */
 840             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 841             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 842             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 843
 844             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 845             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 846             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 847
 848             /**************************
 849              * CALCULATE INTERACTIONS *
 850              **************************/
 851
 852             /* Compute parameters for interactions between i and j atoms */
 853             qq20             = _mm_mul_pd(iq2,jq0);
 854
 855             /* REACTION-FIELD ELECTROSTATICS */
 856             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
 857
 858             fscal            = felec;
 859
 860             /* Update vectorial force */
 861             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 862             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 863             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 864
 865             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 866             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 867             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 868
 869             /**************************
 870              * CALCULATE INTERACTIONS *
 871              **************************/
 872
 873             /* Compute parameters for interactions between i and j atoms */
 874             qq30             = _mm_mul_pd(iq3,jq0);
 875
 876             /* REACTION-FIELD ELECTROSTATICS */
 877             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
 878
 879             fscal            = felec;
 880
 881             /* Update vectorial force */
 882             fix3             = _mm_macc_pd(dx30,fscal,fix3);
 883             fiy3             = _mm_macc_pd(dy30,fscal,fiy3);
 884             fiz3             = _mm_macc_pd(dz30,fscal,fiz3);
 885
 886             fjx0             = _mm_macc_pd(dx30,fscal,fjx0);
 887             fjy0             = _mm_macc_pd(dy30,fscal,fjy0);
 888             fjz0             = _mm_macc_pd(dz30,fscal,fjz0);
 889
 890             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 891
 892             /* Inner loop uses 144 flops */
 893         }
 894
 895         if(jidx<j_index_end)
 896         {
 897
 898             jnrA             = jjnr[jidx];
 899             j_coord_offsetA  = DIM*jnrA;
 900
 901             /* load j atom coordinates */
 902             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 903                                               &jx0,&jy0,&jz0);
 904
 905             /* Calculate displacement vector */
 906             dx00             = _mm_sub_pd(ix0,jx0);
 907             dy00             = _mm_sub_pd(iy0,jy0);
 908             dz00             = _mm_sub_pd(iz0,jz0);
 909             dx10             = _mm_sub_pd(ix1,jx0);
 910             dy10             = _mm_sub_pd(iy1,jy0);
 911             dz10             = _mm_sub_pd(iz1,jz0);
 912             dx20             = _mm_sub_pd(ix2,jx0);
 913             dy20             = _mm_sub_pd(iy2,jy0);
 914             dz20             = _mm_sub_pd(iz2,jz0);
 915             dx30             = _mm_sub_pd(ix3,jx0);
 916             dy30             = _mm_sub_pd(iy3,jy0);
 917             dz30             = _mm_sub_pd(iz3,jz0);
 918
 919             /* Calculate squared distance and things based on it */
 920             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 921             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 922             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 923             rsq30            = gmx_mm_calc_rsq_pd(dx30,dy30,dz30);
 924
 925             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 926             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 927             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 928             rinv30           = gmx_mm_invsqrt_pd(rsq30);
 929
 930             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 931             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 932             rinvsq30         = _mm_mul_pd(rinv30,rinv30);
 933
 934             /* Load parameters for j particles */
 935             jq0              = _mm_load_sd(charge+jnrA+0);
 936             vdwjidx0A        = 2*vdwtype[jnrA+0];
 937
 938             fjx0             = _mm_setzero_pd();
 939             fjy0             = _mm_setzero_pd();
 940             fjz0             = _mm_setzero_pd();
 941
 942             /**************************
 943              * CALCULATE INTERACTIONS *
 944              **************************/
 945
 946             r00              = _mm_mul_pd(rsq00,rinv00);
 947
 948             /* Compute parameters for interactions between i and j atoms */
 949             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 950
 951             /* Calculate table index by multiplying r with table scale and truncate to integer */
 952             rt               = _mm_mul_pd(r00,vftabscale);
 953             vfitab           = _mm_cvttpd_epi32(rt);
 954 #ifdef __XOP__
 955             vfeps            = _mm_frcz_pd(rt);
 956 #else
 957             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 958 #endif
 959             twovfeps         = _mm_add_pd(vfeps,vfeps);
 960             vfitab           = _mm_slli_epi32(vfitab,3);
 961
 962             /* CUBIC SPLINE TABLE DISPERSION */
 963             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 964             F                = _mm_setzero_pd();
 965             GMX_MM_TRANSPOSE2_PD(Y,F);
 966             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 967             H                = _mm_setzero_pd();
 968             GMX_MM_TRANSPOSE2_PD(G,H);
 969             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 970             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 971             fvdw6            = _mm_mul_pd(c6_00,FF);
 972
 973             /* CUBIC SPLINE TABLE REPULSION */
 974             vfitab           = _mm_add_epi32(vfitab,ifour);
 975             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 976             F                = _mm_setzero_pd();
 977             GMX_MM_TRANSPOSE2_PD(Y,F);
 978             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 979             H                = _mm_setzero_pd();
 980             GMX_MM_TRANSPOSE2_PD(G,H);
 981             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 982             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 983             fvdw12           = _mm_mul_pd(c12_00,FF);
 984             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 985
 986             fscal            = fvdw;
 987
 988             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 989
 990             /* Update vectorial force */
 991             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 992             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 993             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 994
 995             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 996             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 997             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq10             = _mm_mul_pd(iq1,jq0);
1005
1006             /* REACTION-FIELD ELECTROSTATICS */
1007             felec            = _mm_mul_pd(qq10,_mm_msub_pd(rinv10,rinvsq10,krf2));
1008
1009             fscal            = felec;
1010
1011             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1012
1013             /* Update vectorial force */
1014             fix1             = _mm_macc_pd(dx10,fscal,fix1);
1015             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
1016             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
1017
1018             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
1019             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
1020             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             /* Compute parameters for interactions between i and j atoms */
1027             qq20             = _mm_mul_pd(iq2,jq0);
1028
1029             /* REACTION-FIELD ELECTROSTATICS */
1030             felec            = _mm_mul_pd(qq20,_mm_msub_pd(rinv20,rinvsq20,krf2));
1031
1032             fscal            = felec;
1033
1034             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
1035
1036             /* Update vectorial force */
1037             fix2             = _mm_macc_pd(dx20,fscal,fix2);
1038             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
1039             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
1040
1041             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
1042             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
1043             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
1044
1045             /**************************
1046              * CALCULATE INTERACTIONS *
1047              **************************/
1048
1049             /* Compute parameters for interactions between i and j atoms */
1050             qq30             = _mm_mul_pd(iq3,jq0);
1051
1052             /* REACTION-FIELD ELECTROSTATICS */
1053             felec            = _mm_mul_pd(qq30,_mm_msub_pd(rinv30,rinvsq30,krf2));
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 }