src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_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_VdwLJSh_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_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              vdwjidx0A,vdwjidx0B;
  69     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  70     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  71     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  72     real             *charge;
  73     int              nvdwtype;
  74     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  75     int              *vdwtype;
  76     real             *vdwparam;
  77     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  78     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  79     __m128d          dummy_mask,cutoff_mask;
  80     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  81     __m128d          one     = _mm_set1_pd(1.0);
  82     __m128d          two     = _mm_set1_pd(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_pd(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96     krf              = _mm_set1_pd(fr->ic->k_rf);
  97     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
  98     crf              = _mm_set1_pd(fr->ic->c_rf);
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 104     rcutoff_scalar   = fr->rcoulomb;
 105     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 106     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 107
 108     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 109     rvdw             = _mm_set1_pd(fr->rvdw);
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     /* Start outer loop over neighborlists */
 120     for(iidx=0; iidx<nri; iidx++)
 121     {
 122         /* Load shift vector for this list */
 123         i_shift_offset   = DIM*shiftidx[iidx];
 124
 125         /* Load limits for loop over neighbors */
 126         j_index_start    = jindex[iidx];
 127         j_index_end      = jindex[iidx+1];
 128
 129         /* Get outer coordinate index */
 130         inr              = iinr[iidx];
 131         i_coord_offset   = DIM*inr;
 132
 133         /* Load i particle coords and add shift vector */
 134         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 135
 136         fix0             = _mm_setzero_pd();
 137         fiy0             = _mm_setzero_pd();
 138         fiz0             = _mm_setzero_pd();
 139
 140         /* Load parameters for i particles */
 141         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 142         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 143
 144         /* Reset potential sums */
 145         velecsum         = _mm_setzero_pd();
 146         vvdwsum          = _mm_setzero_pd();
 147
 148         /* Start inner kernel loop */
 149         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 150         {
 151
 152             /* Get j neighbor index, and coordinate index */
 153             jnrA             = jjnr[jidx];
 154             jnrB             = jjnr[jidx+1];
 155             j_coord_offsetA  = DIM*jnrA;
 156             j_coord_offsetB  = DIM*jnrB;
 157
 158             /* load j atom coordinates */
 159             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 160                                               &jx0,&jy0,&jz0);
 161
 162             /* Calculate displacement vector */
 163             dx00             = _mm_sub_pd(ix0,jx0);
 164             dy00             = _mm_sub_pd(iy0,jy0);
 165             dz00             = _mm_sub_pd(iz0,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 169
 170             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 171
 172             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 173
 174             /* Load parameters for j particles */
 175             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 176             vdwjidx0A        = 2*vdwtype[jnrA+0];
 177             vdwjidx0B        = 2*vdwtype[jnrB+0];
 178
 179             /**************************
 180              * CALCULATE INTERACTIONS *
 181              **************************/
 182
 183             if (gmx_mm_any_lt(rsq00,rcutoff2))
 184             {
 185
 186             /* Compute parameters for interactions between i and j atoms */
 187             qq00             = _mm_mul_pd(iq0,jq0);
 188             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 189                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 190
 191             /* REACTION-FIELD ELECTROSTATICS */
 192             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 193             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 194
 195             /* LENNARD-JONES DISPERSION/REPULSION */
 196
 197             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 198             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 199             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 200             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 201                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 202             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 203
 204             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 205
 206             /* Update potential sum for this i atom from the interaction with this j atom. */
 207             velec            = _mm_and_pd(velec,cutoff_mask);
 208             velecsum         = _mm_add_pd(velecsum,velec);
 209             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 210             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 211
 212             fscal            = _mm_add_pd(felec,fvdw);
 213
 214             fscal            = _mm_and_pd(fscal,cutoff_mask);
 215
 216             /* Update vectorial force */
 217             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 218             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 219             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 220
 221             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 222                                                    _mm_mul_pd(dx00,fscal),
 223                                                    _mm_mul_pd(dy00,fscal),
 224                                                    _mm_mul_pd(dz00,fscal));
 225
 226             }
 227
 228             /* Inner loop uses 57 flops */
 229         }
 230
 231         if(jidx<j_index_end)
 232         {
 233
 234             jnrA             = jjnr[jidx];
 235             j_coord_offsetA  = DIM*jnrA;
 236
 237             /* load j atom coordinates */
 238             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 239                                               &jx0,&jy0,&jz0);
 240
 241             /* Calculate displacement vector */
 242             dx00             = _mm_sub_pd(ix0,jx0);
 243             dy00             = _mm_sub_pd(iy0,jy0);
 244             dz00             = _mm_sub_pd(iz0,jz0);
 245
 246             /* Calculate squared distance and things based on it */
 247             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 248
 249             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 250
 251             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 252
 253             /* Load parameters for j particles */
 254             jq0              = _mm_load_sd(charge+jnrA+0);
 255             vdwjidx0A        = 2*vdwtype[jnrA+0];
 256
 257             /**************************
 258              * CALCULATE INTERACTIONS *
 259              **************************/
 260
 261             if (gmx_mm_any_lt(rsq00,rcutoff2))
 262             {
 263
 264             /* Compute parameters for interactions between i and j atoms */
 265             qq00             = _mm_mul_pd(iq0,jq0);
 266             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 267
 268             /* REACTION-FIELD ELECTROSTATICS */
 269             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 270             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 271
 272             /* LENNARD-JONES DISPERSION/REPULSION */
 273
 274             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 275             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 276             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 277             vvdw             = _mm_msub_pd(_mm_nmacc_pd(c12_00,_mm_mul_pd(sh_vdw_invrcut6,sh_vdw_invrcut6),vvdw12),one_twelfth,
 278                                            _mm_mul_pd(_mm_nmacc_pd( c6_00,sh_vdw_invrcut6,vvdw6),one_sixth));
 279             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 280
 281             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 282
 283             /* Update potential sum for this i atom from the interaction with this j atom. */
 284             velec            = _mm_and_pd(velec,cutoff_mask);
 285             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 286             velecsum         = _mm_add_pd(velecsum,velec);
 287             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 288             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 289             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 290
 291             fscal            = _mm_add_pd(felec,fvdw);
 292
 293             fscal            = _mm_and_pd(fscal,cutoff_mask);
 294
 295             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 296
 297             /* Update vectorial force */
 298             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 299             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 300             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 301
 302             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 303                                                    _mm_mul_pd(dx00,fscal),
 304                                                    _mm_mul_pd(dy00,fscal),
 305                                                    _mm_mul_pd(dz00,fscal));
 306
 307             }
 308
 309             /* Inner loop uses 57 flops */
 310         }
 311
 312         /* End of innermost loop */
 313
 314         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 315                                               f+i_coord_offset,fshift+i_shift_offset);
 316
 317         ggid                        = gid[iidx];
 318         /* Update potential energies */
 319         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 320         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 321
 322         /* Increment number of inner iterations */
 323         inneriter                  += j_index_end - j_index_start;
 324
 325         /* Outer loop uses 9 flops */
 326     }
 327
 328     /* Increment number of outer iterations */
 329     outeriter        += nri;
 330
 331     /* Update outer/inner flops */
 332
 333     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*57);
 334 }
 335 /*
 336  * Gromacs nonbonded kernel:   nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 337  * Electrostatics interaction: ReactionField
 338  * VdW interaction:            LennardJones
 339  * Geometry:                   Particle-Particle
 340  * Calculate force/pot:        Force
 341  */
 342 void
 343 nb_kernel_ElecRFCut_VdwLJSh_GeomP1P1_F_avx_128_fma_double
 344                     (t_nblist * gmx_restrict                nlist,
 345                      rvec * gmx_restrict                    xx,
 346                      rvec * gmx_restrict                    ff,
 347                      t_forcerec * gmx_restrict              fr,
 348                      t_mdatoms * gmx_restrict               mdatoms,
 349                      nb_kernel_data_t * gmx_restrict        kernel_data,
 350                      t_nrnb * gmx_restrict                  nrnb)
 351 {
 352     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 353      * just 0 for non-waters.
 354      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 355      * jnr indices corresponding to data put in the four positions in the SIMD register.
 356      */
 357     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 358     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 359     int              jnrA,jnrB;
 360     int              j_coord_offsetA,j_coord_offsetB;
 361     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 362     real             rcutoff_scalar;
 363     real             *shiftvec,*fshift,*x,*f;
 364     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 365     int              vdwioffset0;
 366     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 367     int              vdwjidx0A,vdwjidx0B;
 368     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 369     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 370     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 371     real             *charge;
 372     int              nvdwtype;
 373     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 374     int              *vdwtype;
 375     real             *vdwparam;
 376     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 377     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 378     __m128d          dummy_mask,cutoff_mask;
 379     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 380     __m128d          one     = _mm_set1_pd(1.0);
 381     __m128d          two     = _mm_set1_pd(2.0);
 382     x                = xx[0];
 383     f                = ff[0];
 384
 385     nri              = nlist->nri;
 386     iinr             = nlist->iinr;
 387     jindex           = nlist->jindex;
 388     jjnr             = nlist->jjnr;
 389     shiftidx         = nlist->shift;
 390     gid              = nlist->gid;
 391     shiftvec         = fr->shift_vec[0];
 392     fshift           = fr->fshift[0];
 393     facel            = _mm_set1_pd(fr->epsfac);
 394     charge           = mdatoms->chargeA;
 395     krf              = _mm_set1_pd(fr->ic->k_rf);
 396     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 397     crf              = _mm_set1_pd(fr->ic->c_rf);
 398     nvdwtype         = fr->ntype;
 399     vdwparam         = fr->nbfp;
 400     vdwtype          = mdatoms->typeA;
 401
 402     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 403     rcutoff_scalar   = fr->rcoulomb;
 404     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 405     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 406
 407     sh_vdw_invrcut6  = _mm_set1_pd(fr->ic->sh_invrc6);
 408     rvdw             = _mm_set1_pd(fr->rvdw);
 409
 410     /* Avoid stupid compiler warnings */
 411     jnrA = jnrB = 0;
 412     j_coord_offsetA = 0;
 413     j_coord_offsetB = 0;
 414
 415     outeriter        = 0;
 416     inneriter        = 0;
 417
 418     /* Start outer loop over neighborlists */
 419     for(iidx=0; iidx<nri; iidx++)
 420     {
 421         /* Load shift vector for this list */
 422         i_shift_offset   = DIM*shiftidx[iidx];
 423
 424         /* Load limits for loop over neighbors */
 425         j_index_start    = jindex[iidx];
 426         j_index_end      = jindex[iidx+1];
 427
 428         /* Get outer coordinate index */
 429         inr              = iinr[iidx];
 430         i_coord_offset   = DIM*inr;
 431
 432         /* Load i particle coords and add shift vector */
 433         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 434
 435         fix0             = _mm_setzero_pd();
 436         fiy0             = _mm_setzero_pd();
 437         fiz0             = _mm_setzero_pd();
 438
 439         /* Load parameters for i particles */
 440         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 441         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 442
 443         /* Start inner kernel loop */
 444         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 445         {
 446
 447             /* Get j neighbor index, and coordinate index */
 448             jnrA             = jjnr[jidx];
 449             jnrB             = jjnr[jidx+1];
 450             j_coord_offsetA  = DIM*jnrA;
 451             j_coord_offsetB  = DIM*jnrB;
 452
 453             /* load j atom coordinates */
 454             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 455                                               &jx0,&jy0,&jz0);
 456
 457             /* Calculate displacement vector */
 458             dx00             = _mm_sub_pd(ix0,jx0);
 459             dy00             = _mm_sub_pd(iy0,jy0);
 460             dz00             = _mm_sub_pd(iz0,jz0);
 461
 462             /* Calculate squared distance and things based on it */
 463             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 464
 465             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 466
 467             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 468
 469             /* Load parameters for j particles */
 470             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 471             vdwjidx0A        = 2*vdwtype[jnrA+0];
 472             vdwjidx0B        = 2*vdwtype[jnrB+0];
 473
 474             /**************************
 475              * CALCULATE INTERACTIONS *
 476              **************************/
 477
 478             if (gmx_mm_any_lt(rsq00,rcutoff2))
 479             {
 480
 481             /* Compute parameters for interactions between i and j atoms */
 482             qq00             = _mm_mul_pd(iq0,jq0);
 483             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 484                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 485
 486             /* REACTION-FIELD ELECTROSTATICS */
 487             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 488
 489             /* LENNARD-JONES DISPERSION/REPULSION */
 490
 491             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 492             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 493
 494             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 495
 496             fscal            = _mm_add_pd(felec,fvdw);
 497
 498             fscal            = _mm_and_pd(fscal,cutoff_mask);
 499
 500             /* Update vectorial force */
 501             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 502             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 503             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 504
 505             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 506                                                    _mm_mul_pd(dx00,fscal),
 507                                                    _mm_mul_pd(dy00,fscal),
 508                                                    _mm_mul_pd(dz00,fscal));
 509
 510             }
 511
 512             /* Inner loop uses 40 flops */
 513         }
 514
 515         if(jidx<j_index_end)
 516         {
 517
 518             jnrA             = jjnr[jidx];
 519             j_coord_offsetA  = DIM*jnrA;
 520
 521             /* load j atom coordinates */
 522             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 523                                               &jx0,&jy0,&jz0);
 524
 525             /* Calculate displacement vector */
 526             dx00             = _mm_sub_pd(ix0,jx0);
 527             dy00             = _mm_sub_pd(iy0,jy0);
 528             dz00             = _mm_sub_pd(iz0,jz0);
 529
 530             /* Calculate squared distance and things based on it */
 531             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 532
 533             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 534
 535             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 536
 537             /* Load parameters for j particles */
 538             jq0              = _mm_load_sd(charge+jnrA+0);
 539             vdwjidx0A        = 2*vdwtype[jnrA+0];
 540
 541             /**************************
 542              * CALCULATE INTERACTIONS *
 543              **************************/
 544
 545             if (gmx_mm_any_lt(rsq00,rcutoff2))
 546             {
 547
 548             /* Compute parameters for interactions between i and j atoms */
 549             qq00             = _mm_mul_pd(iq0,jq0);
 550             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 551
 552             /* REACTION-FIELD ELECTROSTATICS */
 553             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 554
 555             /* LENNARD-JONES DISPERSION/REPULSION */
 556
 557             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 558             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 559
 560             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 561
 562             fscal            = _mm_add_pd(felec,fvdw);
 563
 564             fscal            = _mm_and_pd(fscal,cutoff_mask);
 565
 566             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 567
 568             /* Update vectorial force */
 569             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 570             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 571             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 572
 573             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 574                                                    _mm_mul_pd(dx00,fscal),
 575                                                    _mm_mul_pd(dy00,fscal),
 576                                                    _mm_mul_pd(dz00,fscal));
 577
 578             }
 579
 580             /* Inner loop uses 40 flops */
 581         }
 582
 583         /* End of innermost loop */
 584
 585         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 586                                               f+i_coord_offset,fshift+i_shift_offset);
 587
 588         /* Increment number of inner iterations */
 589         inneriter                  += j_index_end - j_index_start;
 590
 591         /* Outer loop uses 7 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_F,outeriter*7 + inneriter*40);
 600 }