src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwLJ_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomP1P1_VF_sse4_1_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     __m128i          ewitab;
  80     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128d          dummy_mask,cutoff_mask;
  83     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  84     __m128d          one     = _mm_set1_pd(1.0);
  85     __m128d          two     = _mm_set1_pd(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_pd(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     nvdwtype         = fr->ntype;
 100     vdwparam         = fr->nbfp;
 101     vdwtype          = mdatoms->typeA;
 102
 103     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 104     ewtab            = fr->ic->tabq_coul_FDV0;
 105     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 106     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 107
 108     /* Avoid stupid compiler warnings */
 109     jnrA = jnrB = 0;
 110     j_coord_offsetA = 0;
 111     j_coord_offsetB = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     /* Start outer loop over neighborlists */
 117     for(iidx=0; iidx<nri; iidx++)
 118     {
 119         /* Load shift vector for this list */
 120         i_shift_offset   = DIM*shiftidx[iidx];
 121
 122         /* Load limits for loop over neighbors */
 123         j_index_start    = jindex[iidx];
 124         j_index_end      = jindex[iidx+1];
 125
 126         /* Get outer coordinate index */
 127         inr              = iinr[iidx];
 128         i_coord_offset   = DIM*inr;
 129
 130         /* Load i particle coords and add shift vector */
 131         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 132
 133         fix0             = _mm_setzero_pd();
 134         fiy0             = _mm_setzero_pd();
 135         fiz0             = _mm_setzero_pd();
 136
 137         /* Load parameters for i particles */
 138         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 139         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 140
 141         /* Reset potential sums */
 142         velecsum         = _mm_setzero_pd();
 143         vvdwsum          = _mm_setzero_pd();
 144
 145         /* Start inner kernel loop */
 146         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 147         {
 148
 149             /* Get j neighbor index, and coordinate index */
 150             jnrA             = jjnr[jidx];
 151             jnrB             = jjnr[jidx+1];
 152             j_coord_offsetA  = DIM*jnrA;
 153             j_coord_offsetB  = DIM*jnrB;
 154
 155             /* load j atom coordinates */
 156             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 157                                               &jx0,&jy0,&jz0);
 158
 159             /* Calculate displacement vector */
 160             dx00             = _mm_sub_pd(ix0,jx0);
 161             dy00             = _mm_sub_pd(iy0,jy0);
 162             dz00             = _mm_sub_pd(iz0,jz0);
 163
 164             /* Calculate squared distance and things based on it */
 165             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 166
 167             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 168
 169             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 170
 171             /* Load parameters for j particles */
 172             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 173             vdwjidx0A        = 2*vdwtype[jnrA+0];
 174             vdwjidx0B        = 2*vdwtype[jnrB+0];
 175
 176             /**************************
 177              * CALCULATE INTERACTIONS *
 178              **************************/
 179
 180             r00              = _mm_mul_pd(rsq00,rinv00);
 181
 182             /* Compute parameters for interactions between i and j atoms */
 183             qq00             = _mm_mul_pd(iq0,jq0);
 184             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 185                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 186
 187             /* EWALD ELECTROSTATICS */
 188
 189             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 190             ewrt             = _mm_mul_pd(r00,ewtabscale);
 191             ewitab           = _mm_cvttpd_epi32(ewrt);
 192             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 193             ewitab           = _mm_slli_epi32(ewitab,2);
 194             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 195             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 196             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 197             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 198             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 199             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 200             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 201             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 202             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 203             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 204
 205             /* LENNARD-JONES DISPERSION/REPULSION */
 206
 207             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 208             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 209             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 210             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 211             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velecsum         = _mm_add_pd(velecsum,velec);
 215             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 216
 217             fscal            = _mm_add_pd(felec,fvdw);
 218
 219             /* Calculate temporary vectorial force */
 220             tx               = _mm_mul_pd(fscal,dx00);
 221             ty               = _mm_mul_pd(fscal,dy00);
 222             tz               = _mm_mul_pd(fscal,dz00);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm_add_pd(fix0,tx);
 226             fiy0             = _mm_add_pd(fiy0,ty);
 227             fiz0             = _mm_add_pd(fiz0,tz);
 228
 229             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 230
 231             /* Inner loop uses 53 flops */
 232         }
 233
 234         if(jidx<j_index_end)
 235         {
 236
 237             jnrA             = jjnr[jidx];
 238             j_coord_offsetA  = DIM*jnrA;
 239
 240             /* load j atom coordinates */
 241             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 242                                               &jx0,&jy0,&jz0);
 243
 244             /* Calculate displacement vector */
 245             dx00             = _mm_sub_pd(ix0,jx0);
 246             dy00             = _mm_sub_pd(iy0,jy0);
 247             dz00             = _mm_sub_pd(iz0,jz0);
 248
 249             /* Calculate squared distance and things based on it */
 250             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 251
 252             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 253
 254             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 255
 256             /* Load parameters for j particles */
 257             jq0              = _mm_load_sd(charge+jnrA+0);
 258             vdwjidx0A        = 2*vdwtype[jnrA+0];
 259
 260             /**************************
 261              * CALCULATE INTERACTIONS *
 262              **************************/
 263
 264             r00              = _mm_mul_pd(rsq00,rinv00);
 265
 266             /* Compute parameters for interactions between i and j atoms */
 267             qq00             = _mm_mul_pd(iq0,jq0);
 268             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 269
 270             /* EWALD ELECTROSTATICS */
 271
 272             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 273             ewrt             = _mm_mul_pd(r00,ewtabscale);
 274             ewitab           = _mm_cvttpd_epi32(ewrt);
 275             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 276             ewitab           = _mm_slli_epi32(ewitab,2);
 277             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 278             ewtabD           = _mm_setzero_pd();
 279             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 280             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 281             ewtabFn          = _mm_setzero_pd();
 282             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 283             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 284             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 285             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 286             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 287
 288             /* LENNARD-JONES DISPERSION/REPULSION */
 289
 290             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 291             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 292             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 293             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 294             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 298             velecsum         = _mm_add_pd(velecsum,velec);
 299             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 300             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 301
 302             fscal            = _mm_add_pd(felec,fvdw);
 303
 304             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 305
 306             /* Calculate temporary vectorial force */
 307             tx               = _mm_mul_pd(fscal,dx00);
 308             ty               = _mm_mul_pd(fscal,dy00);
 309             tz               = _mm_mul_pd(fscal,dz00);
 310
 311             /* Update vectorial force */
 312             fix0             = _mm_add_pd(fix0,tx);
 313             fiy0             = _mm_add_pd(fiy0,ty);
 314             fiz0             = _mm_add_pd(fiz0,tz);
 315
 316             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 317
 318             /* Inner loop uses 53 flops */
 319         }
 320
 321         /* End of innermost loop */
 322
 323         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 324                                               f+i_coord_offset,fshift+i_shift_offset);
 325
 326         ggid                        = gid[iidx];
 327         /* Update potential energies */
 328         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 329         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 330
 331         /* Increment number of inner iterations */
 332         inneriter                  += j_index_end - j_index_start;
 333
 334         /* Outer loop uses 9 flops */
 335     }
 336
 337     /* Increment number of outer iterations */
 338     outeriter        += nri;
 339
 340     /* Update outer/inner flops */
 341
 342     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*53);
 343 }
 344 /*
 345  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse4_1_double
 346  * Electrostatics interaction: Ewald
 347  * VdW interaction:            LennardJones
 348  * Geometry:                   Particle-Particle
 349  * Calculate force/pot:        Force
 350  */
 351 void
 352 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_sse4_1_double
 353                     (t_nblist * gmx_restrict                nlist,
 354                      rvec * gmx_restrict                    xx,
 355                      rvec * gmx_restrict                    ff,
 356                      t_forcerec * gmx_restrict              fr,
 357                      t_mdatoms * gmx_restrict               mdatoms,
 358                      nb_kernel_data_t * gmx_restrict        kernel_data,
 359                      t_nrnb * gmx_restrict                  nrnb)
 360 {
 361     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 362      * just 0 for non-waters.
 363      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 364      * jnr indices corresponding to data put in the four positions in the SIMD register.
 365      */
 366     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 367     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 368     int              jnrA,jnrB;
 369     int              j_coord_offsetA,j_coord_offsetB;
 370     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 371     real             rcutoff_scalar;
 372     real             *shiftvec,*fshift,*x,*f;
 373     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 374     int              vdwioffset0;
 375     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 376     int              vdwjidx0A,vdwjidx0B;
 377     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 378     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 379     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 380     real             *charge;
 381     int              nvdwtype;
 382     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 383     int              *vdwtype;
 384     real             *vdwparam;
 385     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 386     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 387     __m128i          ewitab;
 388     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 389     real             *ewtab;
 390     __m128d          dummy_mask,cutoff_mask;
 391     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 392     __m128d          one     = _mm_set1_pd(1.0);
 393     __m128d          two     = _mm_set1_pd(2.0);
 394     x                = xx[0];
 395     f                = ff[0];
 396
 397     nri              = nlist->nri;
 398     iinr             = nlist->iinr;
 399     jindex           = nlist->jindex;
 400     jjnr             = nlist->jjnr;
 401     shiftidx         = nlist->shift;
 402     gid              = nlist->gid;
 403     shiftvec         = fr->shift_vec[0];
 404     fshift           = fr->fshift[0];
 405     facel            = _mm_set1_pd(fr->epsfac);
 406     charge           = mdatoms->chargeA;
 407     nvdwtype         = fr->ntype;
 408     vdwparam         = fr->nbfp;
 409     vdwtype          = mdatoms->typeA;
 410
 411     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 412     ewtab            = fr->ic->tabq_coul_F;
 413     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 414     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 415
 416     /* Avoid stupid compiler warnings */
 417     jnrA = jnrB = 0;
 418     j_coord_offsetA = 0;
 419     j_coord_offsetB = 0;
 420
 421     outeriter        = 0;
 422     inneriter        = 0;
 423
 424     /* Start outer loop over neighborlists */
 425     for(iidx=0; iidx<nri; iidx++)
 426     {
 427         /* Load shift vector for this list */
 428         i_shift_offset   = DIM*shiftidx[iidx];
 429
 430         /* Load limits for loop over neighbors */
 431         j_index_start    = jindex[iidx];
 432         j_index_end      = jindex[iidx+1];
 433
 434         /* Get outer coordinate index */
 435         inr              = iinr[iidx];
 436         i_coord_offset   = DIM*inr;
 437
 438         /* Load i particle coords and add shift vector */
 439         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 440
 441         fix0             = _mm_setzero_pd();
 442         fiy0             = _mm_setzero_pd();
 443         fiz0             = _mm_setzero_pd();
 444
 445         /* Load parameters for i particles */
 446         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 447         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 448
 449         /* Start inner kernel loop */
 450         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 451         {
 452
 453             /* Get j neighbor index, and coordinate index */
 454             jnrA             = jjnr[jidx];
 455             jnrB             = jjnr[jidx+1];
 456             j_coord_offsetA  = DIM*jnrA;
 457             j_coord_offsetB  = DIM*jnrB;
 458
 459             /* load j atom coordinates */
 460             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 461                                               &jx0,&jy0,&jz0);
 462
 463             /* Calculate displacement vector */
 464             dx00             = _mm_sub_pd(ix0,jx0);
 465             dy00             = _mm_sub_pd(iy0,jy0);
 466             dz00             = _mm_sub_pd(iz0,jz0);
 467
 468             /* Calculate squared distance and things based on it */
 469             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 470
 471             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 472
 473             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 474
 475             /* Load parameters for j particles */
 476             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 477             vdwjidx0A        = 2*vdwtype[jnrA+0];
 478             vdwjidx0B        = 2*vdwtype[jnrB+0];
 479
 480             /**************************
 481              * CALCULATE INTERACTIONS *
 482              **************************/
 483
 484             r00              = _mm_mul_pd(rsq00,rinv00);
 485
 486             /* Compute parameters for interactions between i and j atoms */
 487             qq00             = _mm_mul_pd(iq0,jq0);
 488             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 489                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 490
 491             /* EWALD ELECTROSTATICS */
 492
 493             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 494             ewrt             = _mm_mul_pd(r00,ewtabscale);
 495             ewitab           = _mm_cvttpd_epi32(ewrt);
 496             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 497             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 498                                          &ewtabF,&ewtabFn);
 499             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 500             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 501
 502             /* LENNARD-JONES DISPERSION/REPULSION */
 503
 504             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 505             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 506
 507             fscal            = _mm_add_pd(felec,fvdw);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_pd(fscal,dx00);
 511             ty               = _mm_mul_pd(fscal,dy00);
 512             tz               = _mm_mul_pd(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_pd(fix0,tx);
 516             fiy0             = _mm_add_pd(fiy0,ty);
 517             fiz0             = _mm_add_pd(fiz0,tz);
 518
 519             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 520
 521             /* Inner loop uses 43 flops */
 522         }
 523
 524         if(jidx<j_index_end)
 525         {
 526
 527             jnrA             = jjnr[jidx];
 528             j_coord_offsetA  = DIM*jnrA;
 529
 530             /* load j atom coordinates */
 531             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 532                                               &jx0,&jy0,&jz0);
 533
 534             /* Calculate displacement vector */
 535             dx00             = _mm_sub_pd(ix0,jx0);
 536             dy00             = _mm_sub_pd(iy0,jy0);
 537             dz00             = _mm_sub_pd(iz0,jz0);
 538
 539             /* Calculate squared distance and things based on it */
 540             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 541
 542             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 543
 544             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 545
 546             /* Load parameters for j particles */
 547             jq0              = _mm_load_sd(charge+jnrA+0);
 548             vdwjidx0A        = 2*vdwtype[jnrA+0];
 549
 550             /**************************
 551              * CALCULATE INTERACTIONS *
 552              **************************/
 553
 554             r00              = _mm_mul_pd(rsq00,rinv00);
 555
 556             /* Compute parameters for interactions between i and j atoms */
 557             qq00             = _mm_mul_pd(iq0,jq0);
 558             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 559
 560             /* EWALD ELECTROSTATICS */
 561
 562             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 563             ewrt             = _mm_mul_pd(r00,ewtabscale);
 564             ewitab           = _mm_cvttpd_epi32(ewrt);
 565             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 566             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 567             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 568             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 569
 570             /* LENNARD-JONES DISPERSION/REPULSION */
 571
 572             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 573             fvdw             = _mm_mul_pd(_mm_sub_pd(_mm_mul_pd(c12_00,rinvsix),c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 574
 575             fscal            = _mm_add_pd(felec,fvdw);
 576
 577             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 578
 579             /* Calculate temporary vectorial force */
 580             tx               = _mm_mul_pd(fscal,dx00);
 581             ty               = _mm_mul_pd(fscal,dy00);
 582             tz               = _mm_mul_pd(fscal,dz00);
 583
 584             /* Update vectorial force */
 585             fix0             = _mm_add_pd(fix0,tx);
 586             fiy0             = _mm_add_pd(fiy0,ty);
 587             fiz0             = _mm_add_pd(fiz0,tz);
 588
 589             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 590
 591             /* Inner loop uses 43 flops */
 592         }
 593
 594         /* End of innermost loop */
 595
 596         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 597                                               f+i_coord_offset,fshift+i_shift_offset);
 598
 599         /* Increment number of inner iterations */
 600         inneriter                  += j_index_end - j_index_start;
 601
 602         /* Outer loop uses 7 flops */
 603     }
 604
 605     /* Increment number of outer iterations */
 606     outeriter        += nri;
 607
 608     /* Update outer/inner flops */
 609
 610     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*43);
 611 }