src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_VdwLJ_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_ElecEw_VdwLJ_GeomP1P1_VF_avx_128_fma_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_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     __m128i          ewitab;
  80     __m128d          ewtabscale,eweps,twoeweps,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 #ifdef __XOP__
 193             eweps            = _mm_frcz_pd(ewrt);
 194 #else
 195             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 196 #endif
 197             twoeweps         = _mm_add_pd(eweps,eweps);
 198             ewitab           = _mm_slli_epi32(ewitab,2);
 199             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 200             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 201             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 202             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 203             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 204             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 205             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 206             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 207             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 208             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 209
 210             /* LENNARD-JONES DISPERSION/REPULSION */
 211
 212             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 213             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 214             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 215             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 216             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 217
 218             /* Update potential sum for this i atom from the interaction with this j atom. */
 219             velecsum         = _mm_add_pd(velecsum,velec);
 220             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 221
 222             fscal            = _mm_add_pd(felec,fvdw);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 226             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 227             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 228
 229             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 230                                                    _mm_mul_pd(dx00,fscal),
 231                                                    _mm_mul_pd(dy00,fscal),
 232                                                    _mm_mul_pd(dz00,fscal));
 233
 234             /* Inner loop uses 56 flops */
 235         }
 236
 237         if(jidx<j_index_end)
 238         {
 239
 240             jnrA             = jjnr[jidx];
 241             j_coord_offsetA  = DIM*jnrA;
 242
 243             /* load j atom coordinates */
 244             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 245                                               &jx0,&jy0,&jz0);
 246
 247             /* Calculate displacement vector */
 248             dx00             = _mm_sub_pd(ix0,jx0);
 249             dy00             = _mm_sub_pd(iy0,jy0);
 250             dz00             = _mm_sub_pd(iz0,jz0);
 251
 252             /* Calculate squared distance and things based on it */
 253             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 254
 255             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 256
 257             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 258
 259             /* Load parameters for j particles */
 260             jq0              = _mm_load_sd(charge+jnrA+0);
 261             vdwjidx0A        = 2*vdwtype[jnrA+0];
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             r00              = _mm_mul_pd(rsq00,rinv00);
 268
 269             /* Compute parameters for interactions between i and j atoms */
 270             qq00             = _mm_mul_pd(iq0,jq0);
 271             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 272
 273             /* EWALD ELECTROSTATICS */
 274
 275             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 276             ewrt             = _mm_mul_pd(r00,ewtabscale);
 277             ewitab           = _mm_cvttpd_epi32(ewrt);
 278 #ifdef __XOP__
 279             eweps            = _mm_frcz_pd(ewrt);
 280 #else
 281             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 282 #endif
 283             twoeweps         = _mm_add_pd(eweps,eweps);
 284             ewitab           = _mm_slli_epi32(ewitab,2);
 285             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 286             ewtabD           = _mm_setzero_pd();
 287             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 288             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 289             ewtabFn          = _mm_setzero_pd();
 290             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 291             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 292             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 293             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 294             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 295
 296             /* LENNARD-JONES DISPERSION/REPULSION */
 297
 298             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 299             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 300             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 301             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 302             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 303
 304             /* Update potential sum for this i atom from the interaction with this j atom. */
 305             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 306             velecsum         = _mm_add_pd(velecsum,velec);
 307             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 308             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 309
 310             fscal            = _mm_add_pd(felec,fvdw);
 311
 312             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 313
 314             /* Update vectorial force */
 315             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 316             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 317             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 318
 319             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 320                                                    _mm_mul_pd(dx00,fscal),
 321                                                    _mm_mul_pd(dy00,fscal),
 322                                                    _mm_mul_pd(dz00,fscal));
 323
 324             /* Inner loop uses 56 flops */
 325         }
 326
 327         /* End of innermost loop */
 328
 329         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 330                                               f+i_coord_offset,fshift+i_shift_offset);
 331
 332         ggid                        = gid[iidx];
 333         /* Update potential energies */
 334         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 335         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 336
 337         /* Increment number of inner iterations */
 338         inneriter                  += j_index_end - j_index_start;
 339
 340         /* Outer loop uses 9 flops */
 341     }
 342
 343     /* Increment number of outer iterations */
 344     outeriter        += nri;
 345
 346     /* Update outer/inner flops */
 347
 348     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*56);
 349 }
 350 /*
 351  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 352  * Electrostatics interaction: Ewald
 353  * VdW interaction:            LennardJones
 354  * Geometry:                   Particle-Particle
 355  * Calculate force/pot:        Force
 356  */
 357 void
 358 nb_kernel_ElecEw_VdwLJ_GeomP1P1_F_avx_128_fma_double
 359                     (t_nblist * gmx_restrict                nlist,
 360                      rvec * gmx_restrict                    xx,
 361                      rvec * gmx_restrict                    ff,
 362                      t_forcerec * gmx_restrict              fr,
 363                      t_mdatoms * gmx_restrict               mdatoms,
 364                      nb_kernel_data_t * gmx_restrict        kernel_data,
 365                      t_nrnb * gmx_restrict                  nrnb)
 366 {
 367     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 368      * just 0 for non-waters.
 369      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 370      * jnr indices corresponding to data put in the four positions in the SIMD register.
 371      */
 372     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 373     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 374     int              jnrA,jnrB;
 375     int              j_coord_offsetA,j_coord_offsetB;
 376     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 377     real             rcutoff_scalar;
 378     real             *shiftvec,*fshift,*x,*f;
 379     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 380     int              vdwioffset0;
 381     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 382     int              vdwjidx0A,vdwjidx0B;
 383     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 384     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 385     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 386     real             *charge;
 387     int              nvdwtype;
 388     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 389     int              *vdwtype;
 390     real             *vdwparam;
 391     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 392     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 393     __m128i          ewitab;
 394     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 395     real             *ewtab;
 396     __m128d          dummy_mask,cutoff_mask;
 397     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 398     __m128d          one     = _mm_set1_pd(1.0);
 399     __m128d          two     = _mm_set1_pd(2.0);
 400     x                = xx[0];
 401     f                = ff[0];
 402
 403     nri              = nlist->nri;
 404     iinr             = nlist->iinr;
 405     jindex           = nlist->jindex;
 406     jjnr             = nlist->jjnr;
 407     shiftidx         = nlist->shift;
 408     gid              = nlist->gid;
 409     shiftvec         = fr->shift_vec[0];
 410     fshift           = fr->fshift[0];
 411     facel            = _mm_set1_pd(fr->epsfac);
 412     charge           = mdatoms->chargeA;
 413     nvdwtype         = fr->ntype;
 414     vdwparam         = fr->nbfp;
 415     vdwtype          = mdatoms->typeA;
 416
 417     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 418     ewtab            = fr->ic->tabq_coul_F;
 419     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 420     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 421
 422     /* Avoid stupid compiler warnings */
 423     jnrA = jnrB = 0;
 424     j_coord_offsetA = 0;
 425     j_coord_offsetB = 0;
 426
 427     outeriter        = 0;
 428     inneriter        = 0;
 429
 430     /* Start outer loop over neighborlists */
 431     for(iidx=0; iidx<nri; iidx++)
 432     {
 433         /* Load shift vector for this list */
 434         i_shift_offset   = DIM*shiftidx[iidx];
 435
 436         /* Load limits for loop over neighbors */
 437         j_index_start    = jindex[iidx];
 438         j_index_end      = jindex[iidx+1];
 439
 440         /* Get outer coordinate index */
 441         inr              = iinr[iidx];
 442         i_coord_offset   = DIM*inr;
 443
 444         /* Load i particle coords and add shift vector */
 445         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 446
 447         fix0             = _mm_setzero_pd();
 448         fiy0             = _mm_setzero_pd();
 449         fiz0             = _mm_setzero_pd();
 450
 451         /* Load parameters for i particles */
 452         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 453         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 454
 455         /* Start inner kernel loop */
 456         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 457         {
 458
 459             /* Get j neighbor index, and coordinate index */
 460             jnrA             = jjnr[jidx];
 461             jnrB             = jjnr[jidx+1];
 462             j_coord_offsetA  = DIM*jnrA;
 463             j_coord_offsetB  = DIM*jnrB;
 464
 465             /* load j atom coordinates */
 466             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 467                                               &jx0,&jy0,&jz0);
 468
 469             /* Calculate displacement vector */
 470             dx00             = _mm_sub_pd(ix0,jx0);
 471             dy00             = _mm_sub_pd(iy0,jy0);
 472             dz00             = _mm_sub_pd(iz0,jz0);
 473
 474             /* Calculate squared distance and things based on it */
 475             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 476
 477             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 478
 479             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 480
 481             /* Load parameters for j particles */
 482             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 483             vdwjidx0A        = 2*vdwtype[jnrA+0];
 484             vdwjidx0B        = 2*vdwtype[jnrB+0];
 485
 486             /**************************
 487              * CALCULATE INTERACTIONS *
 488              **************************/
 489
 490             r00              = _mm_mul_pd(rsq00,rinv00);
 491
 492             /* Compute parameters for interactions between i and j atoms */
 493             qq00             = _mm_mul_pd(iq0,jq0);
 494             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 495                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 496
 497             /* EWALD ELECTROSTATICS */
 498
 499             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 500             ewrt             = _mm_mul_pd(r00,ewtabscale);
 501             ewitab           = _mm_cvttpd_epi32(ewrt);
 502 #ifdef __XOP__
 503             eweps            = _mm_frcz_pd(ewrt);
 504 #else
 505             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 506 #endif
 507             twoeweps         = _mm_add_pd(eweps,eweps);
 508             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 509                                          &ewtabF,&ewtabFn);
 510             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 511             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 512
 513             /* LENNARD-JONES DISPERSION/REPULSION */
 514
 515             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 516             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 517
 518             fscal            = _mm_add_pd(felec,fvdw);
 519
 520             /* Update vectorial force */
 521             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 522             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 523             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 524
 525             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 526                                                    _mm_mul_pd(dx00,fscal),
 527                                                    _mm_mul_pd(dy00,fscal),
 528                                                    _mm_mul_pd(dz00,fscal));
 529
 530             /* Inner loop uses 46 flops */
 531         }
 532
 533         if(jidx<j_index_end)
 534         {
 535
 536             jnrA             = jjnr[jidx];
 537             j_coord_offsetA  = DIM*jnrA;
 538
 539             /* load j atom coordinates */
 540             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 541                                               &jx0,&jy0,&jz0);
 542
 543             /* Calculate displacement vector */
 544             dx00             = _mm_sub_pd(ix0,jx0);
 545             dy00             = _mm_sub_pd(iy0,jy0);
 546             dz00             = _mm_sub_pd(iz0,jz0);
 547
 548             /* Calculate squared distance and things based on it */
 549             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 550
 551             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 552
 553             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 554
 555             /* Load parameters for j particles */
 556             jq0              = _mm_load_sd(charge+jnrA+0);
 557             vdwjidx0A        = 2*vdwtype[jnrA+0];
 558
 559             /**************************
 560              * CALCULATE INTERACTIONS *
 561              **************************/
 562
 563             r00              = _mm_mul_pd(rsq00,rinv00);
 564
 565             /* Compute parameters for interactions between i and j atoms */
 566             qq00             = _mm_mul_pd(iq0,jq0);
 567             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_pd(r00,ewtabscale);
 573             ewitab           = _mm_cvttpd_epi32(ewrt);
 574 #ifdef __XOP__
 575             eweps            = _mm_frcz_pd(ewrt);
 576 #else
 577             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 578 #endif
 579             twoeweps         = _mm_add_pd(eweps,eweps);
 580             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 581             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 582             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 583
 584             /* LENNARD-JONES DISPERSION/REPULSION */
 585
 586             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 587             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 588
 589             fscal            = _mm_add_pd(felec,fvdw);
 590
 591             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 592
 593             /* Update vectorial force */
 594             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 595             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 596             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 597
 598             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 599                                                    _mm_mul_pd(dx00,fscal),
 600                                                    _mm_mul_pd(dy00,fscal),
 601                                                    _mm_mul_pd(dz00,fscal));
 602
 603             /* Inner loop uses 46 flops */
 604         }
 605
 606         /* End of innermost loop */
 607
 608         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 609                                               f+i_coord_offset,fshift+i_shift_offset);
 610
 611         /* Increment number of inner iterations */
 612         inneriter                  += j_index_end - j_index_start;
 613
 614         /* Outer loop uses 7 flops */
 615     }
 616
 617     /* Increment number of outer iterations */
 618     outeriter        += nri;
 619
 620     /* Update outer/inner flops */
 621
 622     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*46);
 623 }