src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCSTab_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_ElecCSTab_VdwLJ_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_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          vfitab;
  80     __m128i          ifour       = _mm_set1_epi32(4);
  81     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
  82     real             *vftab;
  83     __m128d          dummy_mask,cutoff_mask;
  84     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  85     __m128d          one     = _mm_set1_pd(1.0);
  86     __m128d          two     = _mm_set1_pd(2.0);
  87     x                = xx[0];
  88     f                = ff[0];
  89
  90     nri              = nlist->nri;
  91     iinr             = nlist->iinr;
  92     jindex           = nlist->jindex;
  93     jjnr             = nlist->jjnr;
  94     shiftidx         = nlist->shift;
  95     gid              = nlist->gid;
  96     shiftvec         = fr->shift_vec[0];
  97     fshift           = fr->fshift[0];
  98     facel            = _mm_set1_pd(fr->epsfac);
  99     charge           = mdatoms->chargeA;
 100     nvdwtype         = fr->ntype;
 101     vdwparam         = fr->nbfp;
 102     vdwtype          = mdatoms->typeA;
 103
 104     vftab            = kernel_data->table_elec->data;
 105     vftabscale       = _mm_set1_pd(kernel_data->table_elec->scale);
 106
 107     /* Avoid stupid compiler warnings */
 108     jnrA = jnrB = 0;
 109     j_coord_offsetA = 0;
 110     j_coord_offsetB = 0;
 111
 112     outeriter        = 0;
 113     inneriter        = 0;
 114
 115     /* Start outer loop over neighborlists */
 116     for(iidx=0; iidx<nri; iidx++)
 117     {
 118         /* Load shift vector for this list */
 119         i_shift_offset   = DIM*shiftidx[iidx];
 120
 121         /* Load limits for loop over neighbors */
 122         j_index_start    = jindex[iidx];
 123         j_index_end      = jindex[iidx+1];
 124
 125         /* Get outer coordinate index */
 126         inr              = iinr[iidx];
 127         i_coord_offset   = DIM*inr;
 128
 129         /* Load i particle coords and add shift vector */
 130         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 131
 132         fix0             = _mm_setzero_pd();
 133         fiy0             = _mm_setzero_pd();
 134         fiz0             = _mm_setzero_pd();
 135
 136         /* Load parameters for i particles */
 137         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 138         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 139
 140         /* Reset potential sums */
 141         velecsum         = _mm_setzero_pd();
 142         vvdwsum          = _mm_setzero_pd();
 143
 144         /* Start inner kernel loop */
 145         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 146         {
 147
 148             /* Get j neighbor index, and coordinate index */
 149             jnrA             = jjnr[jidx];
 150             jnrB             = jjnr[jidx+1];
 151             j_coord_offsetA  = DIM*jnrA;
 152             j_coord_offsetB  = DIM*jnrB;
 153
 154             /* load j atom coordinates */
 155             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 156                                               &jx0,&jy0,&jz0);
 157
 158             /* Calculate displacement vector */
 159             dx00             = _mm_sub_pd(ix0,jx0);
 160             dy00             = _mm_sub_pd(iy0,jy0);
 161             dz00             = _mm_sub_pd(iz0,jz0);
 162
 163             /* Calculate squared distance and things based on it */
 164             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 165
 166             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 167
 168             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 169
 170             /* Load parameters for j particles */
 171             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 172             vdwjidx0A        = 2*vdwtype[jnrA+0];
 173             vdwjidx0B        = 2*vdwtype[jnrB+0];
 174
 175             /**************************
 176              * CALCULATE INTERACTIONS *
 177              **************************/
 178
 179             r00              = _mm_mul_pd(rsq00,rinv00);
 180
 181             /* Compute parameters for interactions between i and j atoms */
 182             qq00             = _mm_mul_pd(iq0,jq0);
 183             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 184                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 185
 186             /* Calculate table index by multiplying r with table scale and truncate to integer */
 187             rt               = _mm_mul_pd(r00,vftabscale);
 188             vfitab           = _mm_cvttpd_epi32(rt);
 189 #ifdef __XOP__
 190             vfeps            = _mm_frcz_pd(rt);
 191 #else
 192             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 193 #endif
 194             twovfeps         = _mm_add_pd(vfeps,vfeps);
 195             vfitab           = _mm_slli_epi32(vfitab,2);
 196
 197             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 198             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 199             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 200             GMX_MM_TRANSPOSE2_PD(Y,F);
 201             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 202             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 203             GMX_MM_TRANSPOSE2_PD(G,H);
 204             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 205             VV               = _mm_macc_pd(vfeps,Fp,Y);
 206             velec            = _mm_mul_pd(qq00,VV);
 207             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 208             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 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 59 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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 274             rt               = _mm_mul_pd(r00,vftabscale);
 275             vfitab           = _mm_cvttpd_epi32(rt);
 276 #ifdef __XOP__
 277             vfeps            = _mm_frcz_pd(rt);
 278 #else
 279             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 280 #endif
 281             twovfeps         = _mm_add_pd(vfeps,vfeps);
 282             vfitab           = _mm_slli_epi32(vfitab,2);
 283
 284             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 285             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 286             F                = _mm_setzero_pd();
 287             GMX_MM_TRANSPOSE2_PD(Y,F);
 288             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 289             H                = _mm_setzero_pd();
 290             GMX_MM_TRANSPOSE2_PD(G,H);
 291             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 292             VV               = _mm_macc_pd(vfeps,Fp,Y);
 293             velec            = _mm_mul_pd(qq00,VV);
 294             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 295             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 296
 297             /* LENNARD-JONES DISPERSION/REPULSION */
 298
 299             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 300             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 301             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 302             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 303             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 304
 305             /* Update potential sum for this i atom from the interaction with this j atom. */
 306             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 307             velecsum         = _mm_add_pd(velecsum,velec);
 308             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 309             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 310
 311             fscal            = _mm_add_pd(felec,fvdw);
 312
 313             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 314
 315             /* Update vectorial force */
 316             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 317             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 318             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 319
 320             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 321                                                    _mm_mul_pd(dx00,fscal),
 322                                                    _mm_mul_pd(dy00,fscal),
 323                                                    _mm_mul_pd(dz00,fscal));
 324
 325             /* Inner loop uses 59 flops */
 326         }
 327
 328         /* End of innermost loop */
 329
 330         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 331                                               f+i_coord_offset,fshift+i_shift_offset);
 332
 333         ggid                        = gid[iidx];
 334         /* Update potential energies */
 335         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 336         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 337
 338         /* Increment number of inner iterations */
 339         inneriter                  += j_index_end - j_index_start;
 340
 341         /* Outer loop uses 9 flops */
 342     }
 343
 344     /* Increment number of outer iterations */
 345     outeriter        += nri;
 346
 347     /* Update outer/inner flops */
 348
 349     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*59);
 350 }
 351 /*
 352  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
 353  * Electrostatics interaction: CubicSplineTable
 354  * VdW interaction:            LennardJones
 355  * Geometry:                   Particle-Particle
 356  * Calculate force/pot:        Force
 357  */
 358 void
 359 nb_kernel_ElecCSTab_VdwLJ_GeomP1P1_F_avx_128_fma_double
 360                     (t_nblist * gmx_restrict                nlist,
 361                      rvec * gmx_restrict                    xx,
 362                      rvec * gmx_restrict                    ff,
 363                      t_forcerec * gmx_restrict              fr,
 364                      t_mdatoms * gmx_restrict               mdatoms,
 365                      nb_kernel_data_t * gmx_restrict        kernel_data,
 366                      t_nrnb * gmx_restrict                  nrnb)
 367 {
 368     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 369      * just 0 for non-waters.
 370      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 371      * jnr indices corresponding to data put in the four positions in the SIMD register.
 372      */
 373     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 374     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 375     int              jnrA,jnrB;
 376     int              j_coord_offsetA,j_coord_offsetB;
 377     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 378     real             rcutoff_scalar;
 379     real             *shiftvec,*fshift,*x,*f;
 380     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 381     int              vdwioffset0;
 382     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 383     int              vdwjidx0A,vdwjidx0B;
 384     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 385     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 386     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 387     real             *charge;
 388     int              nvdwtype;
 389     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 390     int              *vdwtype;
 391     real             *vdwparam;
 392     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 393     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 394     __m128i          vfitab;
 395     __m128i          ifour       = _mm_set1_epi32(4);
 396     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 397     real             *vftab;
 398     __m128d          dummy_mask,cutoff_mask;
 399     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 400     __m128d          one     = _mm_set1_pd(1.0);
 401     __m128d          two     = _mm_set1_pd(2.0);
 402     x                = xx[0];
 403     f                = ff[0];
 404
 405     nri              = nlist->nri;
 406     iinr             = nlist->iinr;
 407     jindex           = nlist->jindex;
 408     jjnr             = nlist->jjnr;
 409     shiftidx         = nlist->shift;
 410     gid              = nlist->gid;
 411     shiftvec         = fr->shift_vec[0];
 412     fshift           = fr->fshift[0];
 413     facel            = _mm_set1_pd(fr->epsfac);
 414     charge           = mdatoms->chargeA;
 415     nvdwtype         = fr->ntype;
 416     vdwparam         = fr->nbfp;
 417     vdwtype          = mdatoms->typeA;
 418
 419     vftab            = kernel_data->table_elec->data;
 420     vftabscale       = _mm_set1_pd(kernel_data->table_elec->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             /* Calculate table index by multiplying r with table scale and truncate to integer */
 498             rt               = _mm_mul_pd(r00,vftabscale);
 499             vfitab           = _mm_cvttpd_epi32(rt);
 500 #ifdef __XOP__
 501             vfeps            = _mm_frcz_pd(rt);
 502 #else
 503             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 504 #endif
 505             twovfeps         = _mm_add_pd(vfeps,vfeps);
 506             vfitab           = _mm_slli_epi32(vfitab,2);
 507
 508             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 509             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 510             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 511             GMX_MM_TRANSPOSE2_PD(Y,F);
 512             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 513             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 514             GMX_MM_TRANSPOSE2_PD(G,H);
 515             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 516             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 517             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 518
 519             /* LENNARD-JONES DISPERSION/REPULSION */
 520
 521             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 522             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 523
 524             fscal            = _mm_add_pd(felec,fvdw);
 525
 526             /* Update vectorial force */
 527             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 528             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 529             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 530
 531             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 532                                                    _mm_mul_pd(dx00,fscal),
 533                                                    _mm_mul_pd(dy00,fscal),
 534                                                    _mm_mul_pd(dz00,fscal));
 535
 536             /* Inner loop uses 50 flops */
 537         }
 538
 539         if(jidx<j_index_end)
 540         {
 541
 542             jnrA             = jjnr[jidx];
 543             j_coord_offsetA  = DIM*jnrA;
 544
 545             /* load j atom coordinates */
 546             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_pd(ix0,jx0);
 551             dy00             = _mm_sub_pd(iy0,jy0);
 552             dz00             = _mm_sub_pd(iz0,jz0);
 553
 554             /* Calculate squared distance and things based on it */
 555             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 556
 557             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 558
 559             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 560
 561             /* Load parameters for j particles */
 562             jq0              = _mm_load_sd(charge+jnrA+0);
 563             vdwjidx0A        = 2*vdwtype[jnrA+0];
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             r00              = _mm_mul_pd(rsq00,rinv00);
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq00             = _mm_mul_pd(iq0,jq0);
 573             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 574
 575             /* Calculate table index by multiplying r with table scale and truncate to integer */
 576             rt               = _mm_mul_pd(r00,vftabscale);
 577             vfitab           = _mm_cvttpd_epi32(rt);
 578 #ifdef __XOP__
 579             vfeps            = _mm_frcz_pd(rt);
 580 #else
 581             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 582 #endif
 583             twovfeps         = _mm_add_pd(vfeps,vfeps);
 584             vfitab           = _mm_slli_epi32(vfitab,2);
 585
 586             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 587             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 588             F                = _mm_setzero_pd();
 589             GMX_MM_TRANSPOSE2_PD(Y,F);
 590             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 591             H                = _mm_setzero_pd();
 592             GMX_MM_TRANSPOSE2_PD(G,H);
 593             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(vfeps,H,G),F);
 594             FF               = _mm_macc_pd(_mm_macc_pd(twovfeps,H,G),vfeps,Fp);
 595             felec            = _mm_xor_pd(signbit,_mm_mul_pd(_mm_mul_pd(qq00,FF),_mm_mul_pd(vftabscale,rinv00)));
 596
 597             /* LENNARD-JONES DISPERSION/REPULSION */
 598
 599             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 600             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 601
 602             fscal            = _mm_add_pd(felec,fvdw);
 603
 604             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 605
 606             /* Update vectorial force */
 607             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 608             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 609             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 610
 611             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 612                                                    _mm_mul_pd(dx00,fscal),
 613                                                    _mm_mul_pd(dy00,fscal),
 614                                                    _mm_mul_pd(dz00,fscal));
 615
 616             /* Inner loop uses 50 flops */
 617         }
 618
 619         /* End of innermost loop */
 620
 621         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 622                                               f+i_coord_offset,fshift+i_shift_offset);
 623
 624         /* Increment number of inner iterations */
 625         inneriter                  += j_index_end - j_index_start;
 626
 627         /* Outer loop uses 7 flops */
 628     }
 629
 630     /* Increment number of outer iterations */
 631     outeriter        += nri;
 632
 633     /* Update outer/inner flops */
 634
 635     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*50);
 636 }