src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwCSTab_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_ElecCoul_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwCSTab_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_vdw->data;
 105     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->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,3);
 196
 197             /* COULOMB ELECTROSTATICS */
 198             velec            = _mm_mul_pd(qq00,rinv00);
 199             felec            = _mm_mul_pd(velec,rinvsq00);
 200
 201             /* CUBIC SPLINE TABLE DISPERSION */
 202             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 203             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 204             GMX_MM_TRANSPOSE2_PD(Y,F);
 205             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 206             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 207             GMX_MM_TRANSPOSE2_PD(G,H);
 208             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 209             VV               = _mm_macc_pd(vfeps,Fp,Y);
 210             vvdw6            = _mm_mul_pd(c6_00,VV);
 211             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 212             fvdw6            = _mm_mul_pd(c6_00,FF);
 213
 214             /* CUBIC SPLINE TABLE REPULSION */
 215             vfitab           = _mm_add_epi32(vfitab,ifour);
 216             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 217             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 218             GMX_MM_TRANSPOSE2_PD(Y,F);
 219             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 220             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 221             GMX_MM_TRANSPOSE2_PD(G,H);
 222             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 223             VV               = _mm_macc_pd(vfeps,Fp,Y);
 224             vvdw12           = _mm_mul_pd(c12_00,VV);
 225             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 226             fvdw12           = _mm_mul_pd(c12_00,FF);
 227             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 228             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 229
 230             /* Update potential sum for this i atom from the interaction with this j atom. */
 231             velecsum         = _mm_add_pd(velecsum,velec);
 232             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 233
 234             fscal            = _mm_add_pd(felec,fvdw);
 235
 236             /* Update vectorial force */
 237             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 238             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 239             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 240
 241             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 242                                                    _mm_mul_pd(dx00,fscal),
 243                                                    _mm_mul_pd(dy00,fscal),
 244                                                    _mm_mul_pd(dz00,fscal));
 245
 246             /* Inner loop uses 66 flops */
 247         }
 248
 249         if(jidx<j_index_end)
 250         {
 251
 252             jnrA             = jjnr[jidx];
 253             j_coord_offsetA  = DIM*jnrA;
 254
 255             /* load j atom coordinates */
 256             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 257                                               &jx0,&jy0,&jz0);
 258
 259             /* Calculate displacement vector */
 260             dx00             = _mm_sub_pd(ix0,jx0);
 261             dy00             = _mm_sub_pd(iy0,jy0);
 262             dz00             = _mm_sub_pd(iz0,jz0);
 263
 264             /* Calculate squared distance and things based on it */
 265             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 266
 267             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 268
 269             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 270
 271             /* Load parameters for j particles */
 272             jq0              = _mm_load_sd(charge+jnrA+0);
 273             vdwjidx0A        = 2*vdwtype[jnrA+0];
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             r00              = _mm_mul_pd(rsq00,rinv00);
 280
 281             /* Compute parameters for interactions between i and j atoms */
 282             qq00             = _mm_mul_pd(iq0,jq0);
 283             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 284
 285             /* Calculate table index by multiplying r with table scale and truncate to integer */
 286             rt               = _mm_mul_pd(r00,vftabscale);
 287             vfitab           = _mm_cvttpd_epi32(rt);
 288 #ifdef __XOP__
 289             vfeps            = _mm_frcz_pd(rt);
 290 #else
 291             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 292 #endif
 293             twovfeps         = _mm_add_pd(vfeps,vfeps);
 294             vfitab           = _mm_slli_epi32(vfitab,3);
 295
 296             /* COULOMB ELECTROSTATICS */
 297             velec            = _mm_mul_pd(qq00,rinv00);
 298             felec            = _mm_mul_pd(velec,rinvsq00);
 299
 300             /* CUBIC SPLINE TABLE DISPERSION */
 301             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 302             F                = _mm_setzero_pd();
 303             GMX_MM_TRANSPOSE2_PD(Y,F);
 304             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 305             H                = _mm_setzero_pd();
 306             GMX_MM_TRANSPOSE2_PD(G,H);
 307             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 308             VV               = _mm_macc_pd(vfeps,Fp,Y);
 309             vvdw6            = _mm_mul_pd(c6_00,VV);
 310             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 311             fvdw6            = _mm_mul_pd(c6_00,FF);
 312
 313             /* CUBIC SPLINE TABLE REPULSION */
 314             vfitab           = _mm_add_epi32(vfitab,ifour);
 315             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 316             F                = _mm_setzero_pd();
 317             GMX_MM_TRANSPOSE2_PD(Y,F);
 318             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 319             H                = _mm_setzero_pd();
 320             GMX_MM_TRANSPOSE2_PD(G,H);
 321             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 322             VV               = _mm_macc_pd(vfeps,Fp,Y);
 323             vvdw12           = _mm_mul_pd(c12_00,VV);
 324             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 325             fvdw12           = _mm_mul_pd(c12_00,FF);
 326             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 327             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 328
 329             /* Update potential sum for this i atom from the interaction with this j atom. */
 330             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 331             velecsum         = _mm_add_pd(velecsum,velec);
 332             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 333             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 334
 335             fscal            = _mm_add_pd(felec,fvdw);
 336
 337             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 338
 339             /* Update vectorial force */
 340             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 341             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 342             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 343
 344             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 345                                                    _mm_mul_pd(dx00,fscal),
 346                                                    _mm_mul_pd(dy00,fscal),
 347                                                    _mm_mul_pd(dz00,fscal));
 348
 349             /* Inner loop uses 66 flops */
 350         }
 351
 352         /* End of innermost loop */
 353
 354         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 355                                               f+i_coord_offset,fshift+i_shift_offset);
 356
 357         ggid                        = gid[iidx];
 358         /* Update potential energies */
 359         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 360         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 361
 362         /* Increment number of inner iterations */
 363         inneriter                  += j_index_end - j_index_start;
 364
 365         /* Outer loop uses 9 flops */
 366     }
 367
 368     /* Increment number of outer iterations */
 369     outeriter        += nri;
 370
 371     /* Update outer/inner flops */
 372
 373     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*66);
 374 }
 375 /*
 376  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 377  * Electrostatics interaction: Coulomb
 378  * VdW interaction:            CubicSplineTable
 379  * Geometry:                   Particle-Particle
 380  * Calculate force/pot:        Force
 381  */
 382 void
 383 nb_kernel_ElecCoul_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 384                     (t_nblist * gmx_restrict                nlist,
 385                      rvec * gmx_restrict                    xx,
 386                      rvec * gmx_restrict                    ff,
 387                      t_forcerec * gmx_restrict              fr,
 388                      t_mdatoms * gmx_restrict               mdatoms,
 389                      nb_kernel_data_t * gmx_restrict        kernel_data,
 390                      t_nrnb * gmx_restrict                  nrnb)
 391 {
 392     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 393      * just 0 for non-waters.
 394      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 395      * jnr indices corresponding to data put in the four positions in the SIMD register.
 396      */
 397     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 398     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 399     int              jnrA,jnrB;
 400     int              j_coord_offsetA,j_coord_offsetB;
 401     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 402     real             rcutoff_scalar;
 403     real             *shiftvec,*fshift,*x,*f;
 404     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 405     int              vdwioffset0;
 406     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 407     int              vdwjidx0A,vdwjidx0B;
 408     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 409     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 410     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 411     real             *charge;
 412     int              nvdwtype;
 413     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 414     int              *vdwtype;
 415     real             *vdwparam;
 416     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 417     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 418     __m128i          vfitab;
 419     __m128i          ifour       = _mm_set1_epi32(4);
 420     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 421     real             *vftab;
 422     __m128d          dummy_mask,cutoff_mask;
 423     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 424     __m128d          one     = _mm_set1_pd(1.0);
 425     __m128d          two     = _mm_set1_pd(2.0);
 426     x                = xx[0];
 427     f                = ff[0];
 428
 429     nri              = nlist->nri;
 430     iinr             = nlist->iinr;
 431     jindex           = nlist->jindex;
 432     jjnr             = nlist->jjnr;
 433     shiftidx         = nlist->shift;
 434     gid              = nlist->gid;
 435     shiftvec         = fr->shift_vec[0];
 436     fshift           = fr->fshift[0];
 437     facel            = _mm_set1_pd(fr->epsfac);
 438     charge           = mdatoms->chargeA;
 439     nvdwtype         = fr->ntype;
 440     vdwparam         = fr->nbfp;
 441     vdwtype          = mdatoms->typeA;
 442
 443     vftab            = kernel_data->table_vdw->data;
 444     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 445
 446     /* Avoid stupid compiler warnings */
 447     jnrA = jnrB = 0;
 448     j_coord_offsetA = 0;
 449     j_coord_offsetB = 0;
 450
 451     outeriter        = 0;
 452     inneriter        = 0;
 453
 454     /* Start outer loop over neighborlists */
 455     for(iidx=0; iidx<nri; iidx++)
 456     {
 457         /* Load shift vector for this list */
 458         i_shift_offset   = DIM*shiftidx[iidx];
 459
 460         /* Load limits for loop over neighbors */
 461         j_index_start    = jindex[iidx];
 462         j_index_end      = jindex[iidx+1];
 463
 464         /* Get outer coordinate index */
 465         inr              = iinr[iidx];
 466         i_coord_offset   = DIM*inr;
 467
 468         /* Load i particle coords and add shift vector */
 469         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 470
 471         fix0             = _mm_setzero_pd();
 472         fiy0             = _mm_setzero_pd();
 473         fiz0             = _mm_setzero_pd();
 474
 475         /* Load parameters for i particles */
 476         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 477         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 478
 479         /* Start inner kernel loop */
 480         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 481         {
 482
 483             /* Get j neighbor index, and coordinate index */
 484             jnrA             = jjnr[jidx];
 485             jnrB             = jjnr[jidx+1];
 486             j_coord_offsetA  = DIM*jnrA;
 487             j_coord_offsetB  = DIM*jnrB;
 488
 489             /* load j atom coordinates */
 490             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 491                                               &jx0,&jy0,&jz0);
 492
 493             /* Calculate displacement vector */
 494             dx00             = _mm_sub_pd(ix0,jx0);
 495             dy00             = _mm_sub_pd(iy0,jy0);
 496             dz00             = _mm_sub_pd(iz0,jz0);
 497
 498             /* Calculate squared distance and things based on it */
 499             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 500
 501             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 502
 503             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 504
 505             /* Load parameters for j particles */
 506             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 507             vdwjidx0A        = 2*vdwtype[jnrA+0];
 508             vdwjidx0B        = 2*vdwtype[jnrB+0];
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             r00              = _mm_mul_pd(rsq00,rinv00);
 515
 516             /* Compute parameters for interactions between i and j atoms */
 517             qq00             = _mm_mul_pd(iq0,jq0);
 518             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 519                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 520
 521             /* Calculate table index by multiplying r with table scale and truncate to integer */
 522             rt               = _mm_mul_pd(r00,vftabscale);
 523             vfitab           = _mm_cvttpd_epi32(rt);
 524 #ifdef __XOP__
 525             vfeps            = _mm_frcz_pd(rt);
 526 #else
 527             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 528 #endif
 529             twovfeps         = _mm_add_pd(vfeps,vfeps);
 530             vfitab           = _mm_slli_epi32(vfitab,3);
 531
 532             /* COULOMB ELECTROSTATICS */
 533             velec            = _mm_mul_pd(qq00,rinv00);
 534             felec            = _mm_mul_pd(velec,rinvsq00);
 535
 536             /* CUBIC SPLINE TABLE DISPERSION */
 537             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 539             GMX_MM_TRANSPOSE2_PD(Y,F);
 540             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 541             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 542             GMX_MM_TRANSPOSE2_PD(G,H);
 543             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 544             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 545             fvdw6            = _mm_mul_pd(c6_00,FF);
 546
 547             /* CUBIC SPLINE TABLE REPULSION */
 548             vfitab           = _mm_add_epi32(vfitab,ifour);
 549             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 550             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 551             GMX_MM_TRANSPOSE2_PD(Y,F);
 552             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 553             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 554             GMX_MM_TRANSPOSE2_PD(G,H);
 555             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 556             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 557             fvdw12           = _mm_mul_pd(c12_00,FF);
 558             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 559
 560             fscal            = _mm_add_pd(felec,fvdw);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 564             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 565             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 566
 567             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 568                                                    _mm_mul_pd(dx00,fscal),
 569                                                    _mm_mul_pd(dy00,fscal),
 570                                                    _mm_mul_pd(dz00,fscal));
 571
 572             /* Inner loop uses 57 flops */
 573         }
 574
 575         if(jidx<j_index_end)
 576         {
 577
 578             jnrA             = jjnr[jidx];
 579             j_coord_offsetA  = DIM*jnrA;
 580
 581             /* load j atom coordinates */
 582             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 583                                               &jx0,&jy0,&jz0);
 584
 585             /* Calculate displacement vector */
 586             dx00             = _mm_sub_pd(ix0,jx0);
 587             dy00             = _mm_sub_pd(iy0,jy0);
 588             dz00             = _mm_sub_pd(iz0,jz0);
 589
 590             /* Calculate squared distance and things based on it */
 591             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 592
 593             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 594
 595             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 596
 597             /* Load parameters for j particles */
 598             jq0              = _mm_load_sd(charge+jnrA+0);
 599             vdwjidx0A        = 2*vdwtype[jnrA+0];
 600
 601             /**************************
 602              * CALCULATE INTERACTIONS *
 603              **************************/
 604
 605             r00              = _mm_mul_pd(rsq00,rinv00);
 606
 607             /* Compute parameters for interactions between i and j atoms */
 608             qq00             = _mm_mul_pd(iq0,jq0);
 609             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 610
 611             /* Calculate table index by multiplying r with table scale and truncate to integer */
 612             rt               = _mm_mul_pd(r00,vftabscale);
 613             vfitab           = _mm_cvttpd_epi32(rt);
 614 #ifdef __XOP__
 615             vfeps            = _mm_frcz_pd(rt);
 616 #else
 617             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 618 #endif
 619             twovfeps         = _mm_add_pd(vfeps,vfeps);
 620             vfitab           = _mm_slli_epi32(vfitab,3);
 621
 622             /* COULOMB ELECTROSTATICS */
 623             velec            = _mm_mul_pd(qq00,rinv00);
 624             felec            = _mm_mul_pd(velec,rinvsq00);
 625
 626             /* CUBIC SPLINE TABLE DISPERSION */
 627             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 628             F                = _mm_setzero_pd();
 629             GMX_MM_TRANSPOSE2_PD(Y,F);
 630             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 631             H                = _mm_setzero_pd();
 632             GMX_MM_TRANSPOSE2_PD(G,H);
 633             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 634             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 635             fvdw6            = _mm_mul_pd(c6_00,FF);
 636
 637             /* CUBIC SPLINE TABLE REPULSION */
 638             vfitab           = _mm_add_epi32(vfitab,ifour);
 639             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 640             F                = _mm_setzero_pd();
 641             GMX_MM_TRANSPOSE2_PD(Y,F);
 642             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 643             H                = _mm_setzero_pd();
 644             GMX_MM_TRANSPOSE2_PD(G,H);
 645             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 646             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 647             fvdw12           = _mm_mul_pd(c12_00,FF);
 648             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 649
 650             fscal            = _mm_add_pd(felec,fvdw);
 651
 652             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 653
 654             /* Update vectorial force */
 655             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 656             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 657             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 658
 659             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 660                                                    _mm_mul_pd(dx00,fscal),
 661                                                    _mm_mul_pd(dy00,fscal),
 662                                                    _mm_mul_pd(dz00,fscal));
 663
 664             /* Inner loop uses 57 flops */
 665         }
 666
 667         /* End of innermost loop */
 668
 669         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 670                                               f+i_coord_offset,fshift+i_shift_offset);
 671
 672         /* Increment number of inner iterations */
 673         inneriter                  += j_index_end - j_index_start;
 674
 675         /* Outer loop uses 7 flops */
 676     }
 677
 678     /* Increment number of outer iterations */
 679     outeriter        += nri;
 680
 681     /* Update outer/inner flops */
 682
 683     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 684 }