src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecRF_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_ElecRF_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     krf              = _mm_set1_pd(fr->ic->k_rf);
 101     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 102     crf              = _mm_set1_pd(fr->ic->c_rf);
 103     nvdwtype         = fr->ntype;
 104     vdwparam         = fr->nbfp;
 105     vdwtype          = mdatoms->typeA;
 106
 107     vftab            = kernel_data->table_vdw->data;
 108     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 109
 110     /* Avoid stupid compiler warnings */
 111     jnrA = jnrB = 0;
 112     j_coord_offsetA = 0;
 113     j_coord_offsetB = 0;
 114
 115     outeriter        = 0;
 116     inneriter        = 0;
 117
 118     /* Start outer loop over neighborlists */
 119     for(iidx=0; iidx<nri; iidx++)
 120     {
 121         /* Load shift vector for this list */
 122         i_shift_offset   = DIM*shiftidx[iidx];
 123
 124         /* Load limits for loop over neighbors */
 125         j_index_start    = jindex[iidx];
 126         j_index_end      = jindex[iidx+1];
 127
 128         /* Get outer coordinate index */
 129         inr              = iinr[iidx];
 130         i_coord_offset   = DIM*inr;
 131
 132         /* Load i particle coords and add shift vector */
 133         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 134
 135         fix0             = _mm_setzero_pd();
 136         fiy0             = _mm_setzero_pd();
 137         fiz0             = _mm_setzero_pd();
 138
 139         /* Load parameters for i particles */
 140         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 141         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 142
 143         /* Reset potential sums */
 144         velecsum         = _mm_setzero_pd();
 145         vvdwsum          = _mm_setzero_pd();
 146
 147         /* Start inner kernel loop */
 148         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 149         {
 150
 151             /* Get j neighbor index, and coordinate index */
 152             jnrA             = jjnr[jidx];
 153             jnrB             = jjnr[jidx+1];
 154             j_coord_offsetA  = DIM*jnrA;
 155             j_coord_offsetB  = DIM*jnrB;
 156
 157             /* load j atom coordinates */
 158             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 159                                               &jx0,&jy0,&jz0);
 160
 161             /* Calculate displacement vector */
 162             dx00             = _mm_sub_pd(ix0,jx0);
 163             dy00             = _mm_sub_pd(iy0,jy0);
 164             dz00             = _mm_sub_pd(iz0,jz0);
 165
 166             /* Calculate squared distance and things based on it */
 167             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 168
 169             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 170
 171             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 172
 173             /* Load parameters for j particles */
 174             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 175             vdwjidx0A        = 2*vdwtype[jnrA+0];
 176             vdwjidx0B        = 2*vdwtype[jnrB+0];
 177
 178             /**************************
 179              * CALCULATE INTERACTIONS *
 180              **************************/
 181
 182             r00              = _mm_mul_pd(rsq00,rinv00);
 183
 184             /* Compute parameters for interactions between i and j atoms */
 185             qq00             = _mm_mul_pd(iq0,jq0);
 186             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 187                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 188
 189             /* Calculate table index by multiplying r with table scale and truncate to integer */
 190             rt               = _mm_mul_pd(r00,vftabscale);
 191             vfitab           = _mm_cvttpd_epi32(rt);
 192 #ifdef __XOP__
 193             vfeps            = _mm_frcz_pd(rt);
 194 #else
 195             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 196 #endif
 197             twovfeps         = _mm_add_pd(vfeps,vfeps);
 198             vfitab           = _mm_slli_epi32(vfitab,3);
 199
 200             /* REACTION-FIELD ELECTROSTATICS */
 201             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 202             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 203
 204             /* CUBIC SPLINE TABLE DISPERSION */
 205             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 206             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 207             GMX_MM_TRANSPOSE2_PD(Y,F);
 208             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 209             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 210             GMX_MM_TRANSPOSE2_PD(G,H);
 211             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 212             VV               = _mm_macc_pd(vfeps,Fp,Y);
 213             vvdw6            = _mm_mul_pd(c6_00,VV);
 214             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 215             fvdw6            = _mm_mul_pd(c6_00,FF);
 216
 217             /* CUBIC SPLINE TABLE REPULSION */
 218             vfitab           = _mm_add_epi32(vfitab,ifour);
 219             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(Y,F);
 222             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 223             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(G,H);
 225             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 226             VV               = _mm_macc_pd(vfeps,Fp,Y);
 227             vvdw12           = _mm_mul_pd(c12_00,VV);
 228             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 229             fvdw12           = _mm_mul_pd(c12_00,FF);
 230             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 231             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 232
 233             /* Update potential sum for this i atom from the interaction with this j atom. */
 234             velecsum         = _mm_add_pd(velecsum,velec);
 235             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 236
 237             fscal            = _mm_add_pd(felec,fvdw);
 238
 239             /* Update vectorial force */
 240             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 241             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 242             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 243
 244             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 245                                                    _mm_mul_pd(dx00,fscal),
 246                                                    _mm_mul_pd(dy00,fscal),
 247                                                    _mm_mul_pd(dz00,fscal));
 248
 249             /* Inner loop uses 70 flops */
 250         }
 251
 252         if(jidx<j_index_end)
 253         {
 254
 255             jnrA             = jjnr[jidx];
 256             j_coord_offsetA  = DIM*jnrA;
 257
 258             /* load j atom coordinates */
 259             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 260                                               &jx0,&jy0,&jz0);
 261
 262             /* Calculate displacement vector */
 263             dx00             = _mm_sub_pd(ix0,jx0);
 264             dy00             = _mm_sub_pd(iy0,jy0);
 265             dz00             = _mm_sub_pd(iz0,jz0);
 266
 267             /* Calculate squared distance and things based on it */
 268             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 269
 270             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 271
 272             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 273
 274             /* Load parameters for j particles */
 275             jq0              = _mm_load_sd(charge+jnrA+0);
 276             vdwjidx0A        = 2*vdwtype[jnrA+0];
 277
 278             /**************************
 279              * CALCULATE INTERACTIONS *
 280              **************************/
 281
 282             r00              = _mm_mul_pd(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq00             = _mm_mul_pd(iq0,jq0);
 286             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 287
 288             /* Calculate table index by multiplying r with table scale and truncate to integer */
 289             rt               = _mm_mul_pd(r00,vftabscale);
 290             vfitab           = _mm_cvttpd_epi32(rt);
 291 #ifdef __XOP__
 292             vfeps            = _mm_frcz_pd(rt);
 293 #else
 294             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 295 #endif
 296             twovfeps         = _mm_add_pd(vfeps,vfeps);
 297             vfitab           = _mm_slli_epi32(vfitab,3);
 298
 299             /* REACTION-FIELD ELECTROSTATICS */
 300             velec            = _mm_mul_pd(qq00,_mm_sub_pd(_mm_macc_pd(krf,rsq00,rinv00),crf));
 301             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 302
 303             /* CUBIC SPLINE TABLE DISPERSION */
 304             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 305             F                = _mm_setzero_pd();
 306             GMX_MM_TRANSPOSE2_PD(Y,F);
 307             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 308             H                = _mm_setzero_pd();
 309             GMX_MM_TRANSPOSE2_PD(G,H);
 310             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 311             VV               = _mm_macc_pd(vfeps,Fp,Y);
 312             vvdw6            = _mm_mul_pd(c6_00,VV);
 313             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 314             fvdw6            = _mm_mul_pd(c6_00,FF);
 315
 316             /* CUBIC SPLINE TABLE REPULSION */
 317             vfitab           = _mm_add_epi32(vfitab,ifour);
 318             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 319             F                = _mm_setzero_pd();
 320             GMX_MM_TRANSPOSE2_PD(Y,F);
 321             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 322             H                = _mm_setzero_pd();
 323             GMX_MM_TRANSPOSE2_PD(G,H);
 324             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 325             VV               = _mm_macc_pd(vfeps,Fp,Y);
 326             vvdw12           = _mm_mul_pd(c12_00,VV);
 327             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 328             fvdw12           = _mm_mul_pd(c12_00,FF);
 329             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 330             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 331
 332             /* Update potential sum for this i atom from the interaction with this j atom. */
 333             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 334             velecsum         = _mm_add_pd(velecsum,velec);
 335             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 336             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 337
 338             fscal            = _mm_add_pd(felec,fvdw);
 339
 340             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 341
 342             /* Update vectorial force */
 343             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 344             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 345             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 346
 347             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 348                                                    _mm_mul_pd(dx00,fscal),
 349                                                    _mm_mul_pd(dy00,fscal),
 350                                                    _mm_mul_pd(dz00,fscal));
 351
 352             /* Inner loop uses 70 flops */
 353         }
 354
 355         /* End of innermost loop */
 356
 357         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 358                                               f+i_coord_offset,fshift+i_shift_offset);
 359
 360         ggid                        = gid[iidx];
 361         /* Update potential energies */
 362         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 363         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 364
 365         /* Increment number of inner iterations */
 366         inneriter                  += j_index_end - j_index_start;
 367
 368         /* Outer loop uses 9 flops */
 369     }
 370
 371     /* Increment number of outer iterations */
 372     outeriter        += nri;
 373
 374     /* Update outer/inner flops */
 375
 376     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*70);
 377 }
 378 /*
 379  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 380  * Electrostatics interaction: ReactionField
 381  * VdW interaction:            CubicSplineTable
 382  * Geometry:                   Particle-Particle
 383  * Calculate force/pot:        Force
 384  */
 385 void
 386 nb_kernel_ElecRF_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 387                     (t_nblist * gmx_restrict                nlist,
 388                      rvec * gmx_restrict                    xx,
 389                      rvec * gmx_restrict                    ff,
 390                      t_forcerec * gmx_restrict              fr,
 391                      t_mdatoms * gmx_restrict               mdatoms,
 392                      nb_kernel_data_t * gmx_restrict        kernel_data,
 393                      t_nrnb * gmx_restrict                  nrnb)
 394 {
 395     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 396      * just 0 for non-waters.
 397      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 398      * jnr indices corresponding to data put in the four positions in the SIMD register.
 399      */
 400     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 401     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 402     int              jnrA,jnrB;
 403     int              j_coord_offsetA,j_coord_offsetB;
 404     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 405     real             rcutoff_scalar;
 406     real             *shiftvec,*fshift,*x,*f;
 407     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 408     int              vdwioffset0;
 409     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 410     int              vdwjidx0A,vdwjidx0B;
 411     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 412     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 413     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 414     real             *charge;
 415     int              nvdwtype;
 416     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 417     int              *vdwtype;
 418     real             *vdwparam;
 419     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 420     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 421     __m128i          vfitab;
 422     __m128i          ifour       = _mm_set1_epi32(4);
 423     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 424     real             *vftab;
 425     __m128d          dummy_mask,cutoff_mask;
 426     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 427     __m128d          one     = _mm_set1_pd(1.0);
 428     __m128d          two     = _mm_set1_pd(2.0);
 429     x                = xx[0];
 430     f                = ff[0];
 431
 432     nri              = nlist->nri;
 433     iinr             = nlist->iinr;
 434     jindex           = nlist->jindex;
 435     jjnr             = nlist->jjnr;
 436     shiftidx         = nlist->shift;
 437     gid              = nlist->gid;
 438     shiftvec         = fr->shift_vec[0];
 439     fshift           = fr->fshift[0];
 440     facel            = _mm_set1_pd(fr->epsfac);
 441     charge           = mdatoms->chargeA;
 442     krf              = _mm_set1_pd(fr->ic->k_rf);
 443     krf2             = _mm_set1_pd(fr->ic->k_rf*2.0);
 444     crf              = _mm_set1_pd(fr->ic->c_rf);
 445     nvdwtype         = fr->ntype;
 446     vdwparam         = fr->nbfp;
 447     vdwtype          = mdatoms->typeA;
 448
 449     vftab            = kernel_data->table_vdw->data;
 450     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 451
 452     /* Avoid stupid compiler warnings */
 453     jnrA = jnrB = 0;
 454     j_coord_offsetA = 0;
 455     j_coord_offsetB = 0;
 456
 457     outeriter        = 0;
 458     inneriter        = 0;
 459
 460     /* Start outer loop over neighborlists */
 461     for(iidx=0; iidx<nri; iidx++)
 462     {
 463         /* Load shift vector for this list */
 464         i_shift_offset   = DIM*shiftidx[iidx];
 465
 466         /* Load limits for loop over neighbors */
 467         j_index_start    = jindex[iidx];
 468         j_index_end      = jindex[iidx+1];
 469
 470         /* Get outer coordinate index */
 471         inr              = iinr[iidx];
 472         i_coord_offset   = DIM*inr;
 473
 474         /* Load i particle coords and add shift vector */
 475         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 476
 477         fix0             = _mm_setzero_pd();
 478         fiy0             = _mm_setzero_pd();
 479         fiz0             = _mm_setzero_pd();
 480
 481         /* Load parameters for i particles */
 482         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 483         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 484
 485         /* Start inner kernel loop */
 486         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 487         {
 488
 489             /* Get j neighbor index, and coordinate index */
 490             jnrA             = jjnr[jidx];
 491             jnrB             = jjnr[jidx+1];
 492             j_coord_offsetA  = DIM*jnrA;
 493             j_coord_offsetB  = DIM*jnrB;
 494
 495             /* load j atom coordinates */
 496             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 497                                               &jx0,&jy0,&jz0);
 498
 499             /* Calculate displacement vector */
 500             dx00             = _mm_sub_pd(ix0,jx0);
 501             dy00             = _mm_sub_pd(iy0,jy0);
 502             dz00             = _mm_sub_pd(iz0,jz0);
 503
 504             /* Calculate squared distance and things based on it */
 505             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 506
 507             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 508
 509             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 510
 511             /* Load parameters for j particles */
 512             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 513             vdwjidx0A        = 2*vdwtype[jnrA+0];
 514             vdwjidx0B        = 2*vdwtype[jnrB+0];
 515
 516             /**************************
 517              * CALCULATE INTERACTIONS *
 518              **************************/
 519
 520             r00              = _mm_mul_pd(rsq00,rinv00);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq00             = _mm_mul_pd(iq0,jq0);
 524             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 525                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 526
 527             /* Calculate table index by multiplying r with table scale and truncate to integer */
 528             rt               = _mm_mul_pd(r00,vftabscale);
 529             vfitab           = _mm_cvttpd_epi32(rt);
 530 #ifdef __XOP__
 531             vfeps            = _mm_frcz_pd(rt);
 532 #else
 533             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 534 #endif
 535             twovfeps         = _mm_add_pd(vfeps,vfeps);
 536             vfitab           = _mm_slli_epi32(vfitab,3);
 537
 538             /* REACTION-FIELD ELECTROSTATICS */
 539             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 540
 541             /* CUBIC SPLINE TABLE DISPERSION */
 542             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 543             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 544             GMX_MM_TRANSPOSE2_PD(Y,F);
 545             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 546             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 547             GMX_MM_TRANSPOSE2_PD(G,H);
 548             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 549             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 550             fvdw6            = _mm_mul_pd(c6_00,FF);
 551
 552             /* CUBIC SPLINE TABLE REPULSION */
 553             vfitab           = _mm_add_epi32(vfitab,ifour);
 554             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 555             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 556             GMX_MM_TRANSPOSE2_PD(Y,F);
 557             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 558             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 559             GMX_MM_TRANSPOSE2_PD(G,H);
 560             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 561             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 562             fvdw12           = _mm_mul_pd(c12_00,FF);
 563             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 564
 565             fscal            = _mm_add_pd(felec,fvdw);
 566
 567             /* Update vectorial force */
 568             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 569             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 570             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 571
 572             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 573                                                    _mm_mul_pd(dx00,fscal),
 574                                                    _mm_mul_pd(dy00,fscal),
 575                                                    _mm_mul_pd(dz00,fscal));
 576
 577             /* Inner loop uses 57 flops */
 578         }
 579
 580         if(jidx<j_index_end)
 581         {
 582
 583             jnrA             = jjnr[jidx];
 584             j_coord_offsetA  = DIM*jnrA;
 585
 586             /* load j atom coordinates */
 587             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 588                                               &jx0,&jy0,&jz0);
 589
 590             /* Calculate displacement vector */
 591             dx00             = _mm_sub_pd(ix0,jx0);
 592             dy00             = _mm_sub_pd(iy0,jy0);
 593             dz00             = _mm_sub_pd(iz0,jz0);
 594
 595             /* Calculate squared distance and things based on it */
 596             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 597
 598             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 599
 600             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 601
 602             /* Load parameters for j particles */
 603             jq0              = _mm_load_sd(charge+jnrA+0);
 604             vdwjidx0A        = 2*vdwtype[jnrA+0];
 605
 606             /**************************
 607              * CALCULATE INTERACTIONS *
 608              **************************/
 609
 610             r00              = _mm_mul_pd(rsq00,rinv00);
 611
 612             /* Compute parameters for interactions between i and j atoms */
 613             qq00             = _mm_mul_pd(iq0,jq0);
 614             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 615
 616             /* Calculate table index by multiplying r with table scale and truncate to integer */
 617             rt               = _mm_mul_pd(r00,vftabscale);
 618             vfitab           = _mm_cvttpd_epi32(rt);
 619 #ifdef __XOP__
 620             vfeps            = _mm_frcz_pd(rt);
 621 #else
 622             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 623 #endif
 624             twovfeps         = _mm_add_pd(vfeps,vfeps);
 625             vfitab           = _mm_slli_epi32(vfitab,3);
 626
 627             /* REACTION-FIELD ELECTROSTATICS */
 628             felec            = _mm_mul_pd(qq00,_mm_msub_pd(rinv00,rinvsq00,krf2));
 629
 630             /* CUBIC SPLINE TABLE DISPERSION */
 631             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 632             F                = _mm_setzero_pd();
 633             GMX_MM_TRANSPOSE2_PD(Y,F);
 634             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 635             H                = _mm_setzero_pd();
 636             GMX_MM_TRANSPOSE2_PD(G,H);
 637             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 638             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 639             fvdw6            = _mm_mul_pd(c6_00,FF);
 640
 641             /* CUBIC SPLINE TABLE REPULSION */
 642             vfitab           = _mm_add_epi32(vfitab,ifour);
 643             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 644             F                = _mm_setzero_pd();
 645             GMX_MM_TRANSPOSE2_PD(Y,F);
 646             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 647             H                = _mm_setzero_pd();
 648             GMX_MM_TRANSPOSE2_PD(G,H);
 649             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 650             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 651             fvdw12           = _mm_mul_pd(c12_00,FF);
 652             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 653
 654             fscal            = _mm_add_pd(felec,fvdw);
 655
 656             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 657
 658             /* Update vectorial force */
 659             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 660             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 661             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 662
 663             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 664                                                    _mm_mul_pd(dx00,fscal),
 665                                                    _mm_mul_pd(dy00,fscal),
 666                                                    _mm_mul_pd(dz00,fscal));
 667
 668             /* Inner loop uses 57 flops */
 669         }
 670
 671         /* End of innermost loop */
 672
 673         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 674                                               f+i_coord_offset,fshift+i_shift_offset);
 675
 676         /* Increment number of inner iterations */
 677         inneriter                  += j_index_end - j_index_start;
 678
 679         /* Outer loop uses 7 flops */
 680     }
 681
 682     /* Increment number of outer iterations */
 683     outeriter        += nri;
 684
 685     /* Update outer/inner flops */
 686
 687     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*57);
 688 }