src/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecEw_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_ElecEw_VdwCSTab_GeomP1P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_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     __m128i          ewitab;
  84     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  85     real             *ewtab;
  86     __m128d          dummy_mask,cutoff_mask;
  87     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  88     __m128d          one     = _mm_set1_pd(1.0);
  89     __m128d          two     = _mm_set1_pd(2.0);
  90     x                = xx[0];
  91     f                = ff[0];
  92
  93     nri              = nlist->nri;
  94     iinr             = nlist->iinr;
  95     jindex           = nlist->jindex;
  96     jjnr             = nlist->jjnr;
  97     shiftidx         = nlist->shift;
  98     gid              = nlist->gid;
  99     shiftvec         = fr->shift_vec[0];
 100     fshift           = fr->fshift[0];
 101     facel            = _mm_set1_pd(fr->epsfac);
 102     charge           = mdatoms->chargeA;
 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     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 111     ewtab            = fr->ic->tabq_coul_FDV0;
 112     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 113     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 114
 115     /* Avoid stupid compiler warnings */
 116     jnrA = jnrB = 0;
 117     j_coord_offsetA = 0;
 118     j_coord_offsetB = 0;
 119
 120     outeriter        = 0;
 121     inneriter        = 0;
 122
 123     /* Start outer loop over neighborlists */
 124     for(iidx=0; iidx<nri; iidx++)
 125     {
 126         /* Load shift vector for this list */
 127         i_shift_offset   = DIM*shiftidx[iidx];
 128
 129         /* Load limits for loop over neighbors */
 130         j_index_start    = jindex[iidx];
 131         j_index_end      = jindex[iidx+1];
 132
 133         /* Get outer coordinate index */
 134         inr              = iinr[iidx];
 135         i_coord_offset   = DIM*inr;
 136
 137         /* Load i particle coords and add shift vector */
 138         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 139
 140         fix0             = _mm_setzero_pd();
 141         fiy0             = _mm_setzero_pd();
 142         fiz0             = _mm_setzero_pd();
 143
 144         /* Load parameters for i particles */
 145         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 146         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 147
 148         /* Reset potential sums */
 149         velecsum         = _mm_setzero_pd();
 150         vvdwsum          = _mm_setzero_pd();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             j_coord_offsetA  = DIM*jnrA;
 160             j_coord_offsetB  = DIM*jnrB;
 161
 162             /* load j atom coordinates */
 163             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 164                                               &jx0,&jy0,&jz0);
 165
 166             /* Calculate displacement vector */
 167             dx00             = _mm_sub_pd(ix0,jx0);
 168             dy00             = _mm_sub_pd(iy0,jy0);
 169             dz00             = _mm_sub_pd(iz0,jz0);
 170
 171             /* Calculate squared distance and things based on it */
 172             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 173
 174             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 175
 176             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 177
 178             /* Load parameters for j particles */
 179             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 180             vdwjidx0A        = 2*vdwtype[jnrA+0];
 181             vdwjidx0B        = 2*vdwtype[jnrB+0];
 182
 183             /**************************
 184              * CALCULATE INTERACTIONS *
 185              **************************/
 186
 187             r00              = _mm_mul_pd(rsq00,rinv00);
 188
 189             /* Compute parameters for interactions between i and j atoms */
 190             qq00             = _mm_mul_pd(iq0,jq0);
 191             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 192                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 193
 194             /* Calculate table index by multiplying r with table scale and truncate to integer */
 195             rt               = _mm_mul_pd(r00,vftabscale);
 196             vfitab           = _mm_cvttpd_epi32(rt);
 197 #ifdef __XOP__
 198             vfeps            = _mm_frcz_pd(rt);
 199 #else
 200             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 201 #endif
 202             twovfeps         = _mm_add_pd(vfeps,vfeps);
 203             vfitab           = _mm_slli_epi32(vfitab,3);
 204
 205             /* EWALD ELECTROSTATICS */
 206
 207             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 208             ewrt             = _mm_mul_pd(r00,ewtabscale);
 209             ewitab           = _mm_cvttpd_epi32(ewrt);
 210 #ifdef __XOP__
 211             eweps            = _mm_frcz_pd(ewrt);
 212 #else
 213             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 214 #endif
 215             twoeweps         = _mm_add_pd(eweps,eweps);
 216             ewitab           = _mm_slli_epi32(ewitab,2);
 217             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 218             ewtabD           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,1) );
 219             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 220             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 221             ewtabFn          = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,1) +2);
 222             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 223             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 224             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 225             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 226             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 227
 228             /* CUBIC SPLINE TABLE DISPERSION */
 229             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 230             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 231             GMX_MM_TRANSPOSE2_PD(Y,F);
 232             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 233             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 234             GMX_MM_TRANSPOSE2_PD(G,H);
 235             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 236             VV               = _mm_macc_pd(vfeps,Fp,Y);
 237             vvdw6            = _mm_mul_pd(c6_00,VV);
 238             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 239             fvdw6            = _mm_mul_pd(c6_00,FF);
 240
 241             /* CUBIC SPLINE TABLE REPULSION */
 242             vfitab           = _mm_add_epi32(vfitab,ifour);
 243             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 244             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 245             GMX_MM_TRANSPOSE2_PD(Y,F);
 246             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 247             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 248             GMX_MM_TRANSPOSE2_PD(G,H);
 249             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 250             VV               = _mm_macc_pd(vfeps,Fp,Y);
 251             vvdw12           = _mm_mul_pd(c12_00,VV);
 252             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 253             fvdw12           = _mm_mul_pd(c12_00,FF);
 254             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 255             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 256
 257             /* Update potential sum for this i atom from the interaction with this j atom. */
 258             velecsum         = _mm_add_pd(velecsum,velec);
 259             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 260
 261             fscal            = _mm_add_pd(felec,fvdw);
 262
 263             /* Update vectorial force */
 264             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 265             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 266             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 267
 268             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 269                                                    _mm_mul_pd(dx00,fscal),
 270                                                    _mm_mul_pd(dy00,fscal),
 271                                                    _mm_mul_pd(dz00,fscal));
 272
 273             /* Inner loop uses 78 flops */
 274         }
 275
 276         if(jidx<j_index_end)
 277         {
 278
 279             jnrA             = jjnr[jidx];
 280             j_coord_offsetA  = DIM*jnrA;
 281
 282             /* load j atom coordinates */
 283             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 284                                               &jx0,&jy0,&jz0);
 285
 286             /* Calculate displacement vector */
 287             dx00             = _mm_sub_pd(ix0,jx0);
 288             dy00             = _mm_sub_pd(iy0,jy0);
 289             dz00             = _mm_sub_pd(iz0,jz0);
 290
 291             /* Calculate squared distance and things based on it */
 292             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 293
 294             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 295
 296             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 297
 298             /* Load parameters for j particles */
 299             jq0              = _mm_load_sd(charge+jnrA+0);
 300             vdwjidx0A        = 2*vdwtype[jnrA+0];
 301
 302             /**************************
 303              * CALCULATE INTERACTIONS *
 304              **************************/
 305
 306             r00              = _mm_mul_pd(rsq00,rinv00);
 307
 308             /* Compute parameters for interactions between i and j atoms */
 309             qq00             = _mm_mul_pd(iq0,jq0);
 310             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 311
 312             /* Calculate table index by multiplying r with table scale and truncate to integer */
 313             rt               = _mm_mul_pd(r00,vftabscale);
 314             vfitab           = _mm_cvttpd_epi32(rt);
 315 #ifdef __XOP__
 316             vfeps            = _mm_frcz_pd(rt);
 317 #else
 318             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 319 #endif
 320             twovfeps         = _mm_add_pd(vfeps,vfeps);
 321             vfitab           = _mm_slli_epi32(vfitab,3);
 322
 323             /* EWALD ELECTROSTATICS */
 324
 325             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 326             ewrt             = _mm_mul_pd(r00,ewtabscale);
 327             ewitab           = _mm_cvttpd_epi32(ewrt);
 328 #ifdef __XOP__
 329             eweps            = _mm_frcz_pd(ewrt);
 330 #else
 331             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 332 #endif
 333             twoeweps         = _mm_add_pd(eweps,eweps);
 334             ewitab           = _mm_slli_epi32(ewitab,2);
 335             ewtabF           = _mm_load_pd( ewtab + _mm_extract_epi32(ewitab,0) );
 336             ewtabD           = _mm_setzero_pd();
 337             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 338             ewtabV           = _mm_load_sd( ewtab + _mm_extract_epi32(ewitab,0) +2);
 339             ewtabFn          = _mm_setzero_pd();
 340             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 341             felec            = _mm_macc_pd(eweps,ewtabD,ewtabF);
 342             velec            = _mm_nmacc_pd(_mm_mul_pd(ewtabhalfspace,eweps) ,_mm_add_pd(ewtabF,felec), ewtabV);
 343             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 344             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 345
 346             /* CUBIC SPLINE TABLE DISPERSION */
 347             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 348             F                = _mm_setzero_pd();
 349             GMX_MM_TRANSPOSE2_PD(Y,F);
 350             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 351             H                = _mm_setzero_pd();
 352             GMX_MM_TRANSPOSE2_PD(G,H);
 353             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 354             VV               = _mm_macc_pd(vfeps,Fp,Y);
 355             vvdw6            = _mm_mul_pd(c6_00,VV);
 356             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 357             fvdw6            = _mm_mul_pd(c6_00,FF);
 358
 359             /* CUBIC SPLINE TABLE REPULSION */
 360             vfitab           = _mm_add_epi32(vfitab,ifour);
 361             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 362             F                = _mm_setzero_pd();
 363             GMX_MM_TRANSPOSE2_PD(Y,F);
 364             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 365             H                = _mm_setzero_pd();
 366             GMX_MM_TRANSPOSE2_PD(G,H);
 367             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 368             VV               = _mm_macc_pd(vfeps,Fp,Y);
 369             vvdw12           = _mm_mul_pd(c12_00,VV);
 370             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 371             fvdw12           = _mm_mul_pd(c12_00,FF);
 372             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 373             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 379             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 380
 381             fscal            = _mm_add_pd(felec,fvdw);
 382
 383             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 384
 385             /* Update vectorial force */
 386             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 387             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 388             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 389
 390             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 391                                                    _mm_mul_pd(dx00,fscal),
 392                                                    _mm_mul_pd(dy00,fscal),
 393                                                    _mm_mul_pd(dz00,fscal));
 394
 395             /* Inner loop uses 78 flops */
 396         }
 397
 398         /* End of innermost loop */
 399
 400         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 401                                               f+i_coord_offset,fshift+i_shift_offset);
 402
 403         ggid                        = gid[iidx];
 404         /* Update potential energies */
 405         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 406         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 407
 408         /* Increment number of inner iterations */
 409         inneriter                  += j_index_end - j_index_start;
 410
 411         /* Outer loop uses 9 flops */
 412     }
 413
 414     /* Increment number of outer iterations */
 415     outeriter        += nri;
 416
 417     /* Update outer/inner flops */
 418
 419     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*78);
 420 }
 421 /*
 422  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 423  * Electrostatics interaction: Ewald
 424  * VdW interaction:            CubicSplineTable
 425  * Geometry:                   Particle-Particle
 426  * Calculate force/pot:        Force
 427  */
 428 void
 429 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_avx_128_fma_double
 430                     (t_nblist * gmx_restrict                nlist,
 431                      rvec * gmx_restrict                    xx,
 432                      rvec * gmx_restrict                    ff,
 433                      t_forcerec * gmx_restrict              fr,
 434                      t_mdatoms * gmx_restrict               mdatoms,
 435                      nb_kernel_data_t * gmx_restrict        kernel_data,
 436                      t_nrnb * gmx_restrict                  nrnb)
 437 {
 438     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 439      * just 0 for non-waters.
 440      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 441      * jnr indices corresponding to data put in the four positions in the SIMD register.
 442      */
 443     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 444     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 445     int              jnrA,jnrB;
 446     int              j_coord_offsetA,j_coord_offsetB;
 447     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 448     real             rcutoff_scalar;
 449     real             *shiftvec,*fshift,*x,*f;
 450     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 451     int              vdwioffset0;
 452     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 453     int              vdwjidx0A,vdwjidx0B;
 454     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 455     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 456     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 457     real             *charge;
 458     int              nvdwtype;
 459     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 460     int              *vdwtype;
 461     real             *vdwparam;
 462     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 463     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 464     __m128i          vfitab;
 465     __m128i          ifour       = _mm_set1_epi32(4);
 466     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF,twovfeps;
 467     real             *vftab;
 468     __m128i          ewitab;
 469     __m128d          ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 470     real             *ewtab;
 471     __m128d          dummy_mask,cutoff_mask;
 472     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 473     __m128d          one     = _mm_set1_pd(1.0);
 474     __m128d          two     = _mm_set1_pd(2.0);
 475     x                = xx[0];
 476     f                = ff[0];
 477
 478     nri              = nlist->nri;
 479     iinr             = nlist->iinr;
 480     jindex           = nlist->jindex;
 481     jjnr             = nlist->jjnr;
 482     shiftidx         = nlist->shift;
 483     gid              = nlist->gid;
 484     shiftvec         = fr->shift_vec[0];
 485     fshift           = fr->fshift[0];
 486     facel            = _mm_set1_pd(fr->epsfac);
 487     charge           = mdatoms->chargeA;
 488     nvdwtype         = fr->ntype;
 489     vdwparam         = fr->nbfp;
 490     vdwtype          = mdatoms->typeA;
 491
 492     vftab            = kernel_data->table_vdw->data;
 493     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 494
 495     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 496     ewtab            = fr->ic->tabq_coul_F;
 497     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 498     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 499
 500     /* Avoid stupid compiler warnings */
 501     jnrA = jnrB = 0;
 502     j_coord_offsetA = 0;
 503     j_coord_offsetB = 0;
 504
 505     outeriter        = 0;
 506     inneriter        = 0;
 507
 508     /* Start outer loop over neighborlists */
 509     for(iidx=0; iidx<nri; iidx++)
 510     {
 511         /* Load shift vector for this list */
 512         i_shift_offset   = DIM*shiftidx[iidx];
 513
 514         /* Load limits for loop over neighbors */
 515         j_index_start    = jindex[iidx];
 516         j_index_end      = jindex[iidx+1];
 517
 518         /* Get outer coordinate index */
 519         inr              = iinr[iidx];
 520         i_coord_offset   = DIM*inr;
 521
 522         /* Load i particle coords and add shift vector */
 523         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 524
 525         fix0             = _mm_setzero_pd();
 526         fiy0             = _mm_setzero_pd();
 527         fiz0             = _mm_setzero_pd();
 528
 529         /* Load parameters for i particles */
 530         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 531         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 532
 533         /* Start inner kernel loop */
 534         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 535         {
 536
 537             /* Get j neighbor index, and coordinate index */
 538             jnrA             = jjnr[jidx];
 539             jnrB             = jjnr[jidx+1];
 540             j_coord_offsetA  = DIM*jnrA;
 541             j_coord_offsetB  = DIM*jnrB;
 542
 543             /* load j atom coordinates */
 544             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 545                                               &jx0,&jy0,&jz0);
 546
 547             /* Calculate displacement vector */
 548             dx00             = _mm_sub_pd(ix0,jx0);
 549             dy00             = _mm_sub_pd(iy0,jy0);
 550             dz00             = _mm_sub_pd(iz0,jz0);
 551
 552             /* Calculate squared distance and things based on it */
 553             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 554
 555             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 556
 557             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 558
 559             /* Load parameters for j particles */
 560             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 561             vdwjidx0A        = 2*vdwtype[jnrA+0];
 562             vdwjidx0B        = 2*vdwtype[jnrB+0];
 563
 564             /**************************
 565              * CALCULATE INTERACTIONS *
 566              **************************/
 567
 568             r00              = _mm_mul_pd(rsq00,rinv00);
 569
 570             /* Compute parameters for interactions between i and j atoms */
 571             qq00             = _mm_mul_pd(iq0,jq0);
 572             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 573                                          vdwparam+vdwioffset0+vdwjidx0B,&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,3);
 585
 586             /* EWALD ELECTROSTATICS */
 587
 588             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 589             ewrt             = _mm_mul_pd(r00,ewtabscale);
 590             ewitab           = _mm_cvttpd_epi32(ewrt);
 591 #ifdef __XOP__
 592             eweps            = _mm_frcz_pd(ewrt);
 593 #else
 594             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 595 #endif
 596             twoeweps         = _mm_add_pd(eweps,eweps);
 597             gmx_mm_load_2pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),ewtab+_mm_extract_epi32(ewitab,1),
 598                                          &ewtabF,&ewtabFn);
 599             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 600             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 601
 602             /* CUBIC SPLINE TABLE DISPERSION */
 603             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 604             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 605             GMX_MM_TRANSPOSE2_PD(Y,F);
 606             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 607             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 608             GMX_MM_TRANSPOSE2_PD(G,H);
 609             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 610             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 611             fvdw6            = _mm_mul_pd(c6_00,FF);
 612
 613             /* CUBIC SPLINE TABLE REPULSION */
 614             vfitab           = _mm_add_epi32(vfitab,ifour);
 615             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 616             F                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) );
 617             GMX_MM_TRANSPOSE2_PD(Y,F);
 618             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 619             H                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,1) +2);
 620             GMX_MM_TRANSPOSE2_PD(G,H);
 621             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 622             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 623             fvdw12           = _mm_mul_pd(c12_00,FF);
 624             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 625
 626             fscal            = _mm_add_pd(felec,fvdw);
 627
 628             /* Update vectorial force */
 629             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 630             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 631             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 632
 633             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,
 634                                                    _mm_mul_pd(dx00,fscal),
 635                                                    _mm_mul_pd(dy00,fscal),
 636                                                    _mm_mul_pd(dz00,fscal));
 637
 638             /* Inner loop uses 65 flops */
 639         }
 640
 641         if(jidx<j_index_end)
 642         {
 643
 644             jnrA             = jjnr[jidx];
 645             j_coord_offsetA  = DIM*jnrA;
 646
 647             /* load j atom coordinates */
 648             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 649                                               &jx0,&jy0,&jz0);
 650
 651             /* Calculate displacement vector */
 652             dx00             = _mm_sub_pd(ix0,jx0);
 653             dy00             = _mm_sub_pd(iy0,jy0);
 654             dz00             = _mm_sub_pd(iz0,jz0);
 655
 656             /* Calculate squared distance and things based on it */
 657             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 658
 659             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 660
 661             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 662
 663             /* Load parameters for j particles */
 664             jq0              = _mm_load_sd(charge+jnrA+0);
 665             vdwjidx0A        = 2*vdwtype[jnrA+0];
 666
 667             /**************************
 668              * CALCULATE INTERACTIONS *
 669              **************************/
 670
 671             r00              = _mm_mul_pd(rsq00,rinv00);
 672
 673             /* Compute parameters for interactions between i and j atoms */
 674             qq00             = _mm_mul_pd(iq0,jq0);
 675             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 676
 677             /* Calculate table index by multiplying r with table scale and truncate to integer */
 678             rt               = _mm_mul_pd(r00,vftabscale);
 679             vfitab           = _mm_cvttpd_epi32(rt);
 680 #ifdef __XOP__
 681             vfeps            = _mm_frcz_pd(rt);
 682 #else
 683             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 684 #endif
 685             twovfeps         = _mm_add_pd(vfeps,vfeps);
 686             vfitab           = _mm_slli_epi32(vfitab,3);
 687
 688             /* EWALD ELECTROSTATICS */
 689
 690             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 691             ewrt             = _mm_mul_pd(r00,ewtabscale);
 692             ewitab           = _mm_cvttpd_epi32(ewrt);
 693 #ifdef __XOP__
 694             eweps            = _mm_frcz_pd(ewrt);
 695 #else
 696             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 697 #endif
 698             twoeweps         = _mm_add_pd(eweps,eweps);
 699             gmx_mm_load_1pair_swizzle_pd(ewtab+_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 700             felec            = _mm_macc_pd(eweps,ewtabFn,_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF));
 701             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 702
 703             /* CUBIC SPLINE TABLE DISPERSION */
 704             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 705             F                = _mm_setzero_pd();
 706             GMX_MM_TRANSPOSE2_PD(Y,F);
 707             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 708             H                = _mm_setzero_pd();
 709             GMX_MM_TRANSPOSE2_PD(G,H);
 710             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 711             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 712             fvdw6            = _mm_mul_pd(c6_00,FF);
 713
 714             /* CUBIC SPLINE TABLE REPULSION */
 715             vfitab           = _mm_add_epi32(vfitab,ifour);
 716             Y                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) );
 717             F                = _mm_setzero_pd();
 718             GMX_MM_TRANSPOSE2_PD(Y,F);
 719             G                = _mm_load_pd( vftab + _mm_extract_epi32(vfitab,0) +2);
 720             H                = _mm_setzero_pd();
 721             GMX_MM_TRANSPOSE2_PD(G,H);
 722             Fp               = _mm_macc_pd(vfeps,_mm_macc_pd(H,vfeps,G),F);
 723             FF               = _mm_macc_pd(vfeps,_mm_macc_pd(twovfeps,H,G),Fp);
 724             fvdw12           = _mm_mul_pd(c12_00,FF);
 725             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 726
 727             fscal            = _mm_add_pd(felec,fvdw);
 728
 729             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 730
 731             /* Update vectorial force */
 732             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 733             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 734             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 735
 736             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,
 737                                                    _mm_mul_pd(dx00,fscal),
 738                                                    _mm_mul_pd(dy00,fscal),
 739                                                    _mm_mul_pd(dz00,fscal));
 740
 741             /* Inner loop uses 65 flops */
 742         }
 743
 744         /* End of innermost loop */
 745
 746         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 747                                               f+i_coord_offset,fshift+i_shift_offset);
 748
 749         /* Increment number of inner iterations */
 750         inneriter                  += j_index_end - j_index_start;
 751
 752         /* Outer loop uses 7 flops */
 753     }
 754
 755     /* Increment number of outer iterations */
 756     outeriter        += nri;
 757
 758     /* Update outer/inner flops */
 759
 760     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*65);
 761 }