src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEw_VdwCSTab_GeomP1P1_sse4_1_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_double.h"
  34 #include "kernelutil_x86_sse4_1_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_VF_sse4_1_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_sse4_1_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;
  82     real             *vftab;
  83     __m128i          ewitab;
  84     __m128d          ewtabscale,eweps,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             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 198             vfitab           = _mm_slli_epi32(vfitab,3);
 199
 200             /* EWALD ELECTROSTATICS */
 201
 202             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 203             ewrt             = _mm_mul_pd(r00,ewtabscale);
 204             ewitab           = _mm_cvttpd_epi32(ewrt);
 205             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 206             ewitab           = _mm_slli_epi32(ewitab,2);
 207             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 208             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 209             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 210             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 211             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 212             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 213             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 214             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 215             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 216             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 217
 218             /* CUBIC SPLINE TABLE DISPERSION */
 219             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 220             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 221             GMX_MM_TRANSPOSE2_PD(Y,F);
 222             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 223             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 224             GMX_MM_TRANSPOSE2_PD(G,H);
 225             Heps             = _mm_mul_pd(vfeps,H);
 226             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 227             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 228             vvdw6            = _mm_mul_pd(c6_00,VV);
 229             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 230             fvdw6            = _mm_mul_pd(c6_00,FF);
 231
 232             /* CUBIC SPLINE TABLE REPULSION */
 233             vfitab           = _mm_add_epi32(vfitab,ifour);
 234             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 235             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 236             GMX_MM_TRANSPOSE2_PD(Y,F);
 237             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 238             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 239             GMX_MM_TRANSPOSE2_PD(G,H);
 240             Heps             = _mm_mul_pd(vfeps,H);
 241             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 242             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 243             vvdw12           = _mm_mul_pd(c12_00,VV);
 244             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 245             fvdw12           = _mm_mul_pd(c12_00,FF);
 246             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 247             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 248
 249             /* Update potential sum for this i atom from the interaction with this j atom. */
 250             velecsum         = _mm_add_pd(velecsum,velec);
 251             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 252
 253             fscal            = _mm_add_pd(felec,fvdw);
 254
 255             /* Calculate temporary vectorial force */
 256             tx               = _mm_mul_pd(fscal,dx00);
 257             ty               = _mm_mul_pd(fscal,dy00);
 258             tz               = _mm_mul_pd(fscal,dz00);
 259
 260             /* Update vectorial force */
 261             fix0             = _mm_add_pd(fix0,tx);
 262             fiy0             = _mm_add_pd(fiy0,ty);
 263             fiz0             = _mm_add_pd(fiz0,tz);
 264
 265             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 266
 267             /* Inner loop uses 75 flops */
 268         }
 269
 270         if(jidx<j_index_end)
 271         {
 272
 273             jnrA             = jjnr[jidx];
 274             j_coord_offsetA  = DIM*jnrA;
 275
 276             /* load j atom coordinates */
 277             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 278                                               &jx0,&jy0,&jz0);
 279
 280             /* Calculate displacement vector */
 281             dx00             = _mm_sub_pd(ix0,jx0);
 282             dy00             = _mm_sub_pd(iy0,jy0);
 283             dz00             = _mm_sub_pd(iz0,jz0);
 284
 285             /* Calculate squared distance and things based on it */
 286             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 287
 288             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 289
 290             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 291
 292             /* Load parameters for j particles */
 293             jq0              = _mm_load_sd(charge+jnrA+0);
 294             vdwjidx0A        = 2*vdwtype[jnrA+0];
 295
 296             /**************************
 297              * CALCULATE INTERACTIONS *
 298              **************************/
 299
 300             r00              = _mm_mul_pd(rsq00,rinv00);
 301
 302             /* Compute parameters for interactions between i and j atoms */
 303             qq00             = _mm_mul_pd(iq0,jq0);
 304             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 305
 306             /* Calculate table index by multiplying r with table scale and truncate to integer */
 307             rt               = _mm_mul_pd(r00,vftabscale);
 308             vfitab           = _mm_cvttpd_epi32(rt);
 309             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 310             vfitab           = _mm_slli_epi32(vfitab,3);
 311
 312             /* EWALD ELECTROSTATICS */
 313
 314             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 315             ewrt             = _mm_mul_pd(r00,ewtabscale);
 316             ewitab           = _mm_cvttpd_epi32(ewrt);
 317             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 318             ewitab           = _mm_slli_epi32(ewitab,2);
 319             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 320             ewtabD           = _mm_setzero_pd();
 321             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 322             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 323             ewtabFn          = _mm_setzero_pd();
 324             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 325             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 326             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 327             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 328             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 329
 330             /* CUBIC SPLINE TABLE DISPERSION */
 331             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 332             F                = _mm_setzero_pd();
 333             GMX_MM_TRANSPOSE2_PD(Y,F);
 334             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 335             H                = _mm_setzero_pd();
 336             GMX_MM_TRANSPOSE2_PD(G,H);
 337             Heps             = _mm_mul_pd(vfeps,H);
 338             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 339             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 340             vvdw6            = _mm_mul_pd(c6_00,VV);
 341             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 342             fvdw6            = _mm_mul_pd(c6_00,FF);
 343
 344             /* CUBIC SPLINE TABLE REPULSION */
 345             vfitab           = _mm_add_epi32(vfitab,ifour);
 346             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 347             F                = _mm_setzero_pd();
 348             GMX_MM_TRANSPOSE2_PD(Y,F);
 349             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 350             H                = _mm_setzero_pd();
 351             GMX_MM_TRANSPOSE2_PD(G,H);
 352             Heps             = _mm_mul_pd(vfeps,H);
 353             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 354             VV               = _mm_add_pd(Y,_mm_mul_pd(vfeps,Fp));
 355             vvdw12           = _mm_mul_pd(c12_00,VV);
 356             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 357             fvdw12           = _mm_mul_pd(c12_00,FF);
 358             vvdw             = _mm_add_pd(vvdw12,vvdw6);
 359             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 363             velecsum         = _mm_add_pd(velecsum,velec);
 364             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 365             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 366
 367             fscal            = _mm_add_pd(felec,fvdw);
 368
 369             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 370
 371             /* Calculate temporary vectorial force */
 372             tx               = _mm_mul_pd(fscal,dx00);
 373             ty               = _mm_mul_pd(fscal,dy00);
 374             tz               = _mm_mul_pd(fscal,dz00);
 375
 376             /* Update vectorial force */
 377             fix0             = _mm_add_pd(fix0,tx);
 378             fiy0             = _mm_add_pd(fiy0,ty);
 379             fiz0             = _mm_add_pd(fiz0,tz);
 380
 381             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 382
 383             /* Inner loop uses 75 flops */
 384         }
 385
 386         /* End of innermost loop */
 387
 388         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 389                                               f+i_coord_offset,fshift+i_shift_offset);
 390
 391         ggid                        = gid[iidx];
 392         /* Update potential energies */
 393         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 394         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 395
 396         /* Increment number of inner iterations */
 397         inneriter                  += j_index_end - j_index_start;
 398
 399         /* Outer loop uses 9 flops */
 400     }
 401
 402     /* Increment number of outer iterations */
 403     outeriter        += nri;
 404
 405     /* Update outer/inner flops */
 406
 407     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*75);
 408 }
 409 /*
 410  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_double
 411  * Electrostatics interaction: Ewald
 412  * VdW interaction:            CubicSplineTable
 413  * Geometry:                   Particle-Particle
 414  * Calculate force/pot:        Force
 415  */
 416 void
 417 nb_kernel_ElecEw_VdwCSTab_GeomP1P1_F_sse4_1_double
 418                     (t_nblist * gmx_restrict                nlist,
 419                      rvec * gmx_restrict                    xx,
 420                      rvec * gmx_restrict                    ff,
 421                      t_forcerec * gmx_restrict              fr,
 422                      t_mdatoms * gmx_restrict               mdatoms,
 423                      nb_kernel_data_t * gmx_restrict        kernel_data,
 424                      t_nrnb * gmx_restrict                  nrnb)
 425 {
 426     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 427      * just 0 for non-waters.
 428      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 429      * jnr indices corresponding to data put in the four positions in the SIMD register.
 430      */
 431     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 432     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 433     int              jnrA,jnrB;
 434     int              j_coord_offsetA,j_coord_offsetB;
 435     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 436     real             rcutoff_scalar;
 437     real             *shiftvec,*fshift,*x,*f;
 438     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 439     int              vdwioffset0;
 440     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 441     int              vdwjidx0A,vdwjidx0B;
 442     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 443     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 444     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 445     real             *charge;
 446     int              nvdwtype;
 447     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 448     int              *vdwtype;
 449     real             *vdwparam;
 450     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 451     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 452     __m128i          vfitab;
 453     __m128i          ifour       = _mm_set1_epi32(4);
 454     __m128d          rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 455     real             *vftab;
 456     __m128i          ewitab;
 457     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 458     real             *ewtab;
 459     __m128d          dummy_mask,cutoff_mask;
 460     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 461     __m128d          one     = _mm_set1_pd(1.0);
 462     __m128d          two     = _mm_set1_pd(2.0);
 463     x                = xx[0];
 464     f                = ff[0];
 465
 466     nri              = nlist->nri;
 467     iinr             = nlist->iinr;
 468     jindex           = nlist->jindex;
 469     jjnr             = nlist->jjnr;
 470     shiftidx         = nlist->shift;
 471     gid              = nlist->gid;
 472     shiftvec         = fr->shift_vec[0];
 473     fshift           = fr->fshift[0];
 474     facel            = _mm_set1_pd(fr->epsfac);
 475     charge           = mdatoms->chargeA;
 476     nvdwtype         = fr->ntype;
 477     vdwparam         = fr->nbfp;
 478     vdwtype          = mdatoms->typeA;
 479
 480     vftab            = kernel_data->table_vdw->data;
 481     vftabscale       = _mm_set1_pd(kernel_data->table_vdw->scale);
 482
 483     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 484     ewtab            = fr->ic->tabq_coul_F;
 485     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 486     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 487
 488     /* Avoid stupid compiler warnings */
 489     jnrA = jnrB = 0;
 490     j_coord_offsetA = 0;
 491     j_coord_offsetB = 0;
 492
 493     outeriter        = 0;
 494     inneriter        = 0;
 495
 496     /* Start outer loop over neighborlists */
 497     for(iidx=0; iidx<nri; iidx++)
 498     {
 499         /* Load shift vector for this list */
 500         i_shift_offset   = DIM*shiftidx[iidx];
 501
 502         /* Load limits for loop over neighbors */
 503         j_index_start    = jindex[iidx];
 504         j_index_end      = jindex[iidx+1];
 505
 506         /* Get outer coordinate index */
 507         inr              = iinr[iidx];
 508         i_coord_offset   = DIM*inr;
 509
 510         /* Load i particle coords and add shift vector */
 511         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 512
 513         fix0             = _mm_setzero_pd();
 514         fiy0             = _mm_setzero_pd();
 515         fiz0             = _mm_setzero_pd();
 516
 517         /* Load parameters for i particles */
 518         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 519         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 520
 521         /* Start inner kernel loop */
 522         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 523         {
 524
 525             /* Get j neighbor index, and coordinate index */
 526             jnrA             = jjnr[jidx];
 527             jnrB             = jjnr[jidx+1];
 528             j_coord_offsetA  = DIM*jnrA;
 529             j_coord_offsetB  = DIM*jnrB;
 530
 531             /* load j atom coordinates */
 532             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 533                                               &jx0,&jy0,&jz0);
 534
 535             /* Calculate displacement vector */
 536             dx00             = _mm_sub_pd(ix0,jx0);
 537             dy00             = _mm_sub_pd(iy0,jy0);
 538             dz00             = _mm_sub_pd(iz0,jz0);
 539
 540             /* Calculate squared distance and things based on it */
 541             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 542
 543             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 544
 545             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 546
 547             /* Load parameters for j particles */
 548             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 549             vdwjidx0A        = 2*vdwtype[jnrA+0];
 550             vdwjidx0B        = 2*vdwtype[jnrB+0];
 551
 552             /**************************
 553              * CALCULATE INTERACTIONS *
 554              **************************/
 555
 556             r00              = _mm_mul_pd(rsq00,rinv00);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq00             = _mm_mul_pd(iq0,jq0);
 560             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 561                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 562
 563             /* Calculate table index by multiplying r with table scale and truncate to integer */
 564             rt               = _mm_mul_pd(r00,vftabscale);
 565             vfitab           = _mm_cvttpd_epi32(rt);
 566             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 567             vfitab           = _mm_slli_epi32(vfitab,3);
 568
 569             /* EWALD ELECTROSTATICS */
 570
 571             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 572             ewrt             = _mm_mul_pd(r00,ewtabscale);
 573             ewitab           = _mm_cvttpd_epi32(ewrt);
 574             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 575             gmx_mm_load_2pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 576                                          &ewtabF,&ewtabFn);
 577             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 578             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 579
 580             /* CUBIC SPLINE TABLE DISPERSION */
 581             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 582             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 583             GMX_MM_TRANSPOSE2_PD(Y,F);
 584             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 585             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 586             GMX_MM_TRANSPOSE2_PD(G,H);
 587             Heps             = _mm_mul_pd(vfeps,H);
 588             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 589             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 590             fvdw6            = _mm_mul_pd(c6_00,FF);
 591
 592             /* CUBIC SPLINE TABLE REPULSION */
 593             vfitab           = _mm_add_epi32(vfitab,ifour);
 594             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 595             F                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) );
 596             GMX_MM_TRANSPOSE2_PD(Y,F);
 597             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 598             H                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,1) +2);
 599             GMX_MM_TRANSPOSE2_PD(G,H);
 600             Heps             = _mm_mul_pd(vfeps,H);
 601             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 602             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 603             fvdw12           = _mm_mul_pd(c12_00,FF);
 604             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 605
 606             fscal            = _mm_add_pd(felec,fvdw);
 607
 608             /* Calculate temporary vectorial force */
 609             tx               = _mm_mul_pd(fscal,dx00);
 610             ty               = _mm_mul_pd(fscal,dy00);
 611             tz               = _mm_mul_pd(fscal,dz00);
 612
 613             /* Update vectorial force */
 614             fix0             = _mm_add_pd(fix0,tx);
 615             fiy0             = _mm_add_pd(fiy0,ty);
 616             fiz0             = _mm_add_pd(fiz0,tz);
 617
 618             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 619
 620             /* Inner loop uses 62 flops */
 621         }
 622
 623         if(jidx<j_index_end)
 624         {
 625
 626             jnrA             = jjnr[jidx];
 627             j_coord_offsetA  = DIM*jnrA;
 628
 629             /* load j atom coordinates */
 630             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 631                                               &jx0,&jy0,&jz0);
 632
 633             /* Calculate displacement vector */
 634             dx00             = _mm_sub_pd(ix0,jx0);
 635             dy00             = _mm_sub_pd(iy0,jy0);
 636             dz00             = _mm_sub_pd(iz0,jz0);
 637
 638             /* Calculate squared distance and things based on it */
 639             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 640
 641             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 642
 643             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 644
 645             /* Load parameters for j particles */
 646             jq0              = _mm_load_sd(charge+jnrA+0);
 647             vdwjidx0A        = 2*vdwtype[jnrA+0];
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             r00              = _mm_mul_pd(rsq00,rinv00);
 654
 655             /* Compute parameters for interactions between i and j atoms */
 656             qq00             = _mm_mul_pd(iq0,jq0);
 657             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 658
 659             /* Calculate table index by multiplying r with table scale and truncate to integer */
 660             rt               = _mm_mul_pd(r00,vftabscale);
 661             vfitab           = _mm_cvttpd_epi32(rt);
 662             vfeps            = _mm_sub_pd(rt,_mm_round_pd(rt, _MM_FROUND_FLOOR));
 663             vfitab           = _mm_slli_epi32(vfitab,3);
 664
 665             /* EWALD ELECTROSTATICS */
 666
 667             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 668             ewrt             = _mm_mul_pd(r00,ewtabscale);
 669             ewitab           = _mm_cvttpd_epi32(ewrt);
 670             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 671             gmx_mm_load_1pair_swizzle_pd(ewtab+gmx_mm_extract_epi32(ewitab,0),&ewtabF,&ewtabFn);
 672             felec            = _mm_add_pd(_mm_mul_pd( _mm_sub_pd(one,eweps),ewtabF),_mm_mul_pd(eweps,ewtabFn));
 673             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 674
 675             /* CUBIC SPLINE TABLE DISPERSION */
 676             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 677             F                = _mm_setzero_pd();
 678             GMX_MM_TRANSPOSE2_PD(Y,F);
 679             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 680             H                = _mm_setzero_pd();
 681             GMX_MM_TRANSPOSE2_PD(G,H);
 682             Heps             = _mm_mul_pd(vfeps,H);
 683             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 684             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 685             fvdw6            = _mm_mul_pd(c6_00,FF);
 686
 687             /* CUBIC SPLINE TABLE REPULSION */
 688             vfitab           = _mm_add_epi32(vfitab,ifour);
 689             Y                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) );
 690             F                = _mm_setzero_pd();
 691             GMX_MM_TRANSPOSE2_PD(Y,F);
 692             G                = _mm_load_pd( vftab + gmx_mm_extract_epi32(vfitab,0) +2);
 693             H                = _mm_setzero_pd();
 694             GMX_MM_TRANSPOSE2_PD(G,H);
 695             Heps             = _mm_mul_pd(vfeps,H);
 696             Fp               = _mm_add_pd(F,_mm_mul_pd(vfeps,_mm_add_pd(G,Heps)));
 697             FF               = _mm_add_pd(Fp,_mm_mul_pd(vfeps,_mm_add_pd(G,_mm_add_pd(Heps,Heps))));
 698             fvdw12           = _mm_mul_pd(c12_00,FF);
 699             fvdw             = _mm_xor_pd(signbit,_mm_mul_pd(_mm_add_pd(fvdw6,fvdw12),_mm_mul_pd(vftabscale,rinv00)));
 700
 701             fscal            = _mm_add_pd(felec,fvdw);
 702
 703             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 704
 705             /* Calculate temporary vectorial force */
 706             tx               = _mm_mul_pd(fscal,dx00);
 707             ty               = _mm_mul_pd(fscal,dy00);
 708             tz               = _mm_mul_pd(fscal,dz00);
 709
 710             /* Update vectorial force */
 711             fix0             = _mm_add_pd(fix0,tx);
 712             fiy0             = _mm_add_pd(fiy0,ty);
 713             fiz0             = _mm_add_pd(fiz0,tz);
 714
 715             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 716
 717             /* Inner loop uses 62 flops */
 718         }
 719
 720         /* End of innermost loop */
 721
 722         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 723                                               f+i_coord_offset,fshift+i_shift_offset);
 724
 725         /* Increment number of inner iterations */
 726         inneriter                  += j_index_end - j_index_start;
 727
 728         /* Outer loop uses 7 flops */
 729     }
 730
 731     /* Increment number of outer iterations */
 732     outeriter        += nri;
 733
 734     /* Update outer/inner flops */
 735
 736     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*62);
 737 }