src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSw_VdwLJSw_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_ElecEwSw_VdwLJSw_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwLJSw_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          ewitab;
  80     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  83     real             rswitch_scalar,d_scalar;
  84     __m128d          dummy_mask,cutoff_mask;
  85     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  86     __m128d          one     = _mm_set1_pd(1.0);
  87     __m128d          two     = _mm_set1_pd(2.0);
  88     x                = xx[0];
  89     f                = ff[0];
  90
  91     nri              = nlist->nri;
  92     iinr             = nlist->iinr;
  93     jindex           = nlist->jindex;
  94     jjnr             = nlist->jjnr;
  95     shiftidx         = nlist->shift;
  96     gid              = nlist->gid;
  97     shiftvec         = fr->shift_vec[0];
  98     fshift           = fr->fshift[0];
  99     facel            = _mm_set1_pd(fr->epsfac);
 100     charge           = mdatoms->chargeA;
 101     nvdwtype         = fr->ntype;
 102     vdwparam         = fr->nbfp;
 103     vdwtype          = mdatoms->typeA;
 104
 105     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 106     ewtab            = fr->ic->tabq_coul_FDV0;
 107     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 108     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 109
 110     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 111     rcutoff_scalar   = fr->rcoulomb;
 112     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 113     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 114
 115     rswitch_scalar   = fr->rcoulomb_switch;
 116     rswitch          = _mm_set1_pd(rswitch_scalar);
 117     /* Setup switch parameters */
 118     d_scalar         = rcutoff_scalar-rswitch_scalar;
 119     d                = _mm_set1_pd(d_scalar);
 120     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 121     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 122     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 123     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 124     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 125     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 126
 127     /* Avoid stupid compiler warnings */
 128     jnrA = jnrB = 0;
 129     j_coord_offsetA = 0;
 130     j_coord_offsetB = 0;
 131
 132     outeriter        = 0;
 133     inneriter        = 0;
 134
 135     /* Start outer loop over neighborlists */
 136     for(iidx=0; iidx<nri; iidx++)
 137     {
 138         /* Load shift vector for this list */
 139         i_shift_offset   = DIM*shiftidx[iidx];
 140
 141         /* Load limits for loop over neighbors */
 142         j_index_start    = jindex[iidx];
 143         j_index_end      = jindex[iidx+1];
 144
 145         /* Get outer coordinate index */
 146         inr              = iinr[iidx];
 147         i_coord_offset   = DIM*inr;
 148
 149         /* Load i particle coords and add shift vector */
 150         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 151
 152         fix0             = _mm_setzero_pd();
 153         fiy0             = _mm_setzero_pd();
 154         fiz0             = _mm_setzero_pd();
 155
 156         /* Load parameters for i particles */
 157         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 158         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 159
 160         /* Reset potential sums */
 161         velecsum         = _mm_setzero_pd();
 162         vvdwsum          = _mm_setzero_pd();
 163
 164         /* Start inner kernel loop */
 165         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 166         {
 167
 168             /* Get j neighbor index, and coordinate index */
 169             jnrA             = jjnr[jidx];
 170             jnrB             = jjnr[jidx+1];
 171             j_coord_offsetA  = DIM*jnrA;
 172             j_coord_offsetB  = DIM*jnrB;
 173
 174             /* load j atom coordinates */
 175             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 176                                               &jx0,&jy0,&jz0);
 177
 178             /* Calculate displacement vector */
 179             dx00             = _mm_sub_pd(ix0,jx0);
 180             dy00             = _mm_sub_pd(iy0,jy0);
 181             dz00             = _mm_sub_pd(iz0,jz0);
 182
 183             /* Calculate squared distance and things based on it */
 184             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 185
 186             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 187
 188             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 189
 190             /* Load parameters for j particles */
 191             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 192             vdwjidx0A        = 2*vdwtype[jnrA+0];
 193             vdwjidx0B        = 2*vdwtype[jnrB+0];
 194
 195             /**************************
 196              * CALCULATE INTERACTIONS *
 197              **************************/
 198
 199             if (gmx_mm_any_lt(rsq00,rcutoff2))
 200             {
 201
 202             r00              = _mm_mul_pd(rsq00,rinv00);
 203
 204             /* Compute parameters for interactions between i and j atoms */
 205             qq00             = _mm_mul_pd(iq0,jq0);
 206             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 207                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 208
 209             /* EWALD ELECTROSTATICS */
 210
 211             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 212             ewrt             = _mm_mul_pd(r00,ewtabscale);
 213             ewitab           = _mm_cvttpd_epi32(ewrt);
 214             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 215             ewitab           = _mm_slli_epi32(ewitab,2);
 216             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 217             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 218             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 219             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 220             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 221             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 222             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 223             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 224             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 225             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 226
 227             /* LENNARD-JONES DISPERSION/REPULSION */
 228
 229             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 230             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 231             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 232             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 233             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 234
 235             d                = _mm_sub_pd(r00,rswitch);
 236             d                = _mm_max_pd(d,_mm_setzero_pd());
 237             d2               = _mm_mul_pd(d,d);
 238             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 239
 240             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 241
 242             /* Evaluate switch function */
 243             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 244             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 245             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 246             velec            = _mm_mul_pd(velec,sw);
 247             vvdw             = _mm_mul_pd(vvdw,sw);
 248             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 249
 250             /* Update potential sum for this i atom from the interaction with this j atom. */
 251             velec            = _mm_and_pd(velec,cutoff_mask);
 252             velecsum         = _mm_add_pd(velecsum,velec);
 253             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 254             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 255
 256             fscal            = _mm_add_pd(felec,fvdw);
 257
 258             fscal            = _mm_and_pd(fscal,cutoff_mask);
 259
 260             /* Calculate temporary vectorial force */
 261             tx               = _mm_mul_pd(fscal,dx00);
 262             ty               = _mm_mul_pd(fscal,dy00);
 263             tz               = _mm_mul_pd(fscal,dz00);
 264
 265             /* Update vectorial force */
 266             fix0             = _mm_add_pd(fix0,tx);
 267             fiy0             = _mm_add_pd(fiy0,ty);
 268             fiz0             = _mm_add_pd(fiz0,tz);
 269
 270             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 271
 272             }
 273
 274             /* Inner loop uses 83 flops */
 275         }
 276
 277         if(jidx<j_index_end)
 278         {
 279
 280             jnrA             = jjnr[jidx];
 281             j_coord_offsetA  = DIM*jnrA;
 282
 283             /* load j atom coordinates */
 284             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 285                                               &jx0,&jy0,&jz0);
 286
 287             /* Calculate displacement vector */
 288             dx00             = _mm_sub_pd(ix0,jx0);
 289             dy00             = _mm_sub_pd(iy0,jy0);
 290             dz00             = _mm_sub_pd(iz0,jz0);
 291
 292             /* Calculate squared distance and things based on it */
 293             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 294
 295             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 296
 297             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 298
 299             /* Load parameters for j particles */
 300             jq0              = _mm_load_sd(charge+jnrA+0);
 301             vdwjidx0A        = 2*vdwtype[jnrA+0];
 302
 303             /**************************
 304              * CALCULATE INTERACTIONS *
 305              **************************/
 306
 307             if (gmx_mm_any_lt(rsq00,rcutoff2))
 308             {
 309
 310             r00              = _mm_mul_pd(rsq00,rinv00);
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq00             = _mm_mul_pd(iq0,jq0);
 314             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 315
 316             /* EWALD ELECTROSTATICS */
 317
 318             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 319             ewrt             = _mm_mul_pd(r00,ewtabscale);
 320             ewitab           = _mm_cvttpd_epi32(ewrt);
 321             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 322             ewitab           = _mm_slli_epi32(ewitab,2);
 323             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 324             ewtabD           = _mm_setzero_pd();
 325             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 326             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 327             ewtabFn          = _mm_setzero_pd();
 328             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 329             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 330             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 331             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 332             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 333
 334             /* LENNARD-JONES DISPERSION/REPULSION */
 335
 336             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 337             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 338             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 339             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 340             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 341
 342             d                = _mm_sub_pd(r00,rswitch);
 343             d                = _mm_max_pd(d,_mm_setzero_pd());
 344             d2               = _mm_mul_pd(d,d);
 345             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 346
 347             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 348
 349             /* Evaluate switch function */
 350             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 351             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 352             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 353             velec            = _mm_mul_pd(velec,sw);
 354             vvdw             = _mm_mul_pd(vvdw,sw);
 355             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 356
 357             /* Update potential sum for this i atom from the interaction with this j atom. */
 358             velec            = _mm_and_pd(velec,cutoff_mask);
 359             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 360             velecsum         = _mm_add_pd(velecsum,velec);
 361             vvdw             = _mm_and_pd(vvdw,cutoff_mask);
 362             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 363             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 364
 365             fscal            = _mm_add_pd(felec,fvdw);
 366
 367             fscal            = _mm_and_pd(fscal,cutoff_mask);
 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             }
 384
 385             /* Inner loop uses 83 flops */
 386         }
 387
 388         /* End of innermost loop */
 389
 390         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 391                                               f+i_coord_offset,fshift+i_shift_offset);
 392
 393         ggid                        = gid[iidx];
 394         /* Update potential energies */
 395         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 396         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 397
 398         /* Increment number of inner iterations */
 399         inneriter                  += j_index_end - j_index_start;
 400
 401         /* Outer loop uses 9 flops */
 402     }
 403
 404     /* Increment number of outer iterations */
 405     outeriter        += nri;
 406
 407     /* Update outer/inner flops */
 408
 409     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*83);
 410 }
 411 /*
 412  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_double
 413  * Electrostatics interaction: Ewald
 414  * VdW interaction:            LennardJones
 415  * Geometry:                   Particle-Particle
 416  * Calculate force/pot:        Force
 417  */
 418 void
 419 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_sse4_1_double
 420                     (t_nblist * gmx_restrict                nlist,
 421                      rvec * gmx_restrict                    xx,
 422                      rvec * gmx_restrict                    ff,
 423                      t_forcerec * gmx_restrict              fr,
 424                      t_mdatoms * gmx_restrict               mdatoms,
 425                      nb_kernel_data_t * gmx_restrict        kernel_data,
 426                      t_nrnb * gmx_restrict                  nrnb)
 427 {
 428     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 429      * just 0 for non-waters.
 430      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 431      * jnr indices corresponding to data put in the four positions in the SIMD register.
 432      */
 433     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 434     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 435     int              jnrA,jnrB;
 436     int              j_coord_offsetA,j_coord_offsetB;
 437     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 438     real             rcutoff_scalar;
 439     real             *shiftvec,*fshift,*x,*f;
 440     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 441     int              vdwioffset0;
 442     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 443     int              vdwjidx0A,vdwjidx0B;
 444     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 445     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 446     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 447     real             *charge;
 448     int              nvdwtype;
 449     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 450     int              *vdwtype;
 451     real             *vdwparam;
 452     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 453     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 454     __m128i          ewitab;
 455     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 456     real             *ewtab;
 457     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 458     real             rswitch_scalar,d_scalar;
 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     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 481     ewtab            = fr->ic->tabq_coul_FDV0;
 482     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 483     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 484
 485     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 486     rcutoff_scalar   = fr->rcoulomb;
 487     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 488     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 489
 490     rswitch_scalar   = fr->rcoulomb_switch;
 491     rswitch          = _mm_set1_pd(rswitch_scalar);
 492     /* Setup switch parameters */
 493     d_scalar         = rcutoff_scalar-rswitch_scalar;
 494     d                = _mm_set1_pd(d_scalar);
 495     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 496     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 497     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 498     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 499     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 500     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 501
 502     /* Avoid stupid compiler warnings */
 503     jnrA = jnrB = 0;
 504     j_coord_offsetA = 0;
 505     j_coord_offsetB = 0;
 506
 507     outeriter        = 0;
 508     inneriter        = 0;
 509
 510     /* Start outer loop over neighborlists */
 511     for(iidx=0; iidx<nri; iidx++)
 512     {
 513         /* Load shift vector for this list */
 514         i_shift_offset   = DIM*shiftidx[iidx];
 515
 516         /* Load limits for loop over neighbors */
 517         j_index_start    = jindex[iidx];
 518         j_index_end      = jindex[iidx+1];
 519
 520         /* Get outer coordinate index */
 521         inr              = iinr[iidx];
 522         i_coord_offset   = DIM*inr;
 523
 524         /* Load i particle coords and add shift vector */
 525         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 526
 527         fix0             = _mm_setzero_pd();
 528         fiy0             = _mm_setzero_pd();
 529         fiz0             = _mm_setzero_pd();
 530
 531         /* Load parameters for i particles */
 532         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 533         vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 534
 535         /* Start inner kernel loop */
 536         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 537         {
 538
 539             /* Get j neighbor index, and coordinate index */
 540             jnrA             = jjnr[jidx];
 541             jnrB             = jjnr[jidx+1];
 542             j_coord_offsetA  = DIM*jnrA;
 543             j_coord_offsetB  = DIM*jnrB;
 544
 545             /* load j atom coordinates */
 546             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 547                                               &jx0,&jy0,&jz0);
 548
 549             /* Calculate displacement vector */
 550             dx00             = _mm_sub_pd(ix0,jx0);
 551             dy00             = _mm_sub_pd(iy0,jy0);
 552             dz00             = _mm_sub_pd(iz0,jz0);
 553
 554             /* Calculate squared distance and things based on it */
 555             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 556
 557             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 558
 559             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 560
 561             /* Load parameters for j particles */
 562             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 563             vdwjidx0A        = 2*vdwtype[jnrA+0];
 564             vdwjidx0B        = 2*vdwtype[jnrB+0];
 565
 566             /**************************
 567              * CALCULATE INTERACTIONS *
 568              **************************/
 569
 570             if (gmx_mm_any_lt(rsq00,rcutoff2))
 571             {
 572
 573             r00              = _mm_mul_pd(rsq00,rinv00);
 574
 575             /* Compute parameters for interactions between i and j atoms */
 576             qq00             = _mm_mul_pd(iq0,jq0);
 577             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 578                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 579
 580             /* EWALD ELECTROSTATICS */
 581
 582             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 583             ewrt             = _mm_mul_pd(r00,ewtabscale);
 584             ewitab           = _mm_cvttpd_epi32(ewrt);
 585             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 586             ewitab           = _mm_slli_epi32(ewitab,2);
 587             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 588             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 589             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 590             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 591             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 592             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 593             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 594             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 595             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 596             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 597
 598             /* LENNARD-JONES DISPERSION/REPULSION */
 599
 600             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 601             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 602             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 603             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 604             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 605
 606             d                = _mm_sub_pd(r00,rswitch);
 607             d                = _mm_max_pd(d,_mm_setzero_pd());
 608             d2               = _mm_mul_pd(d,d);
 609             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 610
 611             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 612
 613             /* Evaluate switch function */
 614             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 615             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 616             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 617             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 618
 619             fscal            = _mm_add_pd(felec,fvdw);
 620
 621             fscal            = _mm_and_pd(fscal,cutoff_mask);
 622
 623             /* Calculate temporary vectorial force */
 624             tx               = _mm_mul_pd(fscal,dx00);
 625             ty               = _mm_mul_pd(fscal,dy00);
 626             tz               = _mm_mul_pd(fscal,dz00);
 627
 628             /* Update vectorial force */
 629             fix0             = _mm_add_pd(fix0,tx);
 630             fiy0             = _mm_add_pd(fiy0,ty);
 631             fiz0             = _mm_add_pd(fiz0,tz);
 632
 633             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 634
 635             }
 636
 637             /* Inner loop uses 77 flops */
 638         }
 639
 640         if(jidx<j_index_end)
 641         {
 642
 643             jnrA             = jjnr[jidx];
 644             j_coord_offsetA  = DIM*jnrA;
 645
 646             /* load j atom coordinates */
 647             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 648                                               &jx0,&jy0,&jz0);
 649
 650             /* Calculate displacement vector */
 651             dx00             = _mm_sub_pd(ix0,jx0);
 652             dy00             = _mm_sub_pd(iy0,jy0);
 653             dz00             = _mm_sub_pd(iz0,jz0);
 654
 655             /* Calculate squared distance and things based on it */
 656             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 657
 658             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 659
 660             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 661
 662             /* Load parameters for j particles */
 663             jq0              = _mm_load_sd(charge+jnrA+0);
 664             vdwjidx0A        = 2*vdwtype[jnrA+0];
 665
 666             /**************************
 667              * CALCULATE INTERACTIONS *
 668              **************************/
 669
 670             if (gmx_mm_any_lt(rsq00,rcutoff2))
 671             {
 672
 673             r00              = _mm_mul_pd(rsq00,rinv00);
 674
 675             /* Compute parameters for interactions between i and j atoms */
 676             qq00             = _mm_mul_pd(iq0,jq0);
 677             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 678
 679             /* EWALD ELECTROSTATICS */
 680
 681             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 682             ewrt             = _mm_mul_pd(r00,ewtabscale);
 683             ewitab           = _mm_cvttpd_epi32(ewrt);
 684             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 685             ewitab           = _mm_slli_epi32(ewitab,2);
 686             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 687             ewtabD           = _mm_setzero_pd();
 688             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 689             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 690             ewtabFn          = _mm_setzero_pd();
 691             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 692             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 693             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 694             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 695             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 696
 697             /* LENNARD-JONES DISPERSION/REPULSION */
 698
 699             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 700             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 701             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 702             vvdw             = _mm_sub_pd( _mm_mul_pd(vvdw12,one_twelfth) , _mm_mul_pd(vvdw6,one_sixth) );
 703             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 704
 705             d                = _mm_sub_pd(r00,rswitch);
 706             d                = _mm_max_pd(d,_mm_setzero_pd());
 707             d2               = _mm_mul_pd(d,d);
 708             sw               = _mm_add_pd(one,_mm_mul_pd(d2,_mm_mul_pd(d,_mm_add_pd(swV3,_mm_mul_pd(d,_mm_add_pd(swV4,_mm_mul_pd(d,swV5)))))));
 709
 710             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 711
 712             /* Evaluate switch function */
 713             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 714             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 715             fvdw             = _mm_sub_pd( _mm_mul_pd(fvdw,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(vvdw,dsw)) );
 716             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 717
 718             fscal            = _mm_add_pd(felec,fvdw);
 719
 720             fscal            = _mm_and_pd(fscal,cutoff_mask);
 721
 722             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm_mul_pd(fscal,dx00);
 726             ty               = _mm_mul_pd(fscal,dy00);
 727             tz               = _mm_mul_pd(fscal,dz00);
 728
 729             /* Update vectorial force */
 730             fix0             = _mm_add_pd(fix0,tx);
 731             fiy0             = _mm_add_pd(fiy0,ty);
 732             fiz0             = _mm_add_pd(fiz0,tz);
 733
 734             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 735
 736             }
 737
 738             /* Inner loop uses 77 flops */
 739         }
 740
 741         /* End of innermost loop */
 742
 743         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 744                                               f+i_coord_offset,fshift+i_shift_offset);
 745
 746         /* Increment number of inner iterations */
 747         inneriter                  += j_index_end - j_index_start;
 748
 749         /* Outer loop uses 7 flops */
 750     }
 751
 752     /* Increment number of outer iterations */
 753     outeriter        += nri;
 754
 755     /* Update outer/inner flops */
 756
 757     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*77);
 758 }