src/gmxlib/nonbonded/nb_kernel_sse4_1_double/nb_kernel_ElecEwSw_VdwNone_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_VdwNone_GeomP1P1_VF_sse4_1_double
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Particle-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEwSw_VdwNone_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     __m128i          ewitab;
  74     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  75     real             *ewtab;
  76     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
  77     real             rswitch_scalar,d_scalar;
  78     __m128d          dummy_mask,cutoff_mask;
  79     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  80     __m128d          one     = _mm_set1_pd(1.0);
  81     __m128d          two     = _mm_set1_pd(2.0);
  82     x                = xx[0];
  83     f                = ff[0];
  84
  85     nri              = nlist->nri;
  86     iinr             = nlist->iinr;
  87     jindex           = nlist->jindex;
  88     jjnr             = nlist->jjnr;
  89     shiftidx         = nlist->shift;
  90     gid              = nlist->gid;
  91     shiftvec         = fr->shift_vec[0];
  92     fshift           = fr->fshift[0];
  93     facel            = _mm_set1_pd(fr->epsfac);
  94     charge           = mdatoms->chargeA;
  95
  96     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
  97     ewtab            = fr->ic->tabq_coul_FDV0;
  98     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
  99     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 100
 101     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 102     rcutoff_scalar   = fr->rcoulomb;
 103     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 104     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 105
 106     rswitch_scalar   = fr->rcoulomb_switch;
 107     rswitch          = _mm_set1_pd(rswitch_scalar);
 108     /* Setup switch parameters */
 109     d_scalar         = rcutoff_scalar-rswitch_scalar;
 110     d                = _mm_set1_pd(d_scalar);
 111     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 112     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 113     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 114     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 115     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 116     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 117
 118     /* Avoid stupid compiler warnings */
 119     jnrA = jnrB = 0;
 120     j_coord_offsetA = 0;
 121     j_coord_offsetB = 0;
 122
 123     outeriter        = 0;
 124     inneriter        = 0;
 125
 126     /* Start outer loop over neighborlists */
 127     for(iidx=0; iidx<nri; iidx++)
 128     {
 129         /* Load shift vector for this list */
 130         i_shift_offset   = DIM*shiftidx[iidx];
 131
 132         /* Load limits for loop over neighbors */
 133         j_index_start    = jindex[iidx];
 134         j_index_end      = jindex[iidx+1];
 135
 136         /* Get outer coordinate index */
 137         inr              = iinr[iidx];
 138         i_coord_offset   = DIM*inr;
 139
 140         /* Load i particle coords and add shift vector */
 141         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 142
 143         fix0             = _mm_setzero_pd();
 144         fiy0             = _mm_setzero_pd();
 145         fiz0             = _mm_setzero_pd();
 146
 147         /* Load parameters for i particles */
 148         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 149
 150         /* Reset potential sums */
 151         velecsum         = _mm_setzero_pd();
 152
 153         /* Start inner kernel loop */
 154         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 155         {
 156
 157             /* Get j neighbor index, and coordinate index */
 158             jnrA             = jjnr[jidx];
 159             jnrB             = jjnr[jidx+1];
 160             j_coord_offsetA  = DIM*jnrA;
 161             j_coord_offsetB  = DIM*jnrB;
 162
 163             /* load j atom coordinates */
 164             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 165                                               &jx0,&jy0,&jz0);
 166
 167             /* Calculate displacement vector */
 168             dx00             = _mm_sub_pd(ix0,jx0);
 169             dy00             = _mm_sub_pd(iy0,jy0);
 170             dz00             = _mm_sub_pd(iz0,jz0);
 171
 172             /* Calculate squared distance and things based on it */
 173             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 174
 175             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 176
 177             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 178
 179             /* Load parameters for j particles */
 180             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 181
 182             /**************************
 183              * CALCULATE INTERACTIONS *
 184              **************************/
 185
 186             if (gmx_mm_any_lt(rsq00,rcutoff2))
 187             {
 188
 189             r00              = _mm_mul_pd(rsq00,rinv00);
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* EWALD ELECTROSTATICS */
 195
 196             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 197             ewrt             = _mm_mul_pd(r00,ewtabscale);
 198             ewitab           = _mm_cvttpd_epi32(ewrt);
 199             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 200             ewitab           = _mm_slli_epi32(ewitab,2);
 201             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 202             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 203             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 204             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 205             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 206             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 207             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 208             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 209             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 210             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 211
 212             d                = _mm_sub_pd(r00,rswitch);
 213             d                = _mm_max_pd(d,_mm_setzero_pd());
 214             d2               = _mm_mul_pd(d,d);
 215             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)))))));
 216
 217             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 218
 219             /* Evaluate switch function */
 220             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 221             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 222             velec            = _mm_mul_pd(velec,sw);
 223             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 224
 225             /* Update potential sum for this i atom from the interaction with this j atom. */
 226             velec            = _mm_and_pd(velec,cutoff_mask);
 227             velecsum         = _mm_add_pd(velecsum,velec);
 228
 229             fscal            = felec;
 230
 231             fscal            = _mm_and_pd(fscal,cutoff_mask);
 232
 233             /* Calculate temporary vectorial force */
 234             tx               = _mm_mul_pd(fscal,dx00);
 235             ty               = _mm_mul_pd(fscal,dy00);
 236             tz               = _mm_mul_pd(fscal,dz00);
 237
 238             /* Update vectorial force */
 239             fix0             = _mm_add_pd(fix0,tx);
 240             fiy0             = _mm_add_pd(fiy0,ty);
 241             fiz0             = _mm_add_pd(fiz0,tz);
 242
 243             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 244
 245             }
 246
 247             /* Inner loop uses 65 flops */
 248         }
 249
 250         if(jidx<j_index_end)
 251         {
 252
 253             jnrA             = jjnr[jidx];
 254             j_coord_offsetA  = DIM*jnrA;
 255
 256             /* load j atom coordinates */
 257             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 258                                               &jx0,&jy0,&jz0);
 259
 260             /* Calculate displacement vector */
 261             dx00             = _mm_sub_pd(ix0,jx0);
 262             dy00             = _mm_sub_pd(iy0,jy0);
 263             dz00             = _mm_sub_pd(iz0,jz0);
 264
 265             /* Calculate squared distance and things based on it */
 266             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 267
 268             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 269
 270             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 271
 272             /* Load parameters for j particles */
 273             jq0              = _mm_load_sd(charge+jnrA+0);
 274
 275             /**************************
 276              * CALCULATE INTERACTIONS *
 277              **************************/
 278
 279             if (gmx_mm_any_lt(rsq00,rcutoff2))
 280             {
 281
 282             r00              = _mm_mul_pd(rsq00,rinv00);
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq00             = _mm_mul_pd(iq0,jq0);
 286
 287             /* EWALD ELECTROSTATICS */
 288
 289             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 290             ewrt             = _mm_mul_pd(r00,ewtabscale);
 291             ewitab           = _mm_cvttpd_epi32(ewrt);
 292             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 293             ewitab           = _mm_slli_epi32(ewitab,2);
 294             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 295             ewtabD           = _mm_setzero_pd();
 296             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 297             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 298             ewtabFn          = _mm_setzero_pd();
 299             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 300             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 301             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 302             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 303             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 304
 305             d                = _mm_sub_pd(r00,rswitch);
 306             d                = _mm_max_pd(d,_mm_setzero_pd());
 307             d2               = _mm_mul_pd(d,d);
 308             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)))))));
 309
 310             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 311
 312             /* Evaluate switch function */
 313             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 314             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 315             velec            = _mm_mul_pd(velec,sw);
 316             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_and_pd(velec,cutoff_mask);
 320             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_and_pd(fscal,cutoff_mask);
 326
 327             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 328
 329             /* Calculate temporary vectorial force */
 330             tx               = _mm_mul_pd(fscal,dx00);
 331             ty               = _mm_mul_pd(fscal,dy00);
 332             tz               = _mm_mul_pd(fscal,dz00);
 333
 334             /* Update vectorial force */
 335             fix0             = _mm_add_pd(fix0,tx);
 336             fiy0             = _mm_add_pd(fiy0,ty);
 337             fiz0             = _mm_add_pd(fiz0,tz);
 338
 339             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 340
 341             }
 342
 343             /* Inner loop uses 65 flops */
 344         }
 345
 346         /* End of innermost loop */
 347
 348         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 349                                               f+i_coord_offset,fshift+i_shift_offset);
 350
 351         ggid                        = gid[iidx];
 352         /* Update potential energies */
 353         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 354
 355         /* Increment number of inner iterations */
 356         inneriter                  += j_index_end - j_index_start;
 357
 358         /* Outer loop uses 8 flops */
 359     }
 360
 361     /* Increment number of outer iterations */
 362     outeriter        += nri;
 363
 364     /* Update outer/inner flops */
 365
 366     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VF,outeriter*8 + inneriter*65);
 367 }
 368 /*
 369  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse4_1_double
 370  * Electrostatics interaction: Ewald
 371  * VdW interaction:            None
 372  * Geometry:                   Particle-Particle
 373  * Calculate force/pot:        Force
 374  */
 375 void
 376 nb_kernel_ElecEwSw_VdwNone_GeomP1P1_F_sse4_1_double
 377                     (t_nblist * gmx_restrict                nlist,
 378                      rvec * gmx_restrict                    xx,
 379                      rvec * gmx_restrict                    ff,
 380                      t_forcerec * gmx_restrict              fr,
 381                      t_mdatoms * gmx_restrict               mdatoms,
 382                      nb_kernel_data_t * gmx_restrict        kernel_data,
 383                      t_nrnb * gmx_restrict                  nrnb)
 384 {
 385     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 386      * just 0 for non-waters.
 387      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 388      * jnr indices corresponding to data put in the four positions in the SIMD register.
 389      */
 390     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 391     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 392     int              jnrA,jnrB;
 393     int              j_coord_offsetA,j_coord_offsetB;
 394     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 395     real             rcutoff_scalar;
 396     real             *shiftvec,*fshift,*x,*f;
 397     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 398     int              vdwioffset0;
 399     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 400     int              vdwjidx0A,vdwjidx0B;
 401     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 402     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 403     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 404     real             *charge;
 405     __m128i          ewitab;
 406     __m128d          ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 407     real             *ewtab;
 408     __m128d          rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 409     real             rswitch_scalar,d_scalar;
 410     __m128d          dummy_mask,cutoff_mask;
 411     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 412     __m128d          one     = _mm_set1_pd(1.0);
 413     __m128d          two     = _mm_set1_pd(2.0);
 414     x                = xx[0];
 415     f                = ff[0];
 416
 417     nri              = nlist->nri;
 418     iinr             = nlist->iinr;
 419     jindex           = nlist->jindex;
 420     jjnr             = nlist->jjnr;
 421     shiftidx         = nlist->shift;
 422     gid              = nlist->gid;
 423     shiftvec         = fr->shift_vec[0];
 424     fshift           = fr->fshift[0];
 425     facel            = _mm_set1_pd(fr->epsfac);
 426     charge           = mdatoms->chargeA;
 427
 428     sh_ewald         = _mm_set1_pd(fr->ic->sh_ewald);
 429     ewtab            = fr->ic->tabq_coul_FDV0;
 430     ewtabscale       = _mm_set1_pd(fr->ic->tabq_scale);
 431     ewtabhalfspace   = _mm_set1_pd(0.5/fr->ic->tabq_scale);
 432
 433     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 434     rcutoff_scalar   = fr->rcoulomb;
 435     rcutoff          = _mm_set1_pd(rcutoff_scalar);
 436     rcutoff2         = _mm_mul_pd(rcutoff,rcutoff);
 437
 438     rswitch_scalar   = fr->rcoulomb_switch;
 439     rswitch          = _mm_set1_pd(rswitch_scalar);
 440     /* Setup switch parameters */
 441     d_scalar         = rcutoff_scalar-rswitch_scalar;
 442     d                = _mm_set1_pd(d_scalar);
 443     swV3             = _mm_set1_pd(-10.0/(d_scalar*d_scalar*d_scalar));
 444     swV4             = _mm_set1_pd( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 445     swV5             = _mm_set1_pd( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 446     swF2             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar));
 447     swF3             = _mm_set1_pd( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 448     swF4             = _mm_set1_pd(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 449
 450     /* Avoid stupid compiler warnings */
 451     jnrA = jnrB = 0;
 452     j_coord_offsetA = 0;
 453     j_coord_offsetB = 0;
 454
 455     outeriter        = 0;
 456     inneriter        = 0;
 457
 458     /* Start outer loop over neighborlists */
 459     for(iidx=0; iidx<nri; iidx++)
 460     {
 461         /* Load shift vector for this list */
 462         i_shift_offset   = DIM*shiftidx[iidx];
 463
 464         /* Load limits for loop over neighbors */
 465         j_index_start    = jindex[iidx];
 466         j_index_end      = jindex[iidx+1];
 467
 468         /* Get outer coordinate index */
 469         inr              = iinr[iidx];
 470         i_coord_offset   = DIM*inr;
 471
 472         /* Load i particle coords and add shift vector */
 473         gmx_mm_load_shift_and_1rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 474
 475         fix0             = _mm_setzero_pd();
 476         fiy0             = _mm_setzero_pd();
 477         fiz0             = _mm_setzero_pd();
 478
 479         /* Load parameters for i particles */
 480         iq0              = _mm_mul_pd(facel,_mm_load1_pd(charge+inr+0));
 481
 482         /* Start inner kernel loop */
 483         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 484         {
 485
 486             /* Get j neighbor index, and coordinate index */
 487             jnrA             = jjnr[jidx];
 488             jnrB             = jjnr[jidx+1];
 489             j_coord_offsetA  = DIM*jnrA;
 490             j_coord_offsetB  = DIM*jnrB;
 491
 492             /* load j atom coordinates */
 493             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 494                                               &jx0,&jy0,&jz0);
 495
 496             /* Calculate displacement vector */
 497             dx00             = _mm_sub_pd(ix0,jx0);
 498             dy00             = _mm_sub_pd(iy0,jy0);
 499             dz00             = _mm_sub_pd(iz0,jz0);
 500
 501             /* Calculate squared distance and things based on it */
 502             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 503
 504             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 505
 506             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 507
 508             /* Load parameters for j particles */
 509             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 510
 511             /**************************
 512              * CALCULATE INTERACTIONS *
 513              **************************/
 514
 515             if (gmx_mm_any_lt(rsq00,rcutoff2))
 516             {
 517
 518             r00              = _mm_mul_pd(rsq00,rinv00);
 519
 520             /* Compute parameters for interactions between i and j atoms */
 521             qq00             = _mm_mul_pd(iq0,jq0);
 522
 523             /* EWALD ELECTROSTATICS */
 524
 525             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 526             ewrt             = _mm_mul_pd(r00,ewtabscale);
 527             ewitab           = _mm_cvttpd_epi32(ewrt);
 528             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 529             ewitab           = _mm_slli_epi32(ewitab,2);
 530             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 531             ewtabD           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 532             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 533             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 534             ewtabFn          = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,1) +2);
 535             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 536             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 537             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 538             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 539             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 540
 541             d                = _mm_sub_pd(r00,rswitch);
 542             d                = _mm_max_pd(d,_mm_setzero_pd());
 543             d2               = _mm_mul_pd(d,d);
 544             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)))))));
 545
 546             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 547
 548             /* Evaluate switch function */
 549             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 550             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 551             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_and_pd(fscal,cutoff_mask);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_pd(fscal,dx00);
 559             ty               = _mm_mul_pd(fscal,dy00);
 560             tz               = _mm_mul_pd(fscal,dz00);
 561
 562             /* Update vectorial force */
 563             fix0             = _mm_add_pd(fix0,tx);
 564             fiy0             = _mm_add_pd(fiy0,ty);
 565             fiz0             = _mm_add_pd(fiz0,tz);
 566
 567             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,tx,ty,tz);
 568
 569             }
 570
 571             /* Inner loop uses 62 flops */
 572         }
 573
 574         if(jidx<j_index_end)
 575         {
 576
 577             jnrA             = jjnr[jidx];
 578             j_coord_offsetA  = DIM*jnrA;
 579
 580             /* load j atom coordinates */
 581             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 582                                               &jx0,&jy0,&jz0);
 583
 584             /* Calculate displacement vector */
 585             dx00             = _mm_sub_pd(ix0,jx0);
 586             dy00             = _mm_sub_pd(iy0,jy0);
 587             dz00             = _mm_sub_pd(iz0,jz0);
 588
 589             /* Calculate squared distance and things based on it */
 590             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 591
 592             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 593
 594             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 595
 596             /* Load parameters for j particles */
 597             jq0              = _mm_load_sd(charge+jnrA+0);
 598
 599             /**************************
 600              * CALCULATE INTERACTIONS *
 601              **************************/
 602
 603             if (gmx_mm_any_lt(rsq00,rcutoff2))
 604             {
 605
 606             r00              = _mm_mul_pd(rsq00,rinv00);
 607
 608             /* Compute parameters for interactions between i and j atoms */
 609             qq00             = _mm_mul_pd(iq0,jq0);
 610
 611             /* EWALD ELECTROSTATICS */
 612
 613             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 614             ewrt             = _mm_mul_pd(r00,ewtabscale);
 615             ewitab           = _mm_cvttpd_epi32(ewrt);
 616             eweps            = _mm_sub_pd(ewrt,_mm_round_pd(ewrt, _MM_FROUND_FLOOR));
 617             ewitab           = _mm_slli_epi32(ewitab,2);
 618             ewtabF           = _mm_load_pd( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 619             ewtabD           = _mm_setzero_pd();
 620             GMX_MM_TRANSPOSE2_PD(ewtabF,ewtabD);
 621             ewtabV           = _mm_load_sd( ewtab + gmx_mm_extract_epi32(ewitab,0) +2);
 622             ewtabFn          = _mm_setzero_pd();
 623             GMX_MM_TRANSPOSE2_PD(ewtabV,ewtabFn);
 624             felec            = _mm_add_pd(ewtabF,_mm_mul_pd(eweps,ewtabD));
 625             velec            = _mm_sub_pd(ewtabV,_mm_mul_pd(_mm_mul_pd(ewtabhalfspace,eweps),_mm_add_pd(ewtabF,felec)));
 626             velec            = _mm_mul_pd(qq00,_mm_sub_pd(rinv00,velec));
 627             felec            = _mm_mul_pd(_mm_mul_pd(qq00,rinv00),_mm_sub_pd(rinvsq00,felec));
 628
 629             d                = _mm_sub_pd(r00,rswitch);
 630             d                = _mm_max_pd(d,_mm_setzero_pd());
 631             d2               = _mm_mul_pd(d,d);
 632             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)))))));
 633
 634             dsw              = _mm_mul_pd(d2,_mm_add_pd(swF2,_mm_mul_pd(d,_mm_add_pd(swF3,_mm_mul_pd(d,swF4)))));
 635
 636             /* Evaluate switch function */
 637             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 638             felec            = _mm_sub_pd( _mm_mul_pd(felec,sw) , _mm_mul_pd(rinv00,_mm_mul_pd(velec,dsw)) );
 639             cutoff_mask      = _mm_cmplt_pd(rsq00,rcutoff2);
 640
 641             fscal            = felec;
 642
 643             fscal            = _mm_and_pd(fscal,cutoff_mask);
 644
 645             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 646
 647             /* Calculate temporary vectorial force */
 648             tx               = _mm_mul_pd(fscal,dx00);
 649             ty               = _mm_mul_pd(fscal,dy00);
 650             tz               = _mm_mul_pd(fscal,dz00);
 651
 652             /* Update vectorial force */
 653             fix0             = _mm_add_pd(fix0,tx);
 654             fiy0             = _mm_add_pd(fiy0,ty);
 655             fiz0             = _mm_add_pd(fiz0,tz);
 656
 657             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,tx,ty,tz);
 658
 659             }
 660
 661             /* Inner loop uses 62 flops */
 662         }
 663
 664         /* End of innermost loop */
 665
 666         gmx_mm_update_iforce_1atom_swizzle_pd(fix0,fiy0,fiz0,
 667                                               f+i_coord_offset,fshift+i_shift_offset);
 668
 669         /* Increment number of inner iterations */
 670         inneriter                  += j_index_end - j_index_start;
 671
 672         /* Outer loop uses 7 flops */
 673     }
 674
 675     /* Increment number of outer iterations */
 676     outeriter        += nri;
 677
 678     /* Update outer/inner flops */
 679
 680     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_F,outeriter*7 + inneriter*62);
 681 }