src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwNone_GeomW3P1_avx_128_fma_double.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_double kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_avx_128_fma_double.h"
  34 #include "kernelutil_x86_avx_128_fma_double.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_avx_128_fma_double
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB;
  61     int              j_coord_offsetA,j_coord_offsetB;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset0;
  67     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B;
  73     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128d          dummy_mask,cutoff_mask;
  80     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  81     __m128d          one     = _mm_set1_pd(1.0);
  82     __m128d          two     = _mm_set1_pd(2.0);
  83     x                = xx[0];
  84     f                = ff[0];
  85
  86     nri              = nlist->nri;
  87     iinr             = nlist->iinr;
  88     jindex           = nlist->jindex;
  89     jjnr             = nlist->jjnr;
  90     shiftidx         = nlist->shift;
  91     gid              = nlist->gid;
  92     shiftvec         = fr->shift_vec[0];
  93     fshift           = fr->fshift[0];
  94     facel            = _mm_set1_pd(fr->epsfac);
  95     charge           = mdatoms->chargeA;
  96
  97     /* Setup water-specific parameters */
  98     inr              = nlist->iinr[0];
  99     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 100     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 101     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 102
 103     /* Avoid stupid compiler warnings */
 104     jnrA = jnrB = 0;
 105     j_coord_offsetA = 0;
 106     j_coord_offsetB = 0;
 107
 108     outeriter        = 0;
 109     inneriter        = 0;
 110
 111     /* Start outer loop over neighborlists */
 112     for(iidx=0; iidx<nri; iidx++)
 113     {
 114         /* Load shift vector for this list */
 115         i_shift_offset   = DIM*shiftidx[iidx];
 116
 117         /* Load limits for loop over neighbors */
 118         j_index_start    = jindex[iidx];
 119         j_index_end      = jindex[iidx+1];
 120
 121         /* Get outer coordinate index */
 122         inr              = iinr[iidx];
 123         i_coord_offset   = DIM*inr;
 124
 125         /* Load i particle coords and add shift vector */
 126         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 127                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 128
 129         fix0             = _mm_setzero_pd();
 130         fiy0             = _mm_setzero_pd();
 131         fiz0             = _mm_setzero_pd();
 132         fix1             = _mm_setzero_pd();
 133         fiy1             = _mm_setzero_pd();
 134         fiz1             = _mm_setzero_pd();
 135         fix2             = _mm_setzero_pd();
 136         fiy2             = _mm_setzero_pd();
 137         fiz2             = _mm_setzero_pd();
 138
 139         /* Reset potential sums */
 140         velecsum         = _mm_setzero_pd();
 141
 142         /* Start inner kernel loop */
 143         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 144         {
 145
 146             /* Get j neighbor index, and coordinate index */
 147             jnrA             = jjnr[jidx];
 148             jnrB             = jjnr[jidx+1];
 149             j_coord_offsetA  = DIM*jnrA;
 150             j_coord_offsetB  = DIM*jnrB;
 151
 152             /* load j atom coordinates */
 153             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 154                                               &jx0,&jy0,&jz0);
 155
 156             /* Calculate displacement vector */
 157             dx00             = _mm_sub_pd(ix0,jx0);
 158             dy00             = _mm_sub_pd(iy0,jy0);
 159             dz00             = _mm_sub_pd(iz0,jz0);
 160             dx10             = _mm_sub_pd(ix1,jx0);
 161             dy10             = _mm_sub_pd(iy1,jy0);
 162             dz10             = _mm_sub_pd(iz1,jz0);
 163             dx20             = _mm_sub_pd(ix2,jx0);
 164             dy20             = _mm_sub_pd(iy2,jy0);
 165             dz20             = _mm_sub_pd(iz2,jz0);
 166
 167             /* Calculate squared distance and things based on it */
 168             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 169             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 170             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 171
 172             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 173             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 174             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 175
 176             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 177             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 178             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 179
 180             /* Load parameters for j particles */
 181             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 182
 183             fjx0             = _mm_setzero_pd();
 184             fjy0             = _mm_setzero_pd();
 185             fjz0             = _mm_setzero_pd();
 186
 187             /**************************
 188              * CALCULATE INTERACTIONS *
 189              **************************/
 190
 191             /* Compute parameters for interactions between i and j atoms */
 192             qq00             = _mm_mul_pd(iq0,jq0);
 193
 194             /* COULOMB ELECTROSTATICS */
 195             velec            = _mm_mul_pd(qq00,rinv00);
 196             felec            = _mm_mul_pd(velec,rinvsq00);
 197
 198             /* Update potential sum for this i atom from the interaction with this j atom. */
 199             velecsum         = _mm_add_pd(velecsum,velec);
 200
 201             fscal            = felec;
 202
 203             /* Update vectorial force */
 204             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 205             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 206             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 207
 208             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 209             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 210             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 211
 212             /**************************
 213              * CALCULATE INTERACTIONS *
 214              **************************/
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq10             = _mm_mul_pd(iq1,jq0);
 218
 219             /* COULOMB ELECTROSTATICS */
 220             velec            = _mm_mul_pd(qq10,rinv10);
 221             felec            = _mm_mul_pd(velec,rinvsq10);
 222
 223             /* Update potential sum for this i atom from the interaction with this j atom. */
 224             velecsum         = _mm_add_pd(velecsum,velec);
 225
 226             fscal            = felec;
 227
 228             /* Update vectorial force */
 229             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 230             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 231             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 232
 233             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 234             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 235             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq20             = _mm_mul_pd(iq2,jq0);
 243
 244             /* COULOMB ELECTROSTATICS */
 245             velec            = _mm_mul_pd(qq20,rinv20);
 246             felec            = _mm_mul_pd(velec,rinvsq20);
 247
 248             /* Update potential sum for this i atom from the interaction with this j atom. */
 249             velecsum         = _mm_add_pd(velecsum,velec);
 250
 251             fscal            = felec;
 252
 253             /* Update vectorial force */
 254             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 255             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 256             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 257
 258             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 259             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 260             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 261
 262             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 263
 264             /* Inner loop uses 96 flops */
 265         }
 266
 267         if(jidx<j_index_end)
 268         {
 269
 270             jnrA             = jjnr[jidx];
 271             j_coord_offsetA  = DIM*jnrA;
 272
 273             /* load j atom coordinates */
 274             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 275                                               &jx0,&jy0,&jz0);
 276
 277             /* Calculate displacement vector */
 278             dx00             = _mm_sub_pd(ix0,jx0);
 279             dy00             = _mm_sub_pd(iy0,jy0);
 280             dz00             = _mm_sub_pd(iz0,jz0);
 281             dx10             = _mm_sub_pd(ix1,jx0);
 282             dy10             = _mm_sub_pd(iy1,jy0);
 283             dz10             = _mm_sub_pd(iz1,jz0);
 284             dx20             = _mm_sub_pd(ix2,jx0);
 285             dy20             = _mm_sub_pd(iy2,jy0);
 286             dz20             = _mm_sub_pd(iz2,jz0);
 287
 288             /* Calculate squared distance and things based on it */
 289             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 290             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 291             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 292
 293             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 294             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 295             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 296
 297             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 298             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 299             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 300
 301             /* Load parameters for j particles */
 302             jq0              = _mm_load_sd(charge+jnrA+0);
 303
 304             fjx0             = _mm_setzero_pd();
 305             fjy0             = _mm_setzero_pd();
 306             fjz0             = _mm_setzero_pd();
 307
 308             /**************************
 309              * CALCULATE INTERACTIONS *
 310              **************************/
 311
 312             /* Compute parameters for interactions between i and j atoms */
 313             qq00             = _mm_mul_pd(iq0,jq0);
 314
 315             /* COULOMB ELECTROSTATICS */
 316             velec            = _mm_mul_pd(qq00,rinv00);
 317             felec            = _mm_mul_pd(velec,rinvsq00);
 318
 319             /* Update potential sum for this i atom from the interaction with this j atom. */
 320             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 321             velecsum         = _mm_add_pd(velecsum,velec);
 322
 323             fscal            = felec;
 324
 325             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 326
 327             /* Update vectorial force */
 328             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 329             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 330             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 331
 332             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 333             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 334             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 335
 336             /**************************
 337              * CALCULATE INTERACTIONS *
 338              **************************/
 339
 340             /* Compute parameters for interactions between i and j atoms */
 341             qq10             = _mm_mul_pd(iq1,jq0);
 342
 343             /* COULOMB ELECTROSTATICS */
 344             velec            = _mm_mul_pd(qq10,rinv10);
 345             felec            = _mm_mul_pd(velec,rinvsq10);
 346
 347             /* Update potential sum for this i atom from the interaction with this j atom. */
 348             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 349             velecsum         = _mm_add_pd(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 354
 355             /* Update vectorial force */
 356             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 357             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 358             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 359
 360             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 361             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 362             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 363
 364             /**************************
 365              * CALCULATE INTERACTIONS *
 366              **************************/
 367
 368             /* Compute parameters for interactions between i and j atoms */
 369             qq20             = _mm_mul_pd(iq2,jq0);
 370
 371             /* COULOMB ELECTROSTATICS */
 372             velec            = _mm_mul_pd(qq20,rinv20);
 373             felec            = _mm_mul_pd(velec,rinvsq20);
 374
 375             /* Update potential sum for this i atom from the interaction with this j atom. */
 376             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 377             velecsum         = _mm_add_pd(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 382
 383             /* Update vectorial force */
 384             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 385             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 386             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 387
 388             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 389             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 390             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 391
 392             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 393
 394             /* Inner loop uses 96 flops */
 395         }
 396
 397         /* End of innermost loop */
 398
 399         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 400                                               f+i_coord_offset,fshift+i_shift_offset);
 401
 402         ggid                        = gid[iidx];
 403         /* Update potential energies */
 404         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 405
 406         /* Increment number of inner iterations */
 407         inneriter                  += j_index_end - j_index_start;
 408
 409         /* Outer loop uses 19 flops */
 410     }
 411
 412     /* Increment number of outer iterations */
 413     outeriter        += nri;
 414
 415     /* Update outer/inner flops */
 416
 417     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
 418 }
 419 /*
 420  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_128_fma_double
 421  * Electrostatics interaction: Coulomb
 422  * VdW interaction:            None
 423  * Geometry:                   Water3-Particle
 424  * Calculate force/pot:        Force
 425  */
 426 void
 427 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_avx_128_fma_double
 428                     (t_nblist * gmx_restrict                nlist,
 429                      rvec * gmx_restrict                    xx,
 430                      rvec * gmx_restrict                    ff,
 431                      t_forcerec * gmx_restrict              fr,
 432                      t_mdatoms * gmx_restrict               mdatoms,
 433                      nb_kernel_data_t * gmx_restrict        kernel_data,
 434                      t_nrnb * gmx_restrict                  nrnb)
 435 {
 436     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 437      * just 0 for non-waters.
 438      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 439      * jnr indices corresponding to data put in the four positions in the SIMD register.
 440      */
 441     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 442     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 443     int              jnrA,jnrB;
 444     int              j_coord_offsetA,j_coord_offsetB;
 445     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 446     real             rcutoff_scalar;
 447     real             *shiftvec,*fshift,*x,*f;
 448     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 449     int              vdwioffset0;
 450     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 451     int              vdwioffset1;
 452     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 453     int              vdwioffset2;
 454     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 455     int              vdwjidx0A,vdwjidx0B;
 456     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 457     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 458     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 459     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 460     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 461     real             *charge;
 462     __m128d          dummy_mask,cutoff_mask;
 463     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 464     __m128d          one     = _mm_set1_pd(1.0);
 465     __m128d          two     = _mm_set1_pd(2.0);
 466     x                = xx[0];
 467     f                = ff[0];
 468
 469     nri              = nlist->nri;
 470     iinr             = nlist->iinr;
 471     jindex           = nlist->jindex;
 472     jjnr             = nlist->jjnr;
 473     shiftidx         = nlist->shift;
 474     gid              = nlist->gid;
 475     shiftvec         = fr->shift_vec[0];
 476     fshift           = fr->fshift[0];
 477     facel            = _mm_set1_pd(fr->epsfac);
 478     charge           = mdatoms->chargeA;
 479
 480     /* Setup water-specific parameters */
 481     inr              = nlist->iinr[0];
 482     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 483     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 484     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 485
 486     /* Avoid stupid compiler warnings */
 487     jnrA = jnrB = 0;
 488     j_coord_offsetA = 0;
 489     j_coord_offsetB = 0;
 490
 491     outeriter        = 0;
 492     inneriter        = 0;
 493
 494     /* Start outer loop over neighborlists */
 495     for(iidx=0; iidx<nri; iidx++)
 496     {
 497         /* Load shift vector for this list */
 498         i_shift_offset   = DIM*shiftidx[iidx];
 499
 500         /* Load limits for loop over neighbors */
 501         j_index_start    = jindex[iidx];
 502         j_index_end      = jindex[iidx+1];
 503
 504         /* Get outer coordinate index */
 505         inr              = iinr[iidx];
 506         i_coord_offset   = DIM*inr;
 507
 508         /* Load i particle coords and add shift vector */
 509         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 510                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 511
 512         fix0             = _mm_setzero_pd();
 513         fiy0             = _mm_setzero_pd();
 514         fiz0             = _mm_setzero_pd();
 515         fix1             = _mm_setzero_pd();
 516         fiy1             = _mm_setzero_pd();
 517         fiz1             = _mm_setzero_pd();
 518         fix2             = _mm_setzero_pd();
 519         fiy2             = _mm_setzero_pd();
 520         fiz2             = _mm_setzero_pd();
 521
 522         /* Start inner kernel loop */
 523         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 524         {
 525
 526             /* Get j neighbor index, and coordinate index */
 527             jnrA             = jjnr[jidx];
 528             jnrB             = jjnr[jidx+1];
 529             j_coord_offsetA  = DIM*jnrA;
 530             j_coord_offsetB  = DIM*jnrB;
 531
 532             /* load j atom coordinates */
 533             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 534                                               &jx0,&jy0,&jz0);
 535
 536             /* Calculate displacement vector */
 537             dx00             = _mm_sub_pd(ix0,jx0);
 538             dy00             = _mm_sub_pd(iy0,jy0);
 539             dz00             = _mm_sub_pd(iz0,jz0);
 540             dx10             = _mm_sub_pd(ix1,jx0);
 541             dy10             = _mm_sub_pd(iy1,jy0);
 542             dz10             = _mm_sub_pd(iz1,jz0);
 543             dx20             = _mm_sub_pd(ix2,jx0);
 544             dy20             = _mm_sub_pd(iy2,jy0);
 545             dz20             = _mm_sub_pd(iz2,jz0);
 546
 547             /* Calculate squared distance and things based on it */
 548             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 549             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 550             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 551
 552             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 553             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 554             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 555
 556             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 557             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 558             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 559
 560             /* Load parameters for j particles */
 561             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 562
 563             fjx0             = _mm_setzero_pd();
 564             fjy0             = _mm_setzero_pd();
 565             fjz0             = _mm_setzero_pd();
 566
 567             /**************************
 568              * CALCULATE INTERACTIONS *
 569              **************************/
 570
 571             /* Compute parameters for interactions between i and j atoms */
 572             qq00             = _mm_mul_pd(iq0,jq0);
 573
 574             /* COULOMB ELECTROSTATICS */
 575             velec            = _mm_mul_pd(qq00,rinv00);
 576             felec            = _mm_mul_pd(velec,rinvsq00);
 577
 578             fscal            = felec;
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 582             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 583             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 584
 585             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 586             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 587             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             /* Compute parameters for interactions between i and j atoms */
 594             qq10             = _mm_mul_pd(iq1,jq0);
 595
 596             /* COULOMB ELECTROSTATICS */
 597             velec            = _mm_mul_pd(qq10,rinv10);
 598             felec            = _mm_mul_pd(velec,rinvsq10);
 599
 600             fscal            = felec;
 601
 602             /* Update vectorial force */
 603             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 604             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 605             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 606
 607             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 608             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 609             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 610
 611             /**************************
 612              * CALCULATE INTERACTIONS *
 613              **************************/
 614
 615             /* Compute parameters for interactions between i and j atoms */
 616             qq20             = _mm_mul_pd(iq2,jq0);
 617
 618             /* COULOMB ELECTROSTATICS */
 619             velec            = _mm_mul_pd(qq20,rinv20);
 620             felec            = _mm_mul_pd(velec,rinvsq20);
 621
 622             fscal            = felec;
 623
 624             /* Update vectorial force */
 625             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 626             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 627             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 628
 629             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 630             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 631             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 632
 633             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 634
 635             /* Inner loop uses 93 flops */
 636         }
 637
 638         if(jidx<j_index_end)
 639         {
 640
 641             jnrA             = jjnr[jidx];
 642             j_coord_offsetA  = DIM*jnrA;
 643
 644             /* load j atom coordinates */
 645             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 646                                               &jx0,&jy0,&jz0);
 647
 648             /* Calculate displacement vector */
 649             dx00             = _mm_sub_pd(ix0,jx0);
 650             dy00             = _mm_sub_pd(iy0,jy0);
 651             dz00             = _mm_sub_pd(iz0,jz0);
 652             dx10             = _mm_sub_pd(ix1,jx0);
 653             dy10             = _mm_sub_pd(iy1,jy0);
 654             dz10             = _mm_sub_pd(iz1,jz0);
 655             dx20             = _mm_sub_pd(ix2,jx0);
 656             dy20             = _mm_sub_pd(iy2,jy0);
 657             dz20             = _mm_sub_pd(iz2,jz0);
 658
 659             /* Calculate squared distance and things based on it */
 660             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 661             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 662             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 663
 664             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 665             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 666             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 667
 668             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 669             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 670             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 671
 672             /* Load parameters for j particles */
 673             jq0              = _mm_load_sd(charge+jnrA+0);
 674
 675             fjx0             = _mm_setzero_pd();
 676             fjy0             = _mm_setzero_pd();
 677             fjz0             = _mm_setzero_pd();
 678
 679             /**************************
 680              * CALCULATE INTERACTIONS *
 681              **************************/
 682
 683             /* Compute parameters for interactions between i and j atoms */
 684             qq00             = _mm_mul_pd(iq0,jq0);
 685
 686             /* COULOMB ELECTROSTATICS */
 687             velec            = _mm_mul_pd(qq00,rinv00);
 688             felec            = _mm_mul_pd(velec,rinvsq00);
 689
 690             fscal            = felec;
 691
 692             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 693
 694             /* Update vectorial force */
 695             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 696             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 697             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 698
 699             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 700             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 701             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             /* Compute parameters for interactions between i and j atoms */
 708             qq10             = _mm_mul_pd(iq1,jq0);
 709
 710             /* COULOMB ELECTROSTATICS */
 711             velec            = _mm_mul_pd(qq10,rinv10);
 712             felec            = _mm_mul_pd(velec,rinvsq10);
 713
 714             fscal            = felec;
 715
 716             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 717
 718             /* Update vectorial force */
 719             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 720             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 721             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 722
 723             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 724             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 725             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 726
 727             /**************************
 728              * CALCULATE INTERACTIONS *
 729              **************************/
 730
 731             /* Compute parameters for interactions between i and j atoms */
 732             qq20             = _mm_mul_pd(iq2,jq0);
 733
 734             /* COULOMB ELECTROSTATICS */
 735             velec            = _mm_mul_pd(qq20,rinv20);
 736             felec            = _mm_mul_pd(velec,rinvsq20);
 737
 738             fscal            = felec;
 739
 740             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 741
 742             /* Update vectorial force */
 743             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 744             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 745             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 746
 747             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 748             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 749             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 750
 751             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 752
 753             /* Inner loop uses 93 flops */
 754         }
 755
 756         /* End of innermost loop */
 757
 758         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 759                                               f+i_coord_offset,fshift+i_shift_offset);
 760
 761         /* Increment number of inner iterations */
 762         inneriter                  += j_index_end - j_index_start;
 763
 764         /* Outer loop uses 18 flops */
 765     }
 766
 767     /* Increment number of outer iterations */
 768     outeriter        += nri;
 769
 770     /* Update outer/inner flops */
 771
 772     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*93);
 773 }