src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_double/nb_kernel_ElecCoul_VdwLJ_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_VdwLJ_GeomW3P1_VF_avx_128_fma_double
  38  * Electrostatics interaction: Coulomb
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCoul_VdwLJ_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     int              nvdwtype;
  80     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
  84     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
  85     __m128d          dummy_mask,cutoff_mask;
  86     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
  87     __m128d          one     = _mm_set1_pd(1.0);
  88     __m128d          two     = _mm_set1_pd(2.0);
  89     x                = xx[0];
  90     f                = ff[0];
  91
  92     nri              = nlist->nri;
  93     iinr             = nlist->iinr;
  94     jindex           = nlist->jindex;
  95     jjnr             = nlist->jjnr;
  96     shiftidx         = nlist->shift;
  97     gid              = nlist->gid;
  98     shiftvec         = fr->shift_vec[0];
  99     fshift           = fr->fshift[0];
 100     facel            = _mm_set1_pd(fr->epsfac);
 101     charge           = mdatoms->chargeA;
 102     nvdwtype         = fr->ntype;
 103     vdwparam         = fr->nbfp;
 104     vdwtype          = mdatoms->typeA;
 105
 106     /* Setup water-specific parameters */
 107     inr              = nlist->iinr[0];
 108     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 109     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 110     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 111     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 112
 113     /* Avoid stupid compiler warnings */
 114     jnrA = jnrB = 0;
 115     j_coord_offsetA = 0;
 116     j_coord_offsetB = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126
 127         /* Load limits for loop over neighbors */
 128         j_index_start    = jindex[iidx];
 129         j_index_end      = jindex[iidx+1];
 130
 131         /* Get outer coordinate index */
 132         inr              = iinr[iidx];
 133         i_coord_offset   = DIM*inr;
 134
 135         /* Load i particle coords and add shift vector */
 136         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 137                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 138
 139         fix0             = _mm_setzero_pd();
 140         fiy0             = _mm_setzero_pd();
 141         fiz0             = _mm_setzero_pd();
 142         fix1             = _mm_setzero_pd();
 143         fiy1             = _mm_setzero_pd();
 144         fiz1             = _mm_setzero_pd();
 145         fix2             = _mm_setzero_pd();
 146         fiy2             = _mm_setzero_pd();
 147         fiz2             = _mm_setzero_pd();
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_pd();
 151         vvdwsum          = _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             dx10             = _mm_sub_pd(ix1,jx0);
 172             dy10             = _mm_sub_pd(iy1,jy0);
 173             dz10             = _mm_sub_pd(iz1,jz0);
 174             dx20             = _mm_sub_pd(ix2,jx0);
 175             dy20             = _mm_sub_pd(iy2,jy0);
 176             dz20             = _mm_sub_pd(iz2,jz0);
 177
 178             /* Calculate squared distance and things based on it */
 179             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 180             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 181             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 182
 183             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 184             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 185             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 186
 187             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 188             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 189             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 190
 191             /* Load parameters for j particles */
 192             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 193             vdwjidx0A        = 2*vdwtype[jnrA+0];
 194             vdwjidx0B        = 2*vdwtype[jnrB+0];
 195
 196             fjx0             = _mm_setzero_pd();
 197             fjy0             = _mm_setzero_pd();
 198             fjz0             = _mm_setzero_pd();
 199
 200             /**************************
 201              * CALCULATE INTERACTIONS *
 202              **************************/
 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             /* COULOMB ELECTROSTATICS */
 210             velec            = _mm_mul_pd(qq00,rinv00);
 211             felec            = _mm_mul_pd(velec,rinvsq00);
 212
 213             /* LENNARD-JONES DISPERSION/REPULSION */
 214
 215             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 216             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 217             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 218             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 219             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 220
 221             /* Update potential sum for this i atom from the interaction with this j atom. */
 222             velecsum         = _mm_add_pd(velecsum,velec);
 223             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 224
 225             fscal            = _mm_add_pd(felec,fvdw);
 226
 227             /* Update vectorial force */
 228             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 229             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 230             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 231
 232             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 233             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 234             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq10             = _mm_mul_pd(iq1,jq0);
 242
 243             /* COULOMB ELECTROSTATICS */
 244             velec            = _mm_mul_pd(qq10,rinv10);
 245             felec            = _mm_mul_pd(velec,rinvsq10);
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_pd(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             /* Update vectorial force */
 253             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 254             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 255             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 256
 257             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 258             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 259             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 260
 261             /**************************
 262              * CALCULATE INTERACTIONS *
 263              **************************/
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq20             = _mm_mul_pd(iq2,jq0);
 267
 268             /* COULOMB ELECTROSTATICS */
 269             velec            = _mm_mul_pd(qq20,rinv20);
 270             felec            = _mm_mul_pd(velec,rinvsq20);
 271
 272             /* Update potential sum for this i atom from the interaction with this j atom. */
 273             velecsum         = _mm_add_pd(velecsum,velec);
 274
 275             fscal            = felec;
 276
 277             /* Update vectorial force */
 278             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 279             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 280             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 281
 282             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 283             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 284             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 285
 286             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 287
 288             /* Inner loop uses 108 flops */
 289         }
 290
 291         if(jidx<j_index_end)
 292         {
 293
 294             jnrA             = jjnr[jidx];
 295             j_coord_offsetA  = DIM*jnrA;
 296
 297             /* load j atom coordinates */
 298             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 299                                               &jx0,&jy0,&jz0);
 300
 301             /* Calculate displacement vector */
 302             dx00             = _mm_sub_pd(ix0,jx0);
 303             dy00             = _mm_sub_pd(iy0,jy0);
 304             dz00             = _mm_sub_pd(iz0,jz0);
 305             dx10             = _mm_sub_pd(ix1,jx0);
 306             dy10             = _mm_sub_pd(iy1,jy0);
 307             dz10             = _mm_sub_pd(iz1,jz0);
 308             dx20             = _mm_sub_pd(ix2,jx0);
 309             dy20             = _mm_sub_pd(iy2,jy0);
 310             dz20             = _mm_sub_pd(iz2,jz0);
 311
 312             /* Calculate squared distance and things based on it */
 313             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 314             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 315             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 316
 317             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 318             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 319             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 320
 321             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 322             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 323             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 324
 325             /* Load parameters for j particles */
 326             jq0              = _mm_load_sd(charge+jnrA+0);
 327             vdwjidx0A        = 2*vdwtype[jnrA+0];
 328
 329             fjx0             = _mm_setzero_pd();
 330             fjy0             = _mm_setzero_pd();
 331             fjz0             = _mm_setzero_pd();
 332
 333             /**************************
 334              * CALCULATE INTERACTIONS *
 335              **************************/
 336
 337             /* Compute parameters for interactions between i and j atoms */
 338             qq00             = _mm_mul_pd(iq0,jq0);
 339             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 340
 341             /* COULOMB ELECTROSTATICS */
 342             velec            = _mm_mul_pd(qq00,rinv00);
 343             felec            = _mm_mul_pd(velec,rinvsq00);
 344
 345             /* LENNARD-JONES DISPERSION/REPULSION */
 346
 347             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 348             vvdw6            = _mm_mul_pd(c6_00,rinvsix);
 349             vvdw12           = _mm_mul_pd(c12_00,_mm_mul_pd(rinvsix,rinvsix));
 350             vvdw             = _mm_msub_pd( vvdw12,one_twelfth, _mm_mul_pd(vvdw6,one_sixth) );
 351             fvdw             = _mm_mul_pd(_mm_sub_pd(vvdw12,vvdw6),rinvsq00);
 352
 353             /* Update potential sum for this i atom from the interaction with this j atom. */
 354             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 355             velecsum         = _mm_add_pd(velecsum,velec);
 356             vvdw             = _mm_unpacklo_pd(vvdw,_mm_setzero_pd());
 357             vvdwsum          = _mm_add_pd(vvdwsum,vvdw);
 358
 359             fscal            = _mm_add_pd(felec,fvdw);
 360
 361             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 362
 363             /* Update vectorial force */
 364             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 365             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 366             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 367
 368             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 369             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 370             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 371
 372             /**************************
 373              * CALCULATE INTERACTIONS *
 374              **************************/
 375
 376             /* Compute parameters for interactions between i and j atoms */
 377             qq10             = _mm_mul_pd(iq1,jq0);
 378
 379             /* COULOMB ELECTROSTATICS */
 380             velec            = _mm_mul_pd(qq10,rinv10);
 381             felec            = _mm_mul_pd(velec,rinvsq10);
 382
 383             /* Update potential sum for this i atom from the interaction with this j atom. */
 384             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 385             velecsum         = _mm_add_pd(velecsum,velec);
 386
 387             fscal            = felec;
 388
 389             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 390
 391             /* Update vectorial force */
 392             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 393             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 394             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 395
 396             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 397             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 398             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 399
 400             /**************************
 401              * CALCULATE INTERACTIONS *
 402              **************************/
 403
 404             /* Compute parameters for interactions between i and j atoms */
 405             qq20             = _mm_mul_pd(iq2,jq0);
 406
 407             /* COULOMB ELECTROSTATICS */
 408             velec            = _mm_mul_pd(qq20,rinv20);
 409             felec            = _mm_mul_pd(velec,rinvsq20);
 410
 411             /* Update potential sum for this i atom from the interaction with this j atom. */
 412             velec            = _mm_unpacklo_pd(velec,_mm_setzero_pd());
 413             velecsum         = _mm_add_pd(velecsum,velec);
 414
 415             fscal            = felec;
 416
 417             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 418
 419             /* Update vectorial force */
 420             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 421             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 422             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 423
 424             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 425             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 426             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 427
 428             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 429
 430             /* Inner loop uses 108 flops */
 431         }
 432
 433         /* End of innermost loop */
 434
 435         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 436                                               f+i_coord_offset,fshift+i_shift_offset);
 437
 438         ggid                        = gid[iidx];
 439         /* Update potential energies */
 440         gmx_mm_update_1pot_pd(velecsum,kernel_data->energygrp_elec+ggid);
 441         gmx_mm_update_1pot_pd(vvdwsum,kernel_data->energygrp_vdw+ggid);
 442
 443         /* Increment number of inner iterations */
 444         inneriter                  += j_index_end - j_index_start;
 445
 446         /* Outer loop uses 20 flops */
 447     }
 448
 449     /* Increment number of outer iterations */
 450     outeriter        += nri;
 451
 452     /* Update outer/inner flops */
 453
 454     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*108);
 455 }
 456 /*
 457  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_double
 458  * Electrostatics interaction: Coulomb
 459  * VdW interaction:            LennardJones
 460  * Geometry:                   Water3-Particle
 461  * Calculate force/pot:        Force
 462  */
 463 void
 464 nb_kernel_ElecCoul_VdwLJ_GeomW3P1_F_avx_128_fma_double
 465                     (t_nblist * gmx_restrict                nlist,
 466                      rvec * gmx_restrict                    xx,
 467                      rvec * gmx_restrict                    ff,
 468                      t_forcerec * gmx_restrict              fr,
 469                      t_mdatoms * gmx_restrict               mdatoms,
 470                      nb_kernel_data_t * gmx_restrict        kernel_data,
 471                      t_nrnb * gmx_restrict                  nrnb)
 472 {
 473     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 474      * just 0 for non-waters.
 475      * Suffixes A,B refer to j loop unrolling done with SSE double precision, e.g. for the two different
 476      * jnr indices corresponding to data put in the four positions in the SIMD register.
 477      */
 478     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 479     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 480     int              jnrA,jnrB;
 481     int              j_coord_offsetA,j_coord_offsetB;
 482     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 483     real             rcutoff_scalar;
 484     real             *shiftvec,*fshift,*x,*f;
 485     __m128d          tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 486     int              vdwioffset0;
 487     __m128d          ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 488     int              vdwioffset1;
 489     __m128d          ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 490     int              vdwioffset2;
 491     __m128d          ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 492     int              vdwjidx0A,vdwjidx0B;
 493     __m128d          jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 494     __m128d          dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 495     __m128d          dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 496     __m128d          dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 497     __m128d          velec,felec,velecsum,facel,crf,krf,krf2;
 498     real             *charge;
 499     int              nvdwtype;
 500     __m128d          rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 501     int              *vdwtype;
 502     real             *vdwparam;
 503     __m128d          one_sixth   = _mm_set1_pd(1.0/6.0);
 504     __m128d          one_twelfth = _mm_set1_pd(1.0/12.0);
 505     __m128d          dummy_mask,cutoff_mask;
 506     __m128d          signbit   = gmx_mm_castsi128_pd( _mm_set_epi32(0x80000000,0x00000000,0x80000000,0x00000000) );
 507     __m128d          one     = _mm_set1_pd(1.0);
 508     __m128d          two     = _mm_set1_pd(2.0);
 509     x                = xx[0];
 510     f                = ff[0];
 511
 512     nri              = nlist->nri;
 513     iinr             = nlist->iinr;
 514     jindex           = nlist->jindex;
 515     jjnr             = nlist->jjnr;
 516     shiftidx         = nlist->shift;
 517     gid              = nlist->gid;
 518     shiftvec         = fr->shift_vec[0];
 519     fshift           = fr->fshift[0];
 520     facel            = _mm_set1_pd(fr->epsfac);
 521     charge           = mdatoms->chargeA;
 522     nvdwtype         = fr->ntype;
 523     vdwparam         = fr->nbfp;
 524     vdwtype          = mdatoms->typeA;
 525
 526     /* Setup water-specific parameters */
 527     inr              = nlist->iinr[0];
 528     iq0              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+0]));
 529     iq1              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+1]));
 530     iq2              = _mm_mul_pd(facel,_mm_set1_pd(charge[inr+2]));
 531     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 532
 533     /* Avoid stupid compiler warnings */
 534     jnrA = jnrB = 0;
 535     j_coord_offsetA = 0;
 536     j_coord_offsetB = 0;
 537
 538     outeriter        = 0;
 539     inneriter        = 0;
 540
 541     /* Start outer loop over neighborlists */
 542     for(iidx=0; iidx<nri; iidx++)
 543     {
 544         /* Load shift vector for this list */
 545         i_shift_offset   = DIM*shiftidx[iidx];
 546
 547         /* Load limits for loop over neighbors */
 548         j_index_start    = jindex[iidx];
 549         j_index_end      = jindex[iidx+1];
 550
 551         /* Get outer coordinate index */
 552         inr              = iinr[iidx];
 553         i_coord_offset   = DIM*inr;
 554
 555         /* Load i particle coords and add shift vector */
 556         gmx_mm_load_shift_and_3rvec_broadcast_pd(shiftvec+i_shift_offset,x+i_coord_offset,
 557                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 558
 559         fix0             = _mm_setzero_pd();
 560         fiy0             = _mm_setzero_pd();
 561         fiz0             = _mm_setzero_pd();
 562         fix1             = _mm_setzero_pd();
 563         fiy1             = _mm_setzero_pd();
 564         fiz1             = _mm_setzero_pd();
 565         fix2             = _mm_setzero_pd();
 566         fiy2             = _mm_setzero_pd();
 567         fiz2             = _mm_setzero_pd();
 568
 569         /* Start inner kernel loop */
 570         for(jidx=j_index_start; jidx<j_index_end-1; jidx+=2)
 571         {
 572
 573             /* Get j neighbor index, and coordinate index */
 574             jnrA             = jjnr[jidx];
 575             jnrB             = jjnr[jidx+1];
 576             j_coord_offsetA  = DIM*jnrA;
 577             j_coord_offsetB  = DIM*jnrB;
 578
 579             /* load j atom coordinates */
 580             gmx_mm_load_1rvec_2ptr_swizzle_pd(x+j_coord_offsetA,x+j_coord_offsetB,
 581                                               &jx0,&jy0,&jz0);
 582
 583             /* Calculate displacement vector */
 584             dx00             = _mm_sub_pd(ix0,jx0);
 585             dy00             = _mm_sub_pd(iy0,jy0);
 586             dz00             = _mm_sub_pd(iz0,jz0);
 587             dx10             = _mm_sub_pd(ix1,jx0);
 588             dy10             = _mm_sub_pd(iy1,jy0);
 589             dz10             = _mm_sub_pd(iz1,jz0);
 590             dx20             = _mm_sub_pd(ix2,jx0);
 591             dy20             = _mm_sub_pd(iy2,jy0);
 592             dz20             = _mm_sub_pd(iz2,jz0);
 593
 594             /* Calculate squared distance and things based on it */
 595             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 596             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 597             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 598
 599             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 600             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 601             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 602
 603             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 604             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 605             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 606
 607             /* Load parameters for j particles */
 608             jq0              = gmx_mm_load_2real_swizzle_pd(charge+jnrA+0,charge+jnrB+0);
 609             vdwjidx0A        = 2*vdwtype[jnrA+0];
 610             vdwjidx0B        = 2*vdwtype[jnrB+0];
 611
 612             fjx0             = _mm_setzero_pd();
 613             fjy0             = _mm_setzero_pd();
 614             fjz0             = _mm_setzero_pd();
 615
 616             /**************************
 617              * CALCULATE INTERACTIONS *
 618              **************************/
 619
 620             /* Compute parameters for interactions between i and j atoms */
 621             qq00             = _mm_mul_pd(iq0,jq0);
 622             gmx_mm_load_2pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,
 623                                          vdwparam+vdwioffset0+vdwjidx0B,&c6_00,&c12_00);
 624
 625             /* COULOMB ELECTROSTATICS */
 626             velec            = _mm_mul_pd(qq00,rinv00);
 627             felec            = _mm_mul_pd(velec,rinvsq00);
 628
 629             /* LENNARD-JONES DISPERSION/REPULSION */
 630
 631             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 632             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 633
 634             fscal            = _mm_add_pd(felec,fvdw);
 635
 636             /* Update vectorial force */
 637             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 638             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 639             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 640
 641             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 642             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 643             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 644
 645             /**************************
 646              * CALCULATE INTERACTIONS *
 647              **************************/
 648
 649             /* Compute parameters for interactions between i and j atoms */
 650             qq10             = _mm_mul_pd(iq1,jq0);
 651
 652             /* COULOMB ELECTROSTATICS */
 653             velec            = _mm_mul_pd(qq10,rinv10);
 654             felec            = _mm_mul_pd(velec,rinvsq10);
 655
 656             fscal            = felec;
 657
 658             /* Update vectorial force */
 659             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 660             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 661             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 662
 663             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 664             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 665             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 666
 667             /**************************
 668              * CALCULATE INTERACTIONS *
 669              **************************/
 670
 671             /* Compute parameters for interactions between i and j atoms */
 672             qq20             = _mm_mul_pd(iq2,jq0);
 673
 674             /* COULOMB ELECTROSTATICS */
 675             velec            = _mm_mul_pd(qq20,rinv20);
 676             felec            = _mm_mul_pd(velec,rinvsq20);
 677
 678             fscal            = felec;
 679
 680             /* Update vectorial force */
 681             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 682             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 683             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 684
 685             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 686             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 687             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 688
 689             gmx_mm_decrement_1rvec_2ptr_swizzle_pd(f+j_coord_offsetA,f+j_coord_offsetB,fjx0,fjy0,fjz0);
 690
 691             /* Inner loop uses 100 flops */
 692         }
 693
 694         if(jidx<j_index_end)
 695         {
 696
 697             jnrA             = jjnr[jidx];
 698             j_coord_offsetA  = DIM*jnrA;
 699
 700             /* load j atom coordinates */
 701             gmx_mm_load_1rvec_1ptr_swizzle_pd(x+j_coord_offsetA,
 702                                               &jx0,&jy0,&jz0);
 703
 704             /* Calculate displacement vector */
 705             dx00             = _mm_sub_pd(ix0,jx0);
 706             dy00             = _mm_sub_pd(iy0,jy0);
 707             dz00             = _mm_sub_pd(iz0,jz0);
 708             dx10             = _mm_sub_pd(ix1,jx0);
 709             dy10             = _mm_sub_pd(iy1,jy0);
 710             dz10             = _mm_sub_pd(iz1,jz0);
 711             dx20             = _mm_sub_pd(ix2,jx0);
 712             dy20             = _mm_sub_pd(iy2,jy0);
 713             dz20             = _mm_sub_pd(iz2,jz0);
 714
 715             /* Calculate squared distance and things based on it */
 716             rsq00            = gmx_mm_calc_rsq_pd(dx00,dy00,dz00);
 717             rsq10            = gmx_mm_calc_rsq_pd(dx10,dy10,dz10);
 718             rsq20            = gmx_mm_calc_rsq_pd(dx20,dy20,dz20);
 719
 720             rinv00           = gmx_mm_invsqrt_pd(rsq00);
 721             rinv10           = gmx_mm_invsqrt_pd(rsq10);
 722             rinv20           = gmx_mm_invsqrt_pd(rsq20);
 723
 724             rinvsq00         = _mm_mul_pd(rinv00,rinv00);
 725             rinvsq10         = _mm_mul_pd(rinv10,rinv10);
 726             rinvsq20         = _mm_mul_pd(rinv20,rinv20);
 727
 728             /* Load parameters for j particles */
 729             jq0              = _mm_load_sd(charge+jnrA+0);
 730             vdwjidx0A        = 2*vdwtype[jnrA+0];
 731
 732             fjx0             = _mm_setzero_pd();
 733             fjy0             = _mm_setzero_pd();
 734             fjz0             = _mm_setzero_pd();
 735
 736             /**************************
 737              * CALCULATE INTERACTIONS *
 738              **************************/
 739
 740             /* Compute parameters for interactions between i and j atoms */
 741             qq00             = _mm_mul_pd(iq0,jq0);
 742             gmx_mm_load_1pair_swizzle_pd(vdwparam+vdwioffset0+vdwjidx0A,&c6_00,&c12_00);
 743
 744             /* COULOMB ELECTROSTATICS */
 745             velec            = _mm_mul_pd(qq00,rinv00);
 746             felec            = _mm_mul_pd(velec,rinvsq00);
 747
 748             /* LENNARD-JONES DISPERSION/REPULSION */
 749
 750             rinvsix          = _mm_mul_pd(_mm_mul_pd(rinvsq00,rinvsq00),rinvsq00);
 751             fvdw             = _mm_mul_pd(_mm_msub_pd(c12_00,rinvsix,c6_00),_mm_mul_pd(rinvsix,rinvsq00));
 752
 753             fscal            = _mm_add_pd(felec,fvdw);
 754
 755             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 756
 757             /* Update vectorial force */
 758             fix0             = _mm_macc_pd(dx00,fscal,fix0);
 759             fiy0             = _mm_macc_pd(dy00,fscal,fiy0);
 760             fiz0             = _mm_macc_pd(dz00,fscal,fiz0);
 761
 762             fjx0             = _mm_macc_pd(dx00,fscal,fjx0);
 763             fjy0             = _mm_macc_pd(dy00,fscal,fjy0);
 764             fjz0             = _mm_macc_pd(dz00,fscal,fjz0);
 765
 766             /**************************
 767              * CALCULATE INTERACTIONS *
 768              **************************/
 769
 770             /* Compute parameters for interactions between i and j atoms */
 771             qq10             = _mm_mul_pd(iq1,jq0);
 772
 773             /* COULOMB ELECTROSTATICS */
 774             velec            = _mm_mul_pd(qq10,rinv10);
 775             felec            = _mm_mul_pd(velec,rinvsq10);
 776
 777             fscal            = felec;
 778
 779             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 780
 781             /* Update vectorial force */
 782             fix1             = _mm_macc_pd(dx10,fscal,fix1);
 783             fiy1             = _mm_macc_pd(dy10,fscal,fiy1);
 784             fiz1             = _mm_macc_pd(dz10,fscal,fiz1);
 785
 786             fjx0             = _mm_macc_pd(dx10,fscal,fjx0);
 787             fjy0             = _mm_macc_pd(dy10,fscal,fjy0);
 788             fjz0             = _mm_macc_pd(dz10,fscal,fjz0);
 789
 790             /**************************
 791              * CALCULATE INTERACTIONS *
 792              **************************/
 793
 794             /* Compute parameters for interactions between i and j atoms */
 795             qq20             = _mm_mul_pd(iq2,jq0);
 796
 797             /* COULOMB ELECTROSTATICS */
 798             velec            = _mm_mul_pd(qq20,rinv20);
 799             felec            = _mm_mul_pd(velec,rinvsq20);
 800
 801             fscal            = felec;
 802
 803             fscal            = _mm_unpacklo_pd(fscal,_mm_setzero_pd());
 804
 805             /* Update vectorial force */
 806             fix2             = _mm_macc_pd(dx20,fscal,fix2);
 807             fiy2             = _mm_macc_pd(dy20,fscal,fiy2);
 808             fiz2             = _mm_macc_pd(dz20,fscal,fiz2);
 809
 810             fjx0             = _mm_macc_pd(dx20,fscal,fjx0);
 811             fjy0             = _mm_macc_pd(dy20,fscal,fjy0);
 812             fjz0             = _mm_macc_pd(dz20,fscal,fjz0);
 813
 814             gmx_mm_decrement_1rvec_1ptr_swizzle_pd(f+j_coord_offsetA,fjx0,fjy0,fjz0);
 815
 816             /* Inner loop uses 100 flops */
 817         }
 818
 819         /* End of innermost loop */
 820
 821         gmx_mm_update_iforce_3atom_swizzle_pd(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 822                                               f+i_coord_offset,fshift+i_shift_offset);
 823
 824         /* Increment number of inner iterations */
 825         inneriter                  += j_index_end - j_index_start;
 826
 827         /* Outer loop uses 18 flops */
 828     }
 829
 830     /* Increment number of outer iterations */
 831     outeriter        += nri;
 832
 833     /* Update outer/inner flops */
 834
 835     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*100);
 836 }