src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_VdwNone_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_single 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_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_avx_128_fma_single
  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,C,D refer to j loop unrolling done with AVX_128, e.g. for the four 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,jnrC,jnrD;
  61     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102
 103     /* Setup water-specific parameters */
 104     inr              = nlist->iinr[0];
 105     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 106     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 107     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 108
 109     /* Avoid stupid compiler warnings */
 110     jnrA = jnrB = jnrC = jnrD = 0;
 111     j_coord_offsetA = 0;
 112     j_coord_offsetB = 0;
 113     j_coord_offsetC = 0;
 114     j_coord_offsetD = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     for(iidx=0;iidx<4*DIM;iidx++)
 120     {
 121         scratch[iidx] = 0.0;
 122     }
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 140                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 141
 142         fix0             = _mm_setzero_ps();
 143         fiy0             = _mm_setzero_ps();
 144         fiz0             = _mm_setzero_ps();
 145         fix1             = _mm_setzero_ps();
 146         fiy1             = _mm_setzero_ps();
 147         fiz1             = _mm_setzero_ps();
 148         fix2             = _mm_setzero_ps();
 149         fiy2             = _mm_setzero_ps();
 150         fiz2             = _mm_setzero_ps();
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178             dx10             = _mm_sub_ps(ix1,jx0);
 179             dy10             = _mm_sub_ps(iy1,jy0);
 180             dz10             = _mm_sub_ps(iz1,jz0);
 181             dx20             = _mm_sub_ps(ix2,jx0);
 182             dy20             = _mm_sub_ps(iy2,jy0);
 183             dz20             = _mm_sub_ps(iz2,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 187             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 189
 190             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 191             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 192             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 193
 194             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 195             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 196             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201
 202             fjx0             = _mm_setzero_ps();
 203             fjy0             = _mm_setzero_ps();
 204             fjz0             = _mm_setzero_ps();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq00             = _mm_mul_ps(iq0,jq0);
 212
 213             /* REACTION-FIELD ELECTROSTATICS */
 214             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 215             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 216
 217             /* Update potential sum for this i atom from the interaction with this j atom. */
 218             velecsum         = _mm_add_ps(velecsum,velec);
 219
 220             fscal            = felec;
 221
 222              /* Update vectorial force */
 223             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 224             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 225             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 226
 227             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 228             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 229             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 230
 231             /**************************
 232              * CALCULATE INTERACTIONS *
 233              **************************/
 234
 235             /* Compute parameters for interactions between i and j atoms */
 236             qq10             = _mm_mul_ps(iq1,jq0);
 237
 238             /* REACTION-FIELD ELECTROSTATICS */
 239             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 240             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 241
 242             /* Update potential sum for this i atom from the interaction with this j atom. */
 243             velecsum         = _mm_add_ps(velecsum,velec);
 244
 245             fscal            = felec;
 246
 247              /* Update vectorial force */
 248             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 249             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 250             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 251
 252             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 253             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 254             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             /* Compute parameters for interactions between i and j atoms */
 261             qq20             = _mm_mul_ps(iq2,jq0);
 262
 263             /* REACTION-FIELD ELECTROSTATICS */
 264             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 265             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 266
 267             /* Update potential sum for this i atom from the interaction with this j atom. */
 268             velecsum         = _mm_add_ps(velecsum,velec);
 269
 270             fscal            = felec;
 271
 272              /* Update vectorial force */
 273             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 274             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 275             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 276
 277             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 278             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 279             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 280
 281             fjptrA             = f+j_coord_offsetA;
 282             fjptrB             = f+j_coord_offsetB;
 283             fjptrC             = f+j_coord_offsetC;
 284             fjptrD             = f+j_coord_offsetD;
 285
 286             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 287
 288             /* Inner loop uses 105 flops */
 289         }
 290
 291         if(jidx<j_index_end)
 292         {
 293
 294             /* Get j neighbor index, and coordinate index */
 295             jnrlistA         = jjnr[jidx];
 296             jnrlistB         = jjnr[jidx+1];
 297             jnrlistC         = jjnr[jidx+2];
 298             jnrlistD         = jjnr[jidx+3];
 299             /* Sign of each element will be negative for non-real atoms.
 300              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 301              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 302              */
 303             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 304             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 305             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 306             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 307             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 308             j_coord_offsetA  = DIM*jnrA;
 309             j_coord_offsetB  = DIM*jnrB;
 310             j_coord_offsetC  = DIM*jnrC;
 311             j_coord_offsetD  = DIM*jnrD;
 312
 313             /* load j atom coordinates */
 314             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 315                                               x+j_coord_offsetC,x+j_coord_offsetD,
 316                                               &jx0,&jy0,&jz0);
 317
 318             /* Calculate displacement vector */
 319             dx00             = _mm_sub_ps(ix0,jx0);
 320             dy00             = _mm_sub_ps(iy0,jy0);
 321             dz00             = _mm_sub_ps(iz0,jz0);
 322             dx10             = _mm_sub_ps(ix1,jx0);
 323             dy10             = _mm_sub_ps(iy1,jy0);
 324             dz10             = _mm_sub_ps(iz1,jz0);
 325             dx20             = _mm_sub_ps(ix2,jx0);
 326             dy20             = _mm_sub_ps(iy2,jy0);
 327             dz20             = _mm_sub_ps(iz2,jz0);
 328
 329             /* Calculate squared distance and things based on it */
 330             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 331             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 332             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 333
 334             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 335             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 336             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 337
 338             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 339             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 340             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 341
 342             /* Load parameters for j particles */
 343             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 344                                                               charge+jnrC+0,charge+jnrD+0);
 345
 346             fjx0             = _mm_setzero_ps();
 347             fjy0             = _mm_setzero_ps();
 348             fjz0             = _mm_setzero_ps();
 349
 350             /**************************
 351              * CALCULATE INTERACTIONS *
 352              **************************/
 353
 354             /* Compute parameters for interactions between i and j atoms */
 355             qq00             = _mm_mul_ps(iq0,jq0);
 356
 357             /* REACTION-FIELD ELECTROSTATICS */
 358             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 359             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velec            = _mm_andnot_ps(dummy_mask,velec);
 363             velecsum         = _mm_add_ps(velecsum,velec);
 364
 365             fscal            = felec;
 366
 367             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 368
 369              /* Update vectorial force */
 370             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 371             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 372             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 373
 374             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 375             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 376             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 377
 378             /**************************
 379              * CALCULATE INTERACTIONS *
 380              **************************/
 381
 382             /* Compute parameters for interactions between i and j atoms */
 383             qq10             = _mm_mul_ps(iq1,jq0);
 384
 385             /* REACTION-FIELD ELECTROSTATICS */
 386             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 387             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 388
 389             /* Update potential sum for this i atom from the interaction with this j atom. */
 390             velec            = _mm_andnot_ps(dummy_mask,velec);
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 396
 397              /* Update vectorial force */
 398             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 399             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 400             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 401
 402             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 403             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 404             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 405
 406             /**************************
 407              * CALCULATE INTERACTIONS *
 408              **************************/
 409
 410             /* Compute parameters for interactions between i and j atoms */
 411             qq20             = _mm_mul_ps(iq2,jq0);
 412
 413             /* REACTION-FIELD ELECTROSTATICS */
 414             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 415             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 416
 417             /* Update potential sum for this i atom from the interaction with this j atom. */
 418             velec            = _mm_andnot_ps(dummy_mask,velec);
 419             velecsum         = _mm_add_ps(velecsum,velec);
 420
 421             fscal            = felec;
 422
 423             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 424
 425              /* Update vectorial force */
 426             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 427             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 428             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 429
 430             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 431             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 432             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 433
 434             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 435             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 436             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 437             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 438
 439             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 440
 441             /* Inner loop uses 105 flops */
 442         }
 443
 444         /* End of innermost loop */
 445
 446         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 447                                               f+i_coord_offset,fshift+i_shift_offset);
 448
 449         ggid                        = gid[iidx];
 450         /* Update potential energies */
 451         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 452
 453         /* Increment number of inner iterations */
 454         inneriter                  += j_index_end - j_index_start;
 455
 456         /* Outer loop uses 19 flops */
 457     }
 458
 459     /* Increment number of outer iterations */
 460     outeriter        += nri;
 461
 462     /* Update outer/inner flops */
 463
 464     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*105);
 465 }
 466 /*
 467  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_F_avx_128_fma_single
 468  * Electrostatics interaction: ReactionField
 469  * VdW interaction:            None
 470  * Geometry:                   Water3-Particle
 471  * Calculate force/pot:        Force
 472  */
 473 void
 474 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_avx_128_fma_single
 475                     (t_nblist * gmx_restrict                nlist,
 476                      rvec * gmx_restrict                    xx,
 477                      rvec * gmx_restrict                    ff,
 478                      t_forcerec * gmx_restrict              fr,
 479                      t_mdatoms * gmx_restrict               mdatoms,
 480                      nb_kernel_data_t * gmx_restrict        kernel_data,
 481                      t_nrnb * gmx_restrict                  nrnb)
 482 {
 483     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 484      * just 0 for non-waters.
 485      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 486      * jnr indices corresponding to data put in the four positions in the SIMD register.
 487      */
 488     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 489     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 490     int              jnrA,jnrB,jnrC,jnrD;
 491     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 492     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 493     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 494     real             rcutoff_scalar;
 495     real             *shiftvec,*fshift,*x,*f;
 496     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 497     real             scratch[4*DIM];
 498     __m128           fscal,rcutoff,rcutoff2,jidxall;
 499     int              vdwioffset0;
 500     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 501     int              vdwioffset1;
 502     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 503     int              vdwioffset2;
 504     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 505     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 506     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 507     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 508     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 509     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 510     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 511     real             *charge;
 512     __m128           dummy_mask,cutoff_mask;
 513     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 514     __m128           one     = _mm_set1_ps(1.0);
 515     __m128           two     = _mm_set1_ps(2.0);
 516     x                = xx[0];
 517     f                = ff[0];
 518
 519     nri              = nlist->nri;
 520     iinr             = nlist->iinr;
 521     jindex           = nlist->jindex;
 522     jjnr             = nlist->jjnr;
 523     shiftidx         = nlist->shift;
 524     gid              = nlist->gid;
 525     shiftvec         = fr->shift_vec[0];
 526     fshift           = fr->fshift[0];
 527     facel            = _mm_set1_ps(fr->epsfac);
 528     charge           = mdatoms->chargeA;
 529     krf              = _mm_set1_ps(fr->ic->k_rf);
 530     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 531     crf              = _mm_set1_ps(fr->ic->c_rf);
 532
 533     /* Setup water-specific parameters */
 534     inr              = nlist->iinr[0];
 535     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 536     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 537     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 538
 539     /* Avoid stupid compiler warnings */
 540     jnrA = jnrB = jnrC = jnrD = 0;
 541     j_coord_offsetA = 0;
 542     j_coord_offsetB = 0;
 543     j_coord_offsetC = 0;
 544     j_coord_offsetD = 0;
 545
 546     outeriter        = 0;
 547     inneriter        = 0;
 548
 549     for(iidx=0;iidx<4*DIM;iidx++)
 550     {
 551         scratch[iidx] = 0.0;
 552     }
 553
 554     /* Start outer loop over neighborlists */
 555     for(iidx=0; iidx<nri; iidx++)
 556     {
 557         /* Load shift vector for this list */
 558         i_shift_offset   = DIM*shiftidx[iidx];
 559
 560         /* Load limits for loop over neighbors */
 561         j_index_start    = jindex[iidx];
 562         j_index_end      = jindex[iidx+1];
 563
 564         /* Get outer coordinate index */
 565         inr              = iinr[iidx];
 566         i_coord_offset   = DIM*inr;
 567
 568         /* Load i particle coords and add shift vector */
 569         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 570                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 571
 572         fix0             = _mm_setzero_ps();
 573         fiy0             = _mm_setzero_ps();
 574         fiz0             = _mm_setzero_ps();
 575         fix1             = _mm_setzero_ps();
 576         fiy1             = _mm_setzero_ps();
 577         fiz1             = _mm_setzero_ps();
 578         fix2             = _mm_setzero_ps();
 579         fiy2             = _mm_setzero_ps();
 580         fiz2             = _mm_setzero_ps();
 581
 582         /* Start inner kernel loop */
 583         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 584         {
 585
 586             /* Get j neighbor index, and coordinate index */
 587             jnrA             = jjnr[jidx];
 588             jnrB             = jjnr[jidx+1];
 589             jnrC             = jjnr[jidx+2];
 590             jnrD             = jjnr[jidx+3];
 591             j_coord_offsetA  = DIM*jnrA;
 592             j_coord_offsetB  = DIM*jnrB;
 593             j_coord_offsetC  = DIM*jnrC;
 594             j_coord_offsetD  = DIM*jnrD;
 595
 596             /* load j atom coordinates */
 597             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 598                                               x+j_coord_offsetC,x+j_coord_offsetD,
 599                                               &jx0,&jy0,&jz0);
 600
 601             /* Calculate displacement vector */
 602             dx00             = _mm_sub_ps(ix0,jx0);
 603             dy00             = _mm_sub_ps(iy0,jy0);
 604             dz00             = _mm_sub_ps(iz0,jz0);
 605             dx10             = _mm_sub_ps(ix1,jx0);
 606             dy10             = _mm_sub_ps(iy1,jy0);
 607             dz10             = _mm_sub_ps(iz1,jz0);
 608             dx20             = _mm_sub_ps(ix2,jx0);
 609             dy20             = _mm_sub_ps(iy2,jy0);
 610             dz20             = _mm_sub_ps(iz2,jz0);
 611
 612             /* Calculate squared distance and things based on it */
 613             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 614             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 615             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 616
 617             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 618             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 619             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 620
 621             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 622             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 623             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 624
 625             /* Load parameters for j particles */
 626             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 627                                                               charge+jnrC+0,charge+jnrD+0);
 628
 629             fjx0             = _mm_setzero_ps();
 630             fjy0             = _mm_setzero_ps();
 631             fjz0             = _mm_setzero_ps();
 632
 633             /**************************
 634              * CALCULATE INTERACTIONS *
 635              **************************/
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq00             = _mm_mul_ps(iq0,jq0);
 639
 640             /* REACTION-FIELD ELECTROSTATICS */
 641             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 642
 643             fscal            = felec;
 644
 645              /* Update vectorial force */
 646             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 647             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 648             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 649
 650             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 651             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 652             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 653
 654             /**************************
 655              * CALCULATE INTERACTIONS *
 656              **************************/
 657
 658             /* Compute parameters for interactions between i and j atoms */
 659             qq10             = _mm_mul_ps(iq1,jq0);
 660
 661             /* REACTION-FIELD ELECTROSTATICS */
 662             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 663
 664             fscal            = felec;
 665
 666              /* Update vectorial force */
 667             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 668             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 669             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 670
 671             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 672             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 673             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 674
 675             /**************************
 676              * CALCULATE INTERACTIONS *
 677              **************************/
 678
 679             /* Compute parameters for interactions between i and j atoms */
 680             qq20             = _mm_mul_ps(iq2,jq0);
 681
 682             /* REACTION-FIELD ELECTROSTATICS */
 683             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 684
 685             fscal            = felec;
 686
 687              /* Update vectorial force */
 688             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 689             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 690             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 691
 692             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 693             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 694             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 695
 696             fjptrA             = f+j_coord_offsetA;
 697             fjptrB             = f+j_coord_offsetB;
 698             fjptrC             = f+j_coord_offsetC;
 699             fjptrD             = f+j_coord_offsetD;
 700
 701             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 702
 703             /* Inner loop uses 90 flops */
 704         }
 705
 706         if(jidx<j_index_end)
 707         {
 708
 709             /* Get j neighbor index, and coordinate index */
 710             jnrlistA         = jjnr[jidx];
 711             jnrlistB         = jjnr[jidx+1];
 712             jnrlistC         = jjnr[jidx+2];
 713             jnrlistD         = jjnr[jidx+3];
 714             /* Sign of each element will be negative for non-real atoms.
 715              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 716              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 717              */
 718             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 719             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 720             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 721             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 722             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 723             j_coord_offsetA  = DIM*jnrA;
 724             j_coord_offsetB  = DIM*jnrB;
 725             j_coord_offsetC  = DIM*jnrC;
 726             j_coord_offsetD  = DIM*jnrD;
 727
 728             /* load j atom coordinates */
 729             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 730                                               x+j_coord_offsetC,x+j_coord_offsetD,
 731                                               &jx0,&jy0,&jz0);
 732
 733             /* Calculate displacement vector */
 734             dx00             = _mm_sub_ps(ix0,jx0);
 735             dy00             = _mm_sub_ps(iy0,jy0);
 736             dz00             = _mm_sub_ps(iz0,jz0);
 737             dx10             = _mm_sub_ps(ix1,jx0);
 738             dy10             = _mm_sub_ps(iy1,jy0);
 739             dz10             = _mm_sub_ps(iz1,jz0);
 740             dx20             = _mm_sub_ps(ix2,jx0);
 741             dy20             = _mm_sub_ps(iy2,jy0);
 742             dz20             = _mm_sub_ps(iz2,jz0);
 743
 744             /* Calculate squared distance and things based on it */
 745             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 746             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 747             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 748
 749             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 750             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 751             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 752
 753             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 754             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 755             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 756
 757             /* Load parameters for j particles */
 758             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 759                                                               charge+jnrC+0,charge+jnrD+0);
 760
 761             fjx0             = _mm_setzero_ps();
 762             fjy0             = _mm_setzero_ps();
 763             fjz0             = _mm_setzero_ps();
 764
 765             /**************************
 766              * CALCULATE INTERACTIONS *
 767              **************************/
 768
 769             /* Compute parameters for interactions between i and j atoms */
 770             qq00             = _mm_mul_ps(iq0,jq0);
 771
 772             /* REACTION-FIELD ELECTROSTATICS */
 773             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 774
 775             fscal            = felec;
 776
 777             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 778
 779              /* Update vectorial force */
 780             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 781             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 782             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 783
 784             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 785             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 786             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 787
 788             /**************************
 789              * CALCULATE INTERACTIONS *
 790              **************************/
 791
 792             /* Compute parameters for interactions between i and j atoms */
 793             qq10             = _mm_mul_ps(iq1,jq0);
 794
 795             /* REACTION-FIELD ELECTROSTATICS */
 796             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 797
 798             fscal            = felec;
 799
 800             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 801
 802              /* Update vectorial force */
 803             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 804             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 805             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 806
 807             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 808             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 809             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             /* Compute parameters for interactions between i and j atoms */
 816             qq20             = _mm_mul_ps(iq2,jq0);
 817
 818             /* REACTION-FIELD ELECTROSTATICS */
 819             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 820
 821             fscal            = felec;
 822
 823             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 824
 825              /* Update vectorial force */
 826             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 827             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 828             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 829
 830             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 831             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 832             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 833
 834             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 835             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 836             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 837             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 838
 839             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 840
 841             /* Inner loop uses 90 flops */
 842         }
 843
 844         /* End of innermost loop */
 845
 846         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 847                                               f+i_coord_offset,fshift+i_shift_offset);
 848
 849         /* Increment number of inner iterations */
 850         inneriter                  += j_index_end - j_index_start;
 851
 852         /* Outer loop uses 18 flops */
 853     }
 854
 855     /* Increment number of outer iterations */
 856     outeriter        += nri;
 857
 858     /* Update outer/inner flops */
 859
 860     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*90);
 861 }