src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecCoul_VdwNone_GeomW3P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_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_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_VF_sse2_single
  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_sse2_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 SSE, 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           tx,ty,tz,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
 100     /* Setup water-specific parameters */
 101     inr              = nlist->iinr[0];
 102     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 103     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 104     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 105
 106     /* Avoid stupid compiler warnings */
 107     jnrA = jnrB = jnrC = jnrD = 0;
 108     j_coord_offsetA = 0;
 109     j_coord_offsetB = 0;
 110     j_coord_offsetC = 0;
 111     j_coord_offsetD = 0;
 112
 113     outeriter        = 0;
 114     inneriter        = 0;
 115
 116     for(iidx=0;iidx<4*DIM;iidx++)
 117     {
 118         scratch[iidx] = 0.0;
 119     }
 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_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 137                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 138
 139         fix0             = _mm_setzero_ps();
 140         fiy0             = _mm_setzero_ps();
 141         fiz0             = _mm_setzero_ps();
 142         fix1             = _mm_setzero_ps();
 143         fiy1             = _mm_setzero_ps();
 144         fiz1             = _mm_setzero_ps();
 145         fix2             = _mm_setzero_ps();
 146         fiy2             = _mm_setzero_ps();
 147         fiz2             = _mm_setzero_ps();
 148
 149         /* Reset potential sums */
 150         velecsum         = _mm_setzero_ps();
 151
 152         /* Start inner kernel loop */
 153         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 154         {
 155
 156             /* Get j neighbor index, and coordinate index */
 157             jnrA             = jjnr[jidx];
 158             jnrB             = jjnr[jidx+1];
 159             jnrC             = jjnr[jidx+2];
 160             jnrD             = jjnr[jidx+3];
 161             j_coord_offsetA  = DIM*jnrA;
 162             j_coord_offsetB  = DIM*jnrB;
 163             j_coord_offsetC  = DIM*jnrC;
 164             j_coord_offsetD  = DIM*jnrD;
 165
 166             /* load j atom coordinates */
 167             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 168                                               x+j_coord_offsetC,x+j_coord_offsetD,
 169                                               &jx0,&jy0,&jz0);
 170
 171             /* Calculate displacement vector */
 172             dx00             = _mm_sub_ps(ix0,jx0);
 173             dy00             = _mm_sub_ps(iy0,jy0);
 174             dz00             = _mm_sub_ps(iz0,jz0);
 175             dx10             = _mm_sub_ps(ix1,jx0);
 176             dy10             = _mm_sub_ps(iy1,jy0);
 177             dz10             = _mm_sub_ps(iz1,jz0);
 178             dx20             = _mm_sub_ps(ix2,jx0);
 179             dy20             = _mm_sub_ps(iy2,jy0);
 180             dz20             = _mm_sub_ps(iz2,jz0);
 181
 182             /* Calculate squared distance and things based on it */
 183             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 184             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 185             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 186
 187             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 188             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 189             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 190
 191             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 192             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 193             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 194
 195             /* Load parameters for j particles */
 196             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 197                                                               charge+jnrC+0,charge+jnrD+0);
 198
 199             fjx0             = _mm_setzero_ps();
 200             fjy0             = _mm_setzero_ps();
 201             fjz0             = _mm_setzero_ps();
 202
 203             /**************************
 204              * CALCULATE INTERACTIONS *
 205              **************************/
 206
 207             /* Compute parameters for interactions between i and j atoms */
 208             qq00             = _mm_mul_ps(iq0,jq0);
 209
 210             /* COULOMB ELECTROSTATICS */
 211             velec            = _mm_mul_ps(qq00,rinv00);
 212             felec            = _mm_mul_ps(velec,rinvsq00);
 213
 214             /* Update potential sum for this i atom from the interaction with this j atom. */
 215             velecsum         = _mm_add_ps(velecsum,velec);
 216
 217             fscal            = felec;
 218
 219             /* Calculate temporary vectorial force */
 220             tx               = _mm_mul_ps(fscal,dx00);
 221             ty               = _mm_mul_ps(fscal,dy00);
 222             tz               = _mm_mul_ps(fscal,dz00);
 223
 224             /* Update vectorial force */
 225             fix0             = _mm_add_ps(fix0,tx);
 226             fiy0             = _mm_add_ps(fiy0,ty);
 227             fiz0             = _mm_add_ps(fiz0,tz);
 228
 229             fjx0             = _mm_add_ps(fjx0,tx);
 230             fjy0             = _mm_add_ps(fjy0,ty);
 231             fjz0             = _mm_add_ps(fjz0,tz);
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             qq10             = _mm_mul_ps(iq1,jq0);
 239
 240             /* COULOMB ELECTROSTATICS */
 241             velec            = _mm_mul_ps(qq10,rinv10);
 242             felec            = _mm_mul_ps(velec,rinvsq10);
 243
 244             /* Update potential sum for this i atom from the interaction with this j atom. */
 245             velecsum         = _mm_add_ps(velecsum,velec);
 246
 247             fscal            = felec;
 248
 249             /* Calculate temporary vectorial force */
 250             tx               = _mm_mul_ps(fscal,dx10);
 251             ty               = _mm_mul_ps(fscal,dy10);
 252             tz               = _mm_mul_ps(fscal,dz10);
 253
 254             /* Update vectorial force */
 255             fix1             = _mm_add_ps(fix1,tx);
 256             fiy1             = _mm_add_ps(fiy1,ty);
 257             fiz1             = _mm_add_ps(fiz1,tz);
 258
 259             fjx0             = _mm_add_ps(fjx0,tx);
 260             fjy0             = _mm_add_ps(fjy0,ty);
 261             fjz0             = _mm_add_ps(fjz0,tz);
 262
 263             /**************************
 264              * CALCULATE INTERACTIONS *
 265              **************************/
 266
 267             /* Compute parameters for interactions between i and j atoms */
 268             qq20             = _mm_mul_ps(iq2,jq0);
 269
 270             /* COULOMB ELECTROSTATICS */
 271             velec            = _mm_mul_ps(qq20,rinv20);
 272             felec            = _mm_mul_ps(velec,rinvsq20);
 273
 274             /* Update potential sum for this i atom from the interaction with this j atom. */
 275             velecsum         = _mm_add_ps(velecsum,velec);
 276
 277             fscal            = felec;
 278
 279             /* Calculate temporary vectorial force */
 280             tx               = _mm_mul_ps(fscal,dx20);
 281             ty               = _mm_mul_ps(fscal,dy20);
 282             tz               = _mm_mul_ps(fscal,dz20);
 283
 284             /* Update vectorial force */
 285             fix2             = _mm_add_ps(fix2,tx);
 286             fiy2             = _mm_add_ps(fiy2,ty);
 287             fiz2             = _mm_add_ps(fiz2,tz);
 288
 289             fjx0             = _mm_add_ps(fjx0,tx);
 290             fjy0             = _mm_add_ps(fjy0,ty);
 291             fjz0             = _mm_add_ps(fjz0,tz);
 292
 293             fjptrA             = f+j_coord_offsetA;
 294             fjptrB             = f+j_coord_offsetB;
 295             fjptrC             = f+j_coord_offsetC;
 296             fjptrD             = f+j_coord_offsetD;
 297
 298             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 299
 300             /* Inner loop uses 84 flops */
 301         }
 302
 303         if(jidx<j_index_end)
 304         {
 305
 306             /* Get j neighbor index, and coordinate index */
 307             jnrlistA         = jjnr[jidx];
 308             jnrlistB         = jjnr[jidx+1];
 309             jnrlistC         = jjnr[jidx+2];
 310             jnrlistD         = jjnr[jidx+3];
 311             /* Sign of each element will be negative for non-real atoms.
 312              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 313              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 314              */
 315             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 316             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 317             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 318             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 319             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 320             j_coord_offsetA  = DIM*jnrA;
 321             j_coord_offsetB  = DIM*jnrB;
 322             j_coord_offsetC  = DIM*jnrC;
 323             j_coord_offsetD  = DIM*jnrD;
 324
 325             /* load j atom coordinates */
 326             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 327                                               x+j_coord_offsetC,x+j_coord_offsetD,
 328                                               &jx0,&jy0,&jz0);
 329
 330             /* Calculate displacement vector */
 331             dx00             = _mm_sub_ps(ix0,jx0);
 332             dy00             = _mm_sub_ps(iy0,jy0);
 333             dz00             = _mm_sub_ps(iz0,jz0);
 334             dx10             = _mm_sub_ps(ix1,jx0);
 335             dy10             = _mm_sub_ps(iy1,jy0);
 336             dz10             = _mm_sub_ps(iz1,jz0);
 337             dx20             = _mm_sub_ps(ix2,jx0);
 338             dy20             = _mm_sub_ps(iy2,jy0);
 339             dz20             = _mm_sub_ps(iz2,jz0);
 340
 341             /* Calculate squared distance and things based on it */
 342             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 343             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 344             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 345
 346             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 347             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 348             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 349
 350             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 351             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 352             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 353
 354             /* Load parameters for j particles */
 355             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 356                                                               charge+jnrC+0,charge+jnrD+0);
 357
 358             fjx0             = _mm_setzero_ps();
 359             fjy0             = _mm_setzero_ps();
 360             fjz0             = _mm_setzero_ps();
 361
 362             /**************************
 363              * CALCULATE INTERACTIONS *
 364              **************************/
 365
 366             /* Compute parameters for interactions between i and j atoms */
 367             qq00             = _mm_mul_ps(iq0,jq0);
 368
 369             /* COULOMB ELECTROSTATICS */
 370             velec            = _mm_mul_ps(qq00,rinv00);
 371             felec            = _mm_mul_ps(velec,rinvsq00);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_andnot_ps(dummy_mask,velec);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx00);
 383             ty               = _mm_mul_ps(fscal,dy00);
 384             tz               = _mm_mul_ps(fscal,dz00);
 385
 386             /* Update vectorial force */
 387             fix0             = _mm_add_ps(fix0,tx);
 388             fiy0             = _mm_add_ps(fiy0,ty);
 389             fiz0             = _mm_add_ps(fiz0,tz);
 390
 391             fjx0             = _mm_add_ps(fjx0,tx);
 392             fjy0             = _mm_add_ps(fjy0,ty);
 393             fjz0             = _mm_add_ps(fjz0,tz);
 394
 395             /**************************
 396              * CALCULATE INTERACTIONS *
 397              **************************/
 398
 399             /* Compute parameters for interactions between i and j atoms */
 400             qq10             = _mm_mul_ps(iq1,jq0);
 401
 402             /* COULOMB ELECTROSTATICS */
 403             velec            = _mm_mul_ps(qq10,rinv10);
 404             felec            = _mm_mul_ps(velec,rinvsq10);
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velec            = _mm_andnot_ps(dummy_mask,velec);
 408             velecsum         = _mm_add_ps(velecsum,velec);
 409
 410             fscal            = felec;
 411
 412             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 413
 414             /* Calculate temporary vectorial force */
 415             tx               = _mm_mul_ps(fscal,dx10);
 416             ty               = _mm_mul_ps(fscal,dy10);
 417             tz               = _mm_mul_ps(fscal,dz10);
 418
 419             /* Update vectorial force */
 420             fix1             = _mm_add_ps(fix1,tx);
 421             fiy1             = _mm_add_ps(fiy1,ty);
 422             fiz1             = _mm_add_ps(fiz1,tz);
 423
 424             fjx0             = _mm_add_ps(fjx0,tx);
 425             fjy0             = _mm_add_ps(fjy0,ty);
 426             fjz0             = _mm_add_ps(fjz0,tz);
 427
 428             /**************************
 429              * CALCULATE INTERACTIONS *
 430              **************************/
 431
 432             /* Compute parameters for interactions between i and j atoms */
 433             qq20             = _mm_mul_ps(iq2,jq0);
 434
 435             /* COULOMB ELECTROSTATICS */
 436             velec            = _mm_mul_ps(qq20,rinv20);
 437             felec            = _mm_mul_ps(velec,rinvsq20);
 438
 439             /* Update potential sum for this i atom from the interaction with this j atom. */
 440             velec            = _mm_andnot_ps(dummy_mask,velec);
 441             velecsum         = _mm_add_ps(velecsum,velec);
 442
 443             fscal            = felec;
 444
 445             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 446
 447             /* Calculate temporary vectorial force */
 448             tx               = _mm_mul_ps(fscal,dx20);
 449             ty               = _mm_mul_ps(fscal,dy20);
 450             tz               = _mm_mul_ps(fscal,dz20);
 451
 452             /* Update vectorial force */
 453             fix2             = _mm_add_ps(fix2,tx);
 454             fiy2             = _mm_add_ps(fiy2,ty);
 455             fiz2             = _mm_add_ps(fiz2,tz);
 456
 457             fjx0             = _mm_add_ps(fjx0,tx);
 458             fjy0             = _mm_add_ps(fjy0,ty);
 459             fjz0             = _mm_add_ps(fjz0,tz);
 460
 461             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 462             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 463             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 464             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 465
 466             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 467
 468             /* Inner loop uses 84 flops */
 469         }
 470
 471         /* End of innermost loop */
 472
 473         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 474                                               f+i_coord_offset,fshift+i_shift_offset);
 475
 476         ggid                        = gid[iidx];
 477         /* Update potential energies */
 478         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 479
 480         /* Increment number of inner iterations */
 481         inneriter                  += j_index_end - j_index_start;
 482
 483         /* Outer loop uses 19 flops */
 484     }
 485
 486     /* Increment number of outer iterations */
 487     outeriter        += nri;
 488
 489     /* Update outer/inner flops */
 490
 491     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*84);
 492 }
 493 /*
 494  * Gromacs nonbonded kernel:   nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
 495  * Electrostatics interaction: Coulomb
 496  * VdW interaction:            None
 497  * Geometry:                   Water3-Particle
 498  * Calculate force/pot:        Force
 499  */
 500 void
 501 nb_kernel_ElecCoul_VdwNone_GeomW3P1_F_sse2_single
 502                     (t_nblist * gmx_restrict                nlist,
 503                      rvec * gmx_restrict                    xx,
 504                      rvec * gmx_restrict                    ff,
 505                      t_forcerec * gmx_restrict              fr,
 506                      t_mdatoms * gmx_restrict               mdatoms,
 507                      nb_kernel_data_t * gmx_restrict        kernel_data,
 508                      t_nrnb * gmx_restrict                  nrnb)
 509 {
 510     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 511      * just 0 for non-waters.
 512      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 513      * jnr indices corresponding to data put in the four positions in the SIMD register.
 514      */
 515     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 516     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 517     int              jnrA,jnrB,jnrC,jnrD;
 518     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 519     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 520     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 521     real             rcutoff_scalar;
 522     real             *shiftvec,*fshift,*x,*f;
 523     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 524     real             scratch[4*DIM];
 525     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 526     int              vdwioffset0;
 527     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 528     int              vdwioffset1;
 529     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 530     int              vdwioffset2;
 531     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 532     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 533     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 534     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 535     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 536     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 537     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 538     real             *charge;
 539     __m128           dummy_mask,cutoff_mask;
 540     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 541     __m128           one     = _mm_set1_ps(1.0);
 542     __m128           two     = _mm_set1_ps(2.0);
 543     x                = xx[0];
 544     f                = ff[0];
 545
 546     nri              = nlist->nri;
 547     iinr             = nlist->iinr;
 548     jindex           = nlist->jindex;
 549     jjnr             = nlist->jjnr;
 550     shiftidx         = nlist->shift;
 551     gid              = nlist->gid;
 552     shiftvec         = fr->shift_vec[0];
 553     fshift           = fr->fshift[0];
 554     facel            = _mm_set1_ps(fr->epsfac);
 555     charge           = mdatoms->chargeA;
 556
 557     /* Setup water-specific parameters */
 558     inr              = nlist->iinr[0];
 559     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 560     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 561     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 562
 563     /* Avoid stupid compiler warnings */
 564     jnrA = jnrB = jnrC = jnrD = 0;
 565     j_coord_offsetA = 0;
 566     j_coord_offsetB = 0;
 567     j_coord_offsetC = 0;
 568     j_coord_offsetD = 0;
 569
 570     outeriter        = 0;
 571     inneriter        = 0;
 572
 573     for(iidx=0;iidx<4*DIM;iidx++)
 574     {
 575         scratch[iidx] = 0.0;
 576     }
 577
 578     /* Start outer loop over neighborlists */
 579     for(iidx=0; iidx<nri; iidx++)
 580     {
 581         /* Load shift vector for this list */
 582         i_shift_offset   = DIM*shiftidx[iidx];
 583
 584         /* Load limits for loop over neighbors */
 585         j_index_start    = jindex[iidx];
 586         j_index_end      = jindex[iidx+1];
 587
 588         /* Get outer coordinate index */
 589         inr              = iinr[iidx];
 590         i_coord_offset   = DIM*inr;
 591
 592         /* Load i particle coords and add shift vector */
 593         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 594                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 595
 596         fix0             = _mm_setzero_ps();
 597         fiy0             = _mm_setzero_ps();
 598         fiz0             = _mm_setzero_ps();
 599         fix1             = _mm_setzero_ps();
 600         fiy1             = _mm_setzero_ps();
 601         fiz1             = _mm_setzero_ps();
 602         fix2             = _mm_setzero_ps();
 603         fiy2             = _mm_setzero_ps();
 604         fiz2             = _mm_setzero_ps();
 605
 606         /* Start inner kernel loop */
 607         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 608         {
 609
 610             /* Get j neighbor index, and coordinate index */
 611             jnrA             = jjnr[jidx];
 612             jnrB             = jjnr[jidx+1];
 613             jnrC             = jjnr[jidx+2];
 614             jnrD             = jjnr[jidx+3];
 615             j_coord_offsetA  = DIM*jnrA;
 616             j_coord_offsetB  = DIM*jnrB;
 617             j_coord_offsetC  = DIM*jnrC;
 618             j_coord_offsetD  = DIM*jnrD;
 619
 620             /* load j atom coordinates */
 621             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 622                                               x+j_coord_offsetC,x+j_coord_offsetD,
 623                                               &jx0,&jy0,&jz0);
 624
 625             /* Calculate displacement vector */
 626             dx00             = _mm_sub_ps(ix0,jx0);
 627             dy00             = _mm_sub_ps(iy0,jy0);
 628             dz00             = _mm_sub_ps(iz0,jz0);
 629             dx10             = _mm_sub_ps(ix1,jx0);
 630             dy10             = _mm_sub_ps(iy1,jy0);
 631             dz10             = _mm_sub_ps(iz1,jz0);
 632             dx20             = _mm_sub_ps(ix2,jx0);
 633             dy20             = _mm_sub_ps(iy2,jy0);
 634             dz20             = _mm_sub_ps(iz2,jz0);
 635
 636             /* Calculate squared distance and things based on it */
 637             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 638             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 639             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 640
 641             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 642             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 643             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 644
 645             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 646             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 647             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 648
 649             /* Load parameters for j particles */
 650             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 651                                                               charge+jnrC+0,charge+jnrD+0);
 652
 653             fjx0             = _mm_setzero_ps();
 654             fjy0             = _mm_setzero_ps();
 655             fjz0             = _mm_setzero_ps();
 656
 657             /**************************
 658              * CALCULATE INTERACTIONS *
 659              **************************/
 660
 661             /* Compute parameters for interactions between i and j atoms */
 662             qq00             = _mm_mul_ps(iq0,jq0);
 663
 664             /* COULOMB ELECTROSTATICS */
 665             velec            = _mm_mul_ps(qq00,rinv00);
 666             felec            = _mm_mul_ps(velec,rinvsq00);
 667
 668             fscal            = felec;
 669
 670             /* Calculate temporary vectorial force */
 671             tx               = _mm_mul_ps(fscal,dx00);
 672             ty               = _mm_mul_ps(fscal,dy00);
 673             tz               = _mm_mul_ps(fscal,dz00);
 674
 675             /* Update vectorial force */
 676             fix0             = _mm_add_ps(fix0,tx);
 677             fiy0             = _mm_add_ps(fiy0,ty);
 678             fiz0             = _mm_add_ps(fiz0,tz);
 679
 680             fjx0             = _mm_add_ps(fjx0,tx);
 681             fjy0             = _mm_add_ps(fjy0,ty);
 682             fjz0             = _mm_add_ps(fjz0,tz);
 683
 684             /**************************
 685              * CALCULATE INTERACTIONS *
 686              **************************/
 687
 688             /* Compute parameters for interactions between i and j atoms */
 689             qq10             = _mm_mul_ps(iq1,jq0);
 690
 691             /* COULOMB ELECTROSTATICS */
 692             velec            = _mm_mul_ps(qq10,rinv10);
 693             felec            = _mm_mul_ps(velec,rinvsq10);
 694
 695             fscal            = felec;
 696
 697             /* Calculate temporary vectorial force */
 698             tx               = _mm_mul_ps(fscal,dx10);
 699             ty               = _mm_mul_ps(fscal,dy10);
 700             tz               = _mm_mul_ps(fscal,dz10);
 701
 702             /* Update vectorial force */
 703             fix1             = _mm_add_ps(fix1,tx);
 704             fiy1             = _mm_add_ps(fiy1,ty);
 705             fiz1             = _mm_add_ps(fiz1,tz);
 706
 707             fjx0             = _mm_add_ps(fjx0,tx);
 708             fjy0             = _mm_add_ps(fjy0,ty);
 709             fjz0             = _mm_add_ps(fjz0,tz);
 710
 711             /**************************
 712              * CALCULATE INTERACTIONS *
 713              **************************/
 714
 715             /* Compute parameters for interactions between i and j atoms */
 716             qq20             = _mm_mul_ps(iq2,jq0);
 717
 718             /* COULOMB ELECTROSTATICS */
 719             velec            = _mm_mul_ps(qq20,rinv20);
 720             felec            = _mm_mul_ps(velec,rinvsq20);
 721
 722             fscal            = felec;
 723
 724             /* Calculate temporary vectorial force */
 725             tx               = _mm_mul_ps(fscal,dx20);
 726             ty               = _mm_mul_ps(fscal,dy20);
 727             tz               = _mm_mul_ps(fscal,dz20);
 728
 729             /* Update vectorial force */
 730             fix2             = _mm_add_ps(fix2,tx);
 731             fiy2             = _mm_add_ps(fiy2,ty);
 732             fiz2             = _mm_add_ps(fiz2,tz);
 733
 734             fjx0             = _mm_add_ps(fjx0,tx);
 735             fjy0             = _mm_add_ps(fjy0,ty);
 736             fjz0             = _mm_add_ps(fjz0,tz);
 737
 738             fjptrA             = f+j_coord_offsetA;
 739             fjptrB             = f+j_coord_offsetB;
 740             fjptrC             = f+j_coord_offsetC;
 741             fjptrD             = f+j_coord_offsetD;
 742
 743             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 744
 745             /* Inner loop uses 81 flops */
 746         }
 747
 748         if(jidx<j_index_end)
 749         {
 750
 751             /* Get j neighbor index, and coordinate index */
 752             jnrlistA         = jjnr[jidx];
 753             jnrlistB         = jjnr[jidx+1];
 754             jnrlistC         = jjnr[jidx+2];
 755             jnrlistD         = jjnr[jidx+3];
 756             /* Sign of each element will be negative for non-real atoms.
 757              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 758              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 759              */
 760             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 761             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 762             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 763             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 764             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 765             j_coord_offsetA  = DIM*jnrA;
 766             j_coord_offsetB  = DIM*jnrB;
 767             j_coord_offsetC  = DIM*jnrC;
 768             j_coord_offsetD  = DIM*jnrD;
 769
 770             /* load j atom coordinates */
 771             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 772                                               x+j_coord_offsetC,x+j_coord_offsetD,
 773                                               &jx0,&jy0,&jz0);
 774
 775             /* Calculate displacement vector */
 776             dx00             = _mm_sub_ps(ix0,jx0);
 777             dy00             = _mm_sub_ps(iy0,jy0);
 778             dz00             = _mm_sub_ps(iz0,jz0);
 779             dx10             = _mm_sub_ps(ix1,jx0);
 780             dy10             = _mm_sub_ps(iy1,jy0);
 781             dz10             = _mm_sub_ps(iz1,jz0);
 782             dx20             = _mm_sub_ps(ix2,jx0);
 783             dy20             = _mm_sub_ps(iy2,jy0);
 784             dz20             = _mm_sub_ps(iz2,jz0);
 785
 786             /* Calculate squared distance and things based on it */
 787             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 788             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 789             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 790
 791             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 792             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 793             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 794
 795             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 796             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 797             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 798
 799             /* Load parameters for j particles */
 800             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 801                                                               charge+jnrC+0,charge+jnrD+0);
 802
 803             fjx0             = _mm_setzero_ps();
 804             fjy0             = _mm_setzero_ps();
 805             fjz0             = _mm_setzero_ps();
 806
 807             /**************************
 808              * CALCULATE INTERACTIONS *
 809              **************************/
 810
 811             /* Compute parameters for interactions between i and j atoms */
 812             qq00             = _mm_mul_ps(iq0,jq0);
 813
 814             /* COULOMB ELECTROSTATICS */
 815             velec            = _mm_mul_ps(qq00,rinv00);
 816             felec            = _mm_mul_ps(velec,rinvsq00);
 817
 818             fscal            = felec;
 819
 820             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 821
 822             /* Calculate temporary vectorial force */
 823             tx               = _mm_mul_ps(fscal,dx00);
 824             ty               = _mm_mul_ps(fscal,dy00);
 825             tz               = _mm_mul_ps(fscal,dz00);
 826
 827             /* Update vectorial force */
 828             fix0             = _mm_add_ps(fix0,tx);
 829             fiy0             = _mm_add_ps(fiy0,ty);
 830             fiz0             = _mm_add_ps(fiz0,tz);
 831
 832             fjx0             = _mm_add_ps(fjx0,tx);
 833             fjy0             = _mm_add_ps(fjy0,ty);
 834             fjz0             = _mm_add_ps(fjz0,tz);
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             qq10             = _mm_mul_ps(iq1,jq0);
 842
 843             /* COULOMB ELECTROSTATICS */
 844             velec            = _mm_mul_ps(qq10,rinv10);
 845             felec            = _mm_mul_ps(velec,rinvsq10);
 846
 847             fscal            = felec;
 848
 849             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 850
 851             /* Calculate temporary vectorial force */
 852             tx               = _mm_mul_ps(fscal,dx10);
 853             ty               = _mm_mul_ps(fscal,dy10);
 854             tz               = _mm_mul_ps(fscal,dz10);
 855
 856             /* Update vectorial force */
 857             fix1             = _mm_add_ps(fix1,tx);
 858             fiy1             = _mm_add_ps(fiy1,ty);
 859             fiz1             = _mm_add_ps(fiz1,tz);
 860
 861             fjx0             = _mm_add_ps(fjx0,tx);
 862             fjy0             = _mm_add_ps(fjy0,ty);
 863             fjz0             = _mm_add_ps(fjz0,tz);
 864
 865             /**************************
 866              * CALCULATE INTERACTIONS *
 867              **************************/
 868
 869             /* Compute parameters for interactions between i and j atoms */
 870             qq20             = _mm_mul_ps(iq2,jq0);
 871
 872             /* COULOMB ELECTROSTATICS */
 873             velec            = _mm_mul_ps(qq20,rinv20);
 874             felec            = _mm_mul_ps(velec,rinvsq20);
 875
 876             fscal            = felec;
 877
 878             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 879
 880             /* Calculate temporary vectorial force */
 881             tx               = _mm_mul_ps(fscal,dx20);
 882             ty               = _mm_mul_ps(fscal,dy20);
 883             tz               = _mm_mul_ps(fscal,dz20);
 884
 885             /* Update vectorial force */
 886             fix2             = _mm_add_ps(fix2,tx);
 887             fiy2             = _mm_add_ps(fiy2,ty);
 888             fiz2             = _mm_add_ps(fiz2,tz);
 889
 890             fjx0             = _mm_add_ps(fjx0,tx);
 891             fjy0             = _mm_add_ps(fjy0,ty);
 892             fjz0             = _mm_add_ps(fjz0,tz);
 893
 894             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 895             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 896             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 897             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 898
 899             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 900
 901             /* Inner loop uses 81 flops */
 902         }
 903
 904         /* End of innermost loop */
 905
 906         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 907                                               f+i_coord_offset,fshift+i_shift_offset);
 908
 909         /* Increment number of inner iterations */
 910         inneriter                  += j_index_end - j_index_start;
 911
 912         /* Outer loop uses 18 flops */
 913     }
 914
 915     /* Increment number of outer iterations */
 916     outeriter        += nri;
 917
 918     /* Update outer/inner flops */
 919
 920     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
 921 }