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