src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwNone_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse4_1_single.h"
  48 #include "kernelutil_x86_sse4_1_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_single
  52  * Electrostatics interaction: ReactionField
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     krf              = _mm_set1_ps(fr->ic->k_rf);
 114     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 115     crf              = _mm_set1_ps(fr->ic->c_rf);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 120     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 121     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 122
 123     /* Avoid stupid compiler warnings */
 124     jnrA = jnrB = jnrC = jnrD = 0;
 125     j_coord_offsetA = 0;
 126     j_coord_offsetB = 0;
 127     j_coord_offsetC = 0;
 128     j_coord_offsetD = 0;
 129
 130     outeriter        = 0;
 131     inneriter        = 0;
 132
 133     for(iidx=0;iidx<4*DIM;iidx++)
 134     {
 135         scratch[iidx] = 0.0;
 136     }
 137
 138     /* Start outer loop over neighborlists */
 139     for(iidx=0; iidx<nri; iidx++)
 140     {
 141         /* Load shift vector for this list */
 142         i_shift_offset   = DIM*shiftidx[iidx];
 143
 144         /* Load limits for loop over neighbors */
 145         j_index_start    = jindex[iidx];
 146         j_index_end      = jindex[iidx+1];
 147
 148         /* Get outer coordinate index */
 149         inr              = iinr[iidx];
 150         i_coord_offset   = DIM*inr;
 151
 152         /* Load i particle coords and add shift vector */
 153         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 154                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 155
 156         fix0             = _mm_setzero_ps();
 157         fiy0             = _mm_setzero_ps();
 158         fiz0             = _mm_setzero_ps();
 159         fix1             = _mm_setzero_ps();
 160         fiy1             = _mm_setzero_ps();
 161         fiz1             = _mm_setzero_ps();
 162         fix2             = _mm_setzero_ps();
 163         fiy2             = _mm_setzero_ps();
 164         fiz2             = _mm_setzero_ps();
 165
 166         /* Reset potential sums */
 167         velecsum         = _mm_setzero_ps();
 168
 169         /* Start inner kernel loop */
 170         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 171         {
 172
 173             /* Get j neighbor index, and coordinate index */
 174             jnrA             = jjnr[jidx];
 175             jnrB             = jjnr[jidx+1];
 176             jnrC             = jjnr[jidx+2];
 177             jnrD             = jjnr[jidx+3];
 178             j_coord_offsetA  = DIM*jnrA;
 179             j_coord_offsetB  = DIM*jnrB;
 180             j_coord_offsetC  = DIM*jnrC;
 181             j_coord_offsetD  = DIM*jnrD;
 182
 183             /* load j atom coordinates */
 184             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 185                                               x+j_coord_offsetC,x+j_coord_offsetD,
 186                                               &jx0,&jy0,&jz0);
 187
 188             /* Calculate displacement vector */
 189             dx00             = _mm_sub_ps(ix0,jx0);
 190             dy00             = _mm_sub_ps(iy0,jy0);
 191             dz00             = _mm_sub_ps(iz0,jz0);
 192             dx10             = _mm_sub_ps(ix1,jx0);
 193             dy10             = _mm_sub_ps(iy1,jy0);
 194             dz10             = _mm_sub_ps(iz1,jz0);
 195             dx20             = _mm_sub_ps(ix2,jx0);
 196             dy20             = _mm_sub_ps(iy2,jy0);
 197             dz20             = _mm_sub_ps(iz2,jz0);
 198
 199             /* Calculate squared distance and things based on it */
 200             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 201             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 202             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 203
 204             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 205             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 206             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 207
 208             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 209             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 210             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 211
 212             /* Load parameters for j particles */
 213             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 214                                                               charge+jnrC+0,charge+jnrD+0);
 215
 216             fjx0             = _mm_setzero_ps();
 217             fjy0             = _mm_setzero_ps();
 218             fjz0             = _mm_setzero_ps();
 219
 220             /**************************
 221              * CALCULATE INTERACTIONS *
 222              **************************/
 223
 224             /* Compute parameters for interactions between i and j atoms */
 225             qq00             = _mm_mul_ps(iq0,jq0);
 226
 227             /* REACTION-FIELD ELECTROSTATICS */
 228             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 229             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 230
 231             /* Update potential sum for this i atom from the interaction with this j atom. */
 232             velecsum         = _mm_add_ps(velecsum,velec);
 233
 234             fscal            = felec;
 235
 236             /* Calculate temporary vectorial force */
 237             tx               = _mm_mul_ps(fscal,dx00);
 238             ty               = _mm_mul_ps(fscal,dy00);
 239             tz               = _mm_mul_ps(fscal,dz00);
 240
 241             /* Update vectorial force */
 242             fix0             = _mm_add_ps(fix0,tx);
 243             fiy0             = _mm_add_ps(fiy0,ty);
 244             fiz0             = _mm_add_ps(fiz0,tz);
 245
 246             fjx0             = _mm_add_ps(fjx0,tx);
 247             fjy0             = _mm_add_ps(fjy0,ty);
 248             fjz0             = _mm_add_ps(fjz0,tz);
 249
 250             /**************************
 251              * CALCULATE INTERACTIONS *
 252              **************************/
 253
 254             /* Compute parameters for interactions between i and j atoms */
 255             qq10             = _mm_mul_ps(iq1,jq0);
 256
 257             /* REACTION-FIELD ELECTROSTATICS */
 258             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 259             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 260
 261             /* Update potential sum for this i atom from the interaction with this j atom. */
 262             velecsum         = _mm_add_ps(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             /* Calculate temporary vectorial force */
 267             tx               = _mm_mul_ps(fscal,dx10);
 268             ty               = _mm_mul_ps(fscal,dy10);
 269             tz               = _mm_mul_ps(fscal,dz10);
 270
 271             /* Update vectorial force */
 272             fix1             = _mm_add_ps(fix1,tx);
 273             fiy1             = _mm_add_ps(fiy1,ty);
 274             fiz1             = _mm_add_ps(fiz1,tz);
 275
 276             fjx0             = _mm_add_ps(fjx0,tx);
 277             fjy0             = _mm_add_ps(fjy0,ty);
 278             fjz0             = _mm_add_ps(fjz0,tz);
 279
 280             /**************************
 281              * CALCULATE INTERACTIONS *
 282              **************************/
 283
 284             /* Compute parameters for interactions between i and j atoms */
 285             qq20             = _mm_mul_ps(iq2,jq0);
 286
 287             /* REACTION-FIELD ELECTROSTATICS */
 288             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 289             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 290
 291             /* Update potential sum for this i atom from the interaction with this j atom. */
 292             velecsum         = _mm_add_ps(velecsum,velec);
 293
 294             fscal            = felec;
 295
 296             /* Calculate temporary vectorial force */
 297             tx               = _mm_mul_ps(fscal,dx20);
 298             ty               = _mm_mul_ps(fscal,dy20);
 299             tz               = _mm_mul_ps(fscal,dz20);
 300
 301             /* Update vectorial force */
 302             fix2             = _mm_add_ps(fix2,tx);
 303             fiy2             = _mm_add_ps(fiy2,ty);
 304             fiz2             = _mm_add_ps(fiz2,tz);
 305
 306             fjx0             = _mm_add_ps(fjx0,tx);
 307             fjy0             = _mm_add_ps(fjy0,ty);
 308             fjz0             = _mm_add_ps(fjz0,tz);
 309
 310             fjptrA             = f+j_coord_offsetA;
 311             fjptrB             = f+j_coord_offsetB;
 312             fjptrC             = f+j_coord_offsetC;
 313             fjptrD             = f+j_coord_offsetD;
 314
 315             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 316
 317             /* Inner loop uses 96 flops */
 318         }
 319
 320         if(jidx<j_index_end)
 321         {
 322
 323             /* Get j neighbor index, and coordinate index */
 324             jnrlistA         = jjnr[jidx];
 325             jnrlistB         = jjnr[jidx+1];
 326             jnrlistC         = jjnr[jidx+2];
 327             jnrlistD         = jjnr[jidx+3];
 328             /* Sign of each element will be negative for non-real atoms.
 329              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 330              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 331              */
 332             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 333             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 334             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 335             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 336             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 337             j_coord_offsetA  = DIM*jnrA;
 338             j_coord_offsetB  = DIM*jnrB;
 339             j_coord_offsetC  = DIM*jnrC;
 340             j_coord_offsetD  = DIM*jnrD;
 341
 342             /* load j atom coordinates */
 343             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 344                                               x+j_coord_offsetC,x+j_coord_offsetD,
 345                                               &jx0,&jy0,&jz0);
 346
 347             /* Calculate displacement vector */
 348             dx00             = _mm_sub_ps(ix0,jx0);
 349             dy00             = _mm_sub_ps(iy0,jy0);
 350             dz00             = _mm_sub_ps(iz0,jz0);
 351             dx10             = _mm_sub_ps(ix1,jx0);
 352             dy10             = _mm_sub_ps(iy1,jy0);
 353             dz10             = _mm_sub_ps(iz1,jz0);
 354             dx20             = _mm_sub_ps(ix2,jx0);
 355             dy20             = _mm_sub_ps(iy2,jy0);
 356             dz20             = _mm_sub_ps(iz2,jz0);
 357
 358             /* Calculate squared distance and things based on it */
 359             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 360             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 361             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 362
 363             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 364             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 365             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 366
 367             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 368             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 369             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 370
 371             /* Load parameters for j particles */
 372             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 373                                                               charge+jnrC+0,charge+jnrD+0);
 374
 375             fjx0             = _mm_setzero_ps();
 376             fjy0             = _mm_setzero_ps();
 377             fjz0             = _mm_setzero_ps();
 378
 379             /**************************
 380              * CALCULATE INTERACTIONS *
 381              **************************/
 382
 383             /* Compute parameters for interactions between i and j atoms */
 384             qq00             = _mm_mul_ps(iq0,jq0);
 385
 386             /* REACTION-FIELD ELECTROSTATICS */
 387             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 388             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velec            = _mm_andnot_ps(dummy_mask,velec);
 392             velecsum         = _mm_add_ps(velecsum,velec);
 393
 394             fscal            = felec;
 395
 396             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx00);
 400             ty               = _mm_mul_ps(fscal,dy00);
 401             tz               = _mm_mul_ps(fscal,dz00);
 402
 403             /* Update vectorial force */
 404             fix0             = _mm_add_ps(fix0,tx);
 405             fiy0             = _mm_add_ps(fiy0,ty);
 406             fiz0             = _mm_add_ps(fiz0,tz);
 407
 408             fjx0             = _mm_add_ps(fjx0,tx);
 409             fjy0             = _mm_add_ps(fjy0,ty);
 410             fjz0             = _mm_add_ps(fjz0,tz);
 411
 412             /**************************
 413              * CALCULATE INTERACTIONS *
 414              **************************/
 415
 416             /* Compute parameters for interactions between i and j atoms */
 417             qq10             = _mm_mul_ps(iq1,jq0);
 418
 419             /* REACTION-FIELD ELECTROSTATICS */
 420             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 421             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 422
 423             /* Update potential sum for this i atom from the interaction with this j atom. */
 424             velec            = _mm_andnot_ps(dummy_mask,velec);
 425             velecsum         = _mm_add_ps(velecsum,velec);
 426
 427             fscal            = felec;
 428
 429             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 430
 431             /* Calculate temporary vectorial force */
 432             tx               = _mm_mul_ps(fscal,dx10);
 433             ty               = _mm_mul_ps(fscal,dy10);
 434             tz               = _mm_mul_ps(fscal,dz10);
 435
 436             /* Update vectorial force */
 437             fix1             = _mm_add_ps(fix1,tx);
 438             fiy1             = _mm_add_ps(fiy1,ty);
 439             fiz1             = _mm_add_ps(fiz1,tz);
 440
 441             fjx0             = _mm_add_ps(fjx0,tx);
 442             fjy0             = _mm_add_ps(fjy0,ty);
 443             fjz0             = _mm_add_ps(fjz0,tz);
 444
 445             /**************************
 446              * CALCULATE INTERACTIONS *
 447              **************************/
 448
 449             /* Compute parameters for interactions between i and j atoms */
 450             qq20             = _mm_mul_ps(iq2,jq0);
 451
 452             /* REACTION-FIELD ELECTROSTATICS */
 453             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 454             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_andnot_ps(dummy_mask,velec);
 458             velecsum         = _mm_add_ps(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm_mul_ps(fscal,dx20);
 466             ty               = _mm_mul_ps(fscal,dy20);
 467             tz               = _mm_mul_ps(fscal,dz20);
 468
 469             /* Update vectorial force */
 470             fix2             = _mm_add_ps(fix2,tx);
 471             fiy2             = _mm_add_ps(fiy2,ty);
 472             fiz2             = _mm_add_ps(fiz2,tz);
 473
 474             fjx0             = _mm_add_ps(fjx0,tx);
 475             fjy0             = _mm_add_ps(fjy0,ty);
 476             fjz0             = _mm_add_ps(fjz0,tz);
 477
 478             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 479             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 480             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 481             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 482
 483             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 484
 485             /* Inner loop uses 96 flops */
 486         }
 487
 488         /* End of innermost loop */
 489
 490         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 491                                               f+i_coord_offset,fshift+i_shift_offset);
 492
 493         ggid                        = gid[iidx];
 494         /* Update potential energies */
 495         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 496
 497         /* Increment number of inner iterations */
 498         inneriter                  += j_index_end - j_index_start;
 499
 500         /* Outer loop uses 19 flops */
 501     }
 502
 503     /* Increment number of outer iterations */
 504     outeriter        += nri;
 505
 506     /* Update outer/inner flops */
 507
 508     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
 509 }
 510 /*
 511  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
 512  * Electrostatics interaction: ReactionField
 513  * VdW interaction:            None
 514  * Geometry:                   Water3-Particle
 515  * Calculate force/pot:        Force
 516  */
 517 void
 518 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
 519                     (t_nblist                    * gmx_restrict       nlist,
 520                      rvec                        * gmx_restrict          xx,
 521                      rvec                        * gmx_restrict          ff,
 522                      t_forcerec                  * gmx_restrict          fr,
 523                      t_mdatoms                   * gmx_restrict     mdatoms,
 524                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 525                      t_nrnb                      * gmx_restrict        nrnb)
 526 {
 527     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 528      * just 0 for non-waters.
 529      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 530      * jnr indices corresponding to data put in the four positions in the SIMD register.
 531      */
 532     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 533     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 534     int              jnrA,jnrB,jnrC,jnrD;
 535     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 536     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 537     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 538     real             rcutoff_scalar;
 539     real             *shiftvec,*fshift,*x,*f;
 540     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 541     real             scratch[4*DIM];
 542     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 543     int              vdwioffset0;
 544     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 545     int              vdwioffset1;
 546     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 547     int              vdwioffset2;
 548     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 549     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 550     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 551     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 552     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 553     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 554     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 555     real             *charge;
 556     __m128           dummy_mask,cutoff_mask;
 557     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 558     __m128           one     = _mm_set1_ps(1.0);
 559     __m128           two     = _mm_set1_ps(2.0);
 560     x                = xx[0];
 561     f                = ff[0];
 562
 563     nri              = nlist->nri;
 564     iinr             = nlist->iinr;
 565     jindex           = nlist->jindex;
 566     jjnr             = nlist->jjnr;
 567     shiftidx         = nlist->shift;
 568     gid              = nlist->gid;
 569     shiftvec         = fr->shift_vec[0];
 570     fshift           = fr->fshift[0];
 571     facel            = _mm_set1_ps(fr->epsfac);
 572     charge           = mdatoms->chargeA;
 573     krf              = _mm_set1_ps(fr->ic->k_rf);
 574     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 575     crf              = _mm_set1_ps(fr->ic->c_rf);
 576
 577     /* Setup water-specific parameters */
 578     inr              = nlist->iinr[0];
 579     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 580     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 581     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 582
 583     /* Avoid stupid compiler warnings */
 584     jnrA = jnrB = jnrC = jnrD = 0;
 585     j_coord_offsetA = 0;
 586     j_coord_offsetB = 0;
 587     j_coord_offsetC = 0;
 588     j_coord_offsetD = 0;
 589
 590     outeriter        = 0;
 591     inneriter        = 0;
 592
 593     for(iidx=0;iidx<4*DIM;iidx++)
 594     {
 595         scratch[iidx] = 0.0;
 596     }
 597
 598     /* Start outer loop over neighborlists */
 599     for(iidx=0; iidx<nri; iidx++)
 600     {
 601         /* Load shift vector for this list */
 602         i_shift_offset   = DIM*shiftidx[iidx];
 603
 604         /* Load limits for loop over neighbors */
 605         j_index_start    = jindex[iidx];
 606         j_index_end      = jindex[iidx+1];
 607
 608         /* Get outer coordinate index */
 609         inr              = iinr[iidx];
 610         i_coord_offset   = DIM*inr;
 611
 612         /* Load i particle coords and add shift vector */
 613         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 614                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 615
 616         fix0             = _mm_setzero_ps();
 617         fiy0             = _mm_setzero_ps();
 618         fiz0             = _mm_setzero_ps();
 619         fix1             = _mm_setzero_ps();
 620         fiy1             = _mm_setzero_ps();
 621         fiz1             = _mm_setzero_ps();
 622         fix2             = _mm_setzero_ps();
 623         fiy2             = _mm_setzero_ps();
 624         fiz2             = _mm_setzero_ps();
 625
 626         /* Start inner kernel loop */
 627         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 628         {
 629
 630             /* Get j neighbor index, and coordinate index */
 631             jnrA             = jjnr[jidx];
 632             jnrB             = jjnr[jidx+1];
 633             jnrC             = jjnr[jidx+2];
 634             jnrD             = jjnr[jidx+3];
 635             j_coord_offsetA  = DIM*jnrA;
 636             j_coord_offsetB  = DIM*jnrB;
 637             j_coord_offsetC  = DIM*jnrC;
 638             j_coord_offsetD  = DIM*jnrD;
 639
 640             /* load j atom coordinates */
 641             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 642                                               x+j_coord_offsetC,x+j_coord_offsetD,
 643                                               &jx0,&jy0,&jz0);
 644
 645             /* Calculate displacement vector */
 646             dx00             = _mm_sub_ps(ix0,jx0);
 647             dy00             = _mm_sub_ps(iy0,jy0);
 648             dz00             = _mm_sub_ps(iz0,jz0);
 649             dx10             = _mm_sub_ps(ix1,jx0);
 650             dy10             = _mm_sub_ps(iy1,jy0);
 651             dz10             = _mm_sub_ps(iz1,jz0);
 652             dx20             = _mm_sub_ps(ix2,jx0);
 653             dy20             = _mm_sub_ps(iy2,jy0);
 654             dz20             = _mm_sub_ps(iz2,jz0);
 655
 656             /* Calculate squared distance and things based on it */
 657             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 658             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 659             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 660
 661             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 662             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 663             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 664
 665             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 666             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 667             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 668
 669             /* Load parameters for j particles */
 670             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 671                                                               charge+jnrC+0,charge+jnrD+0);
 672
 673             fjx0             = _mm_setzero_ps();
 674             fjy0             = _mm_setzero_ps();
 675             fjz0             = _mm_setzero_ps();
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             /* Compute parameters for interactions between i and j atoms */
 682             qq00             = _mm_mul_ps(iq0,jq0);
 683
 684             /* REACTION-FIELD ELECTROSTATICS */
 685             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 686
 687             fscal            = felec;
 688
 689             /* Calculate temporary vectorial force */
 690             tx               = _mm_mul_ps(fscal,dx00);
 691             ty               = _mm_mul_ps(fscal,dy00);
 692             tz               = _mm_mul_ps(fscal,dz00);
 693
 694             /* Update vectorial force */
 695             fix0             = _mm_add_ps(fix0,tx);
 696             fiy0             = _mm_add_ps(fiy0,ty);
 697             fiz0             = _mm_add_ps(fiz0,tz);
 698
 699             fjx0             = _mm_add_ps(fjx0,tx);
 700             fjy0             = _mm_add_ps(fjy0,ty);
 701             fjz0             = _mm_add_ps(fjz0,tz);
 702
 703             /**************************
 704              * CALCULATE INTERACTIONS *
 705              **************************/
 706
 707             /* Compute parameters for interactions between i and j atoms */
 708             qq10             = _mm_mul_ps(iq1,jq0);
 709
 710             /* REACTION-FIELD ELECTROSTATICS */
 711             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 712
 713             fscal            = felec;
 714
 715             /* Calculate temporary vectorial force */
 716             tx               = _mm_mul_ps(fscal,dx10);
 717             ty               = _mm_mul_ps(fscal,dy10);
 718             tz               = _mm_mul_ps(fscal,dz10);
 719
 720             /* Update vectorial force */
 721             fix1             = _mm_add_ps(fix1,tx);
 722             fiy1             = _mm_add_ps(fiy1,ty);
 723             fiz1             = _mm_add_ps(fiz1,tz);
 724
 725             fjx0             = _mm_add_ps(fjx0,tx);
 726             fjy0             = _mm_add_ps(fjy0,ty);
 727             fjz0             = _mm_add_ps(fjz0,tz);
 728
 729             /**************************
 730              * CALCULATE INTERACTIONS *
 731              **************************/
 732
 733             /* Compute parameters for interactions between i and j atoms */
 734             qq20             = _mm_mul_ps(iq2,jq0);
 735
 736             /* REACTION-FIELD ELECTROSTATICS */
 737             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 738
 739             fscal            = felec;
 740
 741             /* Calculate temporary vectorial force */
 742             tx               = _mm_mul_ps(fscal,dx20);
 743             ty               = _mm_mul_ps(fscal,dy20);
 744             tz               = _mm_mul_ps(fscal,dz20);
 745
 746             /* Update vectorial force */
 747             fix2             = _mm_add_ps(fix2,tx);
 748             fiy2             = _mm_add_ps(fiy2,ty);
 749             fiz2             = _mm_add_ps(fiz2,tz);
 750
 751             fjx0             = _mm_add_ps(fjx0,tx);
 752             fjy0             = _mm_add_ps(fjy0,ty);
 753             fjz0             = _mm_add_ps(fjz0,tz);
 754
 755             fjptrA             = f+j_coord_offsetA;
 756             fjptrB             = f+j_coord_offsetB;
 757             fjptrC             = f+j_coord_offsetC;
 758             fjptrD             = f+j_coord_offsetD;
 759
 760             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 761
 762             /* Inner loop uses 81 flops */
 763         }
 764
 765         if(jidx<j_index_end)
 766         {
 767
 768             /* Get j neighbor index, and coordinate index */
 769             jnrlistA         = jjnr[jidx];
 770             jnrlistB         = jjnr[jidx+1];
 771             jnrlistC         = jjnr[jidx+2];
 772             jnrlistD         = jjnr[jidx+3];
 773             /* Sign of each element will be negative for non-real atoms.
 774              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 775              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 776              */
 777             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 778             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 779             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 780             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 781             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 782             j_coord_offsetA  = DIM*jnrA;
 783             j_coord_offsetB  = DIM*jnrB;
 784             j_coord_offsetC  = DIM*jnrC;
 785             j_coord_offsetD  = DIM*jnrD;
 786
 787             /* load j atom coordinates */
 788             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 789                                               x+j_coord_offsetC,x+j_coord_offsetD,
 790                                               &jx0,&jy0,&jz0);
 791
 792             /* Calculate displacement vector */
 793             dx00             = _mm_sub_ps(ix0,jx0);
 794             dy00             = _mm_sub_ps(iy0,jy0);
 795             dz00             = _mm_sub_ps(iz0,jz0);
 796             dx10             = _mm_sub_ps(ix1,jx0);
 797             dy10             = _mm_sub_ps(iy1,jy0);
 798             dz10             = _mm_sub_ps(iz1,jz0);
 799             dx20             = _mm_sub_ps(ix2,jx0);
 800             dy20             = _mm_sub_ps(iy2,jy0);
 801             dz20             = _mm_sub_ps(iz2,jz0);
 802
 803             /* Calculate squared distance and things based on it */
 804             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 805             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 806             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 807
 808             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 809             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 810             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 811
 812             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 813             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 814             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 815
 816             /* Load parameters for j particles */
 817             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 818                                                               charge+jnrC+0,charge+jnrD+0);
 819
 820             fjx0             = _mm_setzero_ps();
 821             fjy0             = _mm_setzero_ps();
 822             fjz0             = _mm_setzero_ps();
 823
 824             /**************************
 825              * CALCULATE INTERACTIONS *
 826              **************************/
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq00             = _mm_mul_ps(iq0,jq0);
 830
 831             /* REACTION-FIELD ELECTROSTATICS */
 832             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 833
 834             fscal            = felec;
 835
 836             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 837
 838             /* Calculate temporary vectorial force */
 839             tx               = _mm_mul_ps(fscal,dx00);
 840             ty               = _mm_mul_ps(fscal,dy00);
 841             tz               = _mm_mul_ps(fscal,dz00);
 842
 843             /* Update vectorial force */
 844             fix0             = _mm_add_ps(fix0,tx);
 845             fiy0             = _mm_add_ps(fiy0,ty);
 846             fiz0             = _mm_add_ps(fiz0,tz);
 847
 848             fjx0             = _mm_add_ps(fjx0,tx);
 849             fjy0             = _mm_add_ps(fjy0,ty);
 850             fjz0             = _mm_add_ps(fjz0,tz);
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             /* Compute parameters for interactions between i and j atoms */
 857             qq10             = _mm_mul_ps(iq1,jq0);
 858
 859             /* REACTION-FIELD ELECTROSTATICS */
 860             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 861
 862             fscal            = felec;
 863
 864             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 865
 866             /* Calculate temporary vectorial force */
 867             tx               = _mm_mul_ps(fscal,dx10);
 868             ty               = _mm_mul_ps(fscal,dy10);
 869             tz               = _mm_mul_ps(fscal,dz10);
 870
 871             /* Update vectorial force */
 872             fix1             = _mm_add_ps(fix1,tx);
 873             fiy1             = _mm_add_ps(fiy1,ty);
 874             fiz1             = _mm_add_ps(fiz1,tz);
 875
 876             fjx0             = _mm_add_ps(fjx0,tx);
 877             fjy0             = _mm_add_ps(fjy0,ty);
 878             fjz0             = _mm_add_ps(fjz0,tz);
 879
 880             /**************************
 881              * CALCULATE INTERACTIONS *
 882              **************************/
 883
 884             /* Compute parameters for interactions between i and j atoms */
 885             qq20             = _mm_mul_ps(iq2,jq0);
 886
 887             /* REACTION-FIELD ELECTROSTATICS */
 888             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 889
 890             fscal            = felec;
 891
 892             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 893
 894             /* Calculate temporary vectorial force */
 895             tx               = _mm_mul_ps(fscal,dx20);
 896             ty               = _mm_mul_ps(fscal,dy20);
 897             tz               = _mm_mul_ps(fscal,dz20);
 898
 899             /* Update vectorial force */
 900             fix2             = _mm_add_ps(fix2,tx);
 901             fiy2             = _mm_add_ps(fiy2,ty);
 902             fiz2             = _mm_add_ps(fiz2,tz);
 903
 904             fjx0             = _mm_add_ps(fjx0,tx);
 905             fjy0             = _mm_add_ps(fjy0,ty);
 906             fjz0             = _mm_add_ps(fjz0,tz);
 907
 908             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 909             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 910             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 911             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 912
 913             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 914
 915             /* Inner loop uses 81 flops */
 916         }
 917
 918         /* End of innermost loop */
 919
 920         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 921                                               f+i_coord_offset,fshift+i_shift_offset);
 922
 923         /* Increment number of inner iterations */
 924         inneriter                  += j_index_end - j_index_start;
 925
 926         /* Outer loop uses 18 flops */
 927     }
 928
 929     /* Increment number of outer iterations */
 930     outeriter        += nri;
 931
 932     /* Update outer/inner flops */
 933
 934     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*81);
 935 }