src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecRF_VdwNone_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomW3P1_VF_sse4_1_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     krf              = _mm_set1_ps(fr->ic->k_rf);
 100     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 101     crf              = _mm_set1_ps(fr->ic->c_rf);
 102
 103     /* Setup water-specific parameters */
 104     inr              = nlist->iinr[0];
 105     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 106     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 107     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 108
 109     /* Avoid stupid compiler warnings */
 110     jnrA = jnrB = jnrC = jnrD = 0;
 111     j_coord_offsetA = 0;
 112     j_coord_offsetB = 0;
 113     j_coord_offsetC = 0;
 114     j_coord_offsetD = 0;
 115
 116     outeriter        = 0;
 117     inneriter        = 0;
 118
 119     for(iidx=0;iidx<4*DIM;iidx++)
 120     {
 121         scratch[iidx] = 0.0;
 122     }
 123
 124     /* Start outer loop over neighborlists */
 125     for(iidx=0; iidx<nri; iidx++)
 126     {
 127         /* Load shift vector for this list */
 128         i_shift_offset   = DIM*shiftidx[iidx];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 140                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 141
 142         fix0             = _mm_setzero_ps();
 143         fiy0             = _mm_setzero_ps();
 144         fiz0             = _mm_setzero_ps();
 145         fix1             = _mm_setzero_ps();
 146         fiy1             = _mm_setzero_ps();
 147         fiz1             = _mm_setzero_ps();
 148         fix2             = _mm_setzero_ps();
 149         fiy2             = _mm_setzero_ps();
 150         fiz2             = _mm_setzero_ps();
 151
 152         /* Reset potential sums */
 153         velecsum         = _mm_setzero_ps();
 154
 155         /* Start inner kernel loop */
 156         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 157         {
 158
 159             /* Get j neighbor index, and coordinate index */
 160             jnrA             = jjnr[jidx];
 161             jnrB             = jjnr[jidx+1];
 162             jnrC             = jjnr[jidx+2];
 163             jnrD             = jjnr[jidx+3];
 164             j_coord_offsetA  = DIM*jnrA;
 165             j_coord_offsetB  = DIM*jnrB;
 166             j_coord_offsetC  = DIM*jnrC;
 167             j_coord_offsetD  = DIM*jnrD;
 168
 169             /* load j atom coordinates */
 170             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 171                                               x+j_coord_offsetC,x+j_coord_offsetD,
 172                                               &jx0,&jy0,&jz0);
 173
 174             /* Calculate displacement vector */
 175             dx00             = _mm_sub_ps(ix0,jx0);
 176             dy00             = _mm_sub_ps(iy0,jy0);
 177             dz00             = _mm_sub_ps(iz0,jz0);
 178             dx10             = _mm_sub_ps(ix1,jx0);
 179             dy10             = _mm_sub_ps(iy1,jy0);
 180             dz10             = _mm_sub_ps(iz1,jz0);
 181             dx20             = _mm_sub_ps(ix2,jx0);
 182             dy20             = _mm_sub_ps(iy2,jy0);
 183             dz20             = _mm_sub_ps(iz2,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 187             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 188             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 189
 190             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 191             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 192             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 193
 194             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 195             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 196             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 197
 198             /* Load parameters for j particles */
 199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 200                                                               charge+jnrC+0,charge+jnrD+0);
 201
 202             fjx0             = _mm_setzero_ps();
 203             fjy0             = _mm_setzero_ps();
 204             fjz0             = _mm_setzero_ps();
 205
 206             /**************************
 207              * CALCULATE INTERACTIONS *
 208              **************************/
 209
 210             /* Compute parameters for interactions between i and j atoms */
 211             qq00             = _mm_mul_ps(iq0,jq0);
 212
 213             /* REACTION-FIELD ELECTROSTATICS */
 214             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 215             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 216
 217             /* Update potential sum for this i atom from the interaction with this j atom. */
 218             velecsum         = _mm_add_ps(velecsum,velec);
 219
 220             fscal            = felec;
 221
 222             /* Calculate temporary vectorial force */
 223             tx               = _mm_mul_ps(fscal,dx00);
 224             ty               = _mm_mul_ps(fscal,dy00);
 225             tz               = _mm_mul_ps(fscal,dz00);
 226
 227             /* Update vectorial force */
 228             fix0             = _mm_add_ps(fix0,tx);
 229             fiy0             = _mm_add_ps(fiy0,ty);
 230             fiz0             = _mm_add_ps(fiz0,tz);
 231
 232             fjx0             = _mm_add_ps(fjx0,tx);
 233             fjy0             = _mm_add_ps(fjy0,ty);
 234             fjz0             = _mm_add_ps(fjz0,tz);
 235
 236             /**************************
 237              * CALCULATE INTERACTIONS *
 238              **************************/
 239
 240             /* Compute parameters for interactions between i and j atoms */
 241             qq10             = _mm_mul_ps(iq1,jq0);
 242
 243             /* REACTION-FIELD ELECTROSTATICS */
 244             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 245             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 246
 247             /* Update potential sum for this i atom from the interaction with this j atom. */
 248             velecsum         = _mm_add_ps(velecsum,velec);
 249
 250             fscal            = felec;
 251
 252             /* Calculate temporary vectorial force */
 253             tx               = _mm_mul_ps(fscal,dx10);
 254             ty               = _mm_mul_ps(fscal,dy10);
 255             tz               = _mm_mul_ps(fscal,dz10);
 256
 257             /* Update vectorial force */
 258             fix1             = _mm_add_ps(fix1,tx);
 259             fiy1             = _mm_add_ps(fiy1,ty);
 260             fiz1             = _mm_add_ps(fiz1,tz);
 261
 262             fjx0             = _mm_add_ps(fjx0,tx);
 263             fjy0             = _mm_add_ps(fjy0,ty);
 264             fjz0             = _mm_add_ps(fjz0,tz);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq20             = _mm_mul_ps(iq2,jq0);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 275             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 276
 277             /* Update potential sum for this i atom from the interaction with this j atom. */
 278             velecsum         = _mm_add_ps(velecsum,velec);
 279
 280             fscal            = felec;
 281
 282             /* Calculate temporary vectorial force */
 283             tx               = _mm_mul_ps(fscal,dx20);
 284             ty               = _mm_mul_ps(fscal,dy20);
 285             tz               = _mm_mul_ps(fscal,dz20);
 286
 287             /* Update vectorial force */
 288             fix2             = _mm_add_ps(fix2,tx);
 289             fiy2             = _mm_add_ps(fiy2,ty);
 290             fiz2             = _mm_add_ps(fiz2,tz);
 291
 292             fjx0             = _mm_add_ps(fjx0,tx);
 293             fjy0             = _mm_add_ps(fjy0,ty);
 294             fjz0             = _mm_add_ps(fjz0,tz);
 295
 296             fjptrA             = f+j_coord_offsetA;
 297             fjptrB             = f+j_coord_offsetB;
 298             fjptrC             = f+j_coord_offsetC;
 299             fjptrD             = f+j_coord_offsetD;
 300
 301             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 302
 303             /* Inner loop uses 96 flops */
 304         }
 305
 306         if(jidx<j_index_end)
 307         {
 308
 309             /* Get j neighbor index, and coordinate index */
 310             jnrlistA         = jjnr[jidx];
 311             jnrlistB         = jjnr[jidx+1];
 312             jnrlistC         = jjnr[jidx+2];
 313             jnrlistD         = jjnr[jidx+3];
 314             /* Sign of each element will be negative for non-real atoms.
 315              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 316              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 317              */
 318             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 319             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 320             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 321             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 322             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 323             j_coord_offsetA  = DIM*jnrA;
 324             j_coord_offsetB  = DIM*jnrB;
 325             j_coord_offsetC  = DIM*jnrC;
 326             j_coord_offsetD  = DIM*jnrD;
 327
 328             /* load j atom coordinates */
 329             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 330                                               x+j_coord_offsetC,x+j_coord_offsetD,
 331                                               &jx0,&jy0,&jz0);
 332
 333             /* Calculate displacement vector */
 334             dx00             = _mm_sub_ps(ix0,jx0);
 335             dy00             = _mm_sub_ps(iy0,jy0);
 336             dz00             = _mm_sub_ps(iz0,jz0);
 337             dx10             = _mm_sub_ps(ix1,jx0);
 338             dy10             = _mm_sub_ps(iy1,jy0);
 339             dz10             = _mm_sub_ps(iz1,jz0);
 340             dx20             = _mm_sub_ps(ix2,jx0);
 341             dy20             = _mm_sub_ps(iy2,jy0);
 342             dz20             = _mm_sub_ps(iz2,jz0);
 343
 344             /* Calculate squared distance and things based on it */
 345             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 346             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 347             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 348
 349             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 350             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 351             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 352
 353             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 354             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 355             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 356
 357             /* Load parameters for j particles */
 358             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 359                                                               charge+jnrC+0,charge+jnrD+0);
 360
 361             fjx0             = _mm_setzero_ps();
 362             fjy0             = _mm_setzero_ps();
 363             fjz0             = _mm_setzero_ps();
 364
 365             /**************************
 366              * CALCULATE INTERACTIONS *
 367              **************************/
 368
 369             /* Compute parameters for interactions between i and j atoms */
 370             qq00             = _mm_mul_ps(iq0,jq0);
 371
 372             /* REACTION-FIELD ELECTROSTATICS */
 373             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_add_ps(rinv00,_mm_mul_ps(krf,rsq00)),crf));
 374             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velec            = _mm_andnot_ps(dummy_mask,velec);
 378             velecsum         = _mm_add_ps(velecsum,velec);
 379
 380             fscal            = felec;
 381
 382             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 383
 384             /* Calculate temporary vectorial force */
 385             tx               = _mm_mul_ps(fscal,dx00);
 386             ty               = _mm_mul_ps(fscal,dy00);
 387             tz               = _mm_mul_ps(fscal,dz00);
 388
 389             /* Update vectorial force */
 390             fix0             = _mm_add_ps(fix0,tx);
 391             fiy0             = _mm_add_ps(fiy0,ty);
 392             fiz0             = _mm_add_ps(fiz0,tz);
 393
 394             fjx0             = _mm_add_ps(fjx0,tx);
 395             fjy0             = _mm_add_ps(fjy0,ty);
 396             fjz0             = _mm_add_ps(fjz0,tz);
 397
 398             /**************************
 399              * CALCULATE INTERACTIONS *
 400              **************************/
 401
 402             /* Compute parameters for interactions between i and j atoms */
 403             qq10             = _mm_mul_ps(iq1,jq0);
 404
 405             /* REACTION-FIELD ELECTROSTATICS */
 406             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 407             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 408
 409             /* Update potential sum for this i atom from the interaction with this j atom. */
 410             velec            = _mm_andnot_ps(dummy_mask,velec);
 411             velecsum         = _mm_add_ps(velecsum,velec);
 412
 413             fscal            = felec;
 414
 415             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 416
 417             /* Calculate temporary vectorial force */
 418             tx               = _mm_mul_ps(fscal,dx10);
 419             ty               = _mm_mul_ps(fscal,dy10);
 420             tz               = _mm_mul_ps(fscal,dz10);
 421
 422             /* Update vectorial force */
 423             fix1             = _mm_add_ps(fix1,tx);
 424             fiy1             = _mm_add_ps(fiy1,ty);
 425             fiz1             = _mm_add_ps(fiz1,tz);
 426
 427             fjx0             = _mm_add_ps(fjx0,tx);
 428             fjy0             = _mm_add_ps(fjy0,ty);
 429             fjz0             = _mm_add_ps(fjz0,tz);
 430
 431             /**************************
 432              * CALCULATE INTERACTIONS *
 433              **************************/
 434
 435             /* Compute parameters for interactions between i and j atoms */
 436             qq20             = _mm_mul_ps(iq2,jq0);
 437
 438             /* REACTION-FIELD ELECTROSTATICS */
 439             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 440             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 441
 442             /* Update potential sum for this i atom from the interaction with this j atom. */
 443             velec            = _mm_andnot_ps(dummy_mask,velec);
 444             velecsum         = _mm_add_ps(velecsum,velec);
 445
 446             fscal            = felec;
 447
 448             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 449
 450             /* Calculate temporary vectorial force */
 451             tx               = _mm_mul_ps(fscal,dx20);
 452             ty               = _mm_mul_ps(fscal,dy20);
 453             tz               = _mm_mul_ps(fscal,dz20);
 454
 455             /* Update vectorial force */
 456             fix2             = _mm_add_ps(fix2,tx);
 457             fiy2             = _mm_add_ps(fiy2,ty);
 458             fiz2             = _mm_add_ps(fiz2,tz);
 459
 460             fjx0             = _mm_add_ps(fjx0,tx);
 461             fjy0             = _mm_add_ps(fjy0,ty);
 462             fjz0             = _mm_add_ps(fjz0,tz);
 463
 464             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 465             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 466             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 467             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 468
 469             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 470
 471             /* Inner loop uses 96 flops */
 472         }
 473
 474         /* End of innermost loop */
 475
 476         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 477                                               f+i_coord_offset,fshift+i_shift_offset);
 478
 479         ggid                        = gid[iidx];
 480         /* Update potential energies */
 481         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 482
 483         /* Increment number of inner iterations */
 484         inneriter                  += j_index_end - j_index_start;
 485
 486         /* Outer loop uses 19 flops */
 487     }
 488
 489     /* Increment number of outer iterations */
 490     outeriter        += nri;
 491
 492     /* Update outer/inner flops */
 493
 494     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*96);
 495 }
 496 /*
 497  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
 498  * Electrostatics interaction: ReactionField
 499  * VdW interaction:            None
 500  * Geometry:                   Water3-Particle
 501  * Calculate force/pot:        Force
 502  */
 503 void
 504 nb_kernel_ElecRF_VdwNone_GeomW3P1_F_sse4_1_single
 505                     (t_nblist * gmx_restrict                nlist,
 506                      rvec * gmx_restrict                    xx,
 507                      rvec * gmx_restrict                    ff,
 508                      t_forcerec * gmx_restrict              fr,
 509                      t_mdatoms * gmx_restrict               mdatoms,
 510                      nb_kernel_data_t * gmx_restrict        kernel_data,
 511                      t_nrnb * gmx_restrict                  nrnb)
 512 {
 513     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 514      * just 0 for non-waters.
 515      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 516      * jnr indices corresponding to data put in the four positions in the SIMD register.
 517      */
 518     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 519     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 520     int              jnrA,jnrB,jnrC,jnrD;
 521     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 522     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 523     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 524     real             rcutoff_scalar;
 525     real             *shiftvec,*fshift,*x,*f;
 526     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 527     real             scratch[4*DIM];
 528     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 529     int              vdwioffset0;
 530     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 531     int              vdwioffset1;
 532     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 533     int              vdwioffset2;
 534     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 535     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 536     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 537     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 538     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 539     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 540     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 541     real             *charge;
 542     __m128           dummy_mask,cutoff_mask;
 543     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 544     __m128           one     = _mm_set1_ps(1.0);
 545     __m128           two     = _mm_set1_ps(2.0);
 546     x                = xx[0];
 547     f                = ff[0];
 548
 549     nri              = nlist->nri;
 550     iinr             = nlist->iinr;
 551     jindex           = nlist->jindex;
 552     jjnr             = nlist->jjnr;
 553     shiftidx         = nlist->shift;
 554     gid              = nlist->gid;
 555     shiftvec         = fr->shift_vec[0];
 556     fshift           = fr->fshift[0];
 557     facel            = _mm_set1_ps(fr->epsfac);
 558     charge           = mdatoms->chargeA;
 559     krf              = _mm_set1_ps(fr->ic->k_rf);
 560     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 561     crf              = _mm_set1_ps(fr->ic->c_rf);
 562
 563     /* Setup water-specific parameters */
 564     inr              = nlist->iinr[0];
 565     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 566     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 567     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 568
 569     /* Avoid stupid compiler warnings */
 570     jnrA = jnrB = jnrC = jnrD = 0;
 571     j_coord_offsetA = 0;
 572     j_coord_offsetB = 0;
 573     j_coord_offsetC = 0;
 574     j_coord_offsetD = 0;
 575
 576     outeriter        = 0;
 577     inneriter        = 0;
 578
 579     for(iidx=0;iidx<4*DIM;iidx++)
 580     {
 581         scratch[iidx] = 0.0;
 582     }
 583
 584     /* Start outer loop over neighborlists */
 585     for(iidx=0; iidx<nri; iidx++)
 586     {
 587         /* Load shift vector for this list */
 588         i_shift_offset   = DIM*shiftidx[iidx];
 589
 590         /* Load limits for loop over neighbors */
 591         j_index_start    = jindex[iidx];
 592         j_index_end      = jindex[iidx+1];
 593
 594         /* Get outer coordinate index */
 595         inr              = iinr[iidx];
 596         i_coord_offset   = DIM*inr;
 597
 598         /* Load i particle coords and add shift vector */
 599         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 600                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 601
 602         fix0             = _mm_setzero_ps();
 603         fiy0             = _mm_setzero_ps();
 604         fiz0             = _mm_setzero_ps();
 605         fix1             = _mm_setzero_ps();
 606         fiy1             = _mm_setzero_ps();
 607         fiz1             = _mm_setzero_ps();
 608         fix2             = _mm_setzero_ps();
 609         fiy2             = _mm_setzero_ps();
 610         fiz2             = _mm_setzero_ps();
 611
 612         /* Start inner kernel loop */
 613         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 614         {
 615
 616             /* Get j neighbor index, and coordinate index */
 617             jnrA             = jjnr[jidx];
 618             jnrB             = jjnr[jidx+1];
 619             jnrC             = jjnr[jidx+2];
 620             jnrD             = jjnr[jidx+3];
 621             j_coord_offsetA  = DIM*jnrA;
 622             j_coord_offsetB  = DIM*jnrB;
 623             j_coord_offsetC  = DIM*jnrC;
 624             j_coord_offsetD  = DIM*jnrD;
 625
 626             /* load j atom coordinates */
 627             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 628                                               x+j_coord_offsetC,x+j_coord_offsetD,
 629                                               &jx0,&jy0,&jz0);
 630
 631             /* Calculate displacement vector */
 632             dx00             = _mm_sub_ps(ix0,jx0);
 633             dy00             = _mm_sub_ps(iy0,jy0);
 634             dz00             = _mm_sub_ps(iz0,jz0);
 635             dx10             = _mm_sub_ps(ix1,jx0);
 636             dy10             = _mm_sub_ps(iy1,jy0);
 637             dz10             = _mm_sub_ps(iz1,jz0);
 638             dx20             = _mm_sub_ps(ix2,jx0);
 639             dy20             = _mm_sub_ps(iy2,jy0);
 640             dz20             = _mm_sub_ps(iz2,jz0);
 641
 642             /* Calculate squared distance and things based on it */
 643             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 644             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 645             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 646
 647             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 648             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 649             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 650
 651             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 652             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 653             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 654
 655             /* Load parameters for j particles */
 656             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 657                                                               charge+jnrC+0,charge+jnrD+0);
 658
 659             fjx0             = _mm_setzero_ps();
 660             fjy0             = _mm_setzero_ps();
 661             fjz0             = _mm_setzero_ps();
 662
 663             /**************************
 664              * CALCULATE INTERACTIONS *
 665              **************************/
 666
 667             /* Compute parameters for interactions between i and j atoms */
 668             qq00             = _mm_mul_ps(iq0,jq0);
 669
 670             /* REACTION-FIELD ELECTROSTATICS */
 671             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 672
 673             fscal            = felec;
 674
 675             /* Calculate temporary vectorial force */
 676             tx               = _mm_mul_ps(fscal,dx00);
 677             ty               = _mm_mul_ps(fscal,dy00);
 678             tz               = _mm_mul_ps(fscal,dz00);
 679
 680             /* Update vectorial force */
 681             fix0             = _mm_add_ps(fix0,tx);
 682             fiy0             = _mm_add_ps(fiy0,ty);
 683             fiz0             = _mm_add_ps(fiz0,tz);
 684
 685             fjx0             = _mm_add_ps(fjx0,tx);
 686             fjy0             = _mm_add_ps(fjy0,ty);
 687             fjz0             = _mm_add_ps(fjz0,tz);
 688
 689             /**************************
 690              * CALCULATE INTERACTIONS *
 691              **************************/
 692
 693             /* Compute parameters for interactions between i and j atoms */
 694             qq10             = _mm_mul_ps(iq1,jq0);
 695
 696             /* REACTION-FIELD ELECTROSTATICS */
 697             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 698
 699             fscal            = felec;
 700
 701             /* Calculate temporary vectorial force */
 702             tx               = _mm_mul_ps(fscal,dx10);
 703             ty               = _mm_mul_ps(fscal,dy10);
 704             tz               = _mm_mul_ps(fscal,dz10);
 705
 706             /* Update vectorial force */
 707             fix1             = _mm_add_ps(fix1,tx);
 708             fiy1             = _mm_add_ps(fiy1,ty);
 709             fiz1             = _mm_add_ps(fiz1,tz);
 710
 711             fjx0             = _mm_add_ps(fjx0,tx);
 712             fjy0             = _mm_add_ps(fjy0,ty);
 713             fjz0             = _mm_add_ps(fjz0,tz);
 714
 715             /**************************
 716              * CALCULATE INTERACTIONS *
 717              **************************/
 718
 719             /* Compute parameters for interactions between i and j atoms */
 720             qq20             = _mm_mul_ps(iq2,jq0);
 721
 722             /* REACTION-FIELD ELECTROSTATICS */
 723             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 724
 725             fscal            = felec;
 726
 727             /* Calculate temporary vectorial force */
 728             tx               = _mm_mul_ps(fscal,dx20);
 729             ty               = _mm_mul_ps(fscal,dy20);
 730             tz               = _mm_mul_ps(fscal,dz20);
 731
 732             /* Update vectorial force */
 733             fix2             = _mm_add_ps(fix2,tx);
 734             fiy2             = _mm_add_ps(fiy2,ty);
 735             fiz2             = _mm_add_ps(fiz2,tz);
 736
 737             fjx0             = _mm_add_ps(fjx0,tx);
 738             fjy0             = _mm_add_ps(fjy0,ty);
 739             fjz0             = _mm_add_ps(fjz0,tz);
 740
 741             fjptrA             = f+j_coord_offsetA;
 742             fjptrB             = f+j_coord_offsetB;
 743             fjptrC             = f+j_coord_offsetC;
 744             fjptrD             = f+j_coord_offsetD;
 745
 746             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 747
 748             /* Inner loop uses 81 flops */
 749         }
 750
 751         if(jidx<j_index_end)
 752         {
 753
 754             /* Get j neighbor index, and coordinate index */
 755             jnrlistA         = jjnr[jidx];
 756             jnrlistB         = jjnr[jidx+1];
 757             jnrlistC         = jjnr[jidx+2];
 758             jnrlistD         = jjnr[jidx+3];
 759             /* Sign of each element will be negative for non-real atoms.
 760              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 761              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 762              */
 763             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 764             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 765             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 766             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 767             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 768             j_coord_offsetA  = DIM*jnrA;
 769             j_coord_offsetB  = DIM*jnrB;
 770             j_coord_offsetC  = DIM*jnrC;
 771             j_coord_offsetD  = DIM*jnrD;
 772
 773             /* load j atom coordinates */
 774             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 775                                               x+j_coord_offsetC,x+j_coord_offsetD,
 776                                               &jx0,&jy0,&jz0);
 777
 778             /* Calculate displacement vector */
 779             dx00             = _mm_sub_ps(ix0,jx0);
 780             dy00             = _mm_sub_ps(iy0,jy0);
 781             dz00             = _mm_sub_ps(iz0,jz0);
 782             dx10             = _mm_sub_ps(ix1,jx0);
 783             dy10             = _mm_sub_ps(iy1,jy0);
 784             dz10             = _mm_sub_ps(iz1,jz0);
 785             dx20             = _mm_sub_ps(ix2,jx0);
 786             dy20             = _mm_sub_ps(iy2,jy0);
 787             dz20             = _mm_sub_ps(iz2,jz0);
 788
 789             /* Calculate squared distance and things based on it */
 790             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 791             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 792             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 793
 794             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 795             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 796             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 797
 798             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 799             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 800             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 801
 802             /* Load parameters for j particles */
 803             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 804                                                               charge+jnrC+0,charge+jnrD+0);
 805
 806             fjx0             = _mm_setzero_ps();
 807             fjy0             = _mm_setzero_ps();
 808             fjz0             = _mm_setzero_ps();
 809
 810             /**************************
 811              * CALCULATE INTERACTIONS *
 812              **************************/
 813
 814             /* Compute parameters for interactions between i and j atoms */
 815             qq00             = _mm_mul_ps(iq0,jq0);
 816
 817             /* REACTION-FIELD ELECTROSTATICS */
 818             felec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_mul_ps(rinv00,rinvsq00),krf2));
 819
 820             fscal            = felec;
 821
 822             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 823
 824             /* Calculate temporary vectorial force */
 825             tx               = _mm_mul_ps(fscal,dx00);
 826             ty               = _mm_mul_ps(fscal,dy00);
 827             tz               = _mm_mul_ps(fscal,dz00);
 828
 829             /* Update vectorial force */
 830             fix0             = _mm_add_ps(fix0,tx);
 831             fiy0             = _mm_add_ps(fiy0,ty);
 832             fiz0             = _mm_add_ps(fiz0,tz);
 833
 834             fjx0             = _mm_add_ps(fjx0,tx);
 835             fjy0             = _mm_add_ps(fjy0,ty);
 836             fjz0             = _mm_add_ps(fjz0,tz);
 837
 838             /**************************
 839              * CALCULATE INTERACTIONS *
 840              **************************/
 841
 842             /* Compute parameters for interactions between i and j atoms */
 843             qq10             = _mm_mul_ps(iq1,jq0);
 844
 845             /* REACTION-FIELD ELECTROSTATICS */
 846             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 847
 848             fscal            = felec;
 849
 850             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 851
 852             /* Calculate temporary vectorial force */
 853             tx               = _mm_mul_ps(fscal,dx10);
 854             ty               = _mm_mul_ps(fscal,dy10);
 855             tz               = _mm_mul_ps(fscal,dz10);
 856
 857             /* Update vectorial force */
 858             fix1             = _mm_add_ps(fix1,tx);
 859             fiy1             = _mm_add_ps(fiy1,ty);
 860             fiz1             = _mm_add_ps(fiz1,tz);
 861
 862             fjx0             = _mm_add_ps(fjx0,tx);
 863             fjy0             = _mm_add_ps(fjy0,ty);
 864             fjz0             = _mm_add_ps(fjz0,tz);
 865
 866             /**************************
 867              * CALCULATE INTERACTIONS *
 868              **************************/
 869
 870             /* Compute parameters for interactions between i and j atoms */
 871             qq20             = _mm_mul_ps(iq2,jq0);
 872
 873             /* REACTION-FIELD ELECTROSTATICS */
 874             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 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 }