src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecRF_VdwNone_GeomW4P1_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_ElecRF_VdwNone_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_VdwNone_GeomW4P1_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              vdwioffset1;
  70     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  71     int              vdwioffset2;
  72     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  73     int              vdwioffset3;
  74     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  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     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 106     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 107     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 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+DIM,
 140                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 141
 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         fix3             = _mm_setzero_ps();
 149         fiy3             = _mm_setzero_ps();
 150         fiz3             = _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             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             dx30             = _mm_sub_ps(ix3,jx0);
 182             dy30             = _mm_sub_ps(iy3,jy0);
 183             dz30             = _mm_sub_ps(iz3,jz0);
 184
 185             /* Calculate squared distance and things based on it */
 186             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 187             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 188             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 189
 190             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 191             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 192             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 193
 194             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 195             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 196             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 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             /**************************
 203              * CALCULATE INTERACTIONS *
 204              **************************/
 205
 206             /* Compute parameters for interactions between i and j atoms */
 207             qq10             = _mm_mul_ps(iq1,jq0);
 208
 209             /* REACTION-FIELD ELECTROSTATICS */
 210             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 211             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 212
 213             /* Update potential sum for this i atom from the interaction with this j atom. */
 214             velecsum         = _mm_add_ps(velecsum,velec);
 215
 216             fscal            = felec;
 217
 218             /* Calculate temporary vectorial force */
 219             tx               = _mm_mul_ps(fscal,dx10);
 220             ty               = _mm_mul_ps(fscal,dy10);
 221             tz               = _mm_mul_ps(fscal,dz10);
 222
 223             /* Update vectorial force */
 224             fix1             = _mm_add_ps(fix1,tx);
 225             fiy1             = _mm_add_ps(fiy1,ty);
 226             fiz1             = _mm_add_ps(fiz1,tz);
 227
 228             fjptrA             = f+j_coord_offsetA;
 229             fjptrB             = f+j_coord_offsetB;
 230             fjptrC             = f+j_coord_offsetC;
 231             fjptrD             = f+j_coord_offsetD;
 232             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 233
 234             /**************************
 235              * CALCULATE INTERACTIONS *
 236              **************************/
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq20             = _mm_mul_ps(iq2,jq0);
 240
 241             /* REACTION-FIELD ELECTROSTATICS */
 242             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 243             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 244
 245             /* Update potential sum for this i atom from the interaction with this j atom. */
 246             velecsum         = _mm_add_ps(velecsum,velec);
 247
 248             fscal            = felec;
 249
 250             /* Calculate temporary vectorial force */
 251             tx               = _mm_mul_ps(fscal,dx20);
 252             ty               = _mm_mul_ps(fscal,dy20);
 253             tz               = _mm_mul_ps(fscal,dz20);
 254
 255             /* Update vectorial force */
 256             fix2             = _mm_add_ps(fix2,tx);
 257             fiy2             = _mm_add_ps(fiy2,ty);
 258             fiz2             = _mm_add_ps(fiz2,tz);
 259
 260             fjptrA             = f+j_coord_offsetA;
 261             fjptrB             = f+j_coord_offsetB;
 262             fjptrC             = f+j_coord_offsetC;
 263             fjptrD             = f+j_coord_offsetD;
 264             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             /* Compute parameters for interactions between i and j atoms */
 271             qq30             = _mm_mul_ps(iq3,jq0);
 272
 273             /* REACTION-FIELD ELECTROSTATICS */
 274             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 275             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),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,dx30);
 284             ty               = _mm_mul_ps(fscal,dy30);
 285             tz               = _mm_mul_ps(fscal,dz30);
 286
 287             /* Update vectorial force */
 288             fix3             = _mm_add_ps(fix3,tx);
 289             fiy3             = _mm_add_ps(fiy3,ty);
 290             fiz3             = _mm_add_ps(fiz3,tz);
 291
 292             fjptrA             = f+j_coord_offsetA;
 293             fjptrB             = f+j_coord_offsetB;
 294             fjptrC             = f+j_coord_offsetC;
 295             fjptrD             = f+j_coord_offsetD;
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 297
 298             /* Inner loop uses 96 flops */
 299         }
 300
 301         if(jidx<j_index_end)
 302         {
 303
 304             /* Get j neighbor index, and coordinate index */
 305             jnrlistA         = jjnr[jidx];
 306             jnrlistB         = jjnr[jidx+1];
 307             jnrlistC         = jjnr[jidx+2];
 308             jnrlistD         = jjnr[jidx+3];
 309             /* Sign of each element will be negative for non-real atoms.
 310              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 311              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 312              */
 313             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 314             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 315             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 316             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 317             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 318             j_coord_offsetA  = DIM*jnrA;
 319             j_coord_offsetB  = DIM*jnrB;
 320             j_coord_offsetC  = DIM*jnrC;
 321             j_coord_offsetD  = DIM*jnrD;
 322
 323             /* load j atom coordinates */
 324             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 325                                               x+j_coord_offsetC,x+j_coord_offsetD,
 326                                               &jx0,&jy0,&jz0);
 327
 328             /* Calculate displacement vector */
 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             dx30             = _mm_sub_ps(ix3,jx0);
 336             dy30             = _mm_sub_ps(iy3,jy0);
 337             dz30             = _mm_sub_ps(iz3,jz0);
 338
 339             /* Calculate squared distance and things based on it */
 340             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 341             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 342             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 343
 344             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 345             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 346             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 347
 348             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 349             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 350             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 351
 352             /* Load parameters for j particles */
 353             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 354                                                               charge+jnrC+0,charge+jnrD+0);
 355
 356             /**************************
 357              * CALCULATE INTERACTIONS *
 358              **************************/
 359
 360             /* Compute parameters for interactions between i and j atoms */
 361             qq10             = _mm_mul_ps(iq1,jq0);
 362
 363             /* REACTION-FIELD ELECTROSTATICS */
 364             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_add_ps(rinv10,_mm_mul_ps(krf,rsq10)),crf));
 365             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 366
 367             /* Update potential sum for this i atom from the interaction with this j atom. */
 368             velec            = _mm_andnot_ps(dummy_mask,velec);
 369             velecsum         = _mm_add_ps(velecsum,velec);
 370
 371             fscal            = felec;
 372
 373             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 374
 375             /* Calculate temporary vectorial force */
 376             tx               = _mm_mul_ps(fscal,dx10);
 377             ty               = _mm_mul_ps(fscal,dy10);
 378             tz               = _mm_mul_ps(fscal,dz10);
 379
 380             /* Update vectorial force */
 381             fix1             = _mm_add_ps(fix1,tx);
 382             fiy1             = _mm_add_ps(fiy1,ty);
 383             fiz1             = _mm_add_ps(fiz1,tz);
 384
 385             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 386             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 387             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 388             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 389             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 390
 391             /**************************
 392              * CALCULATE INTERACTIONS *
 393              **************************/
 394
 395             /* Compute parameters for interactions between i and j atoms */
 396             qq20             = _mm_mul_ps(iq2,jq0);
 397
 398             /* REACTION-FIELD ELECTROSTATICS */
 399             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_add_ps(rinv20,_mm_mul_ps(krf,rsq20)),crf));
 400             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 401
 402             /* Update potential sum for this i atom from the interaction with this j atom. */
 403             velec            = _mm_andnot_ps(dummy_mask,velec);
 404             velecsum         = _mm_add_ps(velecsum,velec);
 405
 406             fscal            = felec;
 407
 408             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 409
 410             /* Calculate temporary vectorial force */
 411             tx               = _mm_mul_ps(fscal,dx20);
 412             ty               = _mm_mul_ps(fscal,dy20);
 413             tz               = _mm_mul_ps(fscal,dz20);
 414
 415             /* Update vectorial force */
 416             fix2             = _mm_add_ps(fix2,tx);
 417             fiy2             = _mm_add_ps(fiy2,ty);
 418             fiz2             = _mm_add_ps(fiz2,tz);
 419
 420             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 421             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 422             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 423             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 424             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 425
 426             /**************************
 427              * CALCULATE INTERACTIONS *
 428              **************************/
 429
 430             /* Compute parameters for interactions between i and j atoms */
 431             qq30             = _mm_mul_ps(iq3,jq0);
 432
 433             /* REACTION-FIELD ELECTROSTATICS */
 434             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_add_ps(rinv30,_mm_mul_ps(krf,rsq30)),crf));
 435             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 436
 437             /* Update potential sum for this i atom from the interaction with this j atom. */
 438             velec            = _mm_andnot_ps(dummy_mask,velec);
 439             velecsum         = _mm_add_ps(velecsum,velec);
 440
 441             fscal            = felec;
 442
 443             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 444
 445             /* Calculate temporary vectorial force */
 446             tx               = _mm_mul_ps(fscal,dx30);
 447             ty               = _mm_mul_ps(fscal,dy30);
 448             tz               = _mm_mul_ps(fscal,dz30);
 449
 450             /* Update vectorial force */
 451             fix3             = _mm_add_ps(fix3,tx);
 452             fiy3             = _mm_add_ps(fiy3,ty);
 453             fiz3             = _mm_add_ps(fiz3,tz);
 454
 455             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 456             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 457             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 458             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 459             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 460
 461             /* Inner loop uses 96 flops */
 462         }
 463
 464         /* End of innermost loop */
 465
 466         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 467                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 468
 469         ggid                        = gid[iidx];
 470         /* Update potential energies */
 471         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 472
 473         /* Increment number of inner iterations */
 474         inneriter                  += j_index_end - j_index_start;
 475
 476         /* Outer loop uses 19 flops */
 477     }
 478
 479     /* Increment number of outer iterations */
 480     outeriter        += nri;
 481
 482     /* Update outer/inner flops */
 483
 484     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*96);
 485 }
 486 /*
 487  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse2_single
 488  * Electrostatics interaction: ReactionField
 489  * VdW interaction:            None
 490  * Geometry:                   Water4-Particle
 491  * Calculate force/pot:        Force
 492  */
 493 void
 494 nb_kernel_ElecRF_VdwNone_GeomW4P1_F_sse2_single
 495                     (t_nblist * gmx_restrict                nlist,
 496                      rvec * gmx_restrict                    xx,
 497                      rvec * gmx_restrict                    ff,
 498                      t_forcerec * gmx_restrict              fr,
 499                      t_mdatoms * gmx_restrict               mdatoms,
 500                      nb_kernel_data_t * gmx_restrict        kernel_data,
 501                      t_nrnb * gmx_restrict                  nrnb)
 502 {
 503     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 504      * just 0 for non-waters.
 505      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 506      * jnr indices corresponding to data put in the four positions in the SIMD register.
 507      */
 508     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 509     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 510     int              jnrA,jnrB,jnrC,jnrD;
 511     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 512     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 513     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 514     real             rcutoff_scalar;
 515     real             *shiftvec,*fshift,*x,*f;
 516     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 517     real             scratch[4*DIM];
 518     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 519     int              vdwioffset1;
 520     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 521     int              vdwioffset2;
 522     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 523     int              vdwioffset3;
 524     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 525     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 526     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 527     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 528     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 529     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 530     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 531     real             *charge;
 532     __m128           dummy_mask,cutoff_mask;
 533     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 534     __m128           one     = _mm_set1_ps(1.0);
 535     __m128           two     = _mm_set1_ps(2.0);
 536     x                = xx[0];
 537     f                = ff[0];
 538
 539     nri              = nlist->nri;
 540     iinr             = nlist->iinr;
 541     jindex           = nlist->jindex;
 542     jjnr             = nlist->jjnr;
 543     shiftidx         = nlist->shift;
 544     gid              = nlist->gid;
 545     shiftvec         = fr->shift_vec[0];
 546     fshift           = fr->fshift[0];
 547     facel            = _mm_set1_ps(fr->epsfac);
 548     charge           = mdatoms->chargeA;
 549     krf              = _mm_set1_ps(fr->ic->k_rf);
 550     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 551     crf              = _mm_set1_ps(fr->ic->c_rf);
 552
 553     /* Setup water-specific parameters */
 554     inr              = nlist->iinr[0];
 555     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 556     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 557     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 558
 559     /* Avoid stupid compiler warnings */
 560     jnrA = jnrB = jnrC = jnrD = 0;
 561     j_coord_offsetA = 0;
 562     j_coord_offsetB = 0;
 563     j_coord_offsetC = 0;
 564     j_coord_offsetD = 0;
 565
 566     outeriter        = 0;
 567     inneriter        = 0;
 568
 569     for(iidx=0;iidx<4*DIM;iidx++)
 570     {
 571         scratch[iidx] = 0.0;
 572     }
 573
 574     /* Start outer loop over neighborlists */
 575     for(iidx=0; iidx<nri; iidx++)
 576     {
 577         /* Load shift vector for this list */
 578         i_shift_offset   = DIM*shiftidx[iidx];
 579
 580         /* Load limits for loop over neighbors */
 581         j_index_start    = jindex[iidx];
 582         j_index_end      = jindex[iidx+1];
 583
 584         /* Get outer coordinate index */
 585         inr              = iinr[iidx];
 586         i_coord_offset   = DIM*inr;
 587
 588         /* Load i particle coords and add shift vector */
 589         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 590                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 591
 592         fix1             = _mm_setzero_ps();
 593         fiy1             = _mm_setzero_ps();
 594         fiz1             = _mm_setzero_ps();
 595         fix2             = _mm_setzero_ps();
 596         fiy2             = _mm_setzero_ps();
 597         fiz2             = _mm_setzero_ps();
 598         fix3             = _mm_setzero_ps();
 599         fiy3             = _mm_setzero_ps();
 600         fiz3             = _mm_setzero_ps();
 601
 602         /* Start inner kernel loop */
 603         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 604         {
 605
 606             /* Get j neighbor index, and coordinate index */
 607             jnrA             = jjnr[jidx];
 608             jnrB             = jjnr[jidx+1];
 609             jnrC             = jjnr[jidx+2];
 610             jnrD             = jjnr[jidx+3];
 611             j_coord_offsetA  = DIM*jnrA;
 612             j_coord_offsetB  = DIM*jnrB;
 613             j_coord_offsetC  = DIM*jnrC;
 614             j_coord_offsetD  = DIM*jnrD;
 615
 616             /* load j atom coordinates */
 617             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 618                                               x+j_coord_offsetC,x+j_coord_offsetD,
 619                                               &jx0,&jy0,&jz0);
 620
 621             /* Calculate displacement vector */
 622             dx10             = _mm_sub_ps(ix1,jx0);
 623             dy10             = _mm_sub_ps(iy1,jy0);
 624             dz10             = _mm_sub_ps(iz1,jz0);
 625             dx20             = _mm_sub_ps(ix2,jx0);
 626             dy20             = _mm_sub_ps(iy2,jy0);
 627             dz20             = _mm_sub_ps(iz2,jz0);
 628             dx30             = _mm_sub_ps(ix3,jx0);
 629             dy30             = _mm_sub_ps(iy3,jy0);
 630             dz30             = _mm_sub_ps(iz3,jz0);
 631
 632             /* Calculate squared distance and things based on it */
 633             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 634             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 635             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 636
 637             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 638             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 639             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 640
 641             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 642             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 643             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 644
 645             /* Load parameters for j particles */
 646             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 647                                                               charge+jnrC+0,charge+jnrD+0);
 648
 649             /**************************
 650              * CALCULATE INTERACTIONS *
 651              **************************/
 652
 653             /* Compute parameters for interactions between i and j atoms */
 654             qq10             = _mm_mul_ps(iq1,jq0);
 655
 656             /* REACTION-FIELD ELECTROSTATICS */
 657             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 658
 659             fscal            = felec;
 660
 661             /* Calculate temporary vectorial force */
 662             tx               = _mm_mul_ps(fscal,dx10);
 663             ty               = _mm_mul_ps(fscal,dy10);
 664             tz               = _mm_mul_ps(fscal,dz10);
 665
 666             /* Update vectorial force */
 667             fix1             = _mm_add_ps(fix1,tx);
 668             fiy1             = _mm_add_ps(fiy1,ty);
 669             fiz1             = _mm_add_ps(fiz1,tz);
 670
 671             fjptrA             = f+j_coord_offsetA;
 672             fjptrB             = f+j_coord_offsetB;
 673             fjptrC             = f+j_coord_offsetC;
 674             fjptrD             = f+j_coord_offsetD;
 675             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 676
 677             /**************************
 678              * CALCULATE INTERACTIONS *
 679              **************************/
 680
 681             /* Compute parameters for interactions between i and j atoms */
 682             qq20             = _mm_mul_ps(iq2,jq0);
 683
 684             /* REACTION-FIELD ELECTROSTATICS */
 685             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 686
 687             fscal            = felec;
 688
 689             /* Calculate temporary vectorial force */
 690             tx               = _mm_mul_ps(fscal,dx20);
 691             ty               = _mm_mul_ps(fscal,dy20);
 692             tz               = _mm_mul_ps(fscal,dz20);
 693
 694             /* Update vectorial force */
 695             fix2             = _mm_add_ps(fix2,tx);
 696             fiy2             = _mm_add_ps(fiy2,ty);
 697             fiz2             = _mm_add_ps(fiz2,tz);
 698
 699             fjptrA             = f+j_coord_offsetA;
 700             fjptrB             = f+j_coord_offsetB;
 701             fjptrC             = f+j_coord_offsetC;
 702             fjptrD             = f+j_coord_offsetD;
 703             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 704
 705             /**************************
 706              * CALCULATE INTERACTIONS *
 707              **************************/
 708
 709             /* Compute parameters for interactions between i and j atoms */
 710             qq30             = _mm_mul_ps(iq3,jq0);
 711
 712             /* REACTION-FIELD ELECTROSTATICS */
 713             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 714
 715             fscal            = felec;
 716
 717             /* Calculate temporary vectorial force */
 718             tx               = _mm_mul_ps(fscal,dx30);
 719             ty               = _mm_mul_ps(fscal,dy30);
 720             tz               = _mm_mul_ps(fscal,dz30);
 721
 722             /* Update vectorial force */
 723             fix3             = _mm_add_ps(fix3,tx);
 724             fiy3             = _mm_add_ps(fiy3,ty);
 725             fiz3             = _mm_add_ps(fiz3,tz);
 726
 727             fjptrA             = f+j_coord_offsetA;
 728             fjptrB             = f+j_coord_offsetB;
 729             fjptrC             = f+j_coord_offsetC;
 730             fjptrD             = f+j_coord_offsetD;
 731             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 732
 733             /* Inner loop uses 81 flops */
 734         }
 735
 736         if(jidx<j_index_end)
 737         {
 738
 739             /* Get j neighbor index, and coordinate index */
 740             jnrlistA         = jjnr[jidx];
 741             jnrlistB         = jjnr[jidx+1];
 742             jnrlistC         = jjnr[jidx+2];
 743             jnrlistD         = jjnr[jidx+3];
 744             /* Sign of each element will be negative for non-real atoms.
 745              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 746              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 747              */
 748             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 749             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 750             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 751             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 752             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 753             j_coord_offsetA  = DIM*jnrA;
 754             j_coord_offsetB  = DIM*jnrB;
 755             j_coord_offsetC  = DIM*jnrC;
 756             j_coord_offsetD  = DIM*jnrD;
 757
 758             /* load j atom coordinates */
 759             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 760                                               x+j_coord_offsetC,x+j_coord_offsetD,
 761                                               &jx0,&jy0,&jz0);
 762
 763             /* Calculate displacement vector */
 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             dx30             = _mm_sub_ps(ix3,jx0);
 771             dy30             = _mm_sub_ps(iy3,jy0);
 772             dz30             = _mm_sub_ps(iz3,jz0);
 773
 774             /* Calculate squared distance and things based on it */
 775             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 776             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 777             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 778
 779             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 780             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 781             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 782
 783             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 784             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 785             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 786
 787             /* Load parameters for j particles */
 788             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 789                                                               charge+jnrC+0,charge+jnrD+0);
 790
 791             /**************************
 792              * CALCULATE INTERACTIONS *
 793              **************************/
 794
 795             /* Compute parameters for interactions between i and j atoms */
 796             qq10             = _mm_mul_ps(iq1,jq0);
 797
 798             /* REACTION-FIELD ELECTROSTATICS */
 799             felec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_mul_ps(rinv10,rinvsq10),krf2));
 800
 801             fscal            = felec;
 802
 803             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 804
 805             /* Calculate temporary vectorial force */
 806             tx               = _mm_mul_ps(fscal,dx10);
 807             ty               = _mm_mul_ps(fscal,dy10);
 808             tz               = _mm_mul_ps(fscal,dz10);
 809
 810             /* Update vectorial force */
 811             fix1             = _mm_add_ps(fix1,tx);
 812             fiy1             = _mm_add_ps(fiy1,ty);
 813             fiz1             = _mm_add_ps(fiz1,tz);
 814
 815             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 816             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 817             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 818             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 819             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 820
 821             /**************************
 822              * CALCULATE INTERACTIONS *
 823              **************************/
 824
 825             /* Compute parameters for interactions between i and j atoms */
 826             qq20             = _mm_mul_ps(iq2,jq0);
 827
 828             /* REACTION-FIELD ELECTROSTATICS */
 829             felec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_mul_ps(rinv20,rinvsq20),krf2));
 830
 831             fscal            = felec;
 832
 833             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 834
 835             /* Calculate temporary vectorial force */
 836             tx               = _mm_mul_ps(fscal,dx20);
 837             ty               = _mm_mul_ps(fscal,dy20);
 838             tz               = _mm_mul_ps(fscal,dz20);
 839
 840             /* Update vectorial force */
 841             fix2             = _mm_add_ps(fix2,tx);
 842             fiy2             = _mm_add_ps(fiy2,ty);
 843             fiz2             = _mm_add_ps(fiz2,tz);
 844
 845             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 846             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 847             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 848             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 849             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,tx,ty,tz);
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             /* Compute parameters for interactions between i and j atoms */
 856             qq30             = _mm_mul_ps(iq3,jq0);
 857
 858             /* REACTION-FIELD ELECTROSTATICS */
 859             felec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_mul_ps(rinv30,rinvsq30),krf2));
 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,dx30);
 867             ty               = _mm_mul_ps(fscal,dy30);
 868             tz               = _mm_mul_ps(fscal,dz30);
 869
 870             /* Update vectorial force */
 871             fix3             = _mm_add_ps(fix3,tx);
 872             fiy3             = _mm_add_ps(fiy3,ty);
 873             fiz3             = _mm_add_ps(fiz3,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(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 887                                               f+i_coord_offset+DIM,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_W4_F,outeriter*18 + inneriter*81);
 901 }