src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwNone_GeomW4P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014,2015,2017, 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 sse2_single kernel generator.
  37  */
  38 #include "gmxpre.h"
  39
  40 #include "config.h"
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "gromacs/gmxlib/nrnb.h"
  46
  47 #include "kernelutil_x86_sse2_single.h"
  48
  49 /*
  50  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_sse2_single
  51  * Electrostatics interaction: Ewald
  52  * VdW interaction:            None
  53  * Geometry:                   Water4-Particle
  54  * Calculate force/pot:        PotentialAndForce
  55  */
  56 void
  57 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_VF_sse2_single
  58                     (t_nblist                    * gmx_restrict       nlist,
  59                      rvec                        * gmx_restrict          xx,
  60                      rvec                        * gmx_restrict          ff,
  61                      struct t_forcerec           * gmx_restrict          fr,
  62                      t_mdatoms                   * gmx_restrict     mdatoms,
  63                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  64                      t_nrnb                      * gmx_restrict        nrnb)
  65 {
  66     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  67      * just 0 for non-waters.
  68      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  69      * jnr indices corresponding to data put in the four positions in the SIMD register.
  70      */
  71     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  72     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  73     int              jnrA,jnrB,jnrC,jnrD;
  74     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  75     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  76     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  77     real             rcutoff_scalar;
  78     real             *shiftvec,*fshift,*x,*f;
  79     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  80     real             scratch[4*DIM];
  81     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  82     int              vdwioffset1;
  83     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  84     int              vdwioffset2;
  85     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  86     int              vdwioffset3;
  87     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  89     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  91     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  92     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  93     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  94     real             *charge;
  95     __m128i          ewitab;
  96     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  97     real             *ewtab;
  98     __m128           dummy_mask,cutoff_mask;
  99     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 100     __m128           one     = _mm_set1_ps(1.0);
 101     __m128           two     = _mm_set1_ps(2.0);
 102     x                = xx[0];
 103     f                = ff[0];
 104
 105     nri              = nlist->nri;
 106     iinr             = nlist->iinr;
 107     jindex           = nlist->jindex;
 108     jjnr             = nlist->jjnr;
 109     shiftidx         = nlist->shift;
 110     gid              = nlist->gid;
 111     shiftvec         = fr->shift_vec[0];
 112     fshift           = fr->fshift[0];
 113     facel            = _mm_set1_ps(fr->ic->epsfac);
 114     charge           = mdatoms->chargeA;
 115
 116     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 117     ewtab            = fr->ic->tabq_coul_FDV0;
 118     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 119     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 120
 121     /* Setup water-specific parameters */
 122     inr              = nlist->iinr[0];
 123     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 124     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 125     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 126
 127     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 128     rcutoff_scalar   = fr->ic->rcoulomb;
 129     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 130     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 131
 132     /* Avoid stupid compiler warnings */
 133     jnrA = jnrB = jnrC = jnrD = 0;
 134     j_coord_offsetA = 0;
 135     j_coord_offsetB = 0;
 136     j_coord_offsetC = 0;
 137     j_coord_offsetD = 0;
 138
 139     outeriter        = 0;
 140     inneriter        = 0;
 141
 142     for(iidx=0;iidx<4*DIM;iidx++)
 143     {
 144         scratch[iidx] = 0.0;
 145     }
 146
 147     /* Start outer loop over neighborlists */
 148     for(iidx=0; iidx<nri; iidx++)
 149     {
 150         /* Load shift vector for this list */
 151         i_shift_offset   = DIM*shiftidx[iidx];
 152
 153         /* Load limits for loop over neighbors */
 154         j_index_start    = jindex[iidx];
 155         j_index_end      = jindex[iidx+1];
 156
 157         /* Get outer coordinate index */
 158         inr              = iinr[iidx];
 159         i_coord_offset   = DIM*inr;
 160
 161         /* Load i particle coords and add shift vector */
 162         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 163                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 164
 165         fix1             = _mm_setzero_ps();
 166         fiy1             = _mm_setzero_ps();
 167         fiz1             = _mm_setzero_ps();
 168         fix2             = _mm_setzero_ps();
 169         fiy2             = _mm_setzero_ps();
 170         fiz2             = _mm_setzero_ps();
 171         fix3             = _mm_setzero_ps();
 172         fiy3             = _mm_setzero_ps();
 173         fiz3             = _mm_setzero_ps();
 174
 175         /* Reset potential sums */
 176         velecsum         = _mm_setzero_ps();
 177
 178         /* Start inner kernel loop */
 179         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 180         {
 181
 182             /* Get j neighbor index, and coordinate index */
 183             jnrA             = jjnr[jidx];
 184             jnrB             = jjnr[jidx+1];
 185             jnrC             = jjnr[jidx+2];
 186             jnrD             = jjnr[jidx+3];
 187             j_coord_offsetA  = DIM*jnrA;
 188             j_coord_offsetB  = DIM*jnrB;
 189             j_coord_offsetC  = DIM*jnrC;
 190             j_coord_offsetD  = DIM*jnrD;
 191
 192             /* load j atom coordinates */
 193             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 194                                               x+j_coord_offsetC,x+j_coord_offsetD,
 195                                               &jx0,&jy0,&jz0);
 196
 197             /* Calculate displacement vector */
 198             dx10             = _mm_sub_ps(ix1,jx0);
 199             dy10             = _mm_sub_ps(iy1,jy0);
 200             dz10             = _mm_sub_ps(iz1,jz0);
 201             dx20             = _mm_sub_ps(ix2,jx0);
 202             dy20             = _mm_sub_ps(iy2,jy0);
 203             dz20             = _mm_sub_ps(iz2,jz0);
 204             dx30             = _mm_sub_ps(ix3,jx0);
 205             dy30             = _mm_sub_ps(iy3,jy0);
 206             dz30             = _mm_sub_ps(iz3,jz0);
 207
 208             /* Calculate squared distance and things based on it */
 209             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 210             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 211             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 212
 213             rinv10           = sse2_invsqrt_f(rsq10);
 214             rinv20           = sse2_invsqrt_f(rsq20);
 215             rinv30           = sse2_invsqrt_f(rsq30);
 216
 217             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 218             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 219             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224
 225             fjx0             = _mm_setzero_ps();
 226             fjy0             = _mm_setzero_ps();
 227             fjz0             = _mm_setzero_ps();
 228
 229             /**************************
 230              * CALCULATE INTERACTIONS *
 231              **************************/
 232
 233             if (gmx_mm_any_lt(rsq10,rcutoff2))
 234             {
 235
 236             r10              = _mm_mul_ps(rsq10,rinv10);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             qq10             = _mm_mul_ps(iq1,jq0);
 240
 241             /* EWALD ELECTROSTATICS */
 242
 243             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 244             ewrt             = _mm_mul_ps(r10,ewtabscale);
 245             ewitab           = _mm_cvttps_epi32(ewrt);
 246             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 247             ewitab           = _mm_slli_epi32(ewitab,2);
 248             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 249             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 250             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 251             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 252             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 253             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 254             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 255             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 256             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 257
 258             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 259
 260             /* Update potential sum for this i atom from the interaction with this j atom. */
 261             velec            = _mm_and_ps(velec,cutoff_mask);
 262             velecsum         = _mm_add_ps(velecsum,velec);
 263
 264             fscal            = felec;
 265
 266             fscal            = _mm_and_ps(fscal,cutoff_mask);
 267
 268             /* Calculate temporary vectorial force */
 269             tx               = _mm_mul_ps(fscal,dx10);
 270             ty               = _mm_mul_ps(fscal,dy10);
 271             tz               = _mm_mul_ps(fscal,dz10);
 272
 273             /* Update vectorial force */
 274             fix1             = _mm_add_ps(fix1,tx);
 275             fiy1             = _mm_add_ps(fiy1,ty);
 276             fiz1             = _mm_add_ps(fiz1,tz);
 277
 278             fjx0             = _mm_add_ps(fjx0,tx);
 279             fjy0             = _mm_add_ps(fjy0,ty);
 280             fjz0             = _mm_add_ps(fjz0,tz);
 281
 282             }
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             if (gmx_mm_any_lt(rsq20,rcutoff2))
 289             {
 290
 291             r20              = _mm_mul_ps(rsq20,rinv20);
 292
 293             /* Compute parameters for interactions between i and j atoms */
 294             qq20             = _mm_mul_ps(iq2,jq0);
 295
 296             /* EWALD ELECTROSTATICS */
 297
 298             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 299             ewrt             = _mm_mul_ps(r20,ewtabscale);
 300             ewitab           = _mm_cvttps_epi32(ewrt);
 301             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 302             ewitab           = _mm_slli_epi32(ewitab,2);
 303             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 304             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 305             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 306             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 307             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 308             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 309             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 310             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 311             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 312
 313             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velec            = _mm_and_ps(velec,cutoff_mask);
 317             velecsum         = _mm_add_ps(velecsum,velec);
 318
 319             fscal            = felec;
 320
 321             fscal            = _mm_and_ps(fscal,cutoff_mask);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm_mul_ps(fscal,dx20);
 325             ty               = _mm_mul_ps(fscal,dy20);
 326             tz               = _mm_mul_ps(fscal,dz20);
 327
 328             /* Update vectorial force */
 329             fix2             = _mm_add_ps(fix2,tx);
 330             fiy2             = _mm_add_ps(fiy2,ty);
 331             fiz2             = _mm_add_ps(fiz2,tz);
 332
 333             fjx0             = _mm_add_ps(fjx0,tx);
 334             fjy0             = _mm_add_ps(fjy0,ty);
 335             fjz0             = _mm_add_ps(fjz0,tz);
 336
 337             }
 338
 339             /**************************
 340              * CALCULATE INTERACTIONS *
 341              **************************/
 342
 343             if (gmx_mm_any_lt(rsq30,rcutoff2))
 344             {
 345
 346             r30              = _mm_mul_ps(rsq30,rinv30);
 347
 348             /* Compute parameters for interactions between i and j atoms */
 349             qq30             = _mm_mul_ps(iq3,jq0);
 350
 351             /* EWALD ELECTROSTATICS */
 352
 353             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 354             ewrt             = _mm_mul_ps(r30,ewtabscale);
 355             ewitab           = _mm_cvttps_epi32(ewrt);
 356             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 357             ewitab           = _mm_slli_epi32(ewitab,2);
 358             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 359             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 360             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 361             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 362             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 363             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 364             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 365             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 366             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 367
 368             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 369
 370             /* Update potential sum for this i atom from the interaction with this j atom. */
 371             velec            = _mm_and_ps(velec,cutoff_mask);
 372             velecsum         = _mm_add_ps(velecsum,velec);
 373
 374             fscal            = felec;
 375
 376             fscal            = _mm_and_ps(fscal,cutoff_mask);
 377
 378             /* Calculate temporary vectorial force */
 379             tx               = _mm_mul_ps(fscal,dx30);
 380             ty               = _mm_mul_ps(fscal,dy30);
 381             tz               = _mm_mul_ps(fscal,dz30);
 382
 383             /* Update vectorial force */
 384             fix3             = _mm_add_ps(fix3,tx);
 385             fiy3             = _mm_add_ps(fiy3,ty);
 386             fiz3             = _mm_add_ps(fiz3,tz);
 387
 388             fjx0             = _mm_add_ps(fjx0,tx);
 389             fjy0             = _mm_add_ps(fjy0,ty);
 390             fjz0             = _mm_add_ps(fjz0,tz);
 391
 392             }
 393
 394             fjptrA             = f+j_coord_offsetA;
 395             fjptrB             = f+j_coord_offsetB;
 396             fjptrC             = f+j_coord_offsetC;
 397             fjptrD             = f+j_coord_offsetD;
 398
 399             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 400
 401             /* Inner loop uses 138 flops */
 402         }
 403
 404         if(jidx<j_index_end)
 405         {
 406
 407             /* Get j neighbor index, and coordinate index */
 408             jnrlistA         = jjnr[jidx];
 409             jnrlistB         = jjnr[jidx+1];
 410             jnrlistC         = jjnr[jidx+2];
 411             jnrlistD         = jjnr[jidx+3];
 412             /* Sign of each element will be negative for non-real atoms.
 413              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 414              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 415              */
 416             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 417             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 418             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 419             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 420             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 421             j_coord_offsetA  = DIM*jnrA;
 422             j_coord_offsetB  = DIM*jnrB;
 423             j_coord_offsetC  = DIM*jnrC;
 424             j_coord_offsetD  = DIM*jnrD;
 425
 426             /* load j atom coordinates */
 427             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 428                                               x+j_coord_offsetC,x+j_coord_offsetD,
 429                                               &jx0,&jy0,&jz0);
 430
 431             /* Calculate displacement vector */
 432             dx10             = _mm_sub_ps(ix1,jx0);
 433             dy10             = _mm_sub_ps(iy1,jy0);
 434             dz10             = _mm_sub_ps(iz1,jz0);
 435             dx20             = _mm_sub_ps(ix2,jx0);
 436             dy20             = _mm_sub_ps(iy2,jy0);
 437             dz20             = _mm_sub_ps(iz2,jz0);
 438             dx30             = _mm_sub_ps(ix3,jx0);
 439             dy30             = _mm_sub_ps(iy3,jy0);
 440             dz30             = _mm_sub_ps(iz3,jz0);
 441
 442             /* Calculate squared distance and things based on it */
 443             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 444             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 445             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 446
 447             rinv10           = sse2_invsqrt_f(rsq10);
 448             rinv20           = sse2_invsqrt_f(rsq20);
 449             rinv30           = sse2_invsqrt_f(rsq30);
 450
 451             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 452             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 453             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 454
 455             /* Load parameters for j particles */
 456             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 457                                                               charge+jnrC+0,charge+jnrD+0);
 458
 459             fjx0             = _mm_setzero_ps();
 460             fjy0             = _mm_setzero_ps();
 461             fjz0             = _mm_setzero_ps();
 462
 463             /**************************
 464              * CALCULATE INTERACTIONS *
 465              **************************/
 466
 467             if (gmx_mm_any_lt(rsq10,rcutoff2))
 468             {
 469
 470             r10              = _mm_mul_ps(rsq10,rinv10);
 471             r10              = _mm_andnot_ps(dummy_mask,r10);
 472
 473             /* Compute parameters for interactions between i and j atoms */
 474             qq10             = _mm_mul_ps(iq1,jq0);
 475
 476             /* EWALD ELECTROSTATICS */
 477
 478             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 479             ewrt             = _mm_mul_ps(r10,ewtabscale);
 480             ewitab           = _mm_cvttps_epi32(ewrt);
 481             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 482             ewitab           = _mm_slli_epi32(ewitab,2);
 483             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 484             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 485             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 486             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 487             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 488             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 489             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 490             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 491             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 492
 493             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 494
 495             /* Update potential sum for this i atom from the interaction with this j atom. */
 496             velec            = _mm_and_ps(velec,cutoff_mask);
 497             velec            = _mm_andnot_ps(dummy_mask,velec);
 498             velecsum         = _mm_add_ps(velecsum,velec);
 499
 500             fscal            = felec;
 501
 502             fscal            = _mm_and_ps(fscal,cutoff_mask);
 503
 504             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 505
 506             /* Calculate temporary vectorial force */
 507             tx               = _mm_mul_ps(fscal,dx10);
 508             ty               = _mm_mul_ps(fscal,dy10);
 509             tz               = _mm_mul_ps(fscal,dz10);
 510
 511             /* Update vectorial force */
 512             fix1             = _mm_add_ps(fix1,tx);
 513             fiy1             = _mm_add_ps(fiy1,ty);
 514             fiz1             = _mm_add_ps(fiz1,tz);
 515
 516             fjx0             = _mm_add_ps(fjx0,tx);
 517             fjy0             = _mm_add_ps(fjy0,ty);
 518             fjz0             = _mm_add_ps(fjz0,tz);
 519
 520             }
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             if (gmx_mm_any_lt(rsq20,rcutoff2))
 527             {
 528
 529             r20              = _mm_mul_ps(rsq20,rinv20);
 530             r20              = _mm_andnot_ps(dummy_mask,r20);
 531
 532             /* Compute parameters for interactions between i and j atoms */
 533             qq20             = _mm_mul_ps(iq2,jq0);
 534
 535             /* EWALD ELECTROSTATICS */
 536
 537             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 538             ewrt             = _mm_mul_ps(r20,ewtabscale);
 539             ewitab           = _mm_cvttps_epi32(ewrt);
 540             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 541             ewitab           = _mm_slli_epi32(ewitab,2);
 542             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 543             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 544             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 545             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 546             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 547             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 548             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 549             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 550             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 551
 552             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_and_ps(velec,cutoff_mask);
 556             velec            = _mm_andnot_ps(dummy_mask,velec);
 557             velecsum         = _mm_add_ps(velecsum,velec);
 558
 559             fscal            = felec;
 560
 561             fscal            = _mm_and_ps(fscal,cutoff_mask);
 562
 563             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 564
 565             /* Calculate temporary vectorial force */
 566             tx               = _mm_mul_ps(fscal,dx20);
 567             ty               = _mm_mul_ps(fscal,dy20);
 568             tz               = _mm_mul_ps(fscal,dz20);
 569
 570             /* Update vectorial force */
 571             fix2             = _mm_add_ps(fix2,tx);
 572             fiy2             = _mm_add_ps(fiy2,ty);
 573             fiz2             = _mm_add_ps(fiz2,tz);
 574
 575             fjx0             = _mm_add_ps(fjx0,tx);
 576             fjy0             = _mm_add_ps(fjy0,ty);
 577             fjz0             = _mm_add_ps(fjz0,tz);
 578
 579             }
 580
 581             /**************************
 582              * CALCULATE INTERACTIONS *
 583              **************************/
 584
 585             if (gmx_mm_any_lt(rsq30,rcutoff2))
 586             {
 587
 588             r30              = _mm_mul_ps(rsq30,rinv30);
 589             r30              = _mm_andnot_ps(dummy_mask,r30);
 590
 591             /* Compute parameters for interactions between i and j atoms */
 592             qq30             = _mm_mul_ps(iq3,jq0);
 593
 594             /* EWALD ELECTROSTATICS */
 595
 596             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 597             ewrt             = _mm_mul_ps(r30,ewtabscale);
 598             ewitab           = _mm_cvttps_epi32(ewrt);
 599             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 600             ewitab           = _mm_slli_epi32(ewitab,2);
 601             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 602             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 603             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 604             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 605             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 606             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 607             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 608             velec            = _mm_mul_ps(qq30,_mm_sub_ps(_mm_sub_ps(rinv30,sh_ewald),velec));
 609             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 610
 611             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm_and_ps(velec,cutoff_mask);
 615             velec            = _mm_andnot_ps(dummy_mask,velec);
 616             velecsum         = _mm_add_ps(velecsum,velec);
 617
 618             fscal            = felec;
 619
 620             fscal            = _mm_and_ps(fscal,cutoff_mask);
 621
 622             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm_mul_ps(fscal,dx30);
 626             ty               = _mm_mul_ps(fscal,dy30);
 627             tz               = _mm_mul_ps(fscal,dz30);
 628
 629             /* Update vectorial force */
 630             fix3             = _mm_add_ps(fix3,tx);
 631             fiy3             = _mm_add_ps(fiy3,ty);
 632             fiz3             = _mm_add_ps(fiz3,tz);
 633
 634             fjx0             = _mm_add_ps(fjx0,tx);
 635             fjy0             = _mm_add_ps(fjy0,ty);
 636             fjz0             = _mm_add_ps(fjz0,tz);
 637
 638             }
 639
 640             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 641             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 642             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 643             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 644
 645             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 646
 647             /* Inner loop uses 141 flops */
 648         }
 649
 650         /* End of innermost loop */
 651
 652         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 653                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 654
 655         ggid                        = gid[iidx];
 656         /* Update potential energies */
 657         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 658
 659         /* Increment number of inner iterations */
 660         inneriter                  += j_index_end - j_index_start;
 661
 662         /* Outer loop uses 19 flops */
 663     }
 664
 665     /* Increment number of outer iterations */
 666     outeriter        += nri;
 667
 668     /* Update outer/inner flops */
 669
 670     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*19 + inneriter*141);
 671 }
 672 /*
 673  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse2_single
 674  * Electrostatics interaction: Ewald
 675  * VdW interaction:            None
 676  * Geometry:                   Water4-Particle
 677  * Calculate force/pot:        Force
 678  */
 679 void
 680 nb_kernel_ElecEwSh_VdwNone_GeomW4P1_F_sse2_single
 681                     (t_nblist                    * gmx_restrict       nlist,
 682                      rvec                        * gmx_restrict          xx,
 683                      rvec                        * gmx_restrict          ff,
 684                      struct t_forcerec           * gmx_restrict          fr,
 685                      t_mdatoms                   * gmx_restrict     mdatoms,
 686                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 687                      t_nrnb                      * gmx_restrict        nrnb)
 688 {
 689     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 690      * just 0 for non-waters.
 691      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 692      * jnr indices corresponding to data put in the four positions in the SIMD register.
 693      */
 694     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 695     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 696     int              jnrA,jnrB,jnrC,jnrD;
 697     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 698     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 699     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 700     real             rcutoff_scalar;
 701     real             *shiftvec,*fshift,*x,*f;
 702     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 703     real             scratch[4*DIM];
 704     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 705     int              vdwioffset1;
 706     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 707     int              vdwioffset2;
 708     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 709     int              vdwioffset3;
 710     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 711     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 712     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 713     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 714     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 715     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 716     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 717     real             *charge;
 718     __m128i          ewitab;
 719     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 720     real             *ewtab;
 721     __m128           dummy_mask,cutoff_mask;
 722     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 723     __m128           one     = _mm_set1_ps(1.0);
 724     __m128           two     = _mm_set1_ps(2.0);
 725     x                = xx[0];
 726     f                = ff[0];
 727
 728     nri              = nlist->nri;
 729     iinr             = nlist->iinr;
 730     jindex           = nlist->jindex;
 731     jjnr             = nlist->jjnr;
 732     shiftidx         = nlist->shift;
 733     gid              = nlist->gid;
 734     shiftvec         = fr->shift_vec[0];
 735     fshift           = fr->fshift[0];
 736     facel            = _mm_set1_ps(fr->ic->epsfac);
 737     charge           = mdatoms->chargeA;
 738
 739     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 740     ewtab            = fr->ic->tabq_coul_F;
 741     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 742     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 743
 744     /* Setup water-specific parameters */
 745     inr              = nlist->iinr[0];
 746     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 747     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 748     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 749
 750     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 751     rcutoff_scalar   = fr->ic->rcoulomb;
 752     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 753     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 754
 755     /* Avoid stupid compiler warnings */
 756     jnrA = jnrB = jnrC = jnrD = 0;
 757     j_coord_offsetA = 0;
 758     j_coord_offsetB = 0;
 759     j_coord_offsetC = 0;
 760     j_coord_offsetD = 0;
 761
 762     outeriter        = 0;
 763     inneriter        = 0;
 764
 765     for(iidx=0;iidx<4*DIM;iidx++)
 766     {
 767         scratch[iidx] = 0.0;
 768     }
 769
 770     /* Start outer loop over neighborlists */
 771     for(iidx=0; iidx<nri; iidx++)
 772     {
 773         /* Load shift vector for this list */
 774         i_shift_offset   = DIM*shiftidx[iidx];
 775
 776         /* Load limits for loop over neighbors */
 777         j_index_start    = jindex[iidx];
 778         j_index_end      = jindex[iidx+1];
 779
 780         /* Get outer coordinate index */
 781         inr              = iinr[iidx];
 782         i_coord_offset   = DIM*inr;
 783
 784         /* Load i particle coords and add shift vector */
 785         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset+DIM,
 786                                                  &ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 787
 788         fix1             = _mm_setzero_ps();
 789         fiy1             = _mm_setzero_ps();
 790         fiz1             = _mm_setzero_ps();
 791         fix2             = _mm_setzero_ps();
 792         fiy2             = _mm_setzero_ps();
 793         fiz2             = _mm_setzero_ps();
 794         fix3             = _mm_setzero_ps();
 795         fiy3             = _mm_setzero_ps();
 796         fiz3             = _mm_setzero_ps();
 797
 798         /* Start inner kernel loop */
 799         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 800         {
 801
 802             /* Get j neighbor index, and coordinate index */
 803             jnrA             = jjnr[jidx];
 804             jnrB             = jjnr[jidx+1];
 805             jnrC             = jjnr[jidx+2];
 806             jnrD             = jjnr[jidx+3];
 807             j_coord_offsetA  = DIM*jnrA;
 808             j_coord_offsetB  = DIM*jnrB;
 809             j_coord_offsetC  = DIM*jnrC;
 810             j_coord_offsetD  = DIM*jnrD;
 811
 812             /* load j atom coordinates */
 813             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 814                                               x+j_coord_offsetC,x+j_coord_offsetD,
 815                                               &jx0,&jy0,&jz0);
 816
 817             /* Calculate displacement vector */
 818             dx10             = _mm_sub_ps(ix1,jx0);
 819             dy10             = _mm_sub_ps(iy1,jy0);
 820             dz10             = _mm_sub_ps(iz1,jz0);
 821             dx20             = _mm_sub_ps(ix2,jx0);
 822             dy20             = _mm_sub_ps(iy2,jy0);
 823             dz20             = _mm_sub_ps(iz2,jz0);
 824             dx30             = _mm_sub_ps(ix3,jx0);
 825             dy30             = _mm_sub_ps(iy3,jy0);
 826             dz30             = _mm_sub_ps(iz3,jz0);
 827
 828             /* Calculate squared distance and things based on it */
 829             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 830             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 831             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 832
 833             rinv10           = sse2_invsqrt_f(rsq10);
 834             rinv20           = sse2_invsqrt_f(rsq20);
 835             rinv30           = sse2_invsqrt_f(rsq30);
 836
 837             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 838             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 839             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 840
 841             /* Load parameters for j particles */
 842             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 843                                                               charge+jnrC+0,charge+jnrD+0);
 844
 845             fjx0             = _mm_setzero_ps();
 846             fjy0             = _mm_setzero_ps();
 847             fjz0             = _mm_setzero_ps();
 848
 849             /**************************
 850              * CALCULATE INTERACTIONS *
 851              **************************/
 852
 853             if (gmx_mm_any_lt(rsq10,rcutoff2))
 854             {
 855
 856             r10              = _mm_mul_ps(rsq10,rinv10);
 857
 858             /* Compute parameters for interactions between i and j atoms */
 859             qq10             = _mm_mul_ps(iq1,jq0);
 860
 861             /* EWALD ELECTROSTATICS */
 862
 863             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 864             ewrt             = _mm_mul_ps(r10,ewtabscale);
 865             ewitab           = _mm_cvttps_epi32(ewrt);
 866             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 867             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 868                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 869                                          &ewtabF,&ewtabFn);
 870             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 871             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 872
 873             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 874
 875             fscal            = felec;
 876
 877             fscal            = _mm_and_ps(fscal,cutoff_mask);
 878
 879             /* Calculate temporary vectorial force */
 880             tx               = _mm_mul_ps(fscal,dx10);
 881             ty               = _mm_mul_ps(fscal,dy10);
 882             tz               = _mm_mul_ps(fscal,dz10);
 883
 884             /* Update vectorial force */
 885             fix1             = _mm_add_ps(fix1,tx);
 886             fiy1             = _mm_add_ps(fiy1,ty);
 887             fiz1             = _mm_add_ps(fiz1,tz);
 888
 889             fjx0             = _mm_add_ps(fjx0,tx);
 890             fjy0             = _mm_add_ps(fjy0,ty);
 891             fjz0             = _mm_add_ps(fjz0,tz);
 892
 893             }
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             if (gmx_mm_any_lt(rsq20,rcutoff2))
 900             {
 901
 902             r20              = _mm_mul_ps(rsq20,rinv20);
 903
 904             /* Compute parameters for interactions between i and j atoms */
 905             qq20             = _mm_mul_ps(iq2,jq0);
 906
 907             /* EWALD ELECTROSTATICS */
 908
 909             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 910             ewrt             = _mm_mul_ps(r20,ewtabscale);
 911             ewitab           = _mm_cvttps_epi32(ewrt);
 912             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 913             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 914                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 915                                          &ewtabF,&ewtabFn);
 916             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 917             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 918
 919             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 920
 921             fscal            = felec;
 922
 923             fscal            = _mm_and_ps(fscal,cutoff_mask);
 924
 925             /* Calculate temporary vectorial force */
 926             tx               = _mm_mul_ps(fscal,dx20);
 927             ty               = _mm_mul_ps(fscal,dy20);
 928             tz               = _mm_mul_ps(fscal,dz20);
 929
 930             /* Update vectorial force */
 931             fix2             = _mm_add_ps(fix2,tx);
 932             fiy2             = _mm_add_ps(fiy2,ty);
 933             fiz2             = _mm_add_ps(fiz2,tz);
 934
 935             fjx0             = _mm_add_ps(fjx0,tx);
 936             fjy0             = _mm_add_ps(fjy0,ty);
 937             fjz0             = _mm_add_ps(fjz0,tz);
 938
 939             }
 940
 941             /**************************
 942              * CALCULATE INTERACTIONS *
 943              **************************/
 944
 945             if (gmx_mm_any_lt(rsq30,rcutoff2))
 946             {
 947
 948             r30              = _mm_mul_ps(rsq30,rinv30);
 949
 950             /* Compute parameters for interactions between i and j atoms */
 951             qq30             = _mm_mul_ps(iq3,jq0);
 952
 953             /* EWALD ELECTROSTATICS */
 954
 955             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 956             ewrt             = _mm_mul_ps(r30,ewtabscale);
 957             ewitab           = _mm_cvttps_epi32(ewrt);
 958             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 959             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 960                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 961                                          &ewtabF,&ewtabFn);
 962             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 963             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 964
 965             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
 966
 967             fscal            = felec;
 968
 969             fscal            = _mm_and_ps(fscal,cutoff_mask);
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_ps(fscal,dx30);
 973             ty               = _mm_mul_ps(fscal,dy30);
 974             tz               = _mm_mul_ps(fscal,dz30);
 975
 976             /* Update vectorial force */
 977             fix3             = _mm_add_ps(fix3,tx);
 978             fiy3             = _mm_add_ps(fiy3,ty);
 979             fiz3             = _mm_add_ps(fiz3,tz);
 980
 981             fjx0             = _mm_add_ps(fjx0,tx);
 982             fjy0             = _mm_add_ps(fjy0,ty);
 983             fjz0             = _mm_add_ps(fjz0,tz);
 984
 985             }
 986
 987             fjptrA             = f+j_coord_offsetA;
 988             fjptrB             = f+j_coord_offsetB;
 989             fjptrC             = f+j_coord_offsetC;
 990             fjptrD             = f+j_coord_offsetD;
 991
 992             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 993
 994             /* Inner loop uses 117 flops */
 995         }
 996
 997         if(jidx<j_index_end)
 998         {
 999
1000             /* Get j neighbor index, and coordinate index */
1001             jnrlistA         = jjnr[jidx];
1002             jnrlistB         = jjnr[jidx+1];
1003             jnrlistC         = jjnr[jidx+2];
1004             jnrlistD         = jjnr[jidx+3];
1005             /* Sign of each element will be negative for non-real atoms.
1006              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1007              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1008              */
1009             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1010             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1011             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1012             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1013             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1014             j_coord_offsetA  = DIM*jnrA;
1015             j_coord_offsetB  = DIM*jnrB;
1016             j_coord_offsetC  = DIM*jnrC;
1017             j_coord_offsetD  = DIM*jnrD;
1018
1019             /* load j atom coordinates */
1020             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1021                                               x+j_coord_offsetC,x+j_coord_offsetD,
1022                                               &jx0,&jy0,&jz0);
1023
1024             /* Calculate displacement vector */
1025             dx10             = _mm_sub_ps(ix1,jx0);
1026             dy10             = _mm_sub_ps(iy1,jy0);
1027             dz10             = _mm_sub_ps(iz1,jz0);
1028             dx20             = _mm_sub_ps(ix2,jx0);
1029             dy20             = _mm_sub_ps(iy2,jy0);
1030             dz20             = _mm_sub_ps(iz2,jz0);
1031             dx30             = _mm_sub_ps(ix3,jx0);
1032             dy30             = _mm_sub_ps(iy3,jy0);
1033             dz30             = _mm_sub_ps(iz3,jz0);
1034
1035             /* Calculate squared distance and things based on it */
1036             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1037             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1038             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1039
1040             rinv10           = sse2_invsqrt_f(rsq10);
1041             rinv20           = sse2_invsqrt_f(rsq20);
1042             rinv30           = sse2_invsqrt_f(rsq30);
1043
1044             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1045             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1046             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1047
1048             /* Load parameters for j particles */
1049             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1050                                                               charge+jnrC+0,charge+jnrD+0);
1051
1052             fjx0             = _mm_setzero_ps();
1053             fjy0             = _mm_setzero_ps();
1054             fjz0             = _mm_setzero_ps();
1055
1056             /**************************
1057              * CALCULATE INTERACTIONS *
1058              **************************/
1059
1060             if (gmx_mm_any_lt(rsq10,rcutoff2))
1061             {
1062
1063             r10              = _mm_mul_ps(rsq10,rinv10);
1064             r10              = _mm_andnot_ps(dummy_mask,r10);
1065
1066             /* Compute parameters for interactions between i and j atoms */
1067             qq10             = _mm_mul_ps(iq1,jq0);
1068
1069             /* EWALD ELECTROSTATICS */
1070
1071             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1072             ewrt             = _mm_mul_ps(r10,ewtabscale);
1073             ewitab           = _mm_cvttps_epi32(ewrt);
1074             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1075             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1076                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1077                                          &ewtabF,&ewtabFn);
1078             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1079             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1080
1081             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1082
1083             fscal            = felec;
1084
1085             fscal            = _mm_and_ps(fscal,cutoff_mask);
1086
1087             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1088
1089             /* Calculate temporary vectorial force */
1090             tx               = _mm_mul_ps(fscal,dx10);
1091             ty               = _mm_mul_ps(fscal,dy10);
1092             tz               = _mm_mul_ps(fscal,dz10);
1093
1094             /* Update vectorial force */
1095             fix1             = _mm_add_ps(fix1,tx);
1096             fiy1             = _mm_add_ps(fiy1,ty);
1097             fiz1             = _mm_add_ps(fiz1,tz);
1098
1099             fjx0             = _mm_add_ps(fjx0,tx);
1100             fjy0             = _mm_add_ps(fjy0,ty);
1101             fjz0             = _mm_add_ps(fjz0,tz);
1102
1103             }
1104
1105             /**************************
1106              * CALCULATE INTERACTIONS *
1107              **************************/
1108
1109             if (gmx_mm_any_lt(rsq20,rcutoff2))
1110             {
1111
1112             r20              = _mm_mul_ps(rsq20,rinv20);
1113             r20              = _mm_andnot_ps(dummy_mask,r20);
1114
1115             /* Compute parameters for interactions between i and j atoms */
1116             qq20             = _mm_mul_ps(iq2,jq0);
1117
1118             /* EWALD ELECTROSTATICS */
1119
1120             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1121             ewrt             = _mm_mul_ps(r20,ewtabscale);
1122             ewitab           = _mm_cvttps_epi32(ewrt);
1123             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1124             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1125                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1126                                          &ewtabF,&ewtabFn);
1127             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1128             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1129
1130             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1131
1132             fscal            = felec;
1133
1134             fscal            = _mm_and_ps(fscal,cutoff_mask);
1135
1136             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1137
1138             /* Calculate temporary vectorial force */
1139             tx               = _mm_mul_ps(fscal,dx20);
1140             ty               = _mm_mul_ps(fscal,dy20);
1141             tz               = _mm_mul_ps(fscal,dz20);
1142
1143             /* Update vectorial force */
1144             fix2             = _mm_add_ps(fix2,tx);
1145             fiy2             = _mm_add_ps(fiy2,ty);
1146             fiz2             = _mm_add_ps(fiz2,tz);
1147
1148             fjx0             = _mm_add_ps(fjx0,tx);
1149             fjy0             = _mm_add_ps(fjy0,ty);
1150             fjz0             = _mm_add_ps(fjz0,tz);
1151
1152             }
1153
1154             /**************************
1155              * CALCULATE INTERACTIONS *
1156              **************************/
1157
1158             if (gmx_mm_any_lt(rsq30,rcutoff2))
1159             {
1160
1161             r30              = _mm_mul_ps(rsq30,rinv30);
1162             r30              = _mm_andnot_ps(dummy_mask,r30);
1163
1164             /* Compute parameters for interactions between i and j atoms */
1165             qq30             = _mm_mul_ps(iq3,jq0);
1166
1167             /* EWALD ELECTROSTATICS */
1168
1169             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1170             ewrt             = _mm_mul_ps(r30,ewtabscale);
1171             ewitab           = _mm_cvttps_epi32(ewrt);
1172             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1173             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1174                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1175                                          &ewtabF,&ewtabFn);
1176             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1177             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1178
1179             cutoff_mask      = _mm_cmplt_ps(rsq30,rcutoff2);
1180
1181             fscal            = felec;
1182
1183             fscal            = _mm_and_ps(fscal,cutoff_mask);
1184
1185             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1186
1187             /* Calculate temporary vectorial force */
1188             tx               = _mm_mul_ps(fscal,dx30);
1189             ty               = _mm_mul_ps(fscal,dy30);
1190             tz               = _mm_mul_ps(fscal,dz30);
1191
1192             /* Update vectorial force */
1193             fix3             = _mm_add_ps(fix3,tx);
1194             fiy3             = _mm_add_ps(fiy3,ty);
1195             fiz3             = _mm_add_ps(fiz3,tz);
1196
1197             fjx0             = _mm_add_ps(fjx0,tx);
1198             fjy0             = _mm_add_ps(fjy0,ty);
1199             fjz0             = _mm_add_ps(fjz0,tz);
1200
1201             }
1202
1203             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1204             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1205             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1206             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1207
1208             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1209
1210             /* Inner loop uses 120 flops */
1211         }
1212
1213         /* End of innermost loop */
1214
1215         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1216                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1217
1218         /* Increment number of inner iterations */
1219         inneriter                  += j_index_end - j_index_start;
1220
1221         /* Outer loop uses 18 flops */
1222     }
1223
1224     /* Increment number of outer iterations */
1225     outeriter        += nri;
1226
1227     /* Update outer/inner flops */
1228
1229     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*18 + inneriter*120);
1230 }