src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEwSh_VdwNone_GeomW3P1_sse2_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse2_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "gromacs/math/vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gromacs/simd/math_x86_sse2_single.h"
  50 #include "kernelutil_x86_sse2_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_sse2_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            None
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_VF_sse2_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     __m128i          ewitab;
  99     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 100     real             *ewtab;
 101     __m128           dummy_mask,cutoff_mask;
 102     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 103     __m128           one     = _mm_set1_ps(1.0);
 104     __m128           two     = _mm_set1_ps(2.0);
 105     x                = xx[0];
 106     f                = ff[0];
 107
 108     nri              = nlist->nri;
 109     iinr             = nlist->iinr;
 110     jindex           = nlist->jindex;
 111     jjnr             = nlist->jjnr;
 112     shiftidx         = nlist->shift;
 113     gid              = nlist->gid;
 114     shiftvec         = fr->shift_vec[0];
 115     fshift           = fr->fshift[0];
 116     facel            = _mm_set1_ps(fr->epsfac);
 117     charge           = mdatoms->chargeA;
 118
 119     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 120     ewtab            = fr->ic->tabq_coul_FDV0;
 121     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 122     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 123
 124     /* Setup water-specific parameters */
 125     inr              = nlist->iinr[0];
 126     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 127     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 128     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 129
 130     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 131     rcutoff_scalar   = fr->rcoulomb;
 132     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 133     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171         fix1             = _mm_setzero_ps();
 172         fiy1             = _mm_setzero_ps();
 173         fiz1             = _mm_setzero_ps();
 174         fix2             = _mm_setzero_ps();
 175         fiy2             = _mm_setzero_ps();
 176         fiz2             = _mm_setzero_ps();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_ps();
 180
 181         /* Start inner kernel loop */
 182         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 183         {
 184
 185             /* Get j neighbor index, and coordinate index */
 186             jnrA             = jjnr[jidx];
 187             jnrB             = jjnr[jidx+1];
 188             jnrC             = jjnr[jidx+2];
 189             jnrD             = jjnr[jidx+3];
 190             j_coord_offsetA  = DIM*jnrA;
 191             j_coord_offsetB  = DIM*jnrB;
 192             j_coord_offsetC  = DIM*jnrC;
 193             j_coord_offsetD  = DIM*jnrD;
 194
 195             /* load j atom coordinates */
 196             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 197                                               x+j_coord_offsetC,x+j_coord_offsetD,
 198                                               &jx0,&jy0,&jz0);
 199
 200             /* Calculate displacement vector */
 201             dx00             = _mm_sub_ps(ix0,jx0);
 202             dy00             = _mm_sub_ps(iy0,jy0);
 203             dz00             = _mm_sub_ps(iz0,jz0);
 204             dx10             = _mm_sub_ps(ix1,jx0);
 205             dy10             = _mm_sub_ps(iy1,jy0);
 206             dz10             = _mm_sub_ps(iz1,jz0);
 207             dx20             = _mm_sub_ps(ix2,jx0);
 208             dy20             = _mm_sub_ps(iy2,jy0);
 209             dz20             = _mm_sub_ps(iz2,jz0);
 210
 211             /* Calculate squared distance and things based on it */
 212             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 213             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 214             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 215
 216             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 217             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 218             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 219
 220             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 221             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 222             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 223
 224             /* Load parameters for j particles */
 225             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 226                                                               charge+jnrC+0,charge+jnrD+0);
 227
 228             fjx0             = _mm_setzero_ps();
 229             fjy0             = _mm_setzero_ps();
 230             fjz0             = _mm_setzero_ps();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             if (gmx_mm_any_lt(rsq00,rcutoff2))
 237             {
 238
 239             r00              = _mm_mul_ps(rsq00,rinv00);
 240
 241             /* Compute parameters for interactions between i and j atoms */
 242             qq00             = _mm_mul_ps(iq0,jq0);
 243
 244             /* EWALD ELECTROSTATICS */
 245
 246             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 247             ewrt             = _mm_mul_ps(r00,ewtabscale);
 248             ewitab           = _mm_cvttps_epi32(ewrt);
 249             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 250             ewitab           = _mm_slli_epi32(ewitab,2);
 251             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 252             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 253             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 254             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 255             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 256             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 257             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 258             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 259             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 260
 261             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 262
 263             /* Update potential sum for this i atom from the interaction with this j atom. */
 264             velec            = _mm_and_ps(velec,cutoff_mask);
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266
 267             fscal            = felec;
 268
 269             fscal            = _mm_and_ps(fscal,cutoff_mask);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             fjx0             = _mm_add_ps(fjx0,tx);
 282             fjy0             = _mm_add_ps(fjy0,ty);
 283             fjz0             = _mm_add_ps(fjz0,tz);
 284
 285             }
 286
 287             /**************************
 288              * CALCULATE INTERACTIONS *
 289              **************************/
 290
 291             if (gmx_mm_any_lt(rsq10,rcutoff2))
 292             {
 293
 294             r10              = _mm_mul_ps(rsq10,rinv10);
 295
 296             /* Compute parameters for interactions between i and j atoms */
 297             qq10             = _mm_mul_ps(iq1,jq0);
 298
 299             /* EWALD ELECTROSTATICS */
 300
 301             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 302             ewrt             = _mm_mul_ps(r10,ewtabscale);
 303             ewitab           = _mm_cvttps_epi32(ewrt);
 304             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 305             ewitab           = _mm_slli_epi32(ewitab,2);
 306             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 307             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 308             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 309             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 310             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 311             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 312             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 313             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 314             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 315
 316             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 317
 318             /* Update potential sum for this i atom from the interaction with this j atom. */
 319             velec            = _mm_and_ps(velec,cutoff_mask);
 320             velecsum         = _mm_add_ps(velecsum,velec);
 321
 322             fscal            = felec;
 323
 324             fscal            = _mm_and_ps(fscal,cutoff_mask);
 325
 326             /* Calculate temporary vectorial force */
 327             tx               = _mm_mul_ps(fscal,dx10);
 328             ty               = _mm_mul_ps(fscal,dy10);
 329             tz               = _mm_mul_ps(fscal,dz10);
 330
 331             /* Update vectorial force */
 332             fix1             = _mm_add_ps(fix1,tx);
 333             fiy1             = _mm_add_ps(fiy1,ty);
 334             fiz1             = _mm_add_ps(fiz1,tz);
 335
 336             fjx0             = _mm_add_ps(fjx0,tx);
 337             fjy0             = _mm_add_ps(fjy0,ty);
 338             fjz0             = _mm_add_ps(fjz0,tz);
 339
 340             }
 341
 342             /**************************
 343              * CALCULATE INTERACTIONS *
 344              **************************/
 345
 346             if (gmx_mm_any_lt(rsq20,rcutoff2))
 347             {
 348
 349             r20              = _mm_mul_ps(rsq20,rinv20);
 350
 351             /* Compute parameters for interactions between i and j atoms */
 352             qq20             = _mm_mul_ps(iq2,jq0);
 353
 354             /* EWALD ELECTROSTATICS */
 355
 356             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 357             ewrt             = _mm_mul_ps(r20,ewtabscale);
 358             ewitab           = _mm_cvttps_epi32(ewrt);
 359             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 360             ewitab           = _mm_slli_epi32(ewitab,2);
 361             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 362             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 363             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 364             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 365             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 366             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 367             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 368             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 369             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 370
 371             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 372
 373             /* Update potential sum for this i atom from the interaction with this j atom. */
 374             velec            = _mm_and_ps(velec,cutoff_mask);
 375             velecsum         = _mm_add_ps(velecsum,velec);
 376
 377             fscal            = felec;
 378
 379             fscal            = _mm_and_ps(fscal,cutoff_mask);
 380
 381             /* Calculate temporary vectorial force */
 382             tx               = _mm_mul_ps(fscal,dx20);
 383             ty               = _mm_mul_ps(fscal,dy20);
 384             tz               = _mm_mul_ps(fscal,dz20);
 385
 386             /* Update vectorial force */
 387             fix2             = _mm_add_ps(fix2,tx);
 388             fiy2             = _mm_add_ps(fiy2,ty);
 389             fiz2             = _mm_add_ps(fiz2,tz);
 390
 391             fjx0             = _mm_add_ps(fjx0,tx);
 392             fjy0             = _mm_add_ps(fjy0,ty);
 393             fjz0             = _mm_add_ps(fjz0,tz);
 394
 395             }
 396
 397             fjptrA             = f+j_coord_offsetA;
 398             fjptrB             = f+j_coord_offsetB;
 399             fjptrC             = f+j_coord_offsetC;
 400             fjptrD             = f+j_coord_offsetD;
 401
 402             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 403
 404             /* Inner loop uses 138 flops */
 405         }
 406
 407         if(jidx<j_index_end)
 408         {
 409
 410             /* Get j neighbor index, and coordinate index */
 411             jnrlistA         = jjnr[jidx];
 412             jnrlistB         = jjnr[jidx+1];
 413             jnrlistC         = jjnr[jidx+2];
 414             jnrlistD         = jjnr[jidx+3];
 415             /* Sign of each element will be negative for non-real atoms.
 416              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 417              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 418              */
 419             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 420             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 421             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 422             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 423             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 424             j_coord_offsetA  = DIM*jnrA;
 425             j_coord_offsetB  = DIM*jnrB;
 426             j_coord_offsetC  = DIM*jnrC;
 427             j_coord_offsetD  = DIM*jnrD;
 428
 429             /* load j atom coordinates */
 430             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 431                                               x+j_coord_offsetC,x+j_coord_offsetD,
 432                                               &jx0,&jy0,&jz0);
 433
 434             /* Calculate displacement vector */
 435             dx00             = _mm_sub_ps(ix0,jx0);
 436             dy00             = _mm_sub_ps(iy0,jy0);
 437             dz00             = _mm_sub_ps(iz0,jz0);
 438             dx10             = _mm_sub_ps(ix1,jx0);
 439             dy10             = _mm_sub_ps(iy1,jy0);
 440             dz10             = _mm_sub_ps(iz1,jz0);
 441             dx20             = _mm_sub_ps(ix2,jx0);
 442             dy20             = _mm_sub_ps(iy2,jy0);
 443             dz20             = _mm_sub_ps(iz2,jz0);
 444
 445             /* Calculate squared distance and things based on it */
 446             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 447             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 448             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 449
 450             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 451             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 452             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 453
 454             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 455             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 456             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 457
 458             /* Load parameters for j particles */
 459             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 460                                                               charge+jnrC+0,charge+jnrD+0);
 461
 462             fjx0             = _mm_setzero_ps();
 463             fjy0             = _mm_setzero_ps();
 464             fjz0             = _mm_setzero_ps();
 465
 466             /**************************
 467              * CALCULATE INTERACTIONS *
 468              **************************/
 469
 470             if (gmx_mm_any_lt(rsq00,rcutoff2))
 471             {
 472
 473             r00              = _mm_mul_ps(rsq00,rinv00);
 474             r00              = _mm_andnot_ps(dummy_mask,r00);
 475
 476             /* Compute parameters for interactions between i and j atoms */
 477             qq00             = _mm_mul_ps(iq0,jq0);
 478
 479             /* EWALD ELECTROSTATICS */
 480
 481             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 482             ewrt             = _mm_mul_ps(r00,ewtabscale);
 483             ewitab           = _mm_cvttps_epi32(ewrt);
 484             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 485             ewitab           = _mm_slli_epi32(ewitab,2);
 486             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 487             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 488             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 489             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 490             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 491             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 492             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 493             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_sub_ps(rinv00,sh_ewald),velec));
 494             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 495
 496             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             velec            = _mm_and_ps(velec,cutoff_mask);
 500             velec            = _mm_andnot_ps(dummy_mask,velec);
 501             velecsum         = _mm_add_ps(velecsum,velec);
 502
 503             fscal            = felec;
 504
 505             fscal            = _mm_and_ps(fscal,cutoff_mask);
 506
 507             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 508
 509             /* Calculate temporary vectorial force */
 510             tx               = _mm_mul_ps(fscal,dx00);
 511             ty               = _mm_mul_ps(fscal,dy00);
 512             tz               = _mm_mul_ps(fscal,dz00);
 513
 514             /* Update vectorial force */
 515             fix0             = _mm_add_ps(fix0,tx);
 516             fiy0             = _mm_add_ps(fiy0,ty);
 517             fiz0             = _mm_add_ps(fiz0,tz);
 518
 519             fjx0             = _mm_add_ps(fjx0,tx);
 520             fjy0             = _mm_add_ps(fjy0,ty);
 521             fjz0             = _mm_add_ps(fjz0,tz);
 522
 523             }
 524
 525             /**************************
 526              * CALCULATE INTERACTIONS *
 527              **************************/
 528
 529             if (gmx_mm_any_lt(rsq10,rcutoff2))
 530             {
 531
 532             r10              = _mm_mul_ps(rsq10,rinv10);
 533             r10              = _mm_andnot_ps(dummy_mask,r10);
 534
 535             /* Compute parameters for interactions between i and j atoms */
 536             qq10             = _mm_mul_ps(iq1,jq0);
 537
 538             /* EWALD ELECTROSTATICS */
 539
 540             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 541             ewrt             = _mm_mul_ps(r10,ewtabscale);
 542             ewitab           = _mm_cvttps_epi32(ewrt);
 543             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 544             ewitab           = _mm_slli_epi32(ewitab,2);
 545             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 546             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 547             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 548             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 549             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 550             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 551             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 552             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_sub_ps(rinv10,sh_ewald),velec));
 553             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 554
 555             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 556
 557             /* Update potential sum for this i atom from the interaction with this j atom. */
 558             velec            = _mm_and_ps(velec,cutoff_mask);
 559             velec            = _mm_andnot_ps(dummy_mask,velec);
 560             velecsum         = _mm_add_ps(velecsum,velec);
 561
 562             fscal            = felec;
 563
 564             fscal            = _mm_and_ps(fscal,cutoff_mask);
 565
 566             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 567
 568             /* Calculate temporary vectorial force */
 569             tx               = _mm_mul_ps(fscal,dx10);
 570             ty               = _mm_mul_ps(fscal,dy10);
 571             tz               = _mm_mul_ps(fscal,dz10);
 572
 573             /* Update vectorial force */
 574             fix1             = _mm_add_ps(fix1,tx);
 575             fiy1             = _mm_add_ps(fiy1,ty);
 576             fiz1             = _mm_add_ps(fiz1,tz);
 577
 578             fjx0             = _mm_add_ps(fjx0,tx);
 579             fjy0             = _mm_add_ps(fjy0,ty);
 580             fjz0             = _mm_add_ps(fjz0,tz);
 581
 582             }
 583
 584             /**************************
 585              * CALCULATE INTERACTIONS *
 586              **************************/
 587
 588             if (gmx_mm_any_lt(rsq20,rcutoff2))
 589             {
 590
 591             r20              = _mm_mul_ps(rsq20,rinv20);
 592             r20              = _mm_andnot_ps(dummy_mask,r20);
 593
 594             /* Compute parameters for interactions between i and j atoms */
 595             qq20             = _mm_mul_ps(iq2,jq0);
 596
 597             /* EWALD ELECTROSTATICS */
 598
 599             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 600             ewrt             = _mm_mul_ps(r20,ewtabscale);
 601             ewitab           = _mm_cvttps_epi32(ewrt);
 602             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 603             ewitab           = _mm_slli_epi32(ewitab,2);
 604             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 605             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 606             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 607             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 608             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 609             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 610             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 611             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_sub_ps(rinv20,sh_ewald),velec));
 612             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 613
 614             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm_and_ps(velec,cutoff_mask);
 618             velec            = _mm_andnot_ps(dummy_mask,velec);
 619             velecsum         = _mm_add_ps(velecsum,velec);
 620
 621             fscal            = felec;
 622
 623             fscal            = _mm_and_ps(fscal,cutoff_mask);
 624
 625             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 626
 627             /* Calculate temporary vectorial force */
 628             tx               = _mm_mul_ps(fscal,dx20);
 629             ty               = _mm_mul_ps(fscal,dy20);
 630             tz               = _mm_mul_ps(fscal,dz20);
 631
 632             /* Update vectorial force */
 633             fix2             = _mm_add_ps(fix2,tx);
 634             fiy2             = _mm_add_ps(fiy2,ty);
 635             fiz2             = _mm_add_ps(fiz2,tz);
 636
 637             fjx0             = _mm_add_ps(fjx0,tx);
 638             fjy0             = _mm_add_ps(fjy0,ty);
 639             fjz0             = _mm_add_ps(fjz0,tz);
 640
 641             }
 642
 643             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 644             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 645             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 646             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 647
 648             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 649
 650             /* Inner loop uses 141 flops */
 651         }
 652
 653         /* End of innermost loop */
 654
 655         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 656                                               f+i_coord_offset,fshift+i_shift_offset);
 657
 658         ggid                        = gid[iidx];
 659         /* Update potential energies */
 660         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 661
 662         /* Increment number of inner iterations */
 663         inneriter                  += j_index_end - j_index_start;
 664
 665         /* Outer loop uses 19 flops */
 666     }
 667
 668     /* Increment number of outer iterations */
 669     outeriter        += nri;
 670
 671     /* Update outer/inner flops */
 672
 673     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*141);
 674 }
 675 /*
 676  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse2_single
 677  * Electrostatics interaction: Ewald
 678  * VdW interaction:            None
 679  * Geometry:                   Water3-Particle
 680  * Calculate force/pot:        Force
 681  */
 682 void
 683 nb_kernel_ElecEwSh_VdwNone_GeomW3P1_F_sse2_single
 684                     (t_nblist                    * gmx_restrict       nlist,
 685                      rvec                        * gmx_restrict          xx,
 686                      rvec                        * gmx_restrict          ff,
 687                      t_forcerec                  * gmx_restrict          fr,
 688                      t_mdatoms                   * gmx_restrict     mdatoms,
 689                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 690                      t_nrnb                      * gmx_restrict        nrnb)
 691 {
 692     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 693      * just 0 for non-waters.
 694      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 695      * jnr indices corresponding to data put in the four positions in the SIMD register.
 696      */
 697     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 698     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 699     int              jnrA,jnrB,jnrC,jnrD;
 700     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 701     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 702     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 703     real             rcutoff_scalar;
 704     real             *shiftvec,*fshift,*x,*f;
 705     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 706     real             scratch[4*DIM];
 707     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 708     int              vdwioffset0;
 709     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 710     int              vdwioffset1;
 711     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 712     int              vdwioffset2;
 713     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 714     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 715     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 716     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 717     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 718     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 719     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 720     real             *charge;
 721     __m128i          ewitab;
 722     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 723     real             *ewtab;
 724     __m128           dummy_mask,cutoff_mask;
 725     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 726     __m128           one     = _mm_set1_ps(1.0);
 727     __m128           two     = _mm_set1_ps(2.0);
 728     x                = xx[0];
 729     f                = ff[0];
 730
 731     nri              = nlist->nri;
 732     iinr             = nlist->iinr;
 733     jindex           = nlist->jindex;
 734     jjnr             = nlist->jjnr;
 735     shiftidx         = nlist->shift;
 736     gid              = nlist->gid;
 737     shiftvec         = fr->shift_vec[0];
 738     fshift           = fr->fshift[0];
 739     facel            = _mm_set1_ps(fr->epsfac);
 740     charge           = mdatoms->chargeA;
 741
 742     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 743     ewtab            = fr->ic->tabq_coul_F;
 744     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 745     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 746
 747     /* Setup water-specific parameters */
 748     inr              = nlist->iinr[0];
 749     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 750     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 751     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 752
 753     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 754     rcutoff_scalar   = fr->rcoulomb;
 755     rcutoff          = _mm_set1_ps(rcutoff_scalar);
 756     rcutoff2         = _mm_mul_ps(rcutoff,rcutoff);
 757
 758     /* Avoid stupid compiler warnings */
 759     jnrA = jnrB = jnrC = jnrD = 0;
 760     j_coord_offsetA = 0;
 761     j_coord_offsetB = 0;
 762     j_coord_offsetC = 0;
 763     j_coord_offsetD = 0;
 764
 765     outeriter        = 0;
 766     inneriter        = 0;
 767
 768     for(iidx=0;iidx<4*DIM;iidx++)
 769     {
 770         scratch[iidx] = 0.0;
 771     }
 772
 773     /* Start outer loop over neighborlists */
 774     for(iidx=0; iidx<nri; iidx++)
 775     {
 776         /* Load shift vector for this list */
 777         i_shift_offset   = DIM*shiftidx[iidx];
 778
 779         /* Load limits for loop over neighbors */
 780         j_index_start    = jindex[iidx];
 781         j_index_end      = jindex[iidx+1];
 782
 783         /* Get outer coordinate index */
 784         inr              = iinr[iidx];
 785         i_coord_offset   = DIM*inr;
 786
 787         /* Load i particle coords and add shift vector */
 788         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 789                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 790
 791         fix0             = _mm_setzero_ps();
 792         fiy0             = _mm_setzero_ps();
 793         fiz0             = _mm_setzero_ps();
 794         fix1             = _mm_setzero_ps();
 795         fiy1             = _mm_setzero_ps();
 796         fiz1             = _mm_setzero_ps();
 797         fix2             = _mm_setzero_ps();
 798         fiy2             = _mm_setzero_ps();
 799         fiz2             = _mm_setzero_ps();
 800
 801         /* Start inner kernel loop */
 802         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 803         {
 804
 805             /* Get j neighbor index, and coordinate index */
 806             jnrA             = jjnr[jidx];
 807             jnrB             = jjnr[jidx+1];
 808             jnrC             = jjnr[jidx+2];
 809             jnrD             = jjnr[jidx+3];
 810             j_coord_offsetA  = DIM*jnrA;
 811             j_coord_offsetB  = DIM*jnrB;
 812             j_coord_offsetC  = DIM*jnrC;
 813             j_coord_offsetD  = DIM*jnrD;
 814
 815             /* load j atom coordinates */
 816             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 817                                               x+j_coord_offsetC,x+j_coord_offsetD,
 818                                               &jx0,&jy0,&jz0);
 819
 820             /* Calculate displacement vector */
 821             dx00             = _mm_sub_ps(ix0,jx0);
 822             dy00             = _mm_sub_ps(iy0,jy0);
 823             dz00             = _mm_sub_ps(iz0,jz0);
 824             dx10             = _mm_sub_ps(ix1,jx0);
 825             dy10             = _mm_sub_ps(iy1,jy0);
 826             dz10             = _mm_sub_ps(iz1,jz0);
 827             dx20             = _mm_sub_ps(ix2,jx0);
 828             dy20             = _mm_sub_ps(iy2,jy0);
 829             dz20             = _mm_sub_ps(iz2,jz0);
 830
 831             /* Calculate squared distance and things based on it */
 832             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 833             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 834             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 835
 836             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 837             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 838             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 839
 840             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 841             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 842             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 843
 844             /* Load parameters for j particles */
 845             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 846                                                               charge+jnrC+0,charge+jnrD+0);
 847
 848             fjx0             = _mm_setzero_ps();
 849             fjy0             = _mm_setzero_ps();
 850             fjz0             = _mm_setzero_ps();
 851
 852             /**************************
 853              * CALCULATE INTERACTIONS *
 854              **************************/
 855
 856             if (gmx_mm_any_lt(rsq00,rcutoff2))
 857             {
 858
 859             r00              = _mm_mul_ps(rsq00,rinv00);
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq00             = _mm_mul_ps(iq0,jq0);
 863
 864             /* EWALD ELECTROSTATICS */
 865
 866             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 867             ewrt             = _mm_mul_ps(r00,ewtabscale);
 868             ewitab           = _mm_cvttps_epi32(ewrt);
 869             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 870             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 871                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 872                                          &ewtabF,&ewtabFn);
 873             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 874             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 875
 876             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
 877
 878             fscal            = felec;
 879
 880             fscal            = _mm_and_ps(fscal,cutoff_mask);
 881
 882             /* Calculate temporary vectorial force */
 883             tx               = _mm_mul_ps(fscal,dx00);
 884             ty               = _mm_mul_ps(fscal,dy00);
 885             tz               = _mm_mul_ps(fscal,dz00);
 886
 887             /* Update vectorial force */
 888             fix0             = _mm_add_ps(fix0,tx);
 889             fiy0             = _mm_add_ps(fiy0,ty);
 890             fiz0             = _mm_add_ps(fiz0,tz);
 891
 892             fjx0             = _mm_add_ps(fjx0,tx);
 893             fjy0             = _mm_add_ps(fjy0,ty);
 894             fjz0             = _mm_add_ps(fjz0,tz);
 895
 896             }
 897
 898             /**************************
 899              * CALCULATE INTERACTIONS *
 900              **************************/
 901
 902             if (gmx_mm_any_lt(rsq10,rcutoff2))
 903             {
 904
 905             r10              = _mm_mul_ps(rsq10,rinv10);
 906
 907             /* Compute parameters for interactions between i and j atoms */
 908             qq10             = _mm_mul_ps(iq1,jq0);
 909
 910             /* EWALD ELECTROSTATICS */
 911
 912             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 913             ewrt             = _mm_mul_ps(r10,ewtabscale);
 914             ewitab           = _mm_cvttps_epi32(ewrt);
 915             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 916             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 917                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 918                                          &ewtabF,&ewtabFn);
 919             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 920             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 921
 922             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
 923
 924             fscal            = felec;
 925
 926             fscal            = _mm_and_ps(fscal,cutoff_mask);
 927
 928             /* Calculate temporary vectorial force */
 929             tx               = _mm_mul_ps(fscal,dx10);
 930             ty               = _mm_mul_ps(fscal,dy10);
 931             tz               = _mm_mul_ps(fscal,dz10);
 932
 933             /* Update vectorial force */
 934             fix1             = _mm_add_ps(fix1,tx);
 935             fiy1             = _mm_add_ps(fiy1,ty);
 936             fiz1             = _mm_add_ps(fiz1,tz);
 937
 938             fjx0             = _mm_add_ps(fjx0,tx);
 939             fjy0             = _mm_add_ps(fjy0,ty);
 940             fjz0             = _mm_add_ps(fjz0,tz);
 941
 942             }
 943
 944             /**************************
 945              * CALCULATE INTERACTIONS *
 946              **************************/
 947
 948             if (gmx_mm_any_lt(rsq20,rcutoff2))
 949             {
 950
 951             r20              = _mm_mul_ps(rsq20,rinv20);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq20             = _mm_mul_ps(iq2,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 959             ewrt             = _mm_mul_ps(r20,ewtabscale);
 960             ewitab           = _mm_cvttps_epi32(ewrt);
 961             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 962             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 963                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 964                                          &ewtabF,&ewtabFn);
 965             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 966             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 967
 968             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
 969
 970             fscal            = felec;
 971
 972             fscal            = _mm_and_ps(fscal,cutoff_mask);
 973
 974             /* Calculate temporary vectorial force */
 975             tx               = _mm_mul_ps(fscal,dx20);
 976             ty               = _mm_mul_ps(fscal,dy20);
 977             tz               = _mm_mul_ps(fscal,dz20);
 978
 979             /* Update vectorial force */
 980             fix2             = _mm_add_ps(fix2,tx);
 981             fiy2             = _mm_add_ps(fiy2,ty);
 982             fiz2             = _mm_add_ps(fiz2,tz);
 983
 984             fjx0             = _mm_add_ps(fjx0,tx);
 985             fjy0             = _mm_add_ps(fjy0,ty);
 986             fjz0             = _mm_add_ps(fjz0,tz);
 987
 988             }
 989
 990             fjptrA             = f+j_coord_offsetA;
 991             fjptrB             = f+j_coord_offsetB;
 992             fjptrC             = f+j_coord_offsetC;
 993             fjptrD             = f+j_coord_offsetD;
 994
 995             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 996
 997             /* Inner loop uses 117 flops */
 998         }
 999
1000         if(jidx<j_index_end)
1001         {
1002
1003             /* Get j neighbor index, and coordinate index */
1004             jnrlistA         = jjnr[jidx];
1005             jnrlistB         = jjnr[jidx+1];
1006             jnrlistC         = jjnr[jidx+2];
1007             jnrlistD         = jjnr[jidx+3];
1008             /* Sign of each element will be negative for non-real atoms.
1009              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1010              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1011              */
1012             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1013             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1014             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1015             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1016             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1017             j_coord_offsetA  = DIM*jnrA;
1018             j_coord_offsetB  = DIM*jnrB;
1019             j_coord_offsetC  = DIM*jnrC;
1020             j_coord_offsetD  = DIM*jnrD;
1021
1022             /* load j atom coordinates */
1023             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1024                                               x+j_coord_offsetC,x+j_coord_offsetD,
1025                                               &jx0,&jy0,&jz0);
1026
1027             /* Calculate displacement vector */
1028             dx00             = _mm_sub_ps(ix0,jx0);
1029             dy00             = _mm_sub_ps(iy0,jy0);
1030             dz00             = _mm_sub_ps(iz0,jz0);
1031             dx10             = _mm_sub_ps(ix1,jx0);
1032             dy10             = _mm_sub_ps(iy1,jy0);
1033             dz10             = _mm_sub_ps(iz1,jz0);
1034             dx20             = _mm_sub_ps(ix2,jx0);
1035             dy20             = _mm_sub_ps(iy2,jy0);
1036             dz20             = _mm_sub_ps(iz2,jz0);
1037
1038             /* Calculate squared distance and things based on it */
1039             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1040             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1041             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1042
1043             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1044             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1045             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1046
1047             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1048             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1049             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1050
1051             /* Load parameters for j particles */
1052             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1053                                                               charge+jnrC+0,charge+jnrD+0);
1054
1055             fjx0             = _mm_setzero_ps();
1056             fjy0             = _mm_setzero_ps();
1057             fjz0             = _mm_setzero_ps();
1058
1059             /**************************
1060              * CALCULATE INTERACTIONS *
1061              **************************/
1062
1063             if (gmx_mm_any_lt(rsq00,rcutoff2))
1064             {
1065
1066             r00              = _mm_mul_ps(rsq00,rinv00);
1067             r00              = _mm_andnot_ps(dummy_mask,r00);
1068
1069             /* Compute parameters for interactions between i and j atoms */
1070             qq00             = _mm_mul_ps(iq0,jq0);
1071
1072             /* EWALD ELECTROSTATICS */
1073
1074             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1075             ewrt             = _mm_mul_ps(r00,ewtabscale);
1076             ewitab           = _mm_cvttps_epi32(ewrt);
1077             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1078             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1079                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1080                                          &ewtabF,&ewtabFn);
1081             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1082             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1083
1084             cutoff_mask      = _mm_cmplt_ps(rsq00,rcutoff2);
1085
1086             fscal            = felec;
1087
1088             fscal            = _mm_and_ps(fscal,cutoff_mask);
1089
1090             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1091
1092             /* Calculate temporary vectorial force */
1093             tx               = _mm_mul_ps(fscal,dx00);
1094             ty               = _mm_mul_ps(fscal,dy00);
1095             tz               = _mm_mul_ps(fscal,dz00);
1096
1097             /* Update vectorial force */
1098             fix0             = _mm_add_ps(fix0,tx);
1099             fiy0             = _mm_add_ps(fiy0,ty);
1100             fiz0             = _mm_add_ps(fiz0,tz);
1101
1102             fjx0             = _mm_add_ps(fjx0,tx);
1103             fjy0             = _mm_add_ps(fjy0,ty);
1104             fjz0             = _mm_add_ps(fjz0,tz);
1105
1106             }
1107
1108             /**************************
1109              * CALCULATE INTERACTIONS *
1110              **************************/
1111
1112             if (gmx_mm_any_lt(rsq10,rcutoff2))
1113             {
1114
1115             r10              = _mm_mul_ps(rsq10,rinv10);
1116             r10              = _mm_andnot_ps(dummy_mask,r10);
1117
1118             /* Compute parameters for interactions between i and j atoms */
1119             qq10             = _mm_mul_ps(iq1,jq0);
1120
1121             /* EWALD ELECTROSTATICS */
1122
1123             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1124             ewrt             = _mm_mul_ps(r10,ewtabscale);
1125             ewitab           = _mm_cvttps_epi32(ewrt);
1126             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1127             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1128                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1129                                          &ewtabF,&ewtabFn);
1130             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1131             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1132
1133             cutoff_mask      = _mm_cmplt_ps(rsq10,rcutoff2);
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_and_ps(fscal,cutoff_mask);
1138
1139             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1140
1141             /* Calculate temporary vectorial force */
1142             tx               = _mm_mul_ps(fscal,dx10);
1143             ty               = _mm_mul_ps(fscal,dy10);
1144             tz               = _mm_mul_ps(fscal,dz10);
1145
1146             /* Update vectorial force */
1147             fix1             = _mm_add_ps(fix1,tx);
1148             fiy1             = _mm_add_ps(fiy1,ty);
1149             fiz1             = _mm_add_ps(fiz1,tz);
1150
1151             fjx0             = _mm_add_ps(fjx0,tx);
1152             fjy0             = _mm_add_ps(fjy0,ty);
1153             fjz0             = _mm_add_ps(fjz0,tz);
1154
1155             }
1156
1157             /**************************
1158              * CALCULATE INTERACTIONS *
1159              **************************/
1160
1161             if (gmx_mm_any_lt(rsq20,rcutoff2))
1162             {
1163
1164             r20              = _mm_mul_ps(rsq20,rinv20);
1165             r20              = _mm_andnot_ps(dummy_mask,r20);
1166
1167             /* Compute parameters for interactions between i and j atoms */
1168             qq20             = _mm_mul_ps(iq2,jq0);
1169
1170             /* EWALD ELECTROSTATICS */
1171
1172             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1173             ewrt             = _mm_mul_ps(r20,ewtabscale);
1174             ewitab           = _mm_cvttps_epi32(ewrt);
1175             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1176             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1177                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1178                                          &ewtabF,&ewtabFn);
1179             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1180             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1181
1182             cutoff_mask      = _mm_cmplt_ps(rsq20,rcutoff2);
1183
1184             fscal            = felec;
1185
1186             fscal            = _mm_and_ps(fscal,cutoff_mask);
1187
1188             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1189
1190             /* Calculate temporary vectorial force */
1191             tx               = _mm_mul_ps(fscal,dx20);
1192             ty               = _mm_mul_ps(fscal,dy20);
1193             tz               = _mm_mul_ps(fscal,dz20);
1194
1195             /* Update vectorial force */
1196             fix2             = _mm_add_ps(fix2,tx);
1197             fiy2             = _mm_add_ps(fiy2,ty);
1198             fiz2             = _mm_add_ps(fiz2,tz);
1199
1200             fjx0             = _mm_add_ps(fjx0,tx);
1201             fjy0             = _mm_add_ps(fjy0,ty);
1202             fjz0             = _mm_add_ps(fjz0,tz);
1203
1204             }
1205
1206             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1207             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1208             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1209             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1210
1211             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1212
1213             /* Inner loop uses 120 flops */
1214         }
1215
1216         /* End of innermost loop */
1217
1218         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1219                                               f+i_coord_offset,fshift+i_shift_offset);
1220
1221         /* Increment number of inner iterations */
1222         inneriter                  += j_index_end - j_index_start;
1223
1224         /* Outer loop uses 18 flops */
1225     }
1226
1227     /* Increment number of outer iterations */
1228     outeriter        += nri;
1229
1230     /* Update outer/inner flops */
1231
1232     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*120);
1233 }