src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_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 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_sse2_single.h"
  48 #include "kernelutil_x86_sse2_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            None
  54  * Geometry:                   Water3-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEw_VdwNone_GeomW3P1_VF_sse2_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  76     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  77     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  78     real             rcutoff_scalar;
  79     real             *shiftvec,*fshift,*x,*f;
  80     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  81     real             scratch[4*DIM];
  82     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  83     int              vdwioffset0;
  84     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  85     int              vdwioffset1;
  86     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  87     int              vdwioffset2;
  88     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  89     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  90     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  91     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  92     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  93     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  94     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  95     real             *charge;
  96     __m128i          ewitab;
  97     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  98     real             *ewtab;
  99     __m128           dummy_mask,cutoff_mask;
 100     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 101     __m128           one     = _mm_set1_ps(1.0);
 102     __m128           two     = _mm_set1_ps(2.0);
 103     x                = xx[0];
 104     f                = ff[0];
 105
 106     nri              = nlist->nri;
 107     iinr             = nlist->iinr;
 108     jindex           = nlist->jindex;
 109     jjnr             = nlist->jjnr;
 110     shiftidx         = nlist->shift;
 111     gid              = nlist->gid;
 112     shiftvec         = fr->shift_vec[0];
 113     fshift           = fr->fshift[0];
 114     facel            = _mm_set1_ps(fr->epsfac);
 115     charge           = mdatoms->chargeA;
 116
 117     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 118     ewtab            = fr->ic->tabq_coul_FDV0;
 119     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 120     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 121
 122     /* Setup water-specific parameters */
 123     inr              = nlist->iinr[0];
 124     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 125     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 126     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 127
 128     /* Avoid stupid compiler warnings */
 129     jnrA = jnrB = jnrC = jnrD = 0;
 130     j_coord_offsetA = 0;
 131     j_coord_offsetB = 0;
 132     j_coord_offsetC = 0;
 133     j_coord_offsetD = 0;
 134
 135     outeriter        = 0;
 136     inneriter        = 0;
 137
 138     for(iidx=0;iidx<4*DIM;iidx++)
 139     {
 140         scratch[iidx] = 0.0;
 141     }
 142
 143     /* Start outer loop over neighborlists */
 144     for(iidx=0; iidx<nri; iidx++)
 145     {
 146         /* Load shift vector for this list */
 147         i_shift_offset   = DIM*shiftidx[iidx];
 148
 149         /* Load limits for loop over neighbors */
 150         j_index_start    = jindex[iidx];
 151         j_index_end      = jindex[iidx+1];
 152
 153         /* Get outer coordinate index */
 154         inr              = iinr[iidx];
 155         i_coord_offset   = DIM*inr;
 156
 157         /* Load i particle coords and add shift vector */
 158         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 159                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 160
 161         fix0             = _mm_setzero_ps();
 162         fiy0             = _mm_setzero_ps();
 163         fiz0             = _mm_setzero_ps();
 164         fix1             = _mm_setzero_ps();
 165         fiy1             = _mm_setzero_ps();
 166         fiz1             = _mm_setzero_ps();
 167         fix2             = _mm_setzero_ps();
 168         fiy2             = _mm_setzero_ps();
 169         fiz2             = _mm_setzero_ps();
 170
 171         /* Reset potential sums */
 172         velecsum         = _mm_setzero_ps();
 173
 174         /* Start inner kernel loop */
 175         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 176         {
 177
 178             /* Get j neighbor index, and coordinate index */
 179             jnrA             = jjnr[jidx];
 180             jnrB             = jjnr[jidx+1];
 181             jnrC             = jjnr[jidx+2];
 182             jnrD             = jjnr[jidx+3];
 183             j_coord_offsetA  = DIM*jnrA;
 184             j_coord_offsetB  = DIM*jnrB;
 185             j_coord_offsetC  = DIM*jnrC;
 186             j_coord_offsetD  = DIM*jnrD;
 187
 188             /* load j atom coordinates */
 189             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 190                                               x+j_coord_offsetC,x+j_coord_offsetD,
 191                                               &jx0,&jy0,&jz0);
 192
 193             /* Calculate displacement vector */
 194             dx00             = _mm_sub_ps(ix0,jx0);
 195             dy00             = _mm_sub_ps(iy0,jy0);
 196             dz00             = _mm_sub_ps(iz0,jz0);
 197             dx10             = _mm_sub_ps(ix1,jx0);
 198             dy10             = _mm_sub_ps(iy1,jy0);
 199             dz10             = _mm_sub_ps(iz1,jz0);
 200             dx20             = _mm_sub_ps(ix2,jx0);
 201             dy20             = _mm_sub_ps(iy2,jy0);
 202             dz20             = _mm_sub_ps(iz2,jz0);
 203
 204             /* Calculate squared distance and things based on it */
 205             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 206             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 207             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 208
 209             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 210             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 211             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 212
 213             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 214             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 215             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 216
 217             /* Load parameters for j particles */
 218             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 219                                                               charge+jnrC+0,charge+jnrD+0);
 220
 221             fjx0             = _mm_setzero_ps();
 222             fjy0             = _mm_setzero_ps();
 223             fjz0             = _mm_setzero_ps();
 224
 225             /**************************
 226              * CALCULATE INTERACTIONS *
 227              **************************/
 228
 229             r00              = _mm_mul_ps(rsq00,rinv00);
 230
 231             /* Compute parameters for interactions between i and j atoms */
 232             qq00             = _mm_mul_ps(iq0,jq0);
 233
 234             /* EWALD ELECTROSTATICS */
 235
 236             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 237             ewrt             = _mm_mul_ps(r00,ewtabscale);
 238             ewitab           = _mm_cvttps_epi32(ewrt);
 239             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 240             ewitab           = _mm_slli_epi32(ewitab,2);
 241             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 242             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 243             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 244             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 245             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 246             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 247             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 248             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 249             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 250
 251             /* Update potential sum for this i atom from the interaction with this j atom. */
 252             velecsum         = _mm_add_ps(velecsum,velec);
 253
 254             fscal            = felec;
 255
 256             /* Calculate temporary vectorial force */
 257             tx               = _mm_mul_ps(fscal,dx00);
 258             ty               = _mm_mul_ps(fscal,dy00);
 259             tz               = _mm_mul_ps(fscal,dz00);
 260
 261             /* Update vectorial force */
 262             fix0             = _mm_add_ps(fix0,tx);
 263             fiy0             = _mm_add_ps(fiy0,ty);
 264             fiz0             = _mm_add_ps(fiz0,tz);
 265
 266             fjx0             = _mm_add_ps(fjx0,tx);
 267             fjy0             = _mm_add_ps(fjy0,ty);
 268             fjz0             = _mm_add_ps(fjz0,tz);
 269
 270             /**************************
 271              * CALCULATE INTERACTIONS *
 272              **************************/
 273
 274             r10              = _mm_mul_ps(rsq10,rinv10);
 275
 276             /* Compute parameters for interactions between i and j atoms */
 277             qq10             = _mm_mul_ps(iq1,jq0);
 278
 279             /* EWALD ELECTROSTATICS */
 280
 281             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 282             ewrt             = _mm_mul_ps(r10,ewtabscale);
 283             ewitab           = _mm_cvttps_epi32(ewrt);
 284             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 285             ewitab           = _mm_slli_epi32(ewitab,2);
 286             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 287             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 288             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 289             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 290             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 291             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 292             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 293             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 294             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 295
 296             /* Update potential sum for this i atom from the interaction with this j atom. */
 297             velecsum         = _mm_add_ps(velecsum,velec);
 298
 299             fscal            = felec;
 300
 301             /* Calculate temporary vectorial force */
 302             tx               = _mm_mul_ps(fscal,dx10);
 303             ty               = _mm_mul_ps(fscal,dy10);
 304             tz               = _mm_mul_ps(fscal,dz10);
 305
 306             /* Update vectorial force */
 307             fix1             = _mm_add_ps(fix1,tx);
 308             fiy1             = _mm_add_ps(fiy1,ty);
 309             fiz1             = _mm_add_ps(fiz1,tz);
 310
 311             fjx0             = _mm_add_ps(fjx0,tx);
 312             fjy0             = _mm_add_ps(fjy0,ty);
 313             fjz0             = _mm_add_ps(fjz0,tz);
 314
 315             /**************************
 316              * CALCULATE INTERACTIONS *
 317              **************************/
 318
 319             r20              = _mm_mul_ps(rsq20,rinv20);
 320
 321             /* Compute parameters for interactions between i and j atoms */
 322             qq20             = _mm_mul_ps(iq2,jq0);
 323
 324             /* EWALD ELECTROSTATICS */
 325
 326             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 327             ewrt             = _mm_mul_ps(r20,ewtabscale);
 328             ewitab           = _mm_cvttps_epi32(ewrt);
 329             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 330             ewitab           = _mm_slli_epi32(ewitab,2);
 331             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 332             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 333             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 334             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 335             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 336             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 337             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 338             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 339             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 340
 341             /* Update potential sum for this i atom from the interaction with this j atom. */
 342             velecsum         = _mm_add_ps(velecsum,velec);
 343
 344             fscal            = felec;
 345
 346             /* Calculate temporary vectorial force */
 347             tx               = _mm_mul_ps(fscal,dx20);
 348             ty               = _mm_mul_ps(fscal,dy20);
 349             tz               = _mm_mul_ps(fscal,dz20);
 350
 351             /* Update vectorial force */
 352             fix2             = _mm_add_ps(fix2,tx);
 353             fiy2             = _mm_add_ps(fiy2,ty);
 354             fiz2             = _mm_add_ps(fiz2,tz);
 355
 356             fjx0             = _mm_add_ps(fjx0,tx);
 357             fjy0             = _mm_add_ps(fjy0,ty);
 358             fjz0             = _mm_add_ps(fjz0,tz);
 359
 360             fjptrA             = f+j_coord_offsetA;
 361             fjptrB             = f+j_coord_offsetB;
 362             fjptrC             = f+j_coord_offsetC;
 363             fjptrD             = f+j_coord_offsetD;
 364
 365             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 366
 367             /* Inner loop uses 123 flops */
 368         }
 369
 370         if(jidx<j_index_end)
 371         {
 372
 373             /* Get j neighbor index, and coordinate index */
 374             jnrlistA         = jjnr[jidx];
 375             jnrlistB         = jjnr[jidx+1];
 376             jnrlistC         = jjnr[jidx+2];
 377             jnrlistD         = jjnr[jidx+3];
 378             /* Sign of each element will be negative for non-real atoms.
 379              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 380              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 381              */
 382             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 383             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 384             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 385             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 386             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 387             j_coord_offsetA  = DIM*jnrA;
 388             j_coord_offsetB  = DIM*jnrB;
 389             j_coord_offsetC  = DIM*jnrC;
 390             j_coord_offsetD  = DIM*jnrD;
 391
 392             /* load j atom coordinates */
 393             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 394                                               x+j_coord_offsetC,x+j_coord_offsetD,
 395                                               &jx0,&jy0,&jz0);
 396
 397             /* Calculate displacement vector */
 398             dx00             = _mm_sub_ps(ix0,jx0);
 399             dy00             = _mm_sub_ps(iy0,jy0);
 400             dz00             = _mm_sub_ps(iz0,jz0);
 401             dx10             = _mm_sub_ps(ix1,jx0);
 402             dy10             = _mm_sub_ps(iy1,jy0);
 403             dz10             = _mm_sub_ps(iz1,jz0);
 404             dx20             = _mm_sub_ps(ix2,jx0);
 405             dy20             = _mm_sub_ps(iy2,jy0);
 406             dz20             = _mm_sub_ps(iz2,jz0);
 407
 408             /* Calculate squared distance and things based on it */
 409             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 410             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 411             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 412
 413             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 414             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 415             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 416
 417             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 418             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 419             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 420
 421             /* Load parameters for j particles */
 422             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 423                                                               charge+jnrC+0,charge+jnrD+0);
 424
 425             fjx0             = _mm_setzero_ps();
 426             fjy0             = _mm_setzero_ps();
 427             fjz0             = _mm_setzero_ps();
 428
 429             /**************************
 430              * CALCULATE INTERACTIONS *
 431              **************************/
 432
 433             r00              = _mm_mul_ps(rsq00,rinv00);
 434             r00              = _mm_andnot_ps(dummy_mask,r00);
 435
 436             /* Compute parameters for interactions between i and j atoms */
 437             qq00             = _mm_mul_ps(iq0,jq0);
 438
 439             /* EWALD ELECTROSTATICS */
 440
 441             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 442             ewrt             = _mm_mul_ps(r00,ewtabscale);
 443             ewitab           = _mm_cvttps_epi32(ewrt);
 444             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 445             ewitab           = _mm_slli_epi32(ewitab,2);
 446             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 447             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 448             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 449             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 450             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 451             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 452             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 453             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 454             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 455
 456             /* Update potential sum for this i atom from the interaction with this j atom. */
 457             velec            = _mm_andnot_ps(dummy_mask,velec);
 458             velecsum         = _mm_add_ps(velecsum,velec);
 459
 460             fscal            = felec;
 461
 462             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 463
 464             /* Calculate temporary vectorial force */
 465             tx               = _mm_mul_ps(fscal,dx00);
 466             ty               = _mm_mul_ps(fscal,dy00);
 467             tz               = _mm_mul_ps(fscal,dz00);
 468
 469             /* Update vectorial force */
 470             fix0             = _mm_add_ps(fix0,tx);
 471             fiy0             = _mm_add_ps(fiy0,ty);
 472             fiz0             = _mm_add_ps(fiz0,tz);
 473
 474             fjx0             = _mm_add_ps(fjx0,tx);
 475             fjy0             = _mm_add_ps(fjy0,ty);
 476             fjz0             = _mm_add_ps(fjz0,tz);
 477
 478             /**************************
 479              * CALCULATE INTERACTIONS *
 480              **************************/
 481
 482             r10              = _mm_mul_ps(rsq10,rinv10);
 483             r10              = _mm_andnot_ps(dummy_mask,r10);
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             qq10             = _mm_mul_ps(iq1,jq0);
 487
 488             /* EWALD ELECTROSTATICS */
 489
 490             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 491             ewrt             = _mm_mul_ps(r10,ewtabscale);
 492             ewitab           = _mm_cvttps_epi32(ewrt);
 493             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 494             ewitab           = _mm_slli_epi32(ewitab,2);
 495             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 496             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 497             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 498             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 499             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 500             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 501             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 502             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 503             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 504
 505             /* Update potential sum for this i atom from the interaction with this j atom. */
 506             velec            = _mm_andnot_ps(dummy_mask,velec);
 507             velecsum         = _mm_add_ps(velecsum,velec);
 508
 509             fscal            = felec;
 510
 511             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 512
 513             /* Calculate temporary vectorial force */
 514             tx               = _mm_mul_ps(fscal,dx10);
 515             ty               = _mm_mul_ps(fscal,dy10);
 516             tz               = _mm_mul_ps(fscal,dz10);
 517
 518             /* Update vectorial force */
 519             fix1             = _mm_add_ps(fix1,tx);
 520             fiy1             = _mm_add_ps(fiy1,ty);
 521             fiz1             = _mm_add_ps(fiz1,tz);
 522
 523             fjx0             = _mm_add_ps(fjx0,tx);
 524             fjy0             = _mm_add_ps(fjy0,ty);
 525             fjz0             = _mm_add_ps(fjz0,tz);
 526
 527             /**************************
 528              * CALCULATE INTERACTIONS *
 529              **************************/
 530
 531             r20              = _mm_mul_ps(rsq20,rinv20);
 532             r20              = _mm_andnot_ps(dummy_mask,r20);
 533
 534             /* Compute parameters for interactions between i and j atoms */
 535             qq20             = _mm_mul_ps(iq2,jq0);
 536
 537             /* EWALD ELECTROSTATICS */
 538
 539             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 540             ewrt             = _mm_mul_ps(r20,ewtabscale);
 541             ewitab           = _mm_cvttps_epi32(ewrt);
 542             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 543             ewitab           = _mm_slli_epi32(ewitab,2);
 544             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 545             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 546             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 547             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 548             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 549             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 550             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 551             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 552             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 553
 554             /* Update potential sum for this i atom from the interaction with this j atom. */
 555             velec            = _mm_andnot_ps(dummy_mask,velec);
 556             velecsum         = _mm_add_ps(velecsum,velec);
 557
 558             fscal            = felec;
 559
 560             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 561
 562             /* Calculate temporary vectorial force */
 563             tx               = _mm_mul_ps(fscal,dx20);
 564             ty               = _mm_mul_ps(fscal,dy20);
 565             tz               = _mm_mul_ps(fscal,dz20);
 566
 567             /* Update vectorial force */
 568             fix2             = _mm_add_ps(fix2,tx);
 569             fiy2             = _mm_add_ps(fiy2,ty);
 570             fiz2             = _mm_add_ps(fiz2,tz);
 571
 572             fjx0             = _mm_add_ps(fjx0,tx);
 573             fjy0             = _mm_add_ps(fjy0,ty);
 574             fjz0             = _mm_add_ps(fjz0,tz);
 575
 576             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 577             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 578             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 579             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 580
 581             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 582
 583             /* Inner loop uses 126 flops */
 584         }
 585
 586         /* End of innermost loop */
 587
 588         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 589                                               f+i_coord_offset,fshift+i_shift_offset);
 590
 591         ggid                        = gid[iidx];
 592         /* Update potential energies */
 593         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 594
 595         /* Increment number of inner iterations */
 596         inneriter                  += j_index_end - j_index_start;
 597
 598         /* Outer loop uses 19 flops */
 599     }
 600
 601     /* Increment number of outer iterations */
 602     outeriter        += nri;
 603
 604     /* Update outer/inner flops */
 605
 606     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_VF,outeriter*19 + inneriter*126);
 607 }
 608 /*
 609  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_single
 610  * Electrostatics interaction: Ewald
 611  * VdW interaction:            None
 612  * Geometry:                   Water3-Particle
 613  * Calculate force/pot:        Force
 614  */
 615 void
 616 nb_kernel_ElecEw_VdwNone_GeomW3P1_F_sse2_single
 617                     (t_nblist                    * gmx_restrict       nlist,
 618                      rvec                        * gmx_restrict          xx,
 619                      rvec                        * gmx_restrict          ff,
 620                      t_forcerec                  * gmx_restrict          fr,
 621                      t_mdatoms                   * gmx_restrict     mdatoms,
 622                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 623                      t_nrnb                      * gmx_restrict        nrnb)
 624 {
 625     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 626      * just 0 for non-waters.
 627      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 628      * jnr indices corresponding to data put in the four positions in the SIMD register.
 629      */
 630     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 631     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 632     int              jnrA,jnrB,jnrC,jnrD;
 633     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 634     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 635     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 636     real             rcutoff_scalar;
 637     real             *shiftvec,*fshift,*x,*f;
 638     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 639     real             scratch[4*DIM];
 640     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 641     int              vdwioffset0;
 642     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 643     int              vdwioffset1;
 644     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 645     int              vdwioffset2;
 646     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 647     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 648     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 649     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 650     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 651     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 652     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 653     real             *charge;
 654     __m128i          ewitab;
 655     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 656     real             *ewtab;
 657     __m128           dummy_mask,cutoff_mask;
 658     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 659     __m128           one     = _mm_set1_ps(1.0);
 660     __m128           two     = _mm_set1_ps(2.0);
 661     x                = xx[0];
 662     f                = ff[0];
 663
 664     nri              = nlist->nri;
 665     iinr             = nlist->iinr;
 666     jindex           = nlist->jindex;
 667     jjnr             = nlist->jjnr;
 668     shiftidx         = nlist->shift;
 669     gid              = nlist->gid;
 670     shiftvec         = fr->shift_vec[0];
 671     fshift           = fr->fshift[0];
 672     facel            = _mm_set1_ps(fr->epsfac);
 673     charge           = mdatoms->chargeA;
 674
 675     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 676     ewtab            = fr->ic->tabq_coul_F;
 677     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 678     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 679
 680     /* Setup water-specific parameters */
 681     inr              = nlist->iinr[0];
 682     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 683     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 684     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 685
 686     /* Avoid stupid compiler warnings */
 687     jnrA = jnrB = jnrC = jnrD = 0;
 688     j_coord_offsetA = 0;
 689     j_coord_offsetB = 0;
 690     j_coord_offsetC = 0;
 691     j_coord_offsetD = 0;
 692
 693     outeriter        = 0;
 694     inneriter        = 0;
 695
 696     for(iidx=0;iidx<4*DIM;iidx++)
 697     {
 698         scratch[iidx] = 0.0;
 699     }
 700
 701     /* Start outer loop over neighborlists */
 702     for(iidx=0; iidx<nri; iidx++)
 703     {
 704         /* Load shift vector for this list */
 705         i_shift_offset   = DIM*shiftidx[iidx];
 706
 707         /* Load limits for loop over neighbors */
 708         j_index_start    = jindex[iidx];
 709         j_index_end      = jindex[iidx+1];
 710
 711         /* Get outer coordinate index */
 712         inr              = iinr[iidx];
 713         i_coord_offset   = DIM*inr;
 714
 715         /* Load i particle coords and add shift vector */
 716         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 717                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 718
 719         fix0             = _mm_setzero_ps();
 720         fiy0             = _mm_setzero_ps();
 721         fiz0             = _mm_setzero_ps();
 722         fix1             = _mm_setzero_ps();
 723         fiy1             = _mm_setzero_ps();
 724         fiz1             = _mm_setzero_ps();
 725         fix2             = _mm_setzero_ps();
 726         fiy2             = _mm_setzero_ps();
 727         fiz2             = _mm_setzero_ps();
 728
 729         /* Start inner kernel loop */
 730         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 731         {
 732
 733             /* Get j neighbor index, and coordinate index */
 734             jnrA             = jjnr[jidx];
 735             jnrB             = jjnr[jidx+1];
 736             jnrC             = jjnr[jidx+2];
 737             jnrD             = jjnr[jidx+3];
 738             j_coord_offsetA  = DIM*jnrA;
 739             j_coord_offsetB  = DIM*jnrB;
 740             j_coord_offsetC  = DIM*jnrC;
 741             j_coord_offsetD  = DIM*jnrD;
 742
 743             /* load j atom coordinates */
 744             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 745                                               x+j_coord_offsetC,x+j_coord_offsetD,
 746                                               &jx0,&jy0,&jz0);
 747
 748             /* Calculate displacement vector */
 749             dx00             = _mm_sub_ps(ix0,jx0);
 750             dy00             = _mm_sub_ps(iy0,jy0);
 751             dz00             = _mm_sub_ps(iz0,jz0);
 752             dx10             = _mm_sub_ps(ix1,jx0);
 753             dy10             = _mm_sub_ps(iy1,jy0);
 754             dz10             = _mm_sub_ps(iz1,jz0);
 755             dx20             = _mm_sub_ps(ix2,jx0);
 756             dy20             = _mm_sub_ps(iy2,jy0);
 757             dz20             = _mm_sub_ps(iz2,jz0);
 758
 759             /* Calculate squared distance and things based on it */
 760             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 761             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 762             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 763
 764             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 765             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 766             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 767
 768             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 769             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 770             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 771
 772             /* Load parameters for j particles */
 773             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 774                                                               charge+jnrC+0,charge+jnrD+0);
 775
 776             fjx0             = _mm_setzero_ps();
 777             fjy0             = _mm_setzero_ps();
 778             fjz0             = _mm_setzero_ps();
 779
 780             /**************************
 781              * CALCULATE INTERACTIONS *
 782              **************************/
 783
 784             r00              = _mm_mul_ps(rsq00,rinv00);
 785
 786             /* Compute parameters for interactions between i and j atoms */
 787             qq00             = _mm_mul_ps(iq0,jq0);
 788
 789             /* EWALD ELECTROSTATICS */
 790
 791             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 792             ewrt             = _mm_mul_ps(r00,ewtabscale);
 793             ewitab           = _mm_cvttps_epi32(ewrt);
 794             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 795             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 796                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 797                                          &ewtabF,&ewtabFn);
 798             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 799             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 800
 801             fscal            = felec;
 802
 803             /* Calculate temporary vectorial force */
 804             tx               = _mm_mul_ps(fscal,dx00);
 805             ty               = _mm_mul_ps(fscal,dy00);
 806             tz               = _mm_mul_ps(fscal,dz00);
 807
 808             /* Update vectorial force */
 809             fix0             = _mm_add_ps(fix0,tx);
 810             fiy0             = _mm_add_ps(fiy0,ty);
 811             fiz0             = _mm_add_ps(fiz0,tz);
 812
 813             fjx0             = _mm_add_ps(fjx0,tx);
 814             fjy0             = _mm_add_ps(fjy0,ty);
 815             fjz0             = _mm_add_ps(fjz0,tz);
 816
 817             /**************************
 818              * CALCULATE INTERACTIONS *
 819              **************************/
 820
 821             r10              = _mm_mul_ps(rsq10,rinv10);
 822
 823             /* Compute parameters for interactions between i and j atoms */
 824             qq10             = _mm_mul_ps(iq1,jq0);
 825
 826             /* EWALD ELECTROSTATICS */
 827
 828             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 829             ewrt             = _mm_mul_ps(r10,ewtabscale);
 830             ewitab           = _mm_cvttps_epi32(ewrt);
 831             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 832             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 833                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 834                                          &ewtabF,&ewtabFn);
 835             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 836             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 837
 838             fscal            = felec;
 839
 840             /* Calculate temporary vectorial force */
 841             tx               = _mm_mul_ps(fscal,dx10);
 842             ty               = _mm_mul_ps(fscal,dy10);
 843             tz               = _mm_mul_ps(fscal,dz10);
 844
 845             /* Update vectorial force */
 846             fix1             = _mm_add_ps(fix1,tx);
 847             fiy1             = _mm_add_ps(fiy1,ty);
 848             fiz1             = _mm_add_ps(fiz1,tz);
 849
 850             fjx0             = _mm_add_ps(fjx0,tx);
 851             fjy0             = _mm_add_ps(fjy0,ty);
 852             fjz0             = _mm_add_ps(fjz0,tz);
 853
 854             /**************************
 855              * CALCULATE INTERACTIONS *
 856              **************************/
 857
 858             r20              = _mm_mul_ps(rsq20,rinv20);
 859
 860             /* Compute parameters for interactions between i and j atoms */
 861             qq20             = _mm_mul_ps(iq2,jq0);
 862
 863             /* EWALD ELECTROSTATICS */
 864
 865             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 866             ewrt             = _mm_mul_ps(r20,ewtabscale);
 867             ewitab           = _mm_cvttps_epi32(ewrt);
 868             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 869             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 870                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 871                                          &ewtabF,&ewtabFn);
 872             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 873             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 874
 875             fscal            = felec;
 876
 877             /* Calculate temporary vectorial force */
 878             tx               = _mm_mul_ps(fscal,dx20);
 879             ty               = _mm_mul_ps(fscal,dy20);
 880             tz               = _mm_mul_ps(fscal,dz20);
 881
 882             /* Update vectorial force */
 883             fix2             = _mm_add_ps(fix2,tx);
 884             fiy2             = _mm_add_ps(fiy2,ty);
 885             fiz2             = _mm_add_ps(fiz2,tz);
 886
 887             fjx0             = _mm_add_ps(fjx0,tx);
 888             fjy0             = _mm_add_ps(fjy0,ty);
 889             fjz0             = _mm_add_ps(fjz0,tz);
 890
 891             fjptrA             = f+j_coord_offsetA;
 892             fjptrB             = f+j_coord_offsetB;
 893             fjptrC             = f+j_coord_offsetC;
 894             fjptrD             = f+j_coord_offsetD;
 895
 896             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 897
 898             /* Inner loop uses 108 flops */
 899         }
 900
 901         if(jidx<j_index_end)
 902         {
 903
 904             /* Get j neighbor index, and coordinate index */
 905             jnrlistA         = jjnr[jidx];
 906             jnrlistB         = jjnr[jidx+1];
 907             jnrlistC         = jjnr[jidx+2];
 908             jnrlistD         = jjnr[jidx+3];
 909             /* Sign of each element will be negative for non-real atoms.
 910              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 911              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 912              */
 913             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 914             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 915             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 916             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 917             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 918             j_coord_offsetA  = DIM*jnrA;
 919             j_coord_offsetB  = DIM*jnrB;
 920             j_coord_offsetC  = DIM*jnrC;
 921             j_coord_offsetD  = DIM*jnrD;
 922
 923             /* load j atom coordinates */
 924             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 925                                               x+j_coord_offsetC,x+j_coord_offsetD,
 926                                               &jx0,&jy0,&jz0);
 927
 928             /* Calculate displacement vector */
 929             dx00             = _mm_sub_ps(ix0,jx0);
 930             dy00             = _mm_sub_ps(iy0,jy0);
 931             dz00             = _mm_sub_ps(iz0,jz0);
 932             dx10             = _mm_sub_ps(ix1,jx0);
 933             dy10             = _mm_sub_ps(iy1,jy0);
 934             dz10             = _mm_sub_ps(iz1,jz0);
 935             dx20             = _mm_sub_ps(ix2,jx0);
 936             dy20             = _mm_sub_ps(iy2,jy0);
 937             dz20             = _mm_sub_ps(iz2,jz0);
 938
 939             /* Calculate squared distance and things based on it */
 940             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 941             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 942             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 943
 944             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 945             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 946             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 947
 948             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 949             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 950             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 951
 952             /* Load parameters for j particles */
 953             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 954                                                               charge+jnrC+0,charge+jnrD+0);
 955
 956             fjx0             = _mm_setzero_ps();
 957             fjy0             = _mm_setzero_ps();
 958             fjz0             = _mm_setzero_ps();
 959
 960             /**************************
 961              * CALCULATE INTERACTIONS *
 962              **************************/
 963
 964             r00              = _mm_mul_ps(rsq00,rinv00);
 965             r00              = _mm_andnot_ps(dummy_mask,r00);
 966
 967             /* Compute parameters for interactions between i and j atoms */
 968             qq00             = _mm_mul_ps(iq0,jq0);
 969
 970             /* EWALD ELECTROSTATICS */
 971
 972             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 973             ewrt             = _mm_mul_ps(r00,ewtabscale);
 974             ewitab           = _mm_cvttps_epi32(ewrt);
 975             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 976             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 977                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 978                                          &ewtabF,&ewtabFn);
 979             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 980             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 981
 982             fscal            = felec;
 983
 984             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm_mul_ps(fscal,dx00);
 988             ty               = _mm_mul_ps(fscal,dy00);
 989             tz               = _mm_mul_ps(fscal,dz00);
 990
 991             /* Update vectorial force */
 992             fix0             = _mm_add_ps(fix0,tx);
 993             fiy0             = _mm_add_ps(fiy0,ty);
 994             fiz0             = _mm_add_ps(fiz0,tz);
 995
 996             fjx0             = _mm_add_ps(fjx0,tx);
 997             fjy0             = _mm_add_ps(fjy0,ty);
 998             fjz0             = _mm_add_ps(fjz0,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r10              = _mm_mul_ps(rsq10,rinv10);
1005             r10              = _mm_andnot_ps(dummy_mask,r10);
1006
1007             /* Compute parameters for interactions between i and j atoms */
1008             qq10             = _mm_mul_ps(iq1,jq0);
1009
1010             /* EWALD ELECTROSTATICS */
1011
1012             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1013             ewrt             = _mm_mul_ps(r10,ewtabscale);
1014             ewitab           = _mm_cvttps_epi32(ewrt);
1015             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1016             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1017                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1018                                          &ewtabF,&ewtabFn);
1019             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1020             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1021
1022             fscal            = felec;
1023
1024             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1025
1026             /* Calculate temporary vectorial force */
1027             tx               = _mm_mul_ps(fscal,dx10);
1028             ty               = _mm_mul_ps(fscal,dy10);
1029             tz               = _mm_mul_ps(fscal,dz10);
1030
1031             /* Update vectorial force */
1032             fix1             = _mm_add_ps(fix1,tx);
1033             fiy1             = _mm_add_ps(fiy1,ty);
1034             fiz1             = _mm_add_ps(fiz1,tz);
1035
1036             fjx0             = _mm_add_ps(fjx0,tx);
1037             fjy0             = _mm_add_ps(fjy0,ty);
1038             fjz0             = _mm_add_ps(fjz0,tz);
1039
1040             /**************************
1041              * CALCULATE INTERACTIONS *
1042              **************************/
1043
1044             r20              = _mm_mul_ps(rsq20,rinv20);
1045             r20              = _mm_andnot_ps(dummy_mask,r20);
1046
1047             /* Compute parameters for interactions between i and j atoms */
1048             qq20             = _mm_mul_ps(iq2,jq0);
1049
1050             /* EWALD ELECTROSTATICS */
1051
1052             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1053             ewrt             = _mm_mul_ps(r20,ewtabscale);
1054             ewitab           = _mm_cvttps_epi32(ewrt);
1055             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1056             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1057                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1058                                          &ewtabF,&ewtabFn);
1059             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1060             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1061
1062             fscal            = felec;
1063
1064             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1065
1066             /* Calculate temporary vectorial force */
1067             tx               = _mm_mul_ps(fscal,dx20);
1068             ty               = _mm_mul_ps(fscal,dy20);
1069             tz               = _mm_mul_ps(fscal,dz20);
1070
1071             /* Update vectorial force */
1072             fix2             = _mm_add_ps(fix2,tx);
1073             fiy2             = _mm_add_ps(fiy2,ty);
1074             fiz2             = _mm_add_ps(fiz2,tz);
1075
1076             fjx0             = _mm_add_ps(fjx0,tx);
1077             fjy0             = _mm_add_ps(fjy0,ty);
1078             fjz0             = _mm_add_ps(fjz0,tz);
1079
1080             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1081             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1082             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1083             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1084
1085             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1086
1087             /* Inner loop uses 111 flops */
1088         }
1089
1090         /* End of innermost loop */
1091
1092         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1093                                               f+i_coord_offset,fshift+i_shift_offset);
1094
1095         /* Increment number of inner iterations */
1096         inneriter                  += j_index_end - j_index_start;
1097
1098         /* Outer loop uses 18 flops */
1099     }
1100
1101     /* Increment number of outer iterations */
1102     outeriter        += nri;
1103
1104     /* Update outer/inner flops */
1105
1106     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W3_F,outeriter*18 + inneriter*111);
1107 }