src/gromacs/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwNone_GeomW4P1_sse2_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse2_single kernel generator.
   3  *
   4  *                This source code is part of
   5  *
   6  *                 G   R   O   M   A   C   S
   7  *
   8  * Copyright (c) 2001-2012, The GROMACS Development Team
   9  *
  10  * Gromacs is a library for molecular simulation and trajectory analysis,
  11  * written by Erik Lindahl, David van der Spoel, Berk Hess, and others - for
  12  * a full list of developers and information, check out http://www.gromacs.org
  13  *
  14  * This program is free software; you can redistribute it and/or modify it under
  15  * the terms of the GNU Lesser General Public License as published by the Free
  16  * Software Foundation; either version 2 of the License, or (at your option) any
  17  * later version.
  18  *
  19  * To help fund GROMACS development, we humbly ask that you cite
  20  * the papers people have written on it - you can find them on the website.
  21  */
  22 #ifdef HAVE_CONFIG_H
  23 #include <config.h>
  24 #endif
  25
  26 #include <math.h>
  27
  28 #include "../nb_kernel.h"
  29 #include "types/simple.h"
  30 #include "vec.h"
  31 #include "nrnb.h"
  32
  33 #include "gmx_math_x86_sse2_single.h"
  34 #include "kernelutil_x86_sse2_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            None
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwNone_GeomW4P1_VF_sse2_single
  45                     (t_nblist * gmx_restrict                nlist,
  46                      rvec * gmx_restrict                    xx,
  47                      rvec * gmx_restrict                    ff,
  48                      t_forcerec * gmx_restrict              fr,
  49                      t_mdatoms * gmx_restrict               mdatoms,
  50                      nb_kernel_data_t * gmx_restrict        kernel_data,
  51                      t_nrnb * gmx_restrict                  nrnb)
  52 {
  53     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  54      * just 0 for non-waters.
  55      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  56      * jnr indices corresponding to data put in the four positions in the SIMD register.
  57      */
  58     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  59     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  60     int              jnrA,jnrB,jnrC,jnrD;
  61     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  62     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  63     real             shX,shY,shZ,rcutoff_scalar;
  64     real             *shiftvec,*fshift,*x,*f;
  65     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  66     int              vdwioffset1;
  67     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  68     int              vdwioffset2;
  69     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  70     int              vdwioffset3;
  71     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  75     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  76     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     __m128i          ewitab;
  80     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  81     real             *ewtab;
  82     __m128           dummy_mask,cutoff_mask;
  83     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  84     __m128           one     = _mm_set1_ps(1.0);
  85     __m128           two     = _mm_set1_ps(2.0);
  86     x                = xx[0];
  87     f                = ff[0];
  88
  89     nri              = nlist->nri;
  90     iinr             = nlist->iinr;
  91     jindex           = nlist->jindex;
  92     jjnr             = nlist->jjnr;
  93     shiftidx         = nlist->shift;
  94     gid              = nlist->gid;
  95     shiftvec         = fr->shift_vec[0];
  96     fshift           = fr->fshift[0];
  97     facel            = _mm_set1_ps(fr->epsfac);
  98     charge           = mdatoms->chargeA;
  99
 100     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 101     ewtab            = fr->ic->tabq_coul_FDV0;
 102     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 103     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 104
 105     /* Setup water-specific parameters */
 106     inr              = nlist->iinr[0];
 107     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 108     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 109     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 110
 111     /* Avoid stupid compiler warnings */
 112     jnrA = jnrB = jnrC = jnrD = 0;
 113     j_coord_offsetA = 0;
 114     j_coord_offsetB = 0;
 115     j_coord_offsetC = 0;
 116     j_coord_offsetD = 0;
 117
 118     outeriter        = 0;
 119     inneriter        = 0;
 120
 121     /* Start outer loop over neighborlists */
 122     for(iidx=0; iidx<nri; iidx++)
 123     {
 124         /* Load shift vector for this list */
 125         i_shift_offset   = DIM*shiftidx[iidx];
 126         shX              = shiftvec[i_shift_offset+XX];
 127         shY              = shiftvec[i_shift_offset+YY];
 128         shZ              = shiftvec[i_shift_offset+ZZ];
 129
 130         /* Load limits for loop over neighbors */
 131         j_index_start    = jindex[iidx];
 132         j_index_end      = jindex[iidx+1];
 133
 134         /* Get outer coordinate index */
 135         inr              = iinr[iidx];
 136         i_coord_offset   = DIM*inr;
 137
 138         /* Load i particle coords and add shift vector */
 139         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 140         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 141         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 142         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 143         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 144         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 145         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 146         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 147         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 148
 149         fix1             = _mm_setzero_ps();
 150         fiy1             = _mm_setzero_ps();
 151         fiz1             = _mm_setzero_ps();
 152         fix2             = _mm_setzero_ps();
 153         fiy2             = _mm_setzero_ps();
 154         fiz2             = _mm_setzero_ps();
 155         fix3             = _mm_setzero_ps();
 156         fiy3             = _mm_setzero_ps();
 157         fiz3             = _mm_setzero_ps();
 158
 159         /* Reset potential sums */
 160         velecsum         = _mm_setzero_ps();
 161
 162         /* Start inner kernel loop */
 163         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 164         {
 165
 166             /* Get j neighbor index, and coordinate index */
 167             jnrA             = jjnr[jidx];
 168             jnrB             = jjnr[jidx+1];
 169             jnrC             = jjnr[jidx+2];
 170             jnrD             = jjnr[jidx+3];
 171
 172             j_coord_offsetA  = DIM*jnrA;
 173             j_coord_offsetB  = DIM*jnrB;
 174             j_coord_offsetC  = DIM*jnrC;
 175             j_coord_offsetD  = DIM*jnrD;
 176
 177             /* load j atom coordinates */
 178             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 179                                               x+j_coord_offsetC,x+j_coord_offsetD,
 180                                               &jx0,&jy0,&jz0);
 181
 182             /* Calculate displacement vector */
 183             dx10             = _mm_sub_ps(ix1,jx0);
 184             dy10             = _mm_sub_ps(iy1,jy0);
 185             dz10             = _mm_sub_ps(iz1,jz0);
 186             dx20             = _mm_sub_ps(ix2,jx0);
 187             dy20             = _mm_sub_ps(iy2,jy0);
 188             dz20             = _mm_sub_ps(iz2,jz0);
 189             dx30             = _mm_sub_ps(ix3,jx0);
 190             dy30             = _mm_sub_ps(iy3,jy0);
 191             dz30             = _mm_sub_ps(iz3,jz0);
 192
 193             /* Calculate squared distance and things based on it */
 194             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 195             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 196             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 197
 198             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 199             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 200             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 201
 202             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 203             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 204             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 205
 206             /* Load parameters for j particles */
 207             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 208                                                               charge+jnrC+0,charge+jnrD+0);
 209
 210             /**************************
 211              * CALCULATE INTERACTIONS *
 212              **************************/
 213
 214             r10              = _mm_mul_ps(rsq10,rinv10);
 215
 216             /* Compute parameters for interactions between i and j atoms */
 217             qq10             = _mm_mul_ps(iq1,jq0);
 218
 219             /* EWALD ELECTROSTATICS */
 220
 221             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 222             ewrt             = _mm_mul_ps(r10,ewtabscale);
 223             ewitab           = _mm_cvttps_epi32(ewrt);
 224             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 225             ewitab           = _mm_slli_epi32(ewitab,2);
 226             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 227             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 228             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 229             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 230             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 231             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 232             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 233             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 234             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 235
 236             /* Update potential sum for this i atom from the interaction with this j atom. */
 237             velecsum         = _mm_add_ps(velecsum,velec);
 238
 239             fscal            = felec;
 240
 241             /* Calculate temporary vectorial force */
 242             tx               = _mm_mul_ps(fscal,dx10);
 243             ty               = _mm_mul_ps(fscal,dy10);
 244             tz               = _mm_mul_ps(fscal,dz10);
 245
 246             /* Update vectorial force */
 247             fix1             = _mm_add_ps(fix1,tx);
 248             fiy1             = _mm_add_ps(fiy1,ty);
 249             fiz1             = _mm_add_ps(fiz1,tz);
 250
 251             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 252                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 253                                                    tx,ty,tz);
 254
 255             /**************************
 256              * CALCULATE INTERACTIONS *
 257              **************************/
 258
 259             r20              = _mm_mul_ps(rsq20,rinv20);
 260
 261             /* Compute parameters for interactions between i and j atoms */
 262             qq20             = _mm_mul_ps(iq2,jq0);
 263
 264             /* EWALD ELECTROSTATICS */
 265
 266             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 267             ewrt             = _mm_mul_ps(r20,ewtabscale);
 268             ewitab           = _mm_cvttps_epi32(ewrt);
 269             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 270             ewitab           = _mm_slli_epi32(ewitab,2);
 271             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 272             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 273             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 274             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 275             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 276             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 277             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 278             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 279             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 280
 281             /* Update potential sum for this i atom from the interaction with this j atom. */
 282             velecsum         = _mm_add_ps(velecsum,velec);
 283
 284             fscal            = felec;
 285
 286             /* Calculate temporary vectorial force */
 287             tx               = _mm_mul_ps(fscal,dx20);
 288             ty               = _mm_mul_ps(fscal,dy20);
 289             tz               = _mm_mul_ps(fscal,dz20);
 290
 291             /* Update vectorial force */
 292             fix2             = _mm_add_ps(fix2,tx);
 293             fiy2             = _mm_add_ps(fiy2,ty);
 294             fiz2             = _mm_add_ps(fiz2,tz);
 295
 296             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 297                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 298                                                    tx,ty,tz);
 299
 300             /**************************
 301              * CALCULATE INTERACTIONS *
 302              **************************/
 303
 304             r30              = _mm_mul_ps(rsq30,rinv30);
 305
 306             /* Compute parameters for interactions between i and j atoms */
 307             qq30             = _mm_mul_ps(iq3,jq0);
 308
 309             /* EWALD ELECTROSTATICS */
 310
 311             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 312             ewrt             = _mm_mul_ps(r30,ewtabscale);
 313             ewitab           = _mm_cvttps_epi32(ewrt);
 314             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 315             ewitab           = _mm_slli_epi32(ewitab,2);
 316             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 317             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 318             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 319             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 320             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 321             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 322             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 323             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 324             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 325
 326             /* Update potential sum for this i atom from the interaction with this j atom. */
 327             velecsum         = _mm_add_ps(velecsum,velec);
 328
 329             fscal            = felec;
 330
 331             /* Calculate temporary vectorial force */
 332             tx               = _mm_mul_ps(fscal,dx30);
 333             ty               = _mm_mul_ps(fscal,dy30);
 334             tz               = _mm_mul_ps(fscal,dz30);
 335
 336             /* Update vectorial force */
 337             fix3             = _mm_add_ps(fix3,tx);
 338             fiy3             = _mm_add_ps(fiy3,ty);
 339             fiz3             = _mm_add_ps(fiz3,tz);
 340
 341             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 342                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 343                                                    tx,ty,tz);
 344
 345             /* Inner loop uses 123 flops */
 346         }
 347
 348         if(jidx<j_index_end)
 349         {
 350
 351             /* Get j neighbor index, and coordinate index */
 352             jnrA             = jjnr[jidx];
 353             jnrB             = jjnr[jidx+1];
 354             jnrC             = jjnr[jidx+2];
 355             jnrD             = jjnr[jidx+3];
 356
 357             /* Sign of each element will be negative for non-real atoms.
 358              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 359              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 360              */
 361             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 362             jnrA       = (jnrA>=0) ? jnrA : 0;
 363             jnrB       = (jnrB>=0) ? jnrB : 0;
 364             jnrC       = (jnrC>=0) ? jnrC : 0;
 365             jnrD       = (jnrD>=0) ? jnrD : 0;
 366
 367             j_coord_offsetA  = DIM*jnrA;
 368             j_coord_offsetB  = DIM*jnrB;
 369             j_coord_offsetC  = DIM*jnrC;
 370             j_coord_offsetD  = DIM*jnrD;
 371
 372             /* load j atom coordinates */
 373             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 374                                               x+j_coord_offsetC,x+j_coord_offsetD,
 375                                               &jx0,&jy0,&jz0);
 376
 377             /* Calculate displacement vector */
 378             dx10             = _mm_sub_ps(ix1,jx0);
 379             dy10             = _mm_sub_ps(iy1,jy0);
 380             dz10             = _mm_sub_ps(iz1,jz0);
 381             dx20             = _mm_sub_ps(ix2,jx0);
 382             dy20             = _mm_sub_ps(iy2,jy0);
 383             dz20             = _mm_sub_ps(iz2,jz0);
 384             dx30             = _mm_sub_ps(ix3,jx0);
 385             dy30             = _mm_sub_ps(iy3,jy0);
 386             dz30             = _mm_sub_ps(iz3,jz0);
 387
 388             /* Calculate squared distance and things based on it */
 389             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 390             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 391             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 392
 393             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 394             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 395             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 396
 397             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 398             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 399             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 400
 401             /* Load parameters for j particles */
 402             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 403                                                               charge+jnrC+0,charge+jnrD+0);
 404
 405             /**************************
 406              * CALCULATE INTERACTIONS *
 407              **************************/
 408
 409             r10              = _mm_mul_ps(rsq10,rinv10);
 410             r10              = _mm_andnot_ps(dummy_mask,r10);
 411
 412             /* Compute parameters for interactions between i and j atoms */
 413             qq10             = _mm_mul_ps(iq1,jq0);
 414
 415             /* EWALD ELECTROSTATICS */
 416
 417             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 418             ewrt             = _mm_mul_ps(r10,ewtabscale);
 419             ewitab           = _mm_cvttps_epi32(ewrt);
 420             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 421             ewitab           = _mm_slli_epi32(ewitab,2);
 422             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 423             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 424             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 425             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 426             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 427             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 428             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 429             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 430             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 431
 432             /* Update potential sum for this i atom from the interaction with this j atom. */
 433             velec            = _mm_andnot_ps(dummy_mask,velec);
 434             velecsum         = _mm_add_ps(velecsum,velec);
 435
 436             fscal            = felec;
 437
 438             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 439
 440             /* Calculate temporary vectorial force */
 441             tx               = _mm_mul_ps(fscal,dx10);
 442             ty               = _mm_mul_ps(fscal,dy10);
 443             tz               = _mm_mul_ps(fscal,dz10);
 444
 445             /* Update vectorial force */
 446             fix1             = _mm_add_ps(fix1,tx);
 447             fiy1             = _mm_add_ps(fiy1,ty);
 448             fiz1             = _mm_add_ps(fiz1,tz);
 449
 450             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 451                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 452                                                    tx,ty,tz);
 453
 454             /**************************
 455              * CALCULATE INTERACTIONS *
 456              **************************/
 457
 458             r20              = _mm_mul_ps(rsq20,rinv20);
 459             r20              = _mm_andnot_ps(dummy_mask,r20);
 460
 461             /* Compute parameters for interactions between i and j atoms */
 462             qq20             = _mm_mul_ps(iq2,jq0);
 463
 464             /* EWALD ELECTROSTATICS */
 465
 466             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 467             ewrt             = _mm_mul_ps(r20,ewtabscale);
 468             ewitab           = _mm_cvttps_epi32(ewrt);
 469             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 470             ewitab           = _mm_slli_epi32(ewitab,2);
 471             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 472             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 473             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 474             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 475             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 476             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 477             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 478             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 479             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 480
 481             /* Update potential sum for this i atom from the interaction with this j atom. */
 482             velec            = _mm_andnot_ps(dummy_mask,velec);
 483             velecsum         = _mm_add_ps(velecsum,velec);
 484
 485             fscal            = felec;
 486
 487             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm_mul_ps(fscal,dx20);
 491             ty               = _mm_mul_ps(fscal,dy20);
 492             tz               = _mm_mul_ps(fscal,dz20);
 493
 494             /* Update vectorial force */
 495             fix2             = _mm_add_ps(fix2,tx);
 496             fiy2             = _mm_add_ps(fiy2,ty);
 497             fiz2             = _mm_add_ps(fiz2,tz);
 498
 499             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 500                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 501                                                    tx,ty,tz);
 502
 503             /**************************
 504              * CALCULATE INTERACTIONS *
 505              **************************/
 506
 507             r30              = _mm_mul_ps(rsq30,rinv30);
 508             r30              = _mm_andnot_ps(dummy_mask,r30);
 509
 510             /* Compute parameters for interactions between i and j atoms */
 511             qq30             = _mm_mul_ps(iq3,jq0);
 512
 513             /* EWALD ELECTROSTATICS */
 514
 515             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 516             ewrt             = _mm_mul_ps(r30,ewtabscale);
 517             ewitab           = _mm_cvttps_epi32(ewrt);
 518             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 519             ewitab           = _mm_slli_epi32(ewitab,2);
 520             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 521             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 522             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 523             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 524             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 525             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 526             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 527             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 528             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 529
 530             /* Update potential sum for this i atom from the interaction with this j atom. */
 531             velec            = _mm_andnot_ps(dummy_mask,velec);
 532             velecsum         = _mm_add_ps(velecsum,velec);
 533
 534             fscal            = felec;
 535
 536             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 537
 538             /* Calculate temporary vectorial force */
 539             tx               = _mm_mul_ps(fscal,dx30);
 540             ty               = _mm_mul_ps(fscal,dy30);
 541             tz               = _mm_mul_ps(fscal,dz30);
 542
 543             /* Update vectorial force */
 544             fix3             = _mm_add_ps(fix3,tx);
 545             fiy3             = _mm_add_ps(fiy3,ty);
 546             fiz3             = _mm_add_ps(fiz3,tz);
 547
 548             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 549                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 550                                                    tx,ty,tz);
 551
 552             /* Inner loop uses 126 flops */
 553         }
 554
 555         /* End of innermost loop */
 556
 557         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 558                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
 559
 560         ggid                        = gid[iidx];
 561         /* Update potential energies */
 562         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 563
 564         /* Increment number of inner iterations */
 565         inneriter                  += j_index_end - j_index_start;
 566
 567         /* Outer loop uses 28 flops */
 568     }
 569
 570     /* Increment number of outer iterations */
 571     outeriter        += nri;
 572
 573     /* Update outer/inner flops */
 574
 575     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_VF,outeriter*28 + inneriter*126);
 576 }
 577 /*
 578  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
 579  * Electrostatics interaction: Ewald
 580  * VdW interaction:            None
 581  * Geometry:                   Water4-Particle
 582  * Calculate force/pot:        Force
 583  */
 584 void
 585 nb_kernel_ElecEw_VdwNone_GeomW4P1_F_sse2_single
 586                     (t_nblist * gmx_restrict                nlist,
 587                      rvec * gmx_restrict                    xx,
 588                      rvec * gmx_restrict                    ff,
 589                      t_forcerec * gmx_restrict              fr,
 590                      t_mdatoms * gmx_restrict               mdatoms,
 591                      nb_kernel_data_t * gmx_restrict        kernel_data,
 592                      t_nrnb * gmx_restrict                  nrnb)
 593 {
 594     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 595      * just 0 for non-waters.
 596      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 597      * jnr indices corresponding to data put in the four positions in the SIMD register.
 598      */
 599     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 600     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 601     int              jnrA,jnrB,jnrC,jnrD;
 602     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 603     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 604     real             shX,shY,shZ,rcutoff_scalar;
 605     real             *shiftvec,*fshift,*x,*f;
 606     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 607     int              vdwioffset1;
 608     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 609     int              vdwioffset2;
 610     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 611     int              vdwioffset3;
 612     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 613     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 614     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 615     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 616     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 617     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 618     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 619     real             *charge;
 620     __m128i          ewitab;
 621     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 622     real             *ewtab;
 623     __m128           dummy_mask,cutoff_mask;
 624     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 625     __m128           one     = _mm_set1_ps(1.0);
 626     __m128           two     = _mm_set1_ps(2.0);
 627     x                = xx[0];
 628     f                = ff[0];
 629
 630     nri              = nlist->nri;
 631     iinr             = nlist->iinr;
 632     jindex           = nlist->jindex;
 633     jjnr             = nlist->jjnr;
 634     shiftidx         = nlist->shift;
 635     gid              = nlist->gid;
 636     shiftvec         = fr->shift_vec[0];
 637     fshift           = fr->fshift[0];
 638     facel            = _mm_set1_ps(fr->epsfac);
 639     charge           = mdatoms->chargeA;
 640
 641     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 642     ewtab            = fr->ic->tabq_coul_F;
 643     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 644     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 645
 646     /* Setup water-specific parameters */
 647     inr              = nlist->iinr[0];
 648     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 649     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 650     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 651
 652     /* Avoid stupid compiler warnings */
 653     jnrA = jnrB = jnrC = jnrD = 0;
 654     j_coord_offsetA = 0;
 655     j_coord_offsetB = 0;
 656     j_coord_offsetC = 0;
 657     j_coord_offsetD = 0;
 658
 659     outeriter        = 0;
 660     inneriter        = 0;
 661
 662     /* Start outer loop over neighborlists */
 663     for(iidx=0; iidx<nri; iidx++)
 664     {
 665         /* Load shift vector for this list */
 666         i_shift_offset   = DIM*shiftidx[iidx];
 667         shX              = shiftvec[i_shift_offset+XX];
 668         shY              = shiftvec[i_shift_offset+YY];
 669         shZ              = shiftvec[i_shift_offset+ZZ];
 670
 671         /* Load limits for loop over neighbors */
 672         j_index_start    = jindex[iidx];
 673         j_index_end      = jindex[iidx+1];
 674
 675         /* Get outer coordinate index */
 676         inr              = iinr[iidx];
 677         i_coord_offset   = DIM*inr;
 678
 679         /* Load i particle coords and add shift vector */
 680         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 681         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 682         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 683         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 684         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 685         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 686         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 687         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 688         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 689
 690         fix1             = _mm_setzero_ps();
 691         fiy1             = _mm_setzero_ps();
 692         fiz1             = _mm_setzero_ps();
 693         fix2             = _mm_setzero_ps();
 694         fiy2             = _mm_setzero_ps();
 695         fiz2             = _mm_setzero_ps();
 696         fix3             = _mm_setzero_ps();
 697         fiy3             = _mm_setzero_ps();
 698         fiz3             = _mm_setzero_ps();
 699
 700         /* Start inner kernel loop */
 701         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 702         {
 703
 704             /* Get j neighbor index, and coordinate index */
 705             jnrA             = jjnr[jidx];
 706             jnrB             = jjnr[jidx+1];
 707             jnrC             = jjnr[jidx+2];
 708             jnrD             = jjnr[jidx+3];
 709
 710             j_coord_offsetA  = DIM*jnrA;
 711             j_coord_offsetB  = DIM*jnrB;
 712             j_coord_offsetC  = DIM*jnrC;
 713             j_coord_offsetD  = DIM*jnrD;
 714
 715             /* load j atom coordinates */
 716             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 717                                               x+j_coord_offsetC,x+j_coord_offsetD,
 718                                               &jx0,&jy0,&jz0);
 719
 720             /* Calculate displacement vector */
 721             dx10             = _mm_sub_ps(ix1,jx0);
 722             dy10             = _mm_sub_ps(iy1,jy0);
 723             dz10             = _mm_sub_ps(iz1,jz0);
 724             dx20             = _mm_sub_ps(ix2,jx0);
 725             dy20             = _mm_sub_ps(iy2,jy0);
 726             dz20             = _mm_sub_ps(iz2,jz0);
 727             dx30             = _mm_sub_ps(ix3,jx0);
 728             dy30             = _mm_sub_ps(iy3,jy0);
 729             dz30             = _mm_sub_ps(iz3,jz0);
 730
 731             /* Calculate squared distance and things based on it */
 732             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 733             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 734             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 735
 736             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 737             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 738             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 739
 740             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 741             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 742             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 743
 744             /* Load parameters for j particles */
 745             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 746                                                               charge+jnrC+0,charge+jnrD+0);
 747
 748             /**************************
 749              * CALCULATE INTERACTIONS *
 750              **************************/
 751
 752             r10              = _mm_mul_ps(rsq10,rinv10);
 753
 754             /* Compute parameters for interactions between i and j atoms */
 755             qq10             = _mm_mul_ps(iq1,jq0);
 756
 757             /* EWALD ELECTROSTATICS */
 758
 759             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 760             ewrt             = _mm_mul_ps(r10,ewtabscale);
 761             ewitab           = _mm_cvttps_epi32(ewrt);
 762             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 763             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 764                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 765                                          &ewtabF,&ewtabFn);
 766             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 767             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 768
 769             fscal            = felec;
 770
 771             /* Calculate temporary vectorial force */
 772             tx               = _mm_mul_ps(fscal,dx10);
 773             ty               = _mm_mul_ps(fscal,dy10);
 774             tz               = _mm_mul_ps(fscal,dz10);
 775
 776             /* Update vectorial force */
 777             fix1             = _mm_add_ps(fix1,tx);
 778             fiy1             = _mm_add_ps(fiy1,ty);
 779             fiz1             = _mm_add_ps(fiz1,tz);
 780
 781             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 782                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 783                                                    tx,ty,tz);
 784
 785             /**************************
 786              * CALCULATE INTERACTIONS *
 787              **************************/
 788
 789             r20              = _mm_mul_ps(rsq20,rinv20);
 790
 791             /* Compute parameters for interactions between i and j atoms */
 792             qq20             = _mm_mul_ps(iq2,jq0);
 793
 794             /* EWALD ELECTROSTATICS */
 795
 796             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 797             ewrt             = _mm_mul_ps(r20,ewtabscale);
 798             ewitab           = _mm_cvttps_epi32(ewrt);
 799             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 800             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 801                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 802                                          &ewtabF,&ewtabFn);
 803             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 804             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 805
 806             fscal            = felec;
 807
 808             /* Calculate temporary vectorial force */
 809             tx               = _mm_mul_ps(fscal,dx20);
 810             ty               = _mm_mul_ps(fscal,dy20);
 811             tz               = _mm_mul_ps(fscal,dz20);
 812
 813             /* Update vectorial force */
 814             fix2             = _mm_add_ps(fix2,tx);
 815             fiy2             = _mm_add_ps(fiy2,ty);
 816             fiz2             = _mm_add_ps(fiz2,tz);
 817
 818             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 819                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 820                                                    tx,ty,tz);
 821
 822             /**************************
 823              * CALCULATE INTERACTIONS *
 824              **************************/
 825
 826             r30              = _mm_mul_ps(rsq30,rinv30);
 827
 828             /* Compute parameters for interactions between i and j atoms */
 829             qq30             = _mm_mul_ps(iq3,jq0);
 830
 831             /* EWALD ELECTROSTATICS */
 832
 833             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 834             ewrt             = _mm_mul_ps(r30,ewtabscale);
 835             ewitab           = _mm_cvttps_epi32(ewrt);
 836             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 837             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 838                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 839                                          &ewtabF,&ewtabFn);
 840             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 841             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 842
 843             fscal            = felec;
 844
 845             /* Calculate temporary vectorial force */
 846             tx               = _mm_mul_ps(fscal,dx30);
 847             ty               = _mm_mul_ps(fscal,dy30);
 848             tz               = _mm_mul_ps(fscal,dz30);
 849
 850             /* Update vectorial force */
 851             fix3             = _mm_add_ps(fix3,tx);
 852             fiy3             = _mm_add_ps(fiy3,ty);
 853             fiz3             = _mm_add_ps(fiz3,tz);
 854
 855             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 856                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 857                                                    tx,ty,tz);
 858
 859             /* Inner loop uses 108 flops */
 860         }
 861
 862         if(jidx<j_index_end)
 863         {
 864
 865             /* Get j neighbor index, and coordinate index */
 866             jnrA             = jjnr[jidx];
 867             jnrB             = jjnr[jidx+1];
 868             jnrC             = jjnr[jidx+2];
 869             jnrD             = jjnr[jidx+3];
 870
 871             /* Sign of each element will be negative for non-real atoms.
 872              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 873              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 874              */
 875             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 876             jnrA       = (jnrA>=0) ? jnrA : 0;
 877             jnrB       = (jnrB>=0) ? jnrB : 0;
 878             jnrC       = (jnrC>=0) ? jnrC : 0;
 879             jnrD       = (jnrD>=0) ? jnrD : 0;
 880
 881             j_coord_offsetA  = DIM*jnrA;
 882             j_coord_offsetB  = DIM*jnrB;
 883             j_coord_offsetC  = DIM*jnrC;
 884             j_coord_offsetD  = DIM*jnrD;
 885
 886             /* load j atom coordinates */
 887             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 888                                               x+j_coord_offsetC,x+j_coord_offsetD,
 889                                               &jx0,&jy0,&jz0);
 890
 891             /* Calculate displacement vector */
 892             dx10             = _mm_sub_ps(ix1,jx0);
 893             dy10             = _mm_sub_ps(iy1,jy0);
 894             dz10             = _mm_sub_ps(iz1,jz0);
 895             dx20             = _mm_sub_ps(ix2,jx0);
 896             dy20             = _mm_sub_ps(iy2,jy0);
 897             dz20             = _mm_sub_ps(iz2,jz0);
 898             dx30             = _mm_sub_ps(ix3,jx0);
 899             dy30             = _mm_sub_ps(iy3,jy0);
 900             dz30             = _mm_sub_ps(iz3,jz0);
 901
 902             /* Calculate squared distance and things based on it */
 903             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 904             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 905             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 906
 907             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 908             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 909             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 910
 911             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 912             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 913             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 914
 915             /* Load parameters for j particles */
 916             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 917                                                               charge+jnrC+0,charge+jnrD+0);
 918
 919             /**************************
 920              * CALCULATE INTERACTIONS *
 921              **************************/
 922
 923             r10              = _mm_mul_ps(rsq10,rinv10);
 924             r10              = _mm_andnot_ps(dummy_mask,r10);
 925
 926             /* Compute parameters for interactions between i and j atoms */
 927             qq10             = _mm_mul_ps(iq1,jq0);
 928
 929             /* EWALD ELECTROSTATICS */
 930
 931             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 932             ewrt             = _mm_mul_ps(r10,ewtabscale);
 933             ewitab           = _mm_cvttps_epi32(ewrt);
 934             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 935             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 936                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 937                                          &ewtabF,&ewtabFn);
 938             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 939             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 940
 941             fscal            = felec;
 942
 943             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 944
 945             /* Calculate temporary vectorial force */
 946             tx               = _mm_mul_ps(fscal,dx10);
 947             ty               = _mm_mul_ps(fscal,dy10);
 948             tz               = _mm_mul_ps(fscal,dz10);
 949
 950             /* Update vectorial force */
 951             fix1             = _mm_add_ps(fix1,tx);
 952             fiy1             = _mm_add_ps(fiy1,ty);
 953             fiz1             = _mm_add_ps(fiz1,tz);
 954
 955             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 956                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 957                                                    tx,ty,tz);
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r20              = _mm_mul_ps(rsq20,rinv20);
 964             r20              = _mm_andnot_ps(dummy_mask,r20);
 965
 966             /* Compute parameters for interactions between i and j atoms */
 967             qq20             = _mm_mul_ps(iq2,jq0);
 968
 969             /* EWALD ELECTROSTATICS */
 970
 971             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 972             ewrt             = _mm_mul_ps(r20,ewtabscale);
 973             ewitab           = _mm_cvttps_epi32(ewrt);
 974             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 975             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 976                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 977                                          &ewtabF,&ewtabFn);
 978             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 979             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 980
 981             fscal            = felec;
 982
 983             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 984
 985             /* Calculate temporary vectorial force */
 986             tx               = _mm_mul_ps(fscal,dx20);
 987             ty               = _mm_mul_ps(fscal,dy20);
 988             tz               = _mm_mul_ps(fscal,dz20);
 989
 990             /* Update vectorial force */
 991             fix2             = _mm_add_ps(fix2,tx);
 992             fiy2             = _mm_add_ps(fiy2,ty);
 993             fiz2             = _mm_add_ps(fiz2,tz);
 994
 995             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 996                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 997                                                    tx,ty,tz);
 998
 999             /**************************
1000              * CALCULATE INTERACTIONS *
1001              **************************/
1002
1003             r30              = _mm_mul_ps(rsq30,rinv30);
1004             r30              = _mm_andnot_ps(dummy_mask,r30);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq30             = _mm_mul_ps(iq3,jq0);
1008
1009             /* EWALD ELECTROSTATICS */
1010
1011             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1012             ewrt             = _mm_mul_ps(r30,ewtabscale);
1013             ewitab           = _mm_cvttps_epi32(ewrt);
1014             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1015             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1016                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1017                                          &ewtabF,&ewtabFn);
1018             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1019             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1020
1021             fscal            = felec;
1022
1023             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1024
1025             /* Calculate temporary vectorial force */
1026             tx               = _mm_mul_ps(fscal,dx30);
1027             ty               = _mm_mul_ps(fscal,dy30);
1028             tz               = _mm_mul_ps(fscal,dz30);
1029
1030             /* Update vectorial force */
1031             fix3             = _mm_add_ps(fix3,tx);
1032             fiy3             = _mm_add_ps(fiy3,ty);
1033             fiz3             = _mm_add_ps(fiz3,tz);
1034
1035             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1036                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1037                                                    tx,ty,tz);
1038
1039             /* Inner loop uses 111 flops */
1040         }
1041
1042         /* End of innermost loop */
1043
1044         gmx_mm_update_iforce_3atom_swizzle_ps(fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1045                                               f+i_coord_offset+DIM,fshift+i_shift_offset);
1046
1047         /* Increment number of inner iterations */
1048         inneriter                  += j_index_end - j_index_start;
1049
1050         /* Outer loop uses 27 flops */
1051     }
1052
1053     /* Increment number of outer iterations */
1054     outeriter        += nri;
1055
1056     /* Update outer/inner flops */
1057
1058     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_W4_F,outeriter*27 + inneriter*111);
1059 }