src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_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_VdwLJ_GeomW3P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_GeomW3P1_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              vdwioffset0;
  67     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  68     int              vdwioffset1;
  69     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  70     int              vdwioffset2;
  71     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  72     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  73     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  74     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  75     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  76     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  77     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  78     real             *charge;
  79     int              nvdwtype;
  80     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  81     int              *vdwtype;
  82     real             *vdwparam;
  83     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  84     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  85     __m128i          ewitab;
  86     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  87     real             *ewtab;
  88     __m128           dummy_mask,cutoff_mask;
  89     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  90     __m128           one     = _mm_set1_ps(1.0);
  91     __m128           two     = _mm_set1_ps(2.0);
  92     x                = xx[0];
  93     f                = ff[0];
  94
  95     nri              = nlist->nri;
  96     iinr             = nlist->iinr;
  97     jindex           = nlist->jindex;
  98     jjnr             = nlist->jjnr;
  99     shiftidx         = nlist->shift;
 100     gid              = nlist->gid;
 101     shiftvec         = fr->shift_vec[0];
 102     fshift           = fr->fshift[0];
 103     facel            = _mm_set1_ps(fr->epsfac);
 104     charge           = mdatoms->chargeA;
 105     nvdwtype         = fr->ntype;
 106     vdwparam         = fr->nbfp;
 107     vdwtype          = mdatoms->typeA;
 108
 109     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 110     ewtab            = fr->ic->tabq_coul_FDV0;
 111     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 112     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 113
 114     /* Setup water-specific parameters */
 115     inr              = nlist->iinr[0];
 116     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 117     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 118     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 119     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 120
 121     /* Avoid stupid compiler warnings */
 122     jnrA = jnrB = jnrC = jnrD = 0;
 123     j_coord_offsetA = 0;
 124     j_coord_offsetB = 0;
 125     j_coord_offsetC = 0;
 126     j_coord_offsetD = 0;
 127
 128     outeriter        = 0;
 129     inneriter        = 0;
 130
 131     /* Start outer loop over neighborlists */
 132     for(iidx=0; iidx<nri; iidx++)
 133     {
 134         /* Load shift vector for this list */
 135         i_shift_offset   = DIM*shiftidx[iidx];
 136         shX              = shiftvec[i_shift_offset+XX];
 137         shY              = shiftvec[i_shift_offset+YY];
 138         shZ              = shiftvec[i_shift_offset+ZZ];
 139
 140         /* Load limits for loop over neighbors */
 141         j_index_start    = jindex[iidx];
 142         j_index_end      = jindex[iidx+1];
 143
 144         /* Get outer coordinate index */
 145         inr              = iinr[iidx];
 146         i_coord_offset   = DIM*inr;
 147
 148         /* Load i particle coords and add shift vector */
 149         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 150         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 151         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 152         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 153         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 154         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 155         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 156         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 157         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_ps();
 171         vvdwsum          = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182
 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             vdwjidx0A        = 2*vdwtype[jnrA+0];
 221             vdwjidx0B        = 2*vdwtype[jnrB+0];
 222             vdwjidx0C        = 2*vdwtype[jnrC+0];
 223             vdwjidx0D        = 2*vdwtype[jnrD+0];
 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             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 234                                          vdwparam+vdwioffset0+vdwjidx0B,
 235                                          vdwparam+vdwioffset0+vdwjidx0C,
 236                                          vdwparam+vdwioffset0+vdwjidx0D,
 237                                          &c6_00,&c12_00);
 238
 239             /* EWALD ELECTROSTATICS */
 240
 241             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 242             ewrt             = _mm_mul_ps(r00,ewtabscale);
 243             ewitab           = _mm_cvttps_epi32(ewrt);
 244             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 245             ewitab           = _mm_slli_epi32(ewitab,2);
 246             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 247             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 248             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 249             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 250             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 251             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 252             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 253             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 254             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 255
 256             /* LENNARD-JONES DISPERSION/REPULSION */
 257
 258             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 259             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 260             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 261             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 262             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 263
 264             /* Update potential sum for this i atom from the interaction with this j atom. */
 265             velecsum         = _mm_add_ps(velecsum,velec);
 266             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 267
 268             fscal            = _mm_add_ps(felec,fvdw);
 269
 270             /* Calculate temporary vectorial force */
 271             tx               = _mm_mul_ps(fscal,dx00);
 272             ty               = _mm_mul_ps(fscal,dy00);
 273             tz               = _mm_mul_ps(fscal,dz00);
 274
 275             /* Update vectorial force */
 276             fix0             = _mm_add_ps(fix0,tx);
 277             fiy0             = _mm_add_ps(fiy0,ty);
 278             fiz0             = _mm_add_ps(fiz0,tz);
 279
 280             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 281                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 282                                                    tx,ty,tz);
 283
 284             /**************************
 285              * CALCULATE INTERACTIONS *
 286              **************************/
 287
 288             r10              = _mm_mul_ps(rsq10,rinv10);
 289
 290             /* Compute parameters for interactions between i and j atoms */
 291             qq10             = _mm_mul_ps(iq1,jq0);
 292
 293             /* EWALD ELECTROSTATICS */
 294
 295             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 296             ewrt             = _mm_mul_ps(r10,ewtabscale);
 297             ewitab           = _mm_cvttps_epi32(ewrt);
 298             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 299             ewitab           = _mm_slli_epi32(ewitab,2);
 300             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 301             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 302             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 303             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 304             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 305             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 306             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 307             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 308             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm_add_ps(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315             /* Calculate temporary vectorial force */
 316             tx               = _mm_mul_ps(fscal,dx10);
 317             ty               = _mm_mul_ps(fscal,dy10);
 318             tz               = _mm_mul_ps(fscal,dz10);
 319
 320             /* Update vectorial force */
 321             fix1             = _mm_add_ps(fix1,tx);
 322             fiy1             = _mm_add_ps(fiy1,ty);
 323             fiz1             = _mm_add_ps(fiz1,tz);
 324
 325             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 326                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 327                                                    tx,ty,tz);
 328
 329             /**************************
 330              * CALCULATE INTERACTIONS *
 331              **************************/
 332
 333             r20              = _mm_mul_ps(rsq20,rinv20);
 334
 335             /* Compute parameters for interactions between i and j atoms */
 336             qq20             = _mm_mul_ps(iq2,jq0);
 337
 338             /* EWALD ELECTROSTATICS */
 339
 340             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 341             ewrt             = _mm_mul_ps(r20,ewtabscale);
 342             ewitab           = _mm_cvttps_epi32(ewrt);
 343             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 344             ewitab           = _mm_slli_epi32(ewitab,2);
 345             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 346             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 347             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 348             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 349             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 350             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 351             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 352             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 353             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 354
 355             /* Update potential sum for this i atom from the interaction with this j atom. */
 356             velecsum         = _mm_add_ps(velecsum,velec);
 357
 358             fscal            = felec;
 359
 360             /* Calculate temporary vectorial force */
 361             tx               = _mm_mul_ps(fscal,dx20);
 362             ty               = _mm_mul_ps(fscal,dy20);
 363             tz               = _mm_mul_ps(fscal,dz20);
 364
 365             /* Update vectorial force */
 366             fix2             = _mm_add_ps(fix2,tx);
 367             fiy2             = _mm_add_ps(fiy2,ty);
 368             fiz2             = _mm_add_ps(fiz2,tz);
 369
 370             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 371                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 372                                                    tx,ty,tz);
 373
 374             /* Inner loop uses 135 flops */
 375         }
 376
 377         if(jidx<j_index_end)
 378         {
 379
 380             /* Get j neighbor index, and coordinate index */
 381             jnrA             = jjnr[jidx];
 382             jnrB             = jjnr[jidx+1];
 383             jnrC             = jjnr[jidx+2];
 384             jnrD             = jjnr[jidx+3];
 385
 386             /* Sign of each element will be negative for non-real atoms.
 387              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 388              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 389              */
 390             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 391             jnrA       = (jnrA>=0) ? jnrA : 0;
 392             jnrB       = (jnrB>=0) ? jnrB : 0;
 393             jnrC       = (jnrC>=0) ? jnrC : 0;
 394             jnrD       = (jnrD>=0) ? jnrD : 0;
 395
 396             j_coord_offsetA  = DIM*jnrA;
 397             j_coord_offsetB  = DIM*jnrB;
 398             j_coord_offsetC  = DIM*jnrC;
 399             j_coord_offsetD  = DIM*jnrD;
 400
 401             /* load j atom coordinates */
 402             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 403                                               x+j_coord_offsetC,x+j_coord_offsetD,
 404                                               &jx0,&jy0,&jz0);
 405
 406             /* Calculate displacement vector */
 407             dx00             = _mm_sub_ps(ix0,jx0);
 408             dy00             = _mm_sub_ps(iy0,jy0);
 409             dz00             = _mm_sub_ps(iz0,jz0);
 410             dx10             = _mm_sub_ps(ix1,jx0);
 411             dy10             = _mm_sub_ps(iy1,jy0);
 412             dz10             = _mm_sub_ps(iz1,jz0);
 413             dx20             = _mm_sub_ps(ix2,jx0);
 414             dy20             = _mm_sub_ps(iy2,jy0);
 415             dz20             = _mm_sub_ps(iz2,jz0);
 416
 417             /* Calculate squared distance and things based on it */
 418             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 419             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 420             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 421
 422             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 423             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 424             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 425
 426             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 427             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 428             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 429
 430             /* Load parameters for j particles */
 431             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 432                                                               charge+jnrC+0,charge+jnrD+0);
 433             vdwjidx0A        = 2*vdwtype[jnrA+0];
 434             vdwjidx0B        = 2*vdwtype[jnrB+0];
 435             vdwjidx0C        = 2*vdwtype[jnrC+0];
 436             vdwjidx0D        = 2*vdwtype[jnrD+0];
 437
 438             /**************************
 439              * CALCULATE INTERACTIONS *
 440              **************************/
 441
 442             r00              = _mm_mul_ps(rsq00,rinv00);
 443             r00              = _mm_andnot_ps(dummy_mask,r00);
 444
 445             /* Compute parameters for interactions between i and j atoms */
 446             qq00             = _mm_mul_ps(iq0,jq0);
 447             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 448                                          vdwparam+vdwioffset0+vdwjidx0B,
 449                                          vdwparam+vdwioffset0+vdwjidx0C,
 450                                          vdwparam+vdwioffset0+vdwjidx0D,
 451                                          &c6_00,&c12_00);
 452
 453             /* EWALD ELECTROSTATICS */
 454
 455             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 456             ewrt             = _mm_mul_ps(r00,ewtabscale);
 457             ewitab           = _mm_cvttps_epi32(ewrt);
 458             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 459             ewitab           = _mm_slli_epi32(ewitab,2);
 460             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 461             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 462             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 463             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 464             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 465             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 466             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 467             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 468             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 469
 470             /* LENNARD-JONES DISPERSION/REPULSION */
 471
 472             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 473             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 474             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 475             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 476             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 477
 478             /* Update potential sum for this i atom from the interaction with this j atom. */
 479             velec            = _mm_andnot_ps(dummy_mask,velec);
 480             velecsum         = _mm_add_ps(velecsum,velec);
 481             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 482             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 483
 484             fscal            = _mm_add_ps(felec,fvdw);
 485
 486             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 487
 488             /* Calculate temporary vectorial force */
 489             tx               = _mm_mul_ps(fscal,dx00);
 490             ty               = _mm_mul_ps(fscal,dy00);
 491             tz               = _mm_mul_ps(fscal,dz00);
 492
 493             /* Update vectorial force */
 494             fix0             = _mm_add_ps(fix0,tx);
 495             fiy0             = _mm_add_ps(fiy0,ty);
 496             fiz0             = _mm_add_ps(fiz0,tz);
 497
 498             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 499                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 500                                                    tx,ty,tz);
 501
 502             /**************************
 503              * CALCULATE INTERACTIONS *
 504              **************************/
 505
 506             r10              = _mm_mul_ps(rsq10,rinv10);
 507             r10              = _mm_andnot_ps(dummy_mask,r10);
 508
 509             /* Compute parameters for interactions between i and j atoms */
 510             qq10             = _mm_mul_ps(iq1,jq0);
 511
 512             /* EWALD ELECTROSTATICS */
 513
 514             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 515             ewrt             = _mm_mul_ps(r10,ewtabscale);
 516             ewitab           = _mm_cvttps_epi32(ewrt);
 517             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 518             ewitab           = _mm_slli_epi32(ewitab,2);
 519             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 520             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 521             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 522             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 523             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 524             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 525             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 526             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 527             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 528
 529             /* Update potential sum for this i atom from the interaction with this j atom. */
 530             velec            = _mm_andnot_ps(dummy_mask,velec);
 531             velecsum         = _mm_add_ps(velecsum,velec);
 532
 533             fscal            = felec;
 534
 535             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 536
 537             /* Calculate temporary vectorial force */
 538             tx               = _mm_mul_ps(fscal,dx10);
 539             ty               = _mm_mul_ps(fscal,dy10);
 540             tz               = _mm_mul_ps(fscal,dz10);
 541
 542             /* Update vectorial force */
 543             fix1             = _mm_add_ps(fix1,tx);
 544             fiy1             = _mm_add_ps(fiy1,ty);
 545             fiz1             = _mm_add_ps(fiz1,tz);
 546
 547             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 548                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 549                                                    tx,ty,tz);
 550
 551             /**************************
 552              * CALCULATE INTERACTIONS *
 553              **************************/
 554
 555             r20              = _mm_mul_ps(rsq20,rinv20);
 556             r20              = _mm_andnot_ps(dummy_mask,r20);
 557
 558             /* Compute parameters for interactions between i and j atoms */
 559             qq20             = _mm_mul_ps(iq2,jq0);
 560
 561             /* EWALD ELECTROSTATICS */
 562
 563             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 564             ewrt             = _mm_mul_ps(r20,ewtabscale);
 565             ewitab           = _mm_cvttps_epi32(ewrt);
 566             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 567             ewitab           = _mm_slli_epi32(ewitab,2);
 568             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 569             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 570             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 571             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 572             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 573             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 574             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 575             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 576             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 577
 578             /* Update potential sum for this i atom from the interaction with this j atom. */
 579             velec            = _mm_andnot_ps(dummy_mask,velec);
 580             velecsum         = _mm_add_ps(velecsum,velec);
 581
 582             fscal            = felec;
 583
 584             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 585
 586             /* Calculate temporary vectorial force */
 587             tx               = _mm_mul_ps(fscal,dx20);
 588             ty               = _mm_mul_ps(fscal,dy20);
 589             tz               = _mm_mul_ps(fscal,dz20);
 590
 591             /* Update vectorial force */
 592             fix2             = _mm_add_ps(fix2,tx);
 593             fiy2             = _mm_add_ps(fiy2,ty);
 594             fiz2             = _mm_add_ps(fiz2,tz);
 595
 596             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 597                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 598                                                    tx,ty,tz);
 599
 600             /* Inner loop uses 138 flops */
 601         }
 602
 603         /* End of innermost loop */
 604
 605         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 606                                               f+i_coord_offset,fshift+i_shift_offset);
 607
 608         ggid                        = gid[iidx];
 609         /* Update potential energies */
 610         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 611         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 612
 613         /* Increment number of inner iterations */
 614         inneriter                  += j_index_end - j_index_start;
 615
 616         /* Outer loop uses 29 flops */
 617     }
 618
 619     /* Increment number of outer iterations */
 620     outeriter        += nri;
 621
 622     /* Update outer/inner flops */
 623
 624     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*29 + inneriter*138);
 625 }
 626 /*
 627  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_single
 628  * Electrostatics interaction: Ewald
 629  * VdW interaction:            LennardJones
 630  * Geometry:                   Water3-Particle
 631  * Calculate force/pot:        Force
 632  */
 633 void
 634 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse2_single
 635                     (t_nblist * gmx_restrict                nlist,
 636                      rvec * gmx_restrict                    xx,
 637                      rvec * gmx_restrict                    ff,
 638                      t_forcerec * gmx_restrict              fr,
 639                      t_mdatoms * gmx_restrict               mdatoms,
 640                      nb_kernel_data_t * gmx_restrict        kernel_data,
 641                      t_nrnb * gmx_restrict                  nrnb)
 642 {
 643     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 644      * just 0 for non-waters.
 645      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 646      * jnr indices corresponding to data put in the four positions in the SIMD register.
 647      */
 648     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 649     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 650     int              jnrA,jnrB,jnrC,jnrD;
 651     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 652     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 653     real             shX,shY,shZ,rcutoff_scalar;
 654     real             *shiftvec,*fshift,*x,*f;
 655     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 656     int              vdwioffset0;
 657     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 658     int              vdwioffset1;
 659     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 660     int              vdwioffset2;
 661     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 662     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 663     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 664     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 665     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 666     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 667     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 668     real             *charge;
 669     int              nvdwtype;
 670     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 671     int              *vdwtype;
 672     real             *vdwparam;
 673     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 674     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 675     __m128i          ewitab;
 676     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 677     real             *ewtab;
 678     __m128           dummy_mask,cutoff_mask;
 679     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 680     __m128           one     = _mm_set1_ps(1.0);
 681     __m128           two     = _mm_set1_ps(2.0);
 682     x                = xx[0];
 683     f                = ff[0];
 684
 685     nri              = nlist->nri;
 686     iinr             = nlist->iinr;
 687     jindex           = nlist->jindex;
 688     jjnr             = nlist->jjnr;
 689     shiftidx         = nlist->shift;
 690     gid              = nlist->gid;
 691     shiftvec         = fr->shift_vec[0];
 692     fshift           = fr->fshift[0];
 693     facel            = _mm_set1_ps(fr->epsfac);
 694     charge           = mdatoms->chargeA;
 695     nvdwtype         = fr->ntype;
 696     vdwparam         = fr->nbfp;
 697     vdwtype          = mdatoms->typeA;
 698
 699     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 700     ewtab            = fr->ic->tabq_coul_F;
 701     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 702     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 703
 704     /* Setup water-specific parameters */
 705     inr              = nlist->iinr[0];
 706     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 707     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 708     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 709     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 710
 711     /* Avoid stupid compiler warnings */
 712     jnrA = jnrB = jnrC = jnrD = 0;
 713     j_coord_offsetA = 0;
 714     j_coord_offsetB = 0;
 715     j_coord_offsetC = 0;
 716     j_coord_offsetD = 0;
 717
 718     outeriter        = 0;
 719     inneriter        = 0;
 720
 721     /* Start outer loop over neighborlists */
 722     for(iidx=0; iidx<nri; iidx++)
 723     {
 724         /* Load shift vector for this list */
 725         i_shift_offset   = DIM*shiftidx[iidx];
 726         shX              = shiftvec[i_shift_offset+XX];
 727         shY              = shiftvec[i_shift_offset+YY];
 728         shZ              = shiftvec[i_shift_offset+ZZ];
 729
 730         /* Load limits for loop over neighbors */
 731         j_index_start    = jindex[iidx];
 732         j_index_end      = jindex[iidx+1];
 733
 734         /* Get outer coordinate index */
 735         inr              = iinr[iidx];
 736         i_coord_offset   = DIM*inr;
 737
 738         /* Load i particle coords and add shift vector */
 739         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 740         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 741         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 742         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 743         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 744         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 745         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 746         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 747         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 748
 749         fix0             = _mm_setzero_ps();
 750         fiy0             = _mm_setzero_ps();
 751         fiz0             = _mm_setzero_ps();
 752         fix1             = _mm_setzero_ps();
 753         fiy1             = _mm_setzero_ps();
 754         fiz1             = _mm_setzero_ps();
 755         fix2             = _mm_setzero_ps();
 756         fiy2             = _mm_setzero_ps();
 757         fiz2             = _mm_setzero_ps();
 758
 759         /* Start inner kernel loop */
 760         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 761         {
 762
 763             /* Get j neighbor index, and coordinate index */
 764             jnrA             = jjnr[jidx];
 765             jnrB             = jjnr[jidx+1];
 766             jnrC             = jjnr[jidx+2];
 767             jnrD             = jjnr[jidx+3];
 768
 769             j_coord_offsetA  = DIM*jnrA;
 770             j_coord_offsetB  = DIM*jnrB;
 771             j_coord_offsetC  = DIM*jnrC;
 772             j_coord_offsetD  = DIM*jnrD;
 773
 774             /* load j atom coordinates */
 775             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 776                                               x+j_coord_offsetC,x+j_coord_offsetD,
 777                                               &jx0,&jy0,&jz0);
 778
 779             /* Calculate displacement vector */
 780             dx00             = _mm_sub_ps(ix0,jx0);
 781             dy00             = _mm_sub_ps(iy0,jy0);
 782             dz00             = _mm_sub_ps(iz0,jz0);
 783             dx10             = _mm_sub_ps(ix1,jx0);
 784             dy10             = _mm_sub_ps(iy1,jy0);
 785             dz10             = _mm_sub_ps(iz1,jz0);
 786             dx20             = _mm_sub_ps(ix2,jx0);
 787             dy20             = _mm_sub_ps(iy2,jy0);
 788             dz20             = _mm_sub_ps(iz2,jz0);
 789
 790             /* Calculate squared distance and things based on it */
 791             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 792             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 793             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 794
 795             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 796             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 797             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 798
 799             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 800             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 801             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 802
 803             /* Load parameters for j particles */
 804             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 805                                                               charge+jnrC+0,charge+jnrD+0);
 806             vdwjidx0A        = 2*vdwtype[jnrA+0];
 807             vdwjidx0B        = 2*vdwtype[jnrB+0];
 808             vdwjidx0C        = 2*vdwtype[jnrC+0];
 809             vdwjidx0D        = 2*vdwtype[jnrD+0];
 810
 811             /**************************
 812              * CALCULATE INTERACTIONS *
 813              **************************/
 814
 815             r00              = _mm_mul_ps(rsq00,rinv00);
 816
 817             /* Compute parameters for interactions between i and j atoms */
 818             qq00             = _mm_mul_ps(iq0,jq0);
 819             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 820                                          vdwparam+vdwioffset0+vdwjidx0B,
 821                                          vdwparam+vdwioffset0+vdwjidx0C,
 822                                          vdwparam+vdwioffset0+vdwjidx0D,
 823                                          &c6_00,&c12_00);
 824
 825             /* EWALD ELECTROSTATICS */
 826
 827             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 828             ewrt             = _mm_mul_ps(r00,ewtabscale);
 829             ewitab           = _mm_cvttps_epi32(ewrt);
 830             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 831             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 832                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 833                                          &ewtabF,&ewtabFn);
 834             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 835             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 836
 837             /* LENNARD-JONES DISPERSION/REPULSION */
 838
 839             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 840             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 841
 842             fscal            = _mm_add_ps(felec,fvdw);
 843
 844             /* Calculate temporary vectorial force */
 845             tx               = _mm_mul_ps(fscal,dx00);
 846             ty               = _mm_mul_ps(fscal,dy00);
 847             tz               = _mm_mul_ps(fscal,dz00);
 848
 849             /* Update vectorial force */
 850             fix0             = _mm_add_ps(fix0,tx);
 851             fiy0             = _mm_add_ps(fiy0,ty);
 852             fiz0             = _mm_add_ps(fiz0,tz);
 853
 854             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 855                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 856                                                    tx,ty,tz);
 857
 858             /**************************
 859              * CALCULATE INTERACTIONS *
 860              **************************/
 861
 862             r10              = _mm_mul_ps(rsq10,rinv10);
 863
 864             /* Compute parameters for interactions between i and j atoms */
 865             qq10             = _mm_mul_ps(iq1,jq0);
 866
 867             /* EWALD ELECTROSTATICS */
 868
 869             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 870             ewrt             = _mm_mul_ps(r10,ewtabscale);
 871             ewitab           = _mm_cvttps_epi32(ewrt);
 872             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 873             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 874                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 875                                          &ewtabF,&ewtabFn);
 876             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 877             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 878
 879             fscal            = felec;
 880
 881             /* Calculate temporary vectorial force */
 882             tx               = _mm_mul_ps(fscal,dx10);
 883             ty               = _mm_mul_ps(fscal,dy10);
 884             tz               = _mm_mul_ps(fscal,dz10);
 885
 886             /* Update vectorial force */
 887             fix1             = _mm_add_ps(fix1,tx);
 888             fiy1             = _mm_add_ps(fiy1,ty);
 889             fiz1             = _mm_add_ps(fiz1,tz);
 890
 891             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 892                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 893                                                    tx,ty,tz);
 894
 895             /**************************
 896              * CALCULATE INTERACTIONS *
 897              **************************/
 898
 899             r20              = _mm_mul_ps(rsq20,rinv20);
 900
 901             /* Compute parameters for interactions between i and j atoms */
 902             qq20             = _mm_mul_ps(iq2,jq0);
 903
 904             /* EWALD ELECTROSTATICS */
 905
 906             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 907             ewrt             = _mm_mul_ps(r20,ewtabscale);
 908             ewitab           = _mm_cvttps_epi32(ewrt);
 909             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 910             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 911                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 912                                          &ewtabF,&ewtabFn);
 913             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 914             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 915
 916             fscal            = felec;
 917
 918             /* Calculate temporary vectorial force */
 919             tx               = _mm_mul_ps(fscal,dx20);
 920             ty               = _mm_mul_ps(fscal,dy20);
 921             tz               = _mm_mul_ps(fscal,dz20);
 922
 923             /* Update vectorial force */
 924             fix2             = _mm_add_ps(fix2,tx);
 925             fiy2             = _mm_add_ps(fiy2,ty);
 926             fiz2             = _mm_add_ps(fiz2,tz);
 927
 928             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 929                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 930                                                    tx,ty,tz);
 931
 932             /* Inner loop uses 115 flops */
 933         }
 934
 935         if(jidx<j_index_end)
 936         {
 937
 938             /* Get j neighbor index, and coordinate index */
 939             jnrA             = jjnr[jidx];
 940             jnrB             = jjnr[jidx+1];
 941             jnrC             = jjnr[jidx+2];
 942             jnrD             = jjnr[jidx+3];
 943
 944             /* Sign of each element will be negative for non-real atoms.
 945              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 946              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 947              */
 948             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 949             jnrA       = (jnrA>=0) ? jnrA : 0;
 950             jnrB       = (jnrB>=0) ? jnrB : 0;
 951             jnrC       = (jnrC>=0) ? jnrC : 0;
 952             jnrD       = (jnrD>=0) ? jnrD : 0;
 953
 954             j_coord_offsetA  = DIM*jnrA;
 955             j_coord_offsetB  = DIM*jnrB;
 956             j_coord_offsetC  = DIM*jnrC;
 957             j_coord_offsetD  = DIM*jnrD;
 958
 959             /* load j atom coordinates */
 960             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 961                                               x+j_coord_offsetC,x+j_coord_offsetD,
 962                                               &jx0,&jy0,&jz0);
 963
 964             /* Calculate displacement vector */
 965             dx00             = _mm_sub_ps(ix0,jx0);
 966             dy00             = _mm_sub_ps(iy0,jy0);
 967             dz00             = _mm_sub_ps(iz0,jz0);
 968             dx10             = _mm_sub_ps(ix1,jx0);
 969             dy10             = _mm_sub_ps(iy1,jy0);
 970             dz10             = _mm_sub_ps(iz1,jz0);
 971             dx20             = _mm_sub_ps(ix2,jx0);
 972             dy20             = _mm_sub_ps(iy2,jy0);
 973             dz20             = _mm_sub_ps(iz2,jz0);
 974
 975             /* Calculate squared distance and things based on it */
 976             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 977             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 978             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 979
 980             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 981             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 982             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 983
 984             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 985             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 986             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 987
 988             /* Load parameters for j particles */
 989             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 990                                                               charge+jnrC+0,charge+jnrD+0);
 991             vdwjidx0A        = 2*vdwtype[jnrA+0];
 992             vdwjidx0B        = 2*vdwtype[jnrB+0];
 993             vdwjidx0C        = 2*vdwtype[jnrC+0];
 994             vdwjidx0D        = 2*vdwtype[jnrD+0];
 995
 996             /**************************
 997              * CALCULATE INTERACTIONS *
 998              **************************/
 999
1000             r00              = _mm_mul_ps(rsq00,rinv00);
1001             r00              = _mm_andnot_ps(dummy_mask,r00);
1002
1003             /* Compute parameters for interactions between i and j atoms */
1004             qq00             = _mm_mul_ps(iq0,jq0);
1005             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1006                                          vdwparam+vdwioffset0+vdwjidx0B,
1007                                          vdwparam+vdwioffset0+vdwjidx0C,
1008                                          vdwparam+vdwioffset0+vdwjidx0D,
1009                                          &c6_00,&c12_00);
1010
1011             /* EWALD ELECTROSTATICS */
1012
1013             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1014             ewrt             = _mm_mul_ps(r00,ewtabscale);
1015             ewitab           = _mm_cvttps_epi32(ewrt);
1016             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1017             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1018                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1019                                          &ewtabF,&ewtabFn);
1020             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1021             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1022
1023             /* LENNARD-JONES DISPERSION/REPULSION */
1024
1025             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1026             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1027
1028             fscal            = _mm_add_ps(felec,fvdw);
1029
1030             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1031
1032             /* Calculate temporary vectorial force */
1033             tx               = _mm_mul_ps(fscal,dx00);
1034             ty               = _mm_mul_ps(fscal,dy00);
1035             tz               = _mm_mul_ps(fscal,dz00);
1036
1037             /* Update vectorial force */
1038             fix0             = _mm_add_ps(fix0,tx);
1039             fiy0             = _mm_add_ps(fiy0,ty);
1040             fiz0             = _mm_add_ps(fiz0,tz);
1041
1042             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1043                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1044                                                    tx,ty,tz);
1045
1046             /**************************
1047              * CALCULATE INTERACTIONS *
1048              **************************/
1049
1050             r10              = _mm_mul_ps(rsq10,rinv10);
1051             r10              = _mm_andnot_ps(dummy_mask,r10);
1052
1053             /* Compute parameters for interactions between i and j atoms */
1054             qq10             = _mm_mul_ps(iq1,jq0);
1055
1056             /* EWALD ELECTROSTATICS */
1057
1058             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1059             ewrt             = _mm_mul_ps(r10,ewtabscale);
1060             ewitab           = _mm_cvttps_epi32(ewrt);
1061             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1062             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1063                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1064                                          &ewtabF,&ewtabFn);
1065             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1066             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1067
1068             fscal            = felec;
1069
1070             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1071
1072             /* Calculate temporary vectorial force */
1073             tx               = _mm_mul_ps(fscal,dx10);
1074             ty               = _mm_mul_ps(fscal,dy10);
1075             tz               = _mm_mul_ps(fscal,dz10);
1076
1077             /* Update vectorial force */
1078             fix1             = _mm_add_ps(fix1,tx);
1079             fiy1             = _mm_add_ps(fiy1,ty);
1080             fiz1             = _mm_add_ps(fiz1,tz);
1081
1082             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1083                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1084                                                    tx,ty,tz);
1085
1086             /**************************
1087              * CALCULATE INTERACTIONS *
1088              **************************/
1089
1090             r20              = _mm_mul_ps(rsq20,rinv20);
1091             r20              = _mm_andnot_ps(dummy_mask,r20);
1092
1093             /* Compute parameters for interactions between i and j atoms */
1094             qq20             = _mm_mul_ps(iq2,jq0);
1095
1096             /* EWALD ELECTROSTATICS */
1097
1098             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1099             ewrt             = _mm_mul_ps(r20,ewtabscale);
1100             ewitab           = _mm_cvttps_epi32(ewrt);
1101             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1102             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1103                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1104                                          &ewtabF,&ewtabFn);
1105             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1106             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1107
1108             fscal            = felec;
1109
1110             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1111
1112             /* Calculate temporary vectorial force */
1113             tx               = _mm_mul_ps(fscal,dx20);
1114             ty               = _mm_mul_ps(fscal,dy20);
1115             tz               = _mm_mul_ps(fscal,dz20);
1116
1117             /* Update vectorial force */
1118             fix2             = _mm_add_ps(fix2,tx);
1119             fiy2             = _mm_add_ps(fiy2,ty);
1120             fiz2             = _mm_add_ps(fiz2,tz);
1121
1122             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1123                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1124                                                    tx,ty,tz);
1125
1126             /* Inner loop uses 118 flops */
1127         }
1128
1129         /* End of innermost loop */
1130
1131         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1132                                               f+i_coord_offset,fshift+i_shift_offset);
1133
1134         /* Increment number of inner iterations */
1135         inneriter                  += j_index_end - j_index_start;
1136
1137         /* Outer loop uses 27 flops */
1138     }
1139
1140     /* Increment number of outer iterations */
1141     outeriter        += nri;
1142
1143     /* Update outer/inner flops */
1144
1145     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*27 + inneriter*118);
1146 }