src/gmxlib/nonbonded/nb_kernel_sse2_single/nb_kernel_ElecEw_VdwLJ_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_VdwLJ_GeomW4P1_VF_sse2_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwLJ_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              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              vdwioffset3;
  73     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  74     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  75     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  76     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  77     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  78     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  79     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  80     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  81     real             *charge;
  82     int              nvdwtype;
  83     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  84     int              *vdwtype;
  85     real             *vdwparam;
  86     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  87     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  88     __m128i          ewitab;
  89     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  90     real             *ewtab;
  91     __m128           dummy_mask,cutoff_mask;
  92     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  93     __m128           one     = _mm_set1_ps(1.0);
  94     __m128           two     = _mm_set1_ps(2.0);
  95     x                = xx[0];
  96     f                = ff[0];
  97
  98     nri              = nlist->nri;
  99     iinr             = nlist->iinr;
 100     jindex           = nlist->jindex;
 101     jjnr             = nlist->jjnr;
 102     shiftidx         = nlist->shift;
 103     gid              = nlist->gid;
 104     shiftvec         = fr->shift_vec[0];
 105     fshift           = fr->fshift[0];
 106     facel            = _mm_set1_ps(fr->epsfac);
 107     charge           = mdatoms->chargeA;
 108     nvdwtype         = fr->ntype;
 109     vdwparam         = fr->nbfp;
 110     vdwtype          = mdatoms->typeA;
 111
 112     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 113     ewtab            = fr->ic->tabq_coul_FDV0;
 114     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 115     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 116
 117     /* Setup water-specific parameters */
 118     inr              = nlist->iinr[0];
 119     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 120     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 121     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 122     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 123
 124     /* Avoid stupid compiler warnings */
 125     jnrA = jnrB = jnrC = jnrD = 0;
 126     j_coord_offsetA = 0;
 127     j_coord_offsetB = 0;
 128     j_coord_offsetC = 0;
 129     j_coord_offsetD = 0;
 130
 131     outeriter        = 0;
 132     inneriter        = 0;
 133
 134     /* Start outer loop over neighborlists */
 135     for(iidx=0; iidx<nri; iidx++)
 136     {
 137         /* Load shift vector for this list */
 138         i_shift_offset   = DIM*shiftidx[iidx];
 139         shX              = shiftvec[i_shift_offset+XX];
 140         shY              = shiftvec[i_shift_offset+YY];
 141         shZ              = shiftvec[i_shift_offset+ZZ];
 142
 143         /* Load limits for loop over neighbors */
 144         j_index_start    = jindex[iidx];
 145         j_index_end      = jindex[iidx+1];
 146
 147         /* Get outer coordinate index */
 148         inr              = iinr[iidx];
 149         i_coord_offset   = DIM*inr;
 150
 151         /* Load i particle coords and add shift vector */
 152         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 153         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 154         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 155         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 156         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 157         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 158         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 159         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 160         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 161         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 162         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 163         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 164
 165         fix0             = _mm_setzero_ps();
 166         fiy0             = _mm_setzero_ps();
 167         fiz0             = _mm_setzero_ps();
 168         fix1             = _mm_setzero_ps();
 169         fiy1             = _mm_setzero_ps();
 170         fiz1             = _mm_setzero_ps();
 171         fix2             = _mm_setzero_ps();
 172         fiy2             = _mm_setzero_ps();
 173         fiz2             = _mm_setzero_ps();
 174         fix3             = _mm_setzero_ps();
 175         fiy3             = _mm_setzero_ps();
 176         fiz3             = _mm_setzero_ps();
 177
 178         /* Reset potential sums */
 179         velecsum         = _mm_setzero_ps();
 180         vvdwsum          = _mm_setzero_ps();
 181
 182         /* Start inner kernel loop */
 183         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 184         {
 185
 186             /* Get j neighbor index, and coordinate index */
 187             jnrA             = jjnr[jidx];
 188             jnrB             = jjnr[jidx+1];
 189             jnrC             = jjnr[jidx+2];
 190             jnrD             = jjnr[jidx+3];
 191
 192             j_coord_offsetA  = DIM*jnrA;
 193             j_coord_offsetB  = DIM*jnrB;
 194             j_coord_offsetC  = DIM*jnrC;
 195             j_coord_offsetD  = DIM*jnrD;
 196
 197             /* load j atom coordinates */
 198             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 199                                               x+j_coord_offsetC,x+j_coord_offsetD,
 200                                               &jx0,&jy0,&jz0);
 201
 202             /* Calculate displacement vector */
 203             dx00             = _mm_sub_ps(ix0,jx0);
 204             dy00             = _mm_sub_ps(iy0,jy0);
 205             dz00             = _mm_sub_ps(iz0,jz0);
 206             dx10             = _mm_sub_ps(ix1,jx0);
 207             dy10             = _mm_sub_ps(iy1,jy0);
 208             dz10             = _mm_sub_ps(iz1,jz0);
 209             dx20             = _mm_sub_ps(ix2,jx0);
 210             dy20             = _mm_sub_ps(iy2,jy0);
 211             dz20             = _mm_sub_ps(iz2,jz0);
 212             dx30             = _mm_sub_ps(ix3,jx0);
 213             dy30             = _mm_sub_ps(iy3,jy0);
 214             dz30             = _mm_sub_ps(iz3,jz0);
 215
 216             /* Calculate squared distance and things based on it */
 217             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 218             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 219             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 220             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 221
 222             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 223             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 224             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 225
 226             rinvsq00         = gmx_mm_inv_ps(rsq00);
 227             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 228             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 229             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 230
 231             /* Load parameters for j particles */
 232             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 233                                                               charge+jnrC+0,charge+jnrD+0);
 234             vdwjidx0A        = 2*vdwtype[jnrA+0];
 235             vdwjidx0B        = 2*vdwtype[jnrB+0];
 236             vdwjidx0C        = 2*vdwtype[jnrC+0];
 237             vdwjidx0D        = 2*vdwtype[jnrD+0];
 238
 239             /**************************
 240              * CALCULATE INTERACTIONS *
 241              **************************/
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 245                                          vdwparam+vdwioffset0+vdwjidx0B,
 246                                          vdwparam+vdwioffset0+vdwjidx0C,
 247                                          vdwparam+vdwioffset0+vdwjidx0D,
 248                                          &c6_00,&c12_00);
 249
 250             /* LENNARD-JONES DISPERSION/REPULSION */
 251
 252             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 253             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 254             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 255             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 256             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 257
 258             /* Update potential sum for this i atom from the interaction with this j atom. */
 259             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 260
 261             fscal            = fvdw;
 262
 263             /* Calculate temporary vectorial force */
 264             tx               = _mm_mul_ps(fscal,dx00);
 265             ty               = _mm_mul_ps(fscal,dy00);
 266             tz               = _mm_mul_ps(fscal,dz00);
 267
 268             /* Update vectorial force */
 269             fix0             = _mm_add_ps(fix0,tx);
 270             fiy0             = _mm_add_ps(fiy0,ty);
 271             fiz0             = _mm_add_ps(fiz0,tz);
 272
 273             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 274                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 275                                                    tx,ty,tz);
 276
 277             /**************************
 278              * CALCULATE INTERACTIONS *
 279              **************************/
 280
 281             r10              = _mm_mul_ps(rsq10,rinv10);
 282
 283             /* Compute parameters for interactions between i and j atoms */
 284             qq10             = _mm_mul_ps(iq1,jq0);
 285
 286             /* EWALD ELECTROSTATICS */
 287
 288             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 289             ewrt             = _mm_mul_ps(r10,ewtabscale);
 290             ewitab           = _mm_cvttps_epi32(ewrt);
 291             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 292             ewitab           = _mm_slli_epi32(ewitab,2);
 293             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 294             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 295             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 296             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 297             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 298             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 299             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 300             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 301             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 302
 303             /* Update potential sum for this i atom from the interaction with this j atom. */
 304             velecsum         = _mm_add_ps(velecsum,velec);
 305
 306             fscal            = felec;
 307
 308             /* Calculate temporary vectorial force */
 309             tx               = _mm_mul_ps(fscal,dx10);
 310             ty               = _mm_mul_ps(fscal,dy10);
 311             tz               = _mm_mul_ps(fscal,dz10);
 312
 313             /* Update vectorial force */
 314             fix1             = _mm_add_ps(fix1,tx);
 315             fiy1             = _mm_add_ps(fiy1,ty);
 316             fiz1             = _mm_add_ps(fiz1,tz);
 317
 318             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 319                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 320                                                    tx,ty,tz);
 321
 322             /**************************
 323              * CALCULATE INTERACTIONS *
 324              **************************/
 325
 326             r20              = _mm_mul_ps(rsq20,rinv20);
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq20             = _mm_mul_ps(iq2,jq0);
 330
 331             /* EWALD ELECTROSTATICS */
 332
 333             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 334             ewrt             = _mm_mul_ps(r20,ewtabscale);
 335             ewitab           = _mm_cvttps_epi32(ewrt);
 336             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 337             ewitab           = _mm_slli_epi32(ewitab,2);
 338             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 339             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 340             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 341             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 342             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 343             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 344             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 345             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 346             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 347
 348             /* Update potential sum for this i atom from the interaction with this j atom. */
 349             velecsum         = _mm_add_ps(velecsum,velec);
 350
 351             fscal            = felec;
 352
 353             /* Calculate temporary vectorial force */
 354             tx               = _mm_mul_ps(fscal,dx20);
 355             ty               = _mm_mul_ps(fscal,dy20);
 356             tz               = _mm_mul_ps(fscal,dz20);
 357
 358             /* Update vectorial force */
 359             fix2             = _mm_add_ps(fix2,tx);
 360             fiy2             = _mm_add_ps(fiy2,ty);
 361             fiz2             = _mm_add_ps(fiz2,tz);
 362
 363             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 364                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 365                                                    tx,ty,tz);
 366
 367             /**************************
 368              * CALCULATE INTERACTIONS *
 369              **************************/
 370
 371             r30              = _mm_mul_ps(rsq30,rinv30);
 372
 373             /* Compute parameters for interactions between i and j atoms */
 374             qq30             = _mm_mul_ps(iq3,jq0);
 375
 376             /* EWALD ELECTROSTATICS */
 377
 378             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 379             ewrt             = _mm_mul_ps(r30,ewtabscale);
 380             ewitab           = _mm_cvttps_epi32(ewrt);
 381             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 382             ewitab           = _mm_slli_epi32(ewitab,2);
 383             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 384             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 385             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 386             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 387             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 388             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 389             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 390             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 391             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 392
 393             /* Update potential sum for this i atom from the interaction with this j atom. */
 394             velecsum         = _mm_add_ps(velecsum,velec);
 395
 396             fscal            = felec;
 397
 398             /* Calculate temporary vectorial force */
 399             tx               = _mm_mul_ps(fscal,dx30);
 400             ty               = _mm_mul_ps(fscal,dy30);
 401             tz               = _mm_mul_ps(fscal,dz30);
 402
 403             /* Update vectorial force */
 404             fix3             = _mm_add_ps(fix3,tx);
 405             fiy3             = _mm_add_ps(fiy3,ty);
 406             fiz3             = _mm_add_ps(fiz3,tz);
 407
 408             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 409                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 410                                                    tx,ty,tz);
 411
 412             /* Inner loop uses 155 flops */
 413         }
 414
 415         if(jidx<j_index_end)
 416         {
 417
 418             /* Get j neighbor index, and coordinate index */
 419             jnrA             = jjnr[jidx];
 420             jnrB             = jjnr[jidx+1];
 421             jnrC             = jjnr[jidx+2];
 422             jnrD             = jjnr[jidx+3];
 423
 424             /* Sign of each element will be negative for non-real atoms.
 425              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 426              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 427              */
 428             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 429             jnrA       = (jnrA>=0) ? jnrA : 0;
 430             jnrB       = (jnrB>=0) ? jnrB : 0;
 431             jnrC       = (jnrC>=0) ? jnrC : 0;
 432             jnrD       = (jnrD>=0) ? jnrD : 0;
 433
 434             j_coord_offsetA  = DIM*jnrA;
 435             j_coord_offsetB  = DIM*jnrB;
 436             j_coord_offsetC  = DIM*jnrC;
 437             j_coord_offsetD  = DIM*jnrD;
 438
 439             /* load j atom coordinates */
 440             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 441                                               x+j_coord_offsetC,x+j_coord_offsetD,
 442                                               &jx0,&jy0,&jz0);
 443
 444             /* Calculate displacement vector */
 445             dx00             = _mm_sub_ps(ix0,jx0);
 446             dy00             = _mm_sub_ps(iy0,jy0);
 447             dz00             = _mm_sub_ps(iz0,jz0);
 448             dx10             = _mm_sub_ps(ix1,jx0);
 449             dy10             = _mm_sub_ps(iy1,jy0);
 450             dz10             = _mm_sub_ps(iz1,jz0);
 451             dx20             = _mm_sub_ps(ix2,jx0);
 452             dy20             = _mm_sub_ps(iy2,jy0);
 453             dz20             = _mm_sub_ps(iz2,jz0);
 454             dx30             = _mm_sub_ps(ix3,jx0);
 455             dy30             = _mm_sub_ps(iy3,jy0);
 456             dz30             = _mm_sub_ps(iz3,jz0);
 457
 458             /* Calculate squared distance and things based on it */
 459             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 460             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 461             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 462             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 463
 464             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 465             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 466             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 467
 468             rinvsq00         = gmx_mm_inv_ps(rsq00);
 469             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 470             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 471             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 472
 473             /* Load parameters for j particles */
 474             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 475                                                               charge+jnrC+0,charge+jnrD+0);
 476             vdwjidx0A        = 2*vdwtype[jnrA+0];
 477             vdwjidx0B        = 2*vdwtype[jnrB+0];
 478             vdwjidx0C        = 2*vdwtype[jnrC+0];
 479             vdwjidx0D        = 2*vdwtype[jnrD+0];
 480
 481             /**************************
 482              * CALCULATE INTERACTIONS *
 483              **************************/
 484
 485             /* Compute parameters for interactions between i and j atoms */
 486             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 487                                          vdwparam+vdwioffset0+vdwjidx0B,
 488                                          vdwparam+vdwioffset0+vdwjidx0C,
 489                                          vdwparam+vdwioffset0+vdwjidx0D,
 490                                          &c6_00,&c12_00);
 491
 492             /* LENNARD-JONES DISPERSION/REPULSION */
 493
 494             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 495             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 496             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 497             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 498             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 499
 500             /* Update potential sum for this i atom from the interaction with this j atom. */
 501             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 502             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 503
 504             fscal            = fvdw;
 505
 506             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 507
 508             /* Calculate temporary vectorial force */
 509             tx               = _mm_mul_ps(fscal,dx00);
 510             ty               = _mm_mul_ps(fscal,dy00);
 511             tz               = _mm_mul_ps(fscal,dz00);
 512
 513             /* Update vectorial force */
 514             fix0             = _mm_add_ps(fix0,tx);
 515             fiy0             = _mm_add_ps(fiy0,ty);
 516             fiz0             = _mm_add_ps(fiz0,tz);
 517
 518             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 519                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 520                                                    tx,ty,tz);
 521
 522             /**************************
 523              * CALCULATE INTERACTIONS *
 524              **************************/
 525
 526             r10              = _mm_mul_ps(rsq10,rinv10);
 527             r10              = _mm_andnot_ps(dummy_mask,r10);
 528
 529             /* Compute parameters for interactions between i and j atoms */
 530             qq10             = _mm_mul_ps(iq1,jq0);
 531
 532             /* EWALD ELECTROSTATICS */
 533
 534             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 535             ewrt             = _mm_mul_ps(r10,ewtabscale);
 536             ewitab           = _mm_cvttps_epi32(ewrt);
 537             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 538             ewitab           = _mm_slli_epi32(ewitab,2);
 539             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 540             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 541             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 542             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 543             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 544             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 545             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 546             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 547             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 548
 549             /* Update potential sum for this i atom from the interaction with this j atom. */
 550             velec            = _mm_andnot_ps(dummy_mask,velec);
 551             velecsum         = _mm_add_ps(velecsum,velec);
 552
 553             fscal            = felec;
 554
 555             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 556
 557             /* Calculate temporary vectorial force */
 558             tx               = _mm_mul_ps(fscal,dx10);
 559             ty               = _mm_mul_ps(fscal,dy10);
 560             tz               = _mm_mul_ps(fscal,dz10);
 561
 562             /* Update vectorial force */
 563             fix1             = _mm_add_ps(fix1,tx);
 564             fiy1             = _mm_add_ps(fiy1,ty);
 565             fiz1             = _mm_add_ps(fiz1,tz);
 566
 567             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 568                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 569                                                    tx,ty,tz);
 570
 571             /**************************
 572              * CALCULATE INTERACTIONS *
 573              **************************/
 574
 575             r20              = _mm_mul_ps(rsq20,rinv20);
 576             r20              = _mm_andnot_ps(dummy_mask,r20);
 577
 578             /* Compute parameters for interactions between i and j atoms */
 579             qq20             = _mm_mul_ps(iq2,jq0);
 580
 581             /* EWALD ELECTROSTATICS */
 582
 583             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 584             ewrt             = _mm_mul_ps(r20,ewtabscale);
 585             ewitab           = _mm_cvttps_epi32(ewrt);
 586             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 587             ewitab           = _mm_slli_epi32(ewitab,2);
 588             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 589             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 590             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 591             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 592             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 593             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 594             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 595             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 596             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm_andnot_ps(dummy_mask,velec);
 600             velecsum         = _mm_add_ps(velecsum,velec);
 601
 602             fscal            = felec;
 603
 604             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 605
 606             /* Calculate temporary vectorial force */
 607             tx               = _mm_mul_ps(fscal,dx20);
 608             ty               = _mm_mul_ps(fscal,dy20);
 609             tz               = _mm_mul_ps(fscal,dz20);
 610
 611             /* Update vectorial force */
 612             fix2             = _mm_add_ps(fix2,tx);
 613             fiy2             = _mm_add_ps(fiy2,ty);
 614             fiz2             = _mm_add_ps(fiz2,tz);
 615
 616             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 617                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 618                                                    tx,ty,tz);
 619
 620             /**************************
 621              * CALCULATE INTERACTIONS *
 622              **************************/
 623
 624             r30              = _mm_mul_ps(rsq30,rinv30);
 625             r30              = _mm_andnot_ps(dummy_mask,r30);
 626
 627             /* Compute parameters for interactions between i and j atoms */
 628             qq30             = _mm_mul_ps(iq3,jq0);
 629
 630             /* EWALD ELECTROSTATICS */
 631
 632             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 633             ewrt             = _mm_mul_ps(r30,ewtabscale);
 634             ewitab           = _mm_cvttps_epi32(ewrt);
 635             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 636             ewitab           = _mm_slli_epi32(ewitab,2);
 637             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 638             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 639             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 640             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 641             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 642             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 643             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 644             velec            = _mm_mul_ps(qq30,_mm_sub_ps(rinv30,velec));
 645             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
 646
 647             /* Update potential sum for this i atom from the interaction with this j atom. */
 648             velec            = _mm_andnot_ps(dummy_mask,velec);
 649             velecsum         = _mm_add_ps(velecsum,velec);
 650
 651             fscal            = felec;
 652
 653             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 654
 655             /* Calculate temporary vectorial force */
 656             tx               = _mm_mul_ps(fscal,dx30);
 657             ty               = _mm_mul_ps(fscal,dy30);
 658             tz               = _mm_mul_ps(fscal,dz30);
 659
 660             /* Update vectorial force */
 661             fix3             = _mm_add_ps(fix3,tx);
 662             fiy3             = _mm_add_ps(fiy3,ty);
 663             fiz3             = _mm_add_ps(fiz3,tz);
 664
 665             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 666                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 667                                                    tx,ty,tz);
 668
 669             /* Inner loop uses 158 flops */
 670         }
 671
 672         /* End of innermost loop */
 673
 674         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 675                                               f+i_coord_offset,fshift+i_shift_offset);
 676
 677         ggid                        = gid[iidx];
 678         /* Update potential energies */
 679         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 680         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 681
 682         /* Increment number of inner iterations */
 683         inneriter                  += j_index_end - j_index_start;
 684
 685         /* Outer loop uses 38 flops */
 686     }
 687
 688     /* Increment number of outer iterations */
 689     outeriter        += nri;
 690
 691     /* Update outer/inner flops */
 692
 693     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*38 + inneriter*158);
 694 }
 695 /*
 696  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_single
 697  * Electrostatics interaction: Ewald
 698  * VdW interaction:            LennardJones
 699  * Geometry:                   Water4-Particle
 700  * Calculate force/pot:        Force
 701  */
 702 void
 703 nb_kernel_ElecEw_VdwLJ_GeomW4P1_F_sse2_single
 704                     (t_nblist * gmx_restrict                nlist,
 705                      rvec * gmx_restrict                    xx,
 706                      rvec * gmx_restrict                    ff,
 707                      t_forcerec * gmx_restrict              fr,
 708                      t_mdatoms * gmx_restrict               mdatoms,
 709                      nb_kernel_data_t * gmx_restrict        kernel_data,
 710                      t_nrnb * gmx_restrict                  nrnb)
 711 {
 712     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 713      * just 0 for non-waters.
 714      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 715      * jnr indices corresponding to data put in the four positions in the SIMD register.
 716      */
 717     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 718     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 719     int              jnrA,jnrB,jnrC,jnrD;
 720     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 721     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 722     real             shX,shY,shZ,rcutoff_scalar;
 723     real             *shiftvec,*fshift,*x,*f;
 724     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 725     int              vdwioffset0;
 726     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 727     int              vdwioffset1;
 728     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 729     int              vdwioffset2;
 730     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 731     int              vdwioffset3;
 732     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 733     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 734     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 735     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 736     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 737     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 738     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 739     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 740     real             *charge;
 741     int              nvdwtype;
 742     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 743     int              *vdwtype;
 744     real             *vdwparam;
 745     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 746     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 747     __m128i          ewitab;
 748     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 749     real             *ewtab;
 750     __m128           dummy_mask,cutoff_mask;
 751     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 752     __m128           one     = _mm_set1_ps(1.0);
 753     __m128           two     = _mm_set1_ps(2.0);
 754     x                = xx[0];
 755     f                = ff[0];
 756
 757     nri              = nlist->nri;
 758     iinr             = nlist->iinr;
 759     jindex           = nlist->jindex;
 760     jjnr             = nlist->jjnr;
 761     shiftidx         = nlist->shift;
 762     gid              = nlist->gid;
 763     shiftvec         = fr->shift_vec[0];
 764     fshift           = fr->fshift[0];
 765     facel            = _mm_set1_ps(fr->epsfac);
 766     charge           = mdatoms->chargeA;
 767     nvdwtype         = fr->ntype;
 768     vdwparam         = fr->nbfp;
 769     vdwtype          = mdatoms->typeA;
 770
 771     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 772     ewtab            = fr->ic->tabq_coul_F;
 773     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 774     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 775
 776     /* Setup water-specific parameters */
 777     inr              = nlist->iinr[0];
 778     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 779     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 780     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 781     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 782
 783     /* Avoid stupid compiler warnings */
 784     jnrA = jnrB = jnrC = jnrD = 0;
 785     j_coord_offsetA = 0;
 786     j_coord_offsetB = 0;
 787     j_coord_offsetC = 0;
 788     j_coord_offsetD = 0;
 789
 790     outeriter        = 0;
 791     inneriter        = 0;
 792
 793     /* Start outer loop over neighborlists */
 794     for(iidx=0; iidx<nri; iidx++)
 795     {
 796         /* Load shift vector for this list */
 797         i_shift_offset   = DIM*shiftidx[iidx];
 798         shX              = shiftvec[i_shift_offset+XX];
 799         shY              = shiftvec[i_shift_offset+YY];
 800         shZ              = shiftvec[i_shift_offset+ZZ];
 801
 802         /* Load limits for loop over neighbors */
 803         j_index_start    = jindex[iidx];
 804         j_index_end      = jindex[iidx+1];
 805
 806         /* Get outer coordinate index */
 807         inr              = iinr[iidx];
 808         i_coord_offset   = DIM*inr;
 809
 810         /* Load i particle coords and add shift vector */
 811         ix0              = _mm_set1_ps(shX + x[i_coord_offset+DIM*0+XX]);
 812         iy0              = _mm_set1_ps(shY + x[i_coord_offset+DIM*0+YY]);
 813         iz0              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*0+ZZ]);
 814         ix1              = _mm_set1_ps(shX + x[i_coord_offset+DIM*1+XX]);
 815         iy1              = _mm_set1_ps(shY + x[i_coord_offset+DIM*1+YY]);
 816         iz1              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*1+ZZ]);
 817         ix2              = _mm_set1_ps(shX + x[i_coord_offset+DIM*2+XX]);
 818         iy2              = _mm_set1_ps(shY + x[i_coord_offset+DIM*2+YY]);
 819         iz2              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*2+ZZ]);
 820         ix3              = _mm_set1_ps(shX + x[i_coord_offset+DIM*3+XX]);
 821         iy3              = _mm_set1_ps(shY + x[i_coord_offset+DIM*3+YY]);
 822         iz3              = _mm_set1_ps(shZ + x[i_coord_offset+DIM*3+ZZ]);
 823
 824         fix0             = _mm_setzero_ps();
 825         fiy0             = _mm_setzero_ps();
 826         fiz0             = _mm_setzero_ps();
 827         fix1             = _mm_setzero_ps();
 828         fiy1             = _mm_setzero_ps();
 829         fiz1             = _mm_setzero_ps();
 830         fix2             = _mm_setzero_ps();
 831         fiy2             = _mm_setzero_ps();
 832         fiz2             = _mm_setzero_ps();
 833         fix3             = _mm_setzero_ps();
 834         fiy3             = _mm_setzero_ps();
 835         fiz3             = _mm_setzero_ps();
 836
 837         /* Start inner kernel loop */
 838         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 839         {
 840
 841             /* Get j neighbor index, and coordinate index */
 842             jnrA             = jjnr[jidx];
 843             jnrB             = jjnr[jidx+1];
 844             jnrC             = jjnr[jidx+2];
 845             jnrD             = jjnr[jidx+3];
 846
 847             j_coord_offsetA  = DIM*jnrA;
 848             j_coord_offsetB  = DIM*jnrB;
 849             j_coord_offsetC  = DIM*jnrC;
 850             j_coord_offsetD  = DIM*jnrD;
 851
 852             /* load j atom coordinates */
 853             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 854                                               x+j_coord_offsetC,x+j_coord_offsetD,
 855                                               &jx0,&jy0,&jz0);
 856
 857             /* Calculate displacement vector */
 858             dx00             = _mm_sub_ps(ix0,jx0);
 859             dy00             = _mm_sub_ps(iy0,jy0);
 860             dz00             = _mm_sub_ps(iz0,jz0);
 861             dx10             = _mm_sub_ps(ix1,jx0);
 862             dy10             = _mm_sub_ps(iy1,jy0);
 863             dz10             = _mm_sub_ps(iz1,jz0);
 864             dx20             = _mm_sub_ps(ix2,jx0);
 865             dy20             = _mm_sub_ps(iy2,jy0);
 866             dz20             = _mm_sub_ps(iz2,jz0);
 867             dx30             = _mm_sub_ps(ix3,jx0);
 868             dy30             = _mm_sub_ps(iy3,jy0);
 869             dz30             = _mm_sub_ps(iz3,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 873             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 874             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 875             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 876
 877             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 878             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 879             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 880
 881             rinvsq00         = gmx_mm_inv_ps(rsq00);
 882             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 883             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 884             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
 885
 886             /* Load parameters for j particles */
 887             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 888                                                               charge+jnrC+0,charge+jnrD+0);
 889             vdwjidx0A        = 2*vdwtype[jnrA+0];
 890             vdwjidx0B        = 2*vdwtype[jnrB+0];
 891             vdwjidx0C        = 2*vdwtype[jnrC+0];
 892             vdwjidx0D        = 2*vdwtype[jnrD+0];
 893
 894             /**************************
 895              * CALCULATE INTERACTIONS *
 896              **************************/
 897
 898             /* Compute parameters for interactions between i and j atoms */
 899             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 900                                          vdwparam+vdwioffset0+vdwjidx0B,
 901                                          vdwparam+vdwioffset0+vdwjidx0C,
 902                                          vdwparam+vdwioffset0+vdwjidx0D,
 903                                          &c6_00,&c12_00);
 904
 905             /* LENNARD-JONES DISPERSION/REPULSION */
 906
 907             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 908             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 909
 910             fscal            = fvdw;
 911
 912             /* Calculate temporary vectorial force */
 913             tx               = _mm_mul_ps(fscal,dx00);
 914             ty               = _mm_mul_ps(fscal,dy00);
 915             tz               = _mm_mul_ps(fscal,dz00);
 916
 917             /* Update vectorial force */
 918             fix0             = _mm_add_ps(fix0,tx);
 919             fiy0             = _mm_add_ps(fiy0,ty);
 920             fiz0             = _mm_add_ps(fiz0,tz);
 921
 922             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 923                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 924                                                    tx,ty,tz);
 925
 926             /**************************
 927              * CALCULATE INTERACTIONS *
 928              **************************/
 929
 930             r10              = _mm_mul_ps(rsq10,rinv10);
 931
 932             /* Compute parameters for interactions between i and j atoms */
 933             qq10             = _mm_mul_ps(iq1,jq0);
 934
 935             /* EWALD ELECTROSTATICS */
 936
 937             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 938             ewrt             = _mm_mul_ps(r10,ewtabscale);
 939             ewitab           = _mm_cvttps_epi32(ewrt);
 940             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 941             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 942                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 943                                          &ewtabF,&ewtabFn);
 944             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 945             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 946
 947             fscal            = felec;
 948
 949             /* Calculate temporary vectorial force */
 950             tx               = _mm_mul_ps(fscal,dx10);
 951             ty               = _mm_mul_ps(fscal,dy10);
 952             tz               = _mm_mul_ps(fscal,dz10);
 953
 954             /* Update vectorial force */
 955             fix1             = _mm_add_ps(fix1,tx);
 956             fiy1             = _mm_add_ps(fiy1,ty);
 957             fiz1             = _mm_add_ps(fiz1,tz);
 958
 959             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 960                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 961                                                    tx,ty,tz);
 962
 963             /**************************
 964              * CALCULATE INTERACTIONS *
 965              **************************/
 966
 967             r20              = _mm_mul_ps(rsq20,rinv20);
 968
 969             /* Compute parameters for interactions between i and j atoms */
 970             qq20             = _mm_mul_ps(iq2,jq0);
 971
 972             /* EWALD ELECTROSTATICS */
 973
 974             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 975             ewrt             = _mm_mul_ps(r20,ewtabscale);
 976             ewitab           = _mm_cvttps_epi32(ewrt);
 977             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
 978             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
 979                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
 980                                          &ewtabF,&ewtabFn);
 981             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 982             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 983
 984             fscal            = felec;
 985
 986             /* Calculate temporary vectorial force */
 987             tx               = _mm_mul_ps(fscal,dx20);
 988             ty               = _mm_mul_ps(fscal,dy20);
 989             tz               = _mm_mul_ps(fscal,dz20);
 990
 991             /* Update vectorial force */
 992             fix2             = _mm_add_ps(fix2,tx);
 993             fiy2             = _mm_add_ps(fiy2,ty);
 994             fiz2             = _mm_add_ps(fiz2,tz);
 995
 996             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
 997                                                    f+j_coord_offsetC,f+j_coord_offsetD,
 998                                                    tx,ty,tz);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r30              = _mm_mul_ps(rsq30,rinv30);
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             /* Calculate temporary vectorial force */
1024             tx               = _mm_mul_ps(fscal,dx30);
1025             ty               = _mm_mul_ps(fscal,dy30);
1026             tz               = _mm_mul_ps(fscal,dz30);
1027
1028             /* Update vectorial force */
1029             fix3             = _mm_add_ps(fix3,tx);
1030             fiy3             = _mm_add_ps(fiy3,ty);
1031             fiz3             = _mm_add_ps(fiz3,tz);
1032
1033             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1034                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1035                                                    tx,ty,tz);
1036
1037             /* Inner loop uses 135 flops */
1038         }
1039
1040         if(jidx<j_index_end)
1041         {
1042
1043             /* Get j neighbor index, and coordinate index */
1044             jnrA             = jjnr[jidx];
1045             jnrB             = jjnr[jidx+1];
1046             jnrC             = jjnr[jidx+2];
1047             jnrD             = jjnr[jidx+3];
1048
1049             /* Sign of each element will be negative for non-real atoms.
1050              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1051              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1052              */
1053             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1054             jnrA       = (jnrA>=0) ? jnrA : 0;
1055             jnrB       = (jnrB>=0) ? jnrB : 0;
1056             jnrC       = (jnrC>=0) ? jnrC : 0;
1057             jnrD       = (jnrD>=0) ? jnrD : 0;
1058
1059             j_coord_offsetA  = DIM*jnrA;
1060             j_coord_offsetB  = DIM*jnrB;
1061             j_coord_offsetC  = DIM*jnrC;
1062             j_coord_offsetD  = DIM*jnrD;
1063
1064             /* load j atom coordinates */
1065             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1066                                               x+j_coord_offsetC,x+j_coord_offsetD,
1067                                               &jx0,&jy0,&jz0);
1068
1069             /* Calculate displacement vector */
1070             dx00             = _mm_sub_ps(ix0,jx0);
1071             dy00             = _mm_sub_ps(iy0,jy0);
1072             dz00             = _mm_sub_ps(iz0,jz0);
1073             dx10             = _mm_sub_ps(ix1,jx0);
1074             dy10             = _mm_sub_ps(iy1,jy0);
1075             dz10             = _mm_sub_ps(iz1,jz0);
1076             dx20             = _mm_sub_ps(ix2,jx0);
1077             dy20             = _mm_sub_ps(iy2,jy0);
1078             dz20             = _mm_sub_ps(iz2,jz0);
1079             dx30             = _mm_sub_ps(ix3,jx0);
1080             dy30             = _mm_sub_ps(iy3,jy0);
1081             dz30             = _mm_sub_ps(iz3,jz0);
1082
1083             /* Calculate squared distance and things based on it */
1084             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1085             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1086             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1087             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1088
1089             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1090             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1091             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1092
1093             rinvsq00         = gmx_mm_inv_ps(rsq00);
1094             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1095             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1096             rinvsq30         = _mm_mul_ps(rinv30,rinv30);
1097
1098             /* Load parameters for j particles */
1099             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1100                                                               charge+jnrC+0,charge+jnrD+0);
1101             vdwjidx0A        = 2*vdwtype[jnrA+0];
1102             vdwjidx0B        = 2*vdwtype[jnrB+0];
1103             vdwjidx0C        = 2*vdwtype[jnrC+0];
1104             vdwjidx0D        = 2*vdwtype[jnrD+0];
1105
1106             /**************************
1107              * CALCULATE INTERACTIONS *
1108              **************************/
1109
1110             /* Compute parameters for interactions between i and j atoms */
1111             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1112                                          vdwparam+vdwioffset0+vdwjidx0B,
1113                                          vdwparam+vdwioffset0+vdwjidx0C,
1114                                          vdwparam+vdwioffset0+vdwjidx0D,
1115                                          &c6_00,&c12_00);
1116
1117             /* LENNARD-JONES DISPERSION/REPULSION */
1118
1119             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1120             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1121
1122             fscal            = fvdw;
1123
1124             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1125
1126             /* Calculate temporary vectorial force */
1127             tx               = _mm_mul_ps(fscal,dx00);
1128             ty               = _mm_mul_ps(fscal,dy00);
1129             tz               = _mm_mul_ps(fscal,dz00);
1130
1131             /* Update vectorial force */
1132             fix0             = _mm_add_ps(fix0,tx);
1133             fiy0             = _mm_add_ps(fiy0,ty);
1134             fiz0             = _mm_add_ps(fiz0,tz);
1135
1136             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1137                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1138                                                    tx,ty,tz);
1139
1140             /**************************
1141              * CALCULATE INTERACTIONS *
1142              **************************/
1143
1144             r10              = _mm_mul_ps(rsq10,rinv10);
1145             r10              = _mm_andnot_ps(dummy_mask,r10);
1146
1147             /* Compute parameters for interactions between i and j atoms */
1148             qq10             = _mm_mul_ps(iq1,jq0);
1149
1150             /* EWALD ELECTROSTATICS */
1151
1152             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1153             ewrt             = _mm_mul_ps(r10,ewtabscale);
1154             ewitab           = _mm_cvttps_epi32(ewrt);
1155             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1156             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1157                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1158                                          &ewtabF,&ewtabFn);
1159             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1160             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1161
1162             fscal            = felec;
1163
1164             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1165
1166             /* Calculate temporary vectorial force */
1167             tx               = _mm_mul_ps(fscal,dx10);
1168             ty               = _mm_mul_ps(fscal,dy10);
1169             tz               = _mm_mul_ps(fscal,dz10);
1170
1171             /* Update vectorial force */
1172             fix1             = _mm_add_ps(fix1,tx);
1173             fiy1             = _mm_add_ps(fiy1,ty);
1174             fiz1             = _mm_add_ps(fiz1,tz);
1175
1176             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1177                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1178                                                    tx,ty,tz);
1179
1180             /**************************
1181              * CALCULATE INTERACTIONS *
1182              **************************/
1183
1184             r20              = _mm_mul_ps(rsq20,rinv20);
1185             r20              = _mm_andnot_ps(dummy_mask,r20);
1186
1187             /* Compute parameters for interactions between i and j atoms */
1188             qq20             = _mm_mul_ps(iq2,jq0);
1189
1190             /* EWALD ELECTROSTATICS */
1191
1192             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1193             ewrt             = _mm_mul_ps(r20,ewtabscale);
1194             ewitab           = _mm_cvttps_epi32(ewrt);
1195             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1196             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1197                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1198                                          &ewtabF,&ewtabFn);
1199             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1200             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1201
1202             fscal            = felec;
1203
1204             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1205
1206             /* Calculate temporary vectorial force */
1207             tx               = _mm_mul_ps(fscal,dx20);
1208             ty               = _mm_mul_ps(fscal,dy20);
1209             tz               = _mm_mul_ps(fscal,dz20);
1210
1211             /* Update vectorial force */
1212             fix2             = _mm_add_ps(fix2,tx);
1213             fiy2             = _mm_add_ps(fiy2,ty);
1214             fiz2             = _mm_add_ps(fiz2,tz);
1215
1216             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1217                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1218                                                    tx,ty,tz);
1219
1220             /**************************
1221              * CALCULATE INTERACTIONS *
1222              **************************/
1223
1224             r30              = _mm_mul_ps(rsq30,rinv30);
1225             r30              = _mm_andnot_ps(dummy_mask,r30);
1226
1227             /* Compute parameters for interactions between i and j atoms */
1228             qq30             = _mm_mul_ps(iq3,jq0);
1229
1230             /* EWALD ELECTROSTATICS */
1231
1232             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1233             ewrt             = _mm_mul_ps(r30,ewtabscale);
1234             ewitab           = _mm_cvttps_epi32(ewrt);
1235             eweps            = _mm_sub_ps(ewrt,_mm_cvtepi32_ps(ewitab));
1236             gmx_mm_load_4pair_swizzle_ps(ewtab+gmx_mm_extract_epi32(ewitab,0),ewtab+gmx_mm_extract_epi32(ewitab,1),
1237                                          ewtab+gmx_mm_extract_epi32(ewitab,2),ewtab+gmx_mm_extract_epi32(ewitab,3),
1238                                          &ewtabF,&ewtabFn);
1239             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1240             felec            = _mm_mul_ps(_mm_mul_ps(qq30,rinv30),_mm_sub_ps(rinvsq30,felec));
1241
1242             fscal            = felec;
1243
1244             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1245
1246             /* Calculate temporary vectorial force */
1247             tx               = _mm_mul_ps(fscal,dx30);
1248             ty               = _mm_mul_ps(fscal,dy30);
1249             tz               = _mm_mul_ps(fscal,dz30);
1250
1251             /* Update vectorial force */
1252             fix3             = _mm_add_ps(fix3,tx);
1253             fiy3             = _mm_add_ps(fiy3,ty);
1254             fiz3             = _mm_add_ps(fiz3,tz);
1255
1256             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(f+j_coord_offsetA,f+j_coord_offsetB,
1257                                                    f+j_coord_offsetC,f+j_coord_offsetD,
1258                                                    tx,ty,tz);
1259
1260             /* Inner loop uses 138 flops */
1261         }
1262
1263         /* End of innermost loop */
1264
1265         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1266                                               f+i_coord_offset,fshift+i_shift_offset);
1267
1268         /* Increment number of inner iterations */
1269         inneriter                  += j_index_end - j_index_start;
1270
1271         /* Outer loop uses 36 flops */
1272     }
1273
1274     /* Increment number of outer iterations */
1275     outeriter        += nri;
1276
1277     /* Update outer/inner flops */
1278
1279     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*36 + inneriter*138);
1280 }