src/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwLJ_GeomW3P1_sse4_1_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs sse4_1_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_sse4_1_single.h"
  34 #include "kernelutil_x86_sse4_1_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_VF_sse4_1_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_sse4_1_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              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  62     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  63     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  64     real             rcutoff_scalar;
  65     real             *shiftvec,*fshift,*x,*f;
  66     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  67     real             scratch[4*DIM];
  68     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  69     int              vdwioffset0;
  70     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  71     int              vdwioffset1;
  72     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  73     int              vdwioffset2;
  74     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  75     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  76     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  77     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  78     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  79     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  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     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 120     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 121     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 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     for(iidx=0;iidx<4*DIM;iidx++)
 135     {
 136         scratch[iidx] = 0.0;
 137     }
 138
 139     /* Start outer loop over neighborlists */
 140     for(iidx=0; iidx<nri; iidx++)
 141     {
 142         /* Load shift vector for this list */
 143         i_shift_offset   = DIM*shiftidx[iidx];
 144
 145         /* Load limits for loop over neighbors */
 146         j_index_start    = jindex[iidx];
 147         j_index_end      = jindex[iidx+1];
 148
 149         /* Get outer coordinate index */
 150         inr              = iinr[iidx];
 151         i_coord_offset   = DIM*inr;
 152
 153         /* Load i particle coords and add shift vector */
 154         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 155                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 156
 157         fix0             = _mm_setzero_ps();
 158         fiy0             = _mm_setzero_ps();
 159         fiz0             = _mm_setzero_ps();
 160         fix1             = _mm_setzero_ps();
 161         fiy1             = _mm_setzero_ps();
 162         fiz1             = _mm_setzero_ps();
 163         fix2             = _mm_setzero_ps();
 164         fiy2             = _mm_setzero_ps();
 165         fiz2             = _mm_setzero_ps();
 166
 167         /* Reset potential sums */
 168         velecsum         = _mm_setzero_ps();
 169         vvdwsum          = _mm_setzero_ps();
 170
 171         /* Start inner kernel loop */
 172         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 173         {
 174
 175             /* Get j neighbor index, and coordinate index */
 176             jnrA             = jjnr[jidx];
 177             jnrB             = jjnr[jidx+1];
 178             jnrC             = jjnr[jidx+2];
 179             jnrD             = jjnr[jidx+3];
 180             j_coord_offsetA  = DIM*jnrA;
 181             j_coord_offsetB  = DIM*jnrB;
 182             j_coord_offsetC  = DIM*jnrC;
 183             j_coord_offsetD  = DIM*jnrD;
 184
 185             /* load j atom coordinates */
 186             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 187                                               x+j_coord_offsetC,x+j_coord_offsetD,
 188                                               &jx0,&jy0,&jz0);
 189
 190             /* Calculate displacement vector */
 191             dx00             = _mm_sub_ps(ix0,jx0);
 192             dy00             = _mm_sub_ps(iy0,jy0);
 193             dz00             = _mm_sub_ps(iz0,jz0);
 194             dx10             = _mm_sub_ps(ix1,jx0);
 195             dy10             = _mm_sub_ps(iy1,jy0);
 196             dz10             = _mm_sub_ps(iz1,jz0);
 197             dx20             = _mm_sub_ps(ix2,jx0);
 198             dy20             = _mm_sub_ps(iy2,jy0);
 199             dz20             = _mm_sub_ps(iz2,jz0);
 200
 201             /* Calculate squared distance and things based on it */
 202             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 203             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 204             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 205
 206             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 207             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 208             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 209
 210             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 211             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 212             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 213
 214             /* Load parameters for j particles */
 215             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 216                                                               charge+jnrC+0,charge+jnrD+0);
 217             vdwjidx0A        = 2*vdwtype[jnrA+0];
 218             vdwjidx0B        = 2*vdwtype[jnrB+0];
 219             vdwjidx0C        = 2*vdwtype[jnrC+0];
 220             vdwjidx0D        = 2*vdwtype[jnrD+0];
 221
 222             fjx0             = _mm_setzero_ps();
 223             fjy0             = _mm_setzero_ps();
 224             fjz0             = _mm_setzero_ps();
 225
 226             /**************************
 227              * CALCULATE INTERACTIONS *
 228              **************************/
 229
 230             r00              = _mm_mul_ps(rsq00,rinv00);
 231
 232             /* Compute parameters for interactions between i and j atoms */
 233             qq00             = _mm_mul_ps(iq0,jq0);
 234             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 235                                          vdwparam+vdwioffset0+vdwjidx0B,
 236                                          vdwparam+vdwioffset0+vdwjidx0C,
 237                                          vdwparam+vdwioffset0+vdwjidx0D,
 238                                          &c6_00,&c12_00);
 239
 240             /* EWALD ELECTROSTATICS */
 241
 242             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 243             ewrt             = _mm_mul_ps(r00,ewtabscale);
 244             ewitab           = _mm_cvttps_epi32(ewrt);
 245             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 246             ewitab           = _mm_slli_epi32(ewitab,2);
 247             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 248             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 249             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 250             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 251             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 252             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 253             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 254             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 255             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 256
 257             /* LENNARD-JONES DISPERSION/REPULSION */
 258
 259             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 260             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 261             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 262             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 263             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 264
 265             /* Update potential sum for this i atom from the interaction with this j atom. */
 266             velecsum         = _mm_add_ps(velecsum,velec);
 267             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 268
 269             fscal            = _mm_add_ps(felec,fvdw);
 270
 271             /* Calculate temporary vectorial force */
 272             tx               = _mm_mul_ps(fscal,dx00);
 273             ty               = _mm_mul_ps(fscal,dy00);
 274             tz               = _mm_mul_ps(fscal,dz00);
 275
 276             /* Update vectorial force */
 277             fix0             = _mm_add_ps(fix0,tx);
 278             fiy0             = _mm_add_ps(fiy0,ty);
 279             fiz0             = _mm_add_ps(fiz0,tz);
 280
 281             fjx0             = _mm_add_ps(fjx0,tx);
 282             fjy0             = _mm_add_ps(fjy0,ty);
 283             fjz0             = _mm_add_ps(fjz0,tz);
 284
 285             /**************************
 286              * CALCULATE INTERACTIONS *
 287              **************************/
 288
 289             r10              = _mm_mul_ps(rsq10,rinv10);
 290
 291             /* Compute parameters for interactions between i and j atoms */
 292             qq10             = _mm_mul_ps(iq1,jq0);
 293
 294             /* EWALD ELECTROSTATICS */
 295
 296             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 297             ewrt             = _mm_mul_ps(r10,ewtabscale);
 298             ewitab           = _mm_cvttps_epi32(ewrt);
 299             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 300             ewitab           = _mm_slli_epi32(ewitab,2);
 301             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 302             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 303             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 304             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 305             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 306             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 307             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 308             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 309             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 310
 311             /* Update potential sum for this i atom from the interaction with this j atom. */
 312             velecsum         = _mm_add_ps(velecsum,velec);
 313
 314             fscal            = felec;
 315
 316             /* Calculate temporary vectorial force */
 317             tx               = _mm_mul_ps(fscal,dx10);
 318             ty               = _mm_mul_ps(fscal,dy10);
 319             tz               = _mm_mul_ps(fscal,dz10);
 320
 321             /* Update vectorial force */
 322             fix1             = _mm_add_ps(fix1,tx);
 323             fiy1             = _mm_add_ps(fiy1,ty);
 324             fiz1             = _mm_add_ps(fiz1,tz);
 325
 326             fjx0             = _mm_add_ps(fjx0,tx);
 327             fjy0             = _mm_add_ps(fjy0,ty);
 328             fjz0             = _mm_add_ps(fjz0,tz);
 329
 330             /**************************
 331              * CALCULATE INTERACTIONS *
 332              **************************/
 333
 334             r20              = _mm_mul_ps(rsq20,rinv20);
 335
 336             /* Compute parameters for interactions between i and j atoms */
 337             qq20             = _mm_mul_ps(iq2,jq0);
 338
 339             /* EWALD ELECTROSTATICS */
 340
 341             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 342             ewrt             = _mm_mul_ps(r20,ewtabscale);
 343             ewitab           = _mm_cvttps_epi32(ewrt);
 344             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 345             ewitab           = _mm_slli_epi32(ewitab,2);
 346             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 347             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 348             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 349             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 350             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 351             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 352             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 353             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 354             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 355
 356             /* Update potential sum for this i atom from the interaction with this j atom. */
 357             velecsum         = _mm_add_ps(velecsum,velec);
 358
 359             fscal            = felec;
 360
 361             /* Calculate temporary vectorial force */
 362             tx               = _mm_mul_ps(fscal,dx20);
 363             ty               = _mm_mul_ps(fscal,dy20);
 364             tz               = _mm_mul_ps(fscal,dz20);
 365
 366             /* Update vectorial force */
 367             fix2             = _mm_add_ps(fix2,tx);
 368             fiy2             = _mm_add_ps(fiy2,ty);
 369             fiz2             = _mm_add_ps(fiz2,tz);
 370
 371             fjx0             = _mm_add_ps(fjx0,tx);
 372             fjy0             = _mm_add_ps(fjy0,ty);
 373             fjz0             = _mm_add_ps(fjz0,tz);
 374
 375             fjptrA             = f+j_coord_offsetA;
 376             fjptrB             = f+j_coord_offsetB;
 377             fjptrC             = f+j_coord_offsetC;
 378             fjptrD             = f+j_coord_offsetD;
 379
 380             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 381
 382             /* Inner loop uses 135 flops */
 383         }
 384
 385         if(jidx<j_index_end)
 386         {
 387
 388             /* Get j neighbor index, and coordinate index */
 389             jnrlistA         = jjnr[jidx];
 390             jnrlistB         = jjnr[jidx+1];
 391             jnrlistC         = jjnr[jidx+2];
 392             jnrlistD         = jjnr[jidx+3];
 393             /* Sign of each element will be negative for non-real atoms.
 394              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 395              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 396              */
 397             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 398             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 399             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 400             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 401             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 402             j_coord_offsetA  = DIM*jnrA;
 403             j_coord_offsetB  = DIM*jnrB;
 404             j_coord_offsetC  = DIM*jnrC;
 405             j_coord_offsetD  = DIM*jnrD;
 406
 407             /* load j atom coordinates */
 408             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 409                                               x+j_coord_offsetC,x+j_coord_offsetD,
 410                                               &jx0,&jy0,&jz0);
 411
 412             /* Calculate displacement vector */
 413             dx00             = _mm_sub_ps(ix0,jx0);
 414             dy00             = _mm_sub_ps(iy0,jy0);
 415             dz00             = _mm_sub_ps(iz0,jz0);
 416             dx10             = _mm_sub_ps(ix1,jx0);
 417             dy10             = _mm_sub_ps(iy1,jy0);
 418             dz10             = _mm_sub_ps(iz1,jz0);
 419             dx20             = _mm_sub_ps(ix2,jx0);
 420             dy20             = _mm_sub_ps(iy2,jy0);
 421             dz20             = _mm_sub_ps(iz2,jz0);
 422
 423             /* Calculate squared distance and things based on it */
 424             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 425             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 426             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 427
 428             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 429             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 430             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 431
 432             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 433             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 434             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 435
 436             /* Load parameters for j particles */
 437             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 438                                                               charge+jnrC+0,charge+jnrD+0);
 439             vdwjidx0A        = 2*vdwtype[jnrA+0];
 440             vdwjidx0B        = 2*vdwtype[jnrB+0];
 441             vdwjidx0C        = 2*vdwtype[jnrC+0];
 442             vdwjidx0D        = 2*vdwtype[jnrD+0];
 443
 444             fjx0             = _mm_setzero_ps();
 445             fjy0             = _mm_setzero_ps();
 446             fjz0             = _mm_setzero_ps();
 447
 448             /**************************
 449              * CALCULATE INTERACTIONS *
 450              **************************/
 451
 452             r00              = _mm_mul_ps(rsq00,rinv00);
 453             r00              = _mm_andnot_ps(dummy_mask,r00);
 454
 455             /* Compute parameters for interactions between i and j atoms */
 456             qq00             = _mm_mul_ps(iq0,jq0);
 457             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 458                                          vdwparam+vdwioffset0+vdwjidx0B,
 459                                          vdwparam+vdwioffset0+vdwjidx0C,
 460                                          vdwparam+vdwioffset0+vdwjidx0D,
 461                                          &c6_00,&c12_00);
 462
 463             /* EWALD ELECTROSTATICS */
 464
 465             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 466             ewrt             = _mm_mul_ps(r00,ewtabscale);
 467             ewitab           = _mm_cvttps_epi32(ewrt);
 468             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 469             ewitab           = _mm_slli_epi32(ewitab,2);
 470             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 471             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 472             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 473             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 474             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 475             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 476             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 477             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 478             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 479
 480             /* LENNARD-JONES DISPERSION/REPULSION */
 481
 482             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 483             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 484             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 485             vvdw             = _mm_sub_ps( _mm_mul_ps(vvdw12,one_twelfth) , _mm_mul_ps(vvdw6,one_sixth) );
 486             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 487
 488             /* Update potential sum for this i atom from the interaction with this j atom. */
 489             velec            = _mm_andnot_ps(dummy_mask,velec);
 490             velecsum         = _mm_add_ps(velecsum,velec);
 491             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 492             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 493
 494             fscal            = _mm_add_ps(felec,fvdw);
 495
 496             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 497
 498             /* Calculate temporary vectorial force */
 499             tx               = _mm_mul_ps(fscal,dx00);
 500             ty               = _mm_mul_ps(fscal,dy00);
 501             tz               = _mm_mul_ps(fscal,dz00);
 502
 503             /* Update vectorial force */
 504             fix0             = _mm_add_ps(fix0,tx);
 505             fiy0             = _mm_add_ps(fiy0,ty);
 506             fiz0             = _mm_add_ps(fiz0,tz);
 507
 508             fjx0             = _mm_add_ps(fjx0,tx);
 509             fjy0             = _mm_add_ps(fjy0,ty);
 510             fjz0             = _mm_add_ps(fjz0,tz);
 511
 512             /**************************
 513              * CALCULATE INTERACTIONS *
 514              **************************/
 515
 516             r10              = _mm_mul_ps(rsq10,rinv10);
 517             r10              = _mm_andnot_ps(dummy_mask,r10);
 518
 519             /* Compute parameters for interactions between i and j atoms */
 520             qq10             = _mm_mul_ps(iq1,jq0);
 521
 522             /* EWALD ELECTROSTATICS */
 523
 524             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 525             ewrt             = _mm_mul_ps(r10,ewtabscale);
 526             ewitab           = _mm_cvttps_epi32(ewrt);
 527             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 528             ewitab           = _mm_slli_epi32(ewitab,2);
 529             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 530             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 531             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 532             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 533             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 534             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 535             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 536             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 537             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 538
 539             /* Update potential sum for this i atom from the interaction with this j atom. */
 540             velec            = _mm_andnot_ps(dummy_mask,velec);
 541             velecsum         = _mm_add_ps(velecsum,velec);
 542
 543             fscal            = felec;
 544
 545             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 546
 547             /* Calculate temporary vectorial force */
 548             tx               = _mm_mul_ps(fscal,dx10);
 549             ty               = _mm_mul_ps(fscal,dy10);
 550             tz               = _mm_mul_ps(fscal,dz10);
 551
 552             /* Update vectorial force */
 553             fix1             = _mm_add_ps(fix1,tx);
 554             fiy1             = _mm_add_ps(fiy1,ty);
 555             fiz1             = _mm_add_ps(fiz1,tz);
 556
 557             fjx0             = _mm_add_ps(fjx0,tx);
 558             fjy0             = _mm_add_ps(fjy0,ty);
 559             fjz0             = _mm_add_ps(fjz0,tz);
 560
 561             /**************************
 562              * CALCULATE INTERACTIONS *
 563              **************************/
 564
 565             r20              = _mm_mul_ps(rsq20,rinv20);
 566             r20              = _mm_andnot_ps(dummy_mask,r20);
 567
 568             /* Compute parameters for interactions between i and j atoms */
 569             qq20             = _mm_mul_ps(iq2,jq0);
 570
 571             /* EWALD ELECTROSTATICS */
 572
 573             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 574             ewrt             = _mm_mul_ps(r20,ewtabscale);
 575             ewitab           = _mm_cvttps_epi32(ewrt);
 576             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 577             ewitab           = _mm_slli_epi32(ewitab,2);
 578             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 579             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 580             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 581             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 582             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 583             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 584             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 585             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 586             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 587
 588             /* Update potential sum for this i atom from the interaction with this j atom. */
 589             velec            = _mm_andnot_ps(dummy_mask,velec);
 590             velecsum         = _mm_add_ps(velecsum,velec);
 591
 592             fscal            = felec;
 593
 594             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 595
 596             /* Calculate temporary vectorial force */
 597             tx               = _mm_mul_ps(fscal,dx20);
 598             ty               = _mm_mul_ps(fscal,dy20);
 599             tz               = _mm_mul_ps(fscal,dz20);
 600
 601             /* Update vectorial force */
 602             fix2             = _mm_add_ps(fix2,tx);
 603             fiy2             = _mm_add_ps(fiy2,ty);
 604             fiz2             = _mm_add_ps(fiz2,tz);
 605
 606             fjx0             = _mm_add_ps(fjx0,tx);
 607             fjy0             = _mm_add_ps(fjy0,ty);
 608             fjz0             = _mm_add_ps(fjz0,tz);
 609
 610             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 611             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 612             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 613             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 614
 615             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 616
 617             /* Inner loop uses 138 flops */
 618         }
 619
 620         /* End of innermost loop */
 621
 622         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 623                                               f+i_coord_offset,fshift+i_shift_offset);
 624
 625         ggid                        = gid[iidx];
 626         /* Update potential energies */
 627         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 628         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 629
 630         /* Increment number of inner iterations */
 631         inneriter                  += j_index_end - j_index_start;
 632
 633         /* Outer loop uses 20 flops */
 634     }
 635
 636     /* Increment number of outer iterations */
 637     outeriter        += nri;
 638
 639     /* Update outer/inner flops */
 640
 641     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*138);
 642 }
 643 /*
 644  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_single
 645  * Electrostatics interaction: Ewald
 646  * VdW interaction:            LennardJones
 647  * Geometry:                   Water3-Particle
 648  * Calculate force/pot:        Force
 649  */
 650 void
 651 nb_kernel_ElecEw_VdwLJ_GeomW3P1_F_sse4_1_single
 652                     (t_nblist * gmx_restrict                nlist,
 653                      rvec * gmx_restrict                    xx,
 654                      rvec * gmx_restrict                    ff,
 655                      t_forcerec * gmx_restrict              fr,
 656                      t_mdatoms * gmx_restrict               mdatoms,
 657                      nb_kernel_data_t * gmx_restrict        kernel_data,
 658                      t_nrnb * gmx_restrict                  nrnb)
 659 {
 660     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 661      * just 0 for non-waters.
 662      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 663      * jnr indices corresponding to data put in the four positions in the SIMD register.
 664      */
 665     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 666     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 667     int              jnrA,jnrB,jnrC,jnrD;
 668     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 669     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 670     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 671     real             rcutoff_scalar;
 672     real             *shiftvec,*fshift,*x,*f;
 673     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 674     real             scratch[4*DIM];
 675     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 676     int              vdwioffset0;
 677     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 678     int              vdwioffset1;
 679     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 680     int              vdwioffset2;
 681     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 682     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 683     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 684     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 685     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 686     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 687     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 688     real             *charge;
 689     int              nvdwtype;
 690     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 691     int              *vdwtype;
 692     real             *vdwparam;
 693     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 694     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 695     __m128i          ewitab;
 696     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 697     real             *ewtab;
 698     __m128           dummy_mask,cutoff_mask;
 699     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 700     __m128           one     = _mm_set1_ps(1.0);
 701     __m128           two     = _mm_set1_ps(2.0);
 702     x                = xx[0];
 703     f                = ff[0];
 704
 705     nri              = nlist->nri;
 706     iinr             = nlist->iinr;
 707     jindex           = nlist->jindex;
 708     jjnr             = nlist->jjnr;
 709     shiftidx         = nlist->shift;
 710     gid              = nlist->gid;
 711     shiftvec         = fr->shift_vec[0];
 712     fshift           = fr->fshift[0];
 713     facel            = _mm_set1_ps(fr->epsfac);
 714     charge           = mdatoms->chargeA;
 715     nvdwtype         = fr->ntype;
 716     vdwparam         = fr->nbfp;
 717     vdwtype          = mdatoms->typeA;
 718
 719     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 720     ewtab            = fr->ic->tabq_coul_F;
 721     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 722     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 723
 724     /* Setup water-specific parameters */
 725     inr              = nlist->iinr[0];
 726     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 727     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 728     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 729     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 730
 731     /* Avoid stupid compiler warnings */
 732     jnrA = jnrB = jnrC = jnrD = 0;
 733     j_coord_offsetA = 0;
 734     j_coord_offsetB = 0;
 735     j_coord_offsetC = 0;
 736     j_coord_offsetD = 0;
 737
 738     outeriter        = 0;
 739     inneriter        = 0;
 740
 741     for(iidx=0;iidx<4*DIM;iidx++)
 742     {
 743         scratch[iidx] = 0.0;
 744     }
 745
 746     /* Start outer loop over neighborlists */
 747     for(iidx=0; iidx<nri; iidx++)
 748     {
 749         /* Load shift vector for this list */
 750         i_shift_offset   = DIM*shiftidx[iidx];
 751
 752         /* Load limits for loop over neighbors */
 753         j_index_start    = jindex[iidx];
 754         j_index_end      = jindex[iidx+1];
 755
 756         /* Get outer coordinate index */
 757         inr              = iinr[iidx];
 758         i_coord_offset   = DIM*inr;
 759
 760         /* Load i particle coords and add shift vector */
 761         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 762                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 763
 764         fix0             = _mm_setzero_ps();
 765         fiy0             = _mm_setzero_ps();
 766         fiz0             = _mm_setzero_ps();
 767         fix1             = _mm_setzero_ps();
 768         fiy1             = _mm_setzero_ps();
 769         fiz1             = _mm_setzero_ps();
 770         fix2             = _mm_setzero_ps();
 771         fiy2             = _mm_setzero_ps();
 772         fiz2             = _mm_setzero_ps();
 773
 774         /* Start inner kernel loop */
 775         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 776         {
 777
 778             /* Get j neighbor index, and coordinate index */
 779             jnrA             = jjnr[jidx];
 780             jnrB             = jjnr[jidx+1];
 781             jnrC             = jjnr[jidx+2];
 782             jnrD             = jjnr[jidx+3];
 783             j_coord_offsetA  = DIM*jnrA;
 784             j_coord_offsetB  = DIM*jnrB;
 785             j_coord_offsetC  = DIM*jnrC;
 786             j_coord_offsetD  = DIM*jnrD;
 787
 788             /* load j atom coordinates */
 789             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 790                                               x+j_coord_offsetC,x+j_coord_offsetD,
 791                                               &jx0,&jy0,&jz0);
 792
 793             /* Calculate displacement vector */
 794             dx00             = _mm_sub_ps(ix0,jx0);
 795             dy00             = _mm_sub_ps(iy0,jy0);
 796             dz00             = _mm_sub_ps(iz0,jz0);
 797             dx10             = _mm_sub_ps(ix1,jx0);
 798             dy10             = _mm_sub_ps(iy1,jy0);
 799             dz10             = _mm_sub_ps(iz1,jz0);
 800             dx20             = _mm_sub_ps(ix2,jx0);
 801             dy20             = _mm_sub_ps(iy2,jy0);
 802             dz20             = _mm_sub_ps(iz2,jz0);
 803
 804             /* Calculate squared distance and things based on it */
 805             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 806             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 807             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 808
 809             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 810             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 811             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 812
 813             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 814             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 815             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 816
 817             /* Load parameters for j particles */
 818             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 819                                                               charge+jnrC+0,charge+jnrD+0);
 820             vdwjidx0A        = 2*vdwtype[jnrA+0];
 821             vdwjidx0B        = 2*vdwtype[jnrB+0];
 822             vdwjidx0C        = 2*vdwtype[jnrC+0];
 823             vdwjidx0D        = 2*vdwtype[jnrD+0];
 824
 825             fjx0             = _mm_setzero_ps();
 826             fjy0             = _mm_setzero_ps();
 827             fjz0             = _mm_setzero_ps();
 828
 829             /**************************
 830              * CALCULATE INTERACTIONS *
 831              **************************/
 832
 833             r00              = _mm_mul_ps(rsq00,rinv00);
 834
 835             /* Compute parameters for interactions between i and j atoms */
 836             qq00             = _mm_mul_ps(iq0,jq0);
 837             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 838                                          vdwparam+vdwioffset0+vdwjidx0B,
 839                                          vdwparam+vdwioffset0+vdwjidx0C,
 840                                          vdwparam+vdwioffset0+vdwjidx0D,
 841                                          &c6_00,&c12_00);
 842
 843             /* EWALD ELECTROSTATICS */
 844
 845             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 846             ewrt             = _mm_mul_ps(r00,ewtabscale);
 847             ewitab           = _mm_cvttps_epi32(ewrt);
 848             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 849             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 850                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 851                                          &ewtabF,&ewtabFn);
 852             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 853             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 854
 855             /* LENNARD-JONES DISPERSION/REPULSION */
 856
 857             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 858             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 859
 860             fscal            = _mm_add_ps(felec,fvdw);
 861
 862             /* Calculate temporary vectorial force */
 863             tx               = _mm_mul_ps(fscal,dx00);
 864             ty               = _mm_mul_ps(fscal,dy00);
 865             tz               = _mm_mul_ps(fscal,dz00);
 866
 867             /* Update vectorial force */
 868             fix0             = _mm_add_ps(fix0,tx);
 869             fiy0             = _mm_add_ps(fiy0,ty);
 870             fiz0             = _mm_add_ps(fiz0,tz);
 871
 872             fjx0             = _mm_add_ps(fjx0,tx);
 873             fjy0             = _mm_add_ps(fjy0,ty);
 874             fjz0             = _mm_add_ps(fjz0,tz);
 875
 876             /**************************
 877              * CALCULATE INTERACTIONS *
 878              **************************/
 879
 880             r10              = _mm_mul_ps(rsq10,rinv10);
 881
 882             /* Compute parameters for interactions between i and j atoms */
 883             qq10             = _mm_mul_ps(iq1,jq0);
 884
 885             /* EWALD ELECTROSTATICS */
 886
 887             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 888             ewrt             = _mm_mul_ps(r10,ewtabscale);
 889             ewitab           = _mm_cvttps_epi32(ewrt);
 890             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 891             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 892                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 893                                          &ewtabF,&ewtabFn);
 894             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 895             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 896
 897             fscal            = felec;
 898
 899             /* Calculate temporary vectorial force */
 900             tx               = _mm_mul_ps(fscal,dx10);
 901             ty               = _mm_mul_ps(fscal,dy10);
 902             tz               = _mm_mul_ps(fscal,dz10);
 903
 904             /* Update vectorial force */
 905             fix1             = _mm_add_ps(fix1,tx);
 906             fiy1             = _mm_add_ps(fiy1,ty);
 907             fiz1             = _mm_add_ps(fiz1,tz);
 908
 909             fjx0             = _mm_add_ps(fjx0,tx);
 910             fjy0             = _mm_add_ps(fjy0,ty);
 911             fjz0             = _mm_add_ps(fjz0,tz);
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r20              = _mm_mul_ps(rsq20,rinv20);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq20             = _mm_mul_ps(iq2,jq0);
 921
 922             /* EWALD ELECTROSTATICS */
 923
 924             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 925             ewrt             = _mm_mul_ps(r20,ewtabscale);
 926             ewitab           = _mm_cvttps_epi32(ewrt);
 927             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 928             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 929                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 930                                          &ewtabF,&ewtabFn);
 931             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 932             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 933
 934             fscal            = felec;
 935
 936             /* Calculate temporary vectorial force */
 937             tx               = _mm_mul_ps(fscal,dx20);
 938             ty               = _mm_mul_ps(fscal,dy20);
 939             tz               = _mm_mul_ps(fscal,dz20);
 940
 941             /* Update vectorial force */
 942             fix2             = _mm_add_ps(fix2,tx);
 943             fiy2             = _mm_add_ps(fiy2,ty);
 944             fiz2             = _mm_add_ps(fiz2,tz);
 945
 946             fjx0             = _mm_add_ps(fjx0,tx);
 947             fjy0             = _mm_add_ps(fjy0,ty);
 948             fjz0             = _mm_add_ps(fjz0,tz);
 949
 950             fjptrA             = f+j_coord_offsetA;
 951             fjptrB             = f+j_coord_offsetB;
 952             fjptrC             = f+j_coord_offsetC;
 953             fjptrD             = f+j_coord_offsetD;
 954
 955             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 956
 957             /* Inner loop uses 115 flops */
 958         }
 959
 960         if(jidx<j_index_end)
 961         {
 962
 963             /* Get j neighbor index, and coordinate index */
 964             jnrlistA         = jjnr[jidx];
 965             jnrlistB         = jjnr[jidx+1];
 966             jnrlistC         = jjnr[jidx+2];
 967             jnrlistD         = jjnr[jidx+3];
 968             /* Sign of each element will be negative for non-real atoms.
 969              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 970              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 971              */
 972             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 973             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 974             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 975             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 976             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 977             j_coord_offsetA  = DIM*jnrA;
 978             j_coord_offsetB  = DIM*jnrB;
 979             j_coord_offsetC  = DIM*jnrC;
 980             j_coord_offsetD  = DIM*jnrD;
 981
 982             /* load j atom coordinates */
 983             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 984                                               x+j_coord_offsetC,x+j_coord_offsetD,
 985                                               &jx0,&jy0,&jz0);
 986
 987             /* Calculate displacement vector */
 988             dx00             = _mm_sub_ps(ix0,jx0);
 989             dy00             = _mm_sub_ps(iy0,jy0);
 990             dz00             = _mm_sub_ps(iz0,jz0);
 991             dx10             = _mm_sub_ps(ix1,jx0);
 992             dy10             = _mm_sub_ps(iy1,jy0);
 993             dz10             = _mm_sub_ps(iz1,jz0);
 994             dx20             = _mm_sub_ps(ix2,jx0);
 995             dy20             = _mm_sub_ps(iy2,jy0);
 996             dz20             = _mm_sub_ps(iz2,jz0);
 997
 998             /* Calculate squared distance and things based on it */
 999             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1000             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1001             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1002
1003             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1004             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1005             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1006
1007             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1008             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1009             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1010
1011             /* Load parameters for j particles */
1012             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1013                                                               charge+jnrC+0,charge+jnrD+0);
1014             vdwjidx0A        = 2*vdwtype[jnrA+0];
1015             vdwjidx0B        = 2*vdwtype[jnrB+0];
1016             vdwjidx0C        = 2*vdwtype[jnrC+0];
1017             vdwjidx0D        = 2*vdwtype[jnrD+0];
1018
1019             fjx0             = _mm_setzero_ps();
1020             fjy0             = _mm_setzero_ps();
1021             fjz0             = _mm_setzero_ps();
1022
1023             /**************************
1024              * CALCULATE INTERACTIONS *
1025              **************************/
1026
1027             r00              = _mm_mul_ps(rsq00,rinv00);
1028             r00              = _mm_andnot_ps(dummy_mask,r00);
1029
1030             /* Compute parameters for interactions between i and j atoms */
1031             qq00             = _mm_mul_ps(iq0,jq0);
1032             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1033                                          vdwparam+vdwioffset0+vdwjidx0B,
1034                                          vdwparam+vdwioffset0+vdwjidx0C,
1035                                          vdwparam+vdwioffset0+vdwjidx0D,
1036                                          &c6_00,&c12_00);
1037
1038             /* EWALD ELECTROSTATICS */
1039
1040             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1041             ewrt             = _mm_mul_ps(r00,ewtabscale);
1042             ewitab           = _mm_cvttps_epi32(ewrt);
1043             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1044             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1045                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1046                                          &ewtabF,&ewtabFn);
1047             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1048             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1049
1050             /* LENNARD-JONES DISPERSION/REPULSION */
1051
1052             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1053             fvdw             = _mm_mul_ps(_mm_sub_ps(_mm_mul_ps(c12_00,rinvsix),c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1054
1055             fscal            = _mm_add_ps(felec,fvdw);
1056
1057             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1058
1059             /* Calculate temporary vectorial force */
1060             tx               = _mm_mul_ps(fscal,dx00);
1061             ty               = _mm_mul_ps(fscal,dy00);
1062             tz               = _mm_mul_ps(fscal,dz00);
1063
1064             /* Update vectorial force */
1065             fix0             = _mm_add_ps(fix0,tx);
1066             fiy0             = _mm_add_ps(fiy0,ty);
1067             fiz0             = _mm_add_ps(fiz0,tz);
1068
1069             fjx0             = _mm_add_ps(fjx0,tx);
1070             fjy0             = _mm_add_ps(fjy0,ty);
1071             fjz0             = _mm_add_ps(fjz0,tz);
1072
1073             /**************************
1074              * CALCULATE INTERACTIONS *
1075              **************************/
1076
1077             r10              = _mm_mul_ps(rsq10,rinv10);
1078             r10              = _mm_andnot_ps(dummy_mask,r10);
1079
1080             /* Compute parameters for interactions between i and j atoms */
1081             qq10             = _mm_mul_ps(iq1,jq0);
1082
1083             /* EWALD ELECTROSTATICS */
1084
1085             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1086             ewrt             = _mm_mul_ps(r10,ewtabscale);
1087             ewitab           = _mm_cvttps_epi32(ewrt);
1088             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1089             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1090                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1091                                          &ewtabF,&ewtabFn);
1092             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1093             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1094
1095             fscal            = felec;
1096
1097             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1098
1099             /* Calculate temporary vectorial force */
1100             tx               = _mm_mul_ps(fscal,dx10);
1101             ty               = _mm_mul_ps(fscal,dy10);
1102             tz               = _mm_mul_ps(fscal,dz10);
1103
1104             /* Update vectorial force */
1105             fix1             = _mm_add_ps(fix1,tx);
1106             fiy1             = _mm_add_ps(fiy1,ty);
1107             fiz1             = _mm_add_ps(fiz1,tz);
1108
1109             fjx0             = _mm_add_ps(fjx0,tx);
1110             fjy0             = _mm_add_ps(fjy0,ty);
1111             fjz0             = _mm_add_ps(fjz0,tz);
1112
1113             /**************************
1114              * CALCULATE INTERACTIONS *
1115              **************************/
1116
1117             r20              = _mm_mul_ps(rsq20,rinv20);
1118             r20              = _mm_andnot_ps(dummy_mask,r20);
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             qq20             = _mm_mul_ps(iq2,jq0);
1122
1123             /* EWALD ELECTROSTATICS */
1124
1125             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1126             ewrt             = _mm_mul_ps(r20,ewtabscale);
1127             ewitab           = _mm_cvttps_epi32(ewrt);
1128             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1129             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1130                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1131                                          &ewtabF,&ewtabFn);
1132             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1133             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1134
1135             fscal            = felec;
1136
1137             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1138
1139             /* Calculate temporary vectorial force */
1140             tx               = _mm_mul_ps(fscal,dx20);
1141             ty               = _mm_mul_ps(fscal,dy20);
1142             tz               = _mm_mul_ps(fscal,dz20);
1143
1144             /* Update vectorial force */
1145             fix2             = _mm_add_ps(fix2,tx);
1146             fiy2             = _mm_add_ps(fiy2,ty);
1147             fiz2             = _mm_add_ps(fiz2,tz);
1148
1149             fjx0             = _mm_add_ps(fjx0,tx);
1150             fjy0             = _mm_add_ps(fjy0,ty);
1151             fjz0             = _mm_add_ps(fjz0,tz);
1152
1153             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1154             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1155             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1156             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1157
1158             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1159
1160             /* Inner loop uses 118 flops */
1161         }
1162
1163         /* End of innermost loop */
1164
1165         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1166                                               f+i_coord_offset,fshift+i_shift_offset);
1167
1168         /* Increment number of inner iterations */
1169         inneriter                  += j_index_end - j_index_start;
1170
1171         /* Outer loop uses 18 flops */
1172     }
1173
1174     /* Increment number of outer iterations */
1175     outeriter        += nri;
1176
1177     /* Update outer/inner flops */
1178
1179     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1180 }