src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_avx_128_fma_single.c

   1 /*
   2  * Note: this file was generated by the Gromacs avx_128_fma_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_avx_128_fma_single.h"
  34 #include "kernelutil_x86_avx_128_fma_single.h"
  35
  36 /*
  37  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: Ewald
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_avx_128_fma_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 AVX_128, 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           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          vfitab;
  89     __m128i          ifour       = _mm_set1_epi32(4);
  90     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  91     real             *vftab;
  92     __m128i          ewitab;
  93     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
  94     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
  95     real             *ewtab;
  96     __m128           dummy_mask,cutoff_mask;
  97     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  98     __m128           one     = _mm_set1_ps(1.0);
  99     __m128           two     = _mm_set1_ps(2.0);
 100     x                = xx[0];
 101     f                = ff[0];
 102
 103     nri              = nlist->nri;
 104     iinr             = nlist->iinr;
 105     jindex           = nlist->jindex;
 106     jjnr             = nlist->jjnr;
 107     shiftidx         = nlist->shift;
 108     gid              = nlist->gid;
 109     shiftvec         = fr->shift_vec[0];
 110     fshift           = fr->fshift[0];
 111     facel            = _mm_set1_ps(fr->epsfac);
 112     charge           = mdatoms->chargeA;
 113     nvdwtype         = fr->ntype;
 114     vdwparam         = fr->nbfp;
 115     vdwtype          = mdatoms->typeA;
 116
 117     vftab            = kernel_data->table_vdw->data;
 118     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 119
 120     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 121     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 122     beta2            = _mm_mul_ps(beta,beta);
 123     beta3            = _mm_mul_ps(beta,beta2);
 124     ewtab            = fr->ic->tabq_coul_FDV0;
 125     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 126     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 127
 128     /* Setup water-specific parameters */
 129     inr              = nlist->iinr[0];
 130     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 131     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 132     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 133     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 134
 135     /* Avoid stupid compiler warnings */
 136     jnrA = jnrB = jnrC = jnrD = 0;
 137     j_coord_offsetA = 0;
 138     j_coord_offsetB = 0;
 139     j_coord_offsetC = 0;
 140     j_coord_offsetD = 0;
 141
 142     outeriter        = 0;
 143     inneriter        = 0;
 144
 145     for(iidx=0;iidx<4*DIM;iidx++)
 146     {
 147         scratch[iidx] = 0.0;
 148     }
 149
 150     /* Start outer loop over neighborlists */
 151     for(iidx=0; iidx<nri; iidx++)
 152     {
 153         /* Load shift vector for this list */
 154         i_shift_offset   = DIM*shiftidx[iidx];
 155
 156         /* Load limits for loop over neighbors */
 157         j_index_start    = jindex[iidx];
 158         j_index_end      = jindex[iidx+1];
 159
 160         /* Get outer coordinate index */
 161         inr              = iinr[iidx];
 162         i_coord_offset   = DIM*inr;
 163
 164         /* Load i particle coords and add shift vector */
 165         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 166                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 167
 168         fix0             = _mm_setzero_ps();
 169         fiy0             = _mm_setzero_ps();
 170         fiz0             = _mm_setzero_ps();
 171         fix1             = _mm_setzero_ps();
 172         fiy1             = _mm_setzero_ps();
 173         fiz1             = _mm_setzero_ps();
 174         fix2             = _mm_setzero_ps();
 175         fiy2             = _mm_setzero_ps();
 176         fiz2             = _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             j_coord_offsetA  = DIM*jnrA;
 192             j_coord_offsetB  = DIM*jnrB;
 193             j_coord_offsetC  = DIM*jnrC;
 194             j_coord_offsetD  = DIM*jnrD;
 195
 196             /* load j atom coordinates */
 197             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 198                                               x+j_coord_offsetC,x+j_coord_offsetD,
 199                                               &jx0,&jy0,&jz0);
 200
 201             /* Calculate displacement vector */
 202             dx00             = _mm_sub_ps(ix0,jx0);
 203             dy00             = _mm_sub_ps(iy0,jy0);
 204             dz00             = _mm_sub_ps(iz0,jz0);
 205             dx10             = _mm_sub_ps(ix1,jx0);
 206             dy10             = _mm_sub_ps(iy1,jy0);
 207             dz10             = _mm_sub_ps(iz1,jz0);
 208             dx20             = _mm_sub_ps(ix2,jx0);
 209             dy20             = _mm_sub_ps(iy2,jy0);
 210             dz20             = _mm_sub_ps(iz2,jz0);
 211
 212             /* Calculate squared distance and things based on it */
 213             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 214             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 215             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 216
 217             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 218             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 219             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 220
 221             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 222             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 223             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 224
 225             /* Load parameters for j particles */
 226             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 227                                                               charge+jnrC+0,charge+jnrD+0);
 228             vdwjidx0A        = 2*vdwtype[jnrA+0];
 229             vdwjidx0B        = 2*vdwtype[jnrB+0];
 230             vdwjidx0C        = 2*vdwtype[jnrC+0];
 231             vdwjidx0D        = 2*vdwtype[jnrD+0];
 232
 233             fjx0             = _mm_setzero_ps();
 234             fjy0             = _mm_setzero_ps();
 235             fjz0             = _mm_setzero_ps();
 236
 237             /**************************
 238              * CALCULATE INTERACTIONS *
 239              **************************/
 240
 241             r00              = _mm_mul_ps(rsq00,rinv00);
 242
 243             /* Compute parameters for interactions between i and j atoms */
 244             qq00             = _mm_mul_ps(iq0,jq0);
 245             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 246                                          vdwparam+vdwioffset0+vdwjidx0B,
 247                                          vdwparam+vdwioffset0+vdwjidx0C,
 248                                          vdwparam+vdwioffset0+vdwjidx0D,
 249                                          &c6_00,&c12_00);
 250
 251             /* Calculate table index by multiplying r with table scale and truncate to integer */
 252             rt               = _mm_mul_ps(r00,vftabscale);
 253             vfitab           = _mm_cvttps_epi32(rt);
 254 #ifdef __XOP__
 255             vfeps            = _mm_frcz_ps(rt);
 256 #else
 257             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 258 #endif
 259             twovfeps         = _mm_add_ps(vfeps,vfeps);
 260             vfitab           = _mm_slli_epi32(vfitab,3);
 261
 262             /* EWALD ELECTROSTATICS */
 263
 264             /* Analytical PME correction */
 265             zeta2            = _mm_mul_ps(beta2,rsq00);
 266             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 267             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 268             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 269             felec            = _mm_mul_ps(qq00,felec);
 270             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 271             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 272             velec            = _mm_mul_ps(qq00,velec);
 273
 274             /* CUBIC SPLINE TABLE DISPERSION */
 275             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 276             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 277             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 278             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 279             _MM_TRANSPOSE4_PS(Y,F,G,H);
 280             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 281             VV               = _mm_macc_ps(vfeps,Fp,Y);
 282             vvdw6            = _mm_mul_ps(c6_00,VV);
 283             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 284             fvdw6            = _mm_mul_ps(c6_00,FF);
 285
 286             /* CUBIC SPLINE TABLE REPULSION */
 287             vfitab           = _mm_add_epi32(vfitab,ifour);
 288             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 289             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 290             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 291             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 292             _MM_TRANSPOSE4_PS(Y,F,G,H);
 293             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 294             VV               = _mm_macc_ps(vfeps,Fp,Y);
 295             vvdw12           = _mm_mul_ps(c12_00,VV);
 296             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 297             fvdw12           = _mm_mul_ps(c12_00,FF);
 298             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 299             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 300
 301             /* Update potential sum for this i atom from the interaction with this j atom. */
 302             velecsum         = _mm_add_ps(velecsum,velec);
 303             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 304
 305             fscal            = _mm_add_ps(felec,fvdw);
 306
 307              /* Update vectorial force */
 308             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 309             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 310             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 311
 312             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 313             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 314             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 315
 316             /**************************
 317              * CALCULATE INTERACTIONS *
 318              **************************/
 319
 320             r10              = _mm_mul_ps(rsq10,rinv10);
 321
 322             /* Compute parameters for interactions between i and j atoms */
 323             qq10             = _mm_mul_ps(iq1,jq0);
 324
 325             /* EWALD ELECTROSTATICS */
 326
 327             /* Analytical PME correction */
 328             zeta2            = _mm_mul_ps(beta2,rsq10);
 329             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 330             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 331             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 332             felec            = _mm_mul_ps(qq10,felec);
 333             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 334             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 335             velec            = _mm_mul_ps(qq10,velec);
 336
 337             /* Update potential sum for this i atom from the interaction with this j atom. */
 338             velecsum         = _mm_add_ps(velecsum,velec);
 339
 340             fscal            = felec;
 341
 342              /* Update vectorial force */
 343             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 344             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 345             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 346
 347             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 348             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 349             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 350
 351             /**************************
 352              * CALCULATE INTERACTIONS *
 353              **************************/
 354
 355             r20              = _mm_mul_ps(rsq20,rinv20);
 356
 357             /* Compute parameters for interactions between i and j atoms */
 358             qq20             = _mm_mul_ps(iq2,jq0);
 359
 360             /* EWALD ELECTROSTATICS */
 361
 362             /* Analytical PME correction */
 363             zeta2            = _mm_mul_ps(beta2,rsq20);
 364             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 365             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 366             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 367             felec            = _mm_mul_ps(qq20,felec);
 368             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 369             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 370             velec            = _mm_mul_ps(qq20,velec);
 371
 372             /* Update potential sum for this i atom from the interaction with this j atom. */
 373             velecsum         = _mm_add_ps(velecsum,velec);
 374
 375             fscal            = felec;
 376
 377              /* Update vectorial force */
 378             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 379             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 380             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 381
 382             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 383             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 384             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 385
 386             fjptrA             = f+j_coord_offsetA;
 387             fjptrB             = f+j_coord_offsetB;
 388             fjptrC             = f+j_coord_offsetC;
 389             fjptrD             = f+j_coord_offsetD;
 390
 391             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 392
 393             /* Inner loop uses 121 flops */
 394         }
 395
 396         if(jidx<j_index_end)
 397         {
 398
 399             /* Get j neighbor index, and coordinate index */
 400             jnrlistA         = jjnr[jidx];
 401             jnrlistB         = jjnr[jidx+1];
 402             jnrlistC         = jjnr[jidx+2];
 403             jnrlistD         = jjnr[jidx+3];
 404             /* Sign of each element will be negative for non-real atoms.
 405              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 406              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 407              */
 408             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 409             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 410             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 411             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 412             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 413             j_coord_offsetA  = DIM*jnrA;
 414             j_coord_offsetB  = DIM*jnrB;
 415             j_coord_offsetC  = DIM*jnrC;
 416             j_coord_offsetD  = DIM*jnrD;
 417
 418             /* load j atom coordinates */
 419             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 420                                               x+j_coord_offsetC,x+j_coord_offsetD,
 421                                               &jx0,&jy0,&jz0);
 422
 423             /* Calculate displacement vector */
 424             dx00             = _mm_sub_ps(ix0,jx0);
 425             dy00             = _mm_sub_ps(iy0,jy0);
 426             dz00             = _mm_sub_ps(iz0,jz0);
 427             dx10             = _mm_sub_ps(ix1,jx0);
 428             dy10             = _mm_sub_ps(iy1,jy0);
 429             dz10             = _mm_sub_ps(iz1,jz0);
 430             dx20             = _mm_sub_ps(ix2,jx0);
 431             dy20             = _mm_sub_ps(iy2,jy0);
 432             dz20             = _mm_sub_ps(iz2,jz0);
 433
 434             /* Calculate squared distance and things based on it */
 435             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 436             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 437             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 438
 439             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 440             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 441             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 442
 443             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 444             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 445             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 446
 447             /* Load parameters for j particles */
 448             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 449                                                               charge+jnrC+0,charge+jnrD+0);
 450             vdwjidx0A        = 2*vdwtype[jnrA+0];
 451             vdwjidx0B        = 2*vdwtype[jnrB+0];
 452             vdwjidx0C        = 2*vdwtype[jnrC+0];
 453             vdwjidx0D        = 2*vdwtype[jnrD+0];
 454
 455             fjx0             = _mm_setzero_ps();
 456             fjy0             = _mm_setzero_ps();
 457             fjz0             = _mm_setzero_ps();
 458
 459             /**************************
 460              * CALCULATE INTERACTIONS *
 461              **************************/
 462
 463             r00              = _mm_mul_ps(rsq00,rinv00);
 464             r00              = _mm_andnot_ps(dummy_mask,r00);
 465
 466             /* Compute parameters for interactions between i and j atoms */
 467             qq00             = _mm_mul_ps(iq0,jq0);
 468             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 469                                          vdwparam+vdwioffset0+vdwjidx0B,
 470                                          vdwparam+vdwioffset0+vdwjidx0C,
 471                                          vdwparam+vdwioffset0+vdwjidx0D,
 472                                          &c6_00,&c12_00);
 473
 474             /* Calculate table index by multiplying r with table scale and truncate to integer */
 475             rt               = _mm_mul_ps(r00,vftabscale);
 476             vfitab           = _mm_cvttps_epi32(rt);
 477 #ifdef __XOP__
 478             vfeps            = _mm_frcz_ps(rt);
 479 #else
 480             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 481 #endif
 482             twovfeps         = _mm_add_ps(vfeps,vfeps);
 483             vfitab           = _mm_slli_epi32(vfitab,3);
 484
 485             /* EWALD ELECTROSTATICS */
 486
 487             /* Analytical PME correction */
 488             zeta2            = _mm_mul_ps(beta2,rsq00);
 489             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 490             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 491             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 492             felec            = _mm_mul_ps(qq00,felec);
 493             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 494             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv00);
 495             velec            = _mm_mul_ps(qq00,velec);
 496
 497             /* CUBIC SPLINE TABLE DISPERSION */
 498             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 499             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 500             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 501             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 502             _MM_TRANSPOSE4_PS(Y,F,G,H);
 503             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 504             VV               = _mm_macc_ps(vfeps,Fp,Y);
 505             vvdw6            = _mm_mul_ps(c6_00,VV);
 506             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 507             fvdw6            = _mm_mul_ps(c6_00,FF);
 508
 509             /* CUBIC SPLINE TABLE REPULSION */
 510             vfitab           = _mm_add_epi32(vfitab,ifour);
 511             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 512             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 513             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 514             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 515             _MM_TRANSPOSE4_PS(Y,F,G,H);
 516             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 517             VV               = _mm_macc_ps(vfeps,Fp,Y);
 518             vvdw12           = _mm_mul_ps(c12_00,VV);
 519             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 520             fvdw12           = _mm_mul_ps(c12_00,FF);
 521             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 522             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 523
 524             /* Update potential sum for this i atom from the interaction with this j atom. */
 525             velec            = _mm_andnot_ps(dummy_mask,velec);
 526             velecsum         = _mm_add_ps(velecsum,velec);
 527             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 528             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 529
 530             fscal            = _mm_add_ps(felec,fvdw);
 531
 532             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 533
 534              /* Update vectorial force */
 535             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 536             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 537             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 538
 539             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 540             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 541             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 542
 543             /**************************
 544              * CALCULATE INTERACTIONS *
 545              **************************/
 546
 547             r10              = _mm_mul_ps(rsq10,rinv10);
 548             r10              = _mm_andnot_ps(dummy_mask,r10);
 549
 550             /* Compute parameters for interactions between i and j atoms */
 551             qq10             = _mm_mul_ps(iq1,jq0);
 552
 553             /* EWALD ELECTROSTATICS */
 554
 555             /* Analytical PME correction */
 556             zeta2            = _mm_mul_ps(beta2,rsq10);
 557             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 558             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 559             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 560             felec            = _mm_mul_ps(qq10,felec);
 561             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 562             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv10);
 563             velec            = _mm_mul_ps(qq10,velec);
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm_andnot_ps(dummy_mask,velec);
 567             velecsum         = _mm_add_ps(velecsum,velec);
 568
 569             fscal            = felec;
 570
 571             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 572
 573              /* Update vectorial force */
 574             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 575             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 576             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 577
 578             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 579             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 580             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 581
 582             /**************************
 583              * CALCULATE INTERACTIONS *
 584              **************************/
 585
 586             r20              = _mm_mul_ps(rsq20,rinv20);
 587             r20              = _mm_andnot_ps(dummy_mask,r20);
 588
 589             /* Compute parameters for interactions between i and j atoms */
 590             qq20             = _mm_mul_ps(iq2,jq0);
 591
 592             /* EWALD ELECTROSTATICS */
 593
 594             /* Analytical PME correction */
 595             zeta2            = _mm_mul_ps(beta2,rsq20);
 596             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 597             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 598             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 599             felec            = _mm_mul_ps(qq20,felec);
 600             pmecorrV         = gmx_mm_pmecorrV_ps(zeta2);
 601             velec            = _mm_nmacc_ps(pmecorrV,beta,rinv20);
 602             velec            = _mm_mul_ps(qq20,velec);
 603
 604             /* Update potential sum for this i atom from the interaction with this j atom. */
 605             velec            = _mm_andnot_ps(dummy_mask,velec);
 606             velecsum         = _mm_add_ps(velecsum,velec);
 607
 608             fscal            = felec;
 609
 610             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 611
 612              /* Update vectorial force */
 613             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 614             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 615             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 616
 617             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 618             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 619             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 620
 621             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 622             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 623             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 624             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 625
 626             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 627
 628             /* Inner loop uses 124 flops */
 629         }
 630
 631         /* End of innermost loop */
 632
 633         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 634                                               f+i_coord_offset,fshift+i_shift_offset);
 635
 636         ggid                        = gid[iidx];
 637         /* Update potential energies */
 638         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 639         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 640
 641         /* Increment number of inner iterations */
 642         inneriter                  += j_index_end - j_index_start;
 643
 644         /* Outer loop uses 20 flops */
 645     }
 646
 647     /* Increment number of outer iterations */
 648     outeriter        += nri;
 649
 650     /* Update outer/inner flops */
 651
 652     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*124);
 653 }
 654 /*
 655  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 656  * Electrostatics interaction: Ewald
 657  * VdW interaction:            CubicSplineTable
 658  * Geometry:                   Water3-Particle
 659  * Calculate force/pot:        Force
 660  */
 661 void
 662 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 663                     (t_nblist * gmx_restrict                nlist,
 664                      rvec * gmx_restrict                    xx,
 665                      rvec * gmx_restrict                    ff,
 666                      t_forcerec * gmx_restrict              fr,
 667                      t_mdatoms * gmx_restrict               mdatoms,
 668                      nb_kernel_data_t * gmx_restrict        kernel_data,
 669                      t_nrnb * gmx_restrict                  nrnb)
 670 {
 671     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 672      * just 0 for non-waters.
 673      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 674      * jnr indices corresponding to data put in the four positions in the SIMD register.
 675      */
 676     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 677     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 678     int              jnrA,jnrB,jnrC,jnrD;
 679     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 680     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 681     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 682     real             rcutoff_scalar;
 683     real             *shiftvec,*fshift,*x,*f;
 684     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 685     real             scratch[4*DIM];
 686     __m128           fscal,rcutoff,rcutoff2,jidxall;
 687     int              vdwioffset0;
 688     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 689     int              vdwioffset1;
 690     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 691     int              vdwioffset2;
 692     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 693     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 694     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 695     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 696     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 697     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 698     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 699     real             *charge;
 700     int              nvdwtype;
 701     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 702     int              *vdwtype;
 703     real             *vdwparam;
 704     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 705     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 706     __m128i          vfitab;
 707     __m128i          ifour       = _mm_set1_epi32(4);
 708     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 709     real             *vftab;
 710     __m128i          ewitab;
 711     __m128           ewtabscale,eweps,twoeweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 712     __m128           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 713     real             *ewtab;
 714     __m128           dummy_mask,cutoff_mask;
 715     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 716     __m128           one     = _mm_set1_ps(1.0);
 717     __m128           two     = _mm_set1_ps(2.0);
 718     x                = xx[0];
 719     f                = ff[0];
 720
 721     nri              = nlist->nri;
 722     iinr             = nlist->iinr;
 723     jindex           = nlist->jindex;
 724     jjnr             = nlist->jjnr;
 725     shiftidx         = nlist->shift;
 726     gid              = nlist->gid;
 727     shiftvec         = fr->shift_vec[0];
 728     fshift           = fr->fshift[0];
 729     facel            = _mm_set1_ps(fr->epsfac);
 730     charge           = mdatoms->chargeA;
 731     nvdwtype         = fr->ntype;
 732     vdwparam         = fr->nbfp;
 733     vdwtype          = mdatoms->typeA;
 734
 735     vftab            = kernel_data->table_vdw->data;
 736     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 737
 738     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 739     beta             = _mm_set1_ps(fr->ic->ewaldcoeff);
 740     beta2            = _mm_mul_ps(beta,beta);
 741     beta3            = _mm_mul_ps(beta,beta2);
 742     ewtab            = fr->ic->tabq_coul_F;
 743     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 744     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 745
 746     /* Setup water-specific parameters */
 747     inr              = nlist->iinr[0];
 748     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 749     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 750     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 751     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 752
 753     /* Avoid stupid compiler warnings */
 754     jnrA = jnrB = jnrC = jnrD = 0;
 755     j_coord_offsetA = 0;
 756     j_coord_offsetB = 0;
 757     j_coord_offsetC = 0;
 758     j_coord_offsetD = 0;
 759
 760     outeriter        = 0;
 761     inneriter        = 0;
 762
 763     for(iidx=0;iidx<4*DIM;iidx++)
 764     {
 765         scratch[iidx] = 0.0;
 766     }
 767
 768     /* Start outer loop over neighborlists */
 769     for(iidx=0; iidx<nri; iidx++)
 770     {
 771         /* Load shift vector for this list */
 772         i_shift_offset   = DIM*shiftidx[iidx];
 773
 774         /* Load limits for loop over neighbors */
 775         j_index_start    = jindex[iidx];
 776         j_index_end      = jindex[iidx+1];
 777
 778         /* Get outer coordinate index */
 779         inr              = iinr[iidx];
 780         i_coord_offset   = DIM*inr;
 781
 782         /* Load i particle coords and add shift vector */
 783         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 784                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 785
 786         fix0             = _mm_setzero_ps();
 787         fiy0             = _mm_setzero_ps();
 788         fiz0             = _mm_setzero_ps();
 789         fix1             = _mm_setzero_ps();
 790         fiy1             = _mm_setzero_ps();
 791         fiz1             = _mm_setzero_ps();
 792         fix2             = _mm_setzero_ps();
 793         fiy2             = _mm_setzero_ps();
 794         fiz2             = _mm_setzero_ps();
 795
 796         /* Start inner kernel loop */
 797         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 798         {
 799
 800             /* Get j neighbor index, and coordinate index */
 801             jnrA             = jjnr[jidx];
 802             jnrB             = jjnr[jidx+1];
 803             jnrC             = jjnr[jidx+2];
 804             jnrD             = jjnr[jidx+3];
 805             j_coord_offsetA  = DIM*jnrA;
 806             j_coord_offsetB  = DIM*jnrB;
 807             j_coord_offsetC  = DIM*jnrC;
 808             j_coord_offsetD  = DIM*jnrD;
 809
 810             /* load j atom coordinates */
 811             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 812                                               x+j_coord_offsetC,x+j_coord_offsetD,
 813                                               &jx0,&jy0,&jz0);
 814
 815             /* Calculate displacement vector */
 816             dx00             = _mm_sub_ps(ix0,jx0);
 817             dy00             = _mm_sub_ps(iy0,jy0);
 818             dz00             = _mm_sub_ps(iz0,jz0);
 819             dx10             = _mm_sub_ps(ix1,jx0);
 820             dy10             = _mm_sub_ps(iy1,jy0);
 821             dz10             = _mm_sub_ps(iz1,jz0);
 822             dx20             = _mm_sub_ps(ix2,jx0);
 823             dy20             = _mm_sub_ps(iy2,jy0);
 824             dz20             = _mm_sub_ps(iz2,jz0);
 825
 826             /* Calculate squared distance and things based on it */
 827             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 828             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 829             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 830
 831             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 832             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 833             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 834
 835             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 836             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 837             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 838
 839             /* Load parameters for j particles */
 840             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 841                                                               charge+jnrC+0,charge+jnrD+0);
 842             vdwjidx0A        = 2*vdwtype[jnrA+0];
 843             vdwjidx0B        = 2*vdwtype[jnrB+0];
 844             vdwjidx0C        = 2*vdwtype[jnrC+0];
 845             vdwjidx0D        = 2*vdwtype[jnrD+0];
 846
 847             fjx0             = _mm_setzero_ps();
 848             fjy0             = _mm_setzero_ps();
 849             fjz0             = _mm_setzero_ps();
 850
 851             /**************************
 852              * CALCULATE INTERACTIONS *
 853              **************************/
 854
 855             r00              = _mm_mul_ps(rsq00,rinv00);
 856
 857             /* Compute parameters for interactions between i and j atoms */
 858             qq00             = _mm_mul_ps(iq0,jq0);
 859             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 860                                          vdwparam+vdwioffset0+vdwjidx0B,
 861                                          vdwparam+vdwioffset0+vdwjidx0C,
 862                                          vdwparam+vdwioffset0+vdwjidx0D,
 863                                          &c6_00,&c12_00);
 864
 865             /* Calculate table index by multiplying r with table scale and truncate to integer */
 866             rt               = _mm_mul_ps(r00,vftabscale);
 867             vfitab           = _mm_cvttps_epi32(rt);
 868 #ifdef __XOP__
 869             vfeps            = _mm_frcz_ps(rt);
 870 #else
 871             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 872 #endif
 873             twovfeps         = _mm_add_ps(vfeps,vfeps);
 874             vfitab           = _mm_slli_epi32(vfitab,3);
 875
 876             /* EWALD ELECTROSTATICS */
 877
 878             /* Analytical PME correction */
 879             zeta2            = _mm_mul_ps(beta2,rsq00);
 880             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
 881             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 882             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 883             felec            = _mm_mul_ps(qq00,felec);
 884
 885             /* CUBIC SPLINE TABLE DISPERSION */
 886             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 887             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 888             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 889             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 890             _MM_TRANSPOSE4_PS(Y,F,G,H);
 891             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 892             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 893             fvdw6            = _mm_mul_ps(c6_00,FF);
 894
 895             /* CUBIC SPLINE TABLE REPULSION */
 896             vfitab           = _mm_add_epi32(vfitab,ifour);
 897             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 898             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 899             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 900             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 901             _MM_TRANSPOSE4_PS(Y,F,G,H);
 902             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 903             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 904             fvdw12           = _mm_mul_ps(c12_00,FF);
 905             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 906
 907             fscal            = _mm_add_ps(felec,fvdw);
 908
 909              /* Update vectorial force */
 910             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 911             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 912             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 913
 914             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 915             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 916             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 917
 918             /**************************
 919              * CALCULATE INTERACTIONS *
 920              **************************/
 921
 922             r10              = _mm_mul_ps(rsq10,rinv10);
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq10             = _mm_mul_ps(iq1,jq0);
 926
 927             /* EWALD ELECTROSTATICS */
 928
 929             /* Analytical PME correction */
 930             zeta2            = _mm_mul_ps(beta2,rsq10);
 931             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
 932             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 933             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 934             felec            = _mm_mul_ps(qq10,felec);
 935
 936             fscal            = felec;
 937
 938              /* Update vectorial force */
 939             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 940             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 941             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 942
 943             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 944             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 945             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 946
 947             /**************************
 948              * CALCULATE INTERACTIONS *
 949              **************************/
 950
 951             r20              = _mm_mul_ps(rsq20,rinv20);
 952
 953             /* Compute parameters for interactions between i and j atoms */
 954             qq20             = _mm_mul_ps(iq2,jq0);
 955
 956             /* EWALD ELECTROSTATICS */
 957
 958             /* Analytical PME correction */
 959             zeta2            = _mm_mul_ps(beta2,rsq20);
 960             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
 961             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
 962             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
 963             felec            = _mm_mul_ps(qq20,felec);
 964
 965             fscal            = felec;
 966
 967              /* Update vectorial force */
 968             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 969             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 970             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 971
 972             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 973             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 974             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 975
 976             fjptrA             = f+j_coord_offsetA;
 977             fjptrB             = f+j_coord_offsetB;
 978             fjptrC             = f+j_coord_offsetC;
 979             fjptrD             = f+j_coord_offsetD;
 980
 981             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 982
 983             /* Inner loop uses 110 flops */
 984         }
 985
 986         if(jidx<j_index_end)
 987         {
 988
 989             /* Get j neighbor index, and coordinate index */
 990             jnrlistA         = jjnr[jidx];
 991             jnrlistB         = jjnr[jidx+1];
 992             jnrlistC         = jjnr[jidx+2];
 993             jnrlistD         = jjnr[jidx+3];
 994             /* Sign of each element will be negative for non-real atoms.
 995              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 996              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 997              */
 998             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 999             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1000             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1001             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1002             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1003             j_coord_offsetA  = DIM*jnrA;
1004             j_coord_offsetB  = DIM*jnrB;
1005             j_coord_offsetC  = DIM*jnrC;
1006             j_coord_offsetD  = DIM*jnrD;
1007
1008             /* load j atom coordinates */
1009             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1010                                               x+j_coord_offsetC,x+j_coord_offsetD,
1011                                               &jx0,&jy0,&jz0);
1012
1013             /* Calculate displacement vector */
1014             dx00             = _mm_sub_ps(ix0,jx0);
1015             dy00             = _mm_sub_ps(iy0,jy0);
1016             dz00             = _mm_sub_ps(iz0,jz0);
1017             dx10             = _mm_sub_ps(ix1,jx0);
1018             dy10             = _mm_sub_ps(iy1,jy0);
1019             dz10             = _mm_sub_ps(iz1,jz0);
1020             dx20             = _mm_sub_ps(ix2,jx0);
1021             dy20             = _mm_sub_ps(iy2,jy0);
1022             dz20             = _mm_sub_ps(iz2,jz0);
1023
1024             /* Calculate squared distance and things based on it */
1025             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1026             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1027             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1028
1029             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1030             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1031             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1032
1033             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1034             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1035             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1036
1037             /* Load parameters for j particles */
1038             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1039                                                               charge+jnrC+0,charge+jnrD+0);
1040             vdwjidx0A        = 2*vdwtype[jnrA+0];
1041             vdwjidx0B        = 2*vdwtype[jnrB+0];
1042             vdwjidx0C        = 2*vdwtype[jnrC+0];
1043             vdwjidx0D        = 2*vdwtype[jnrD+0];
1044
1045             fjx0             = _mm_setzero_ps();
1046             fjy0             = _mm_setzero_ps();
1047             fjz0             = _mm_setzero_ps();
1048
1049             /**************************
1050              * CALCULATE INTERACTIONS *
1051              **************************/
1052
1053             r00              = _mm_mul_ps(rsq00,rinv00);
1054             r00              = _mm_andnot_ps(dummy_mask,r00);
1055
1056             /* Compute parameters for interactions between i and j atoms */
1057             qq00             = _mm_mul_ps(iq0,jq0);
1058             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1059                                          vdwparam+vdwioffset0+vdwjidx0B,
1060                                          vdwparam+vdwioffset0+vdwjidx0C,
1061                                          vdwparam+vdwioffset0+vdwjidx0D,
1062                                          &c6_00,&c12_00);
1063
1064             /* Calculate table index by multiplying r with table scale and truncate to integer */
1065             rt               = _mm_mul_ps(r00,vftabscale);
1066             vfitab           = _mm_cvttps_epi32(rt);
1067 #ifdef __XOP__
1068             vfeps            = _mm_frcz_ps(rt);
1069 #else
1070             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1071 #endif
1072             twovfeps         = _mm_add_ps(vfeps,vfeps);
1073             vfitab           = _mm_slli_epi32(vfitab,3);
1074
1075             /* EWALD ELECTROSTATICS */
1076
1077             /* Analytical PME correction */
1078             zeta2            = _mm_mul_ps(beta2,rsq00);
1079             rinv3            = _mm_mul_ps(rinvsq00,rinv00);
1080             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1081             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1082             felec            = _mm_mul_ps(qq00,felec);
1083
1084             /* CUBIC SPLINE TABLE DISPERSION */
1085             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1086             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1087             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1088             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1089             _MM_TRANSPOSE4_PS(Y,F,G,H);
1090             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1091             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1092             fvdw6            = _mm_mul_ps(c6_00,FF);
1093
1094             /* CUBIC SPLINE TABLE REPULSION */
1095             vfitab           = _mm_add_epi32(vfitab,ifour);
1096             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1097             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1098             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1099             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1100             _MM_TRANSPOSE4_PS(Y,F,G,H);
1101             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1102             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1103             fvdw12           = _mm_mul_ps(c12_00,FF);
1104             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1105
1106             fscal            = _mm_add_ps(felec,fvdw);
1107
1108             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1109
1110              /* Update vectorial force */
1111             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1112             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1113             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1114
1115             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1116             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1117             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1118
1119             /**************************
1120              * CALCULATE INTERACTIONS *
1121              **************************/
1122
1123             r10              = _mm_mul_ps(rsq10,rinv10);
1124             r10              = _mm_andnot_ps(dummy_mask,r10);
1125
1126             /* Compute parameters for interactions between i and j atoms */
1127             qq10             = _mm_mul_ps(iq1,jq0);
1128
1129             /* EWALD ELECTROSTATICS */
1130
1131             /* Analytical PME correction */
1132             zeta2            = _mm_mul_ps(beta2,rsq10);
1133             rinv3            = _mm_mul_ps(rinvsq10,rinv10);
1134             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1135             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1136             felec            = _mm_mul_ps(qq10,felec);
1137
1138             fscal            = felec;
1139
1140             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1141
1142              /* Update vectorial force */
1143             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1144             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1145             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1146
1147             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1148             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1149             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1150
1151             /**************************
1152              * CALCULATE INTERACTIONS *
1153              **************************/
1154
1155             r20              = _mm_mul_ps(rsq20,rinv20);
1156             r20              = _mm_andnot_ps(dummy_mask,r20);
1157
1158             /* Compute parameters for interactions between i and j atoms */
1159             qq20             = _mm_mul_ps(iq2,jq0);
1160
1161             /* EWALD ELECTROSTATICS */
1162
1163             /* Analytical PME correction */
1164             zeta2            = _mm_mul_ps(beta2,rsq20);
1165             rinv3            = _mm_mul_ps(rinvsq20,rinv20);
1166             pmecorrF         = gmx_mm_pmecorrF_ps(zeta2);
1167             felec            = _mm_macc_ps(pmecorrF,beta3,rinv3);
1168             felec            = _mm_mul_ps(qq20,felec);
1169
1170             fscal            = felec;
1171
1172             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1173
1174              /* Update vectorial force */
1175             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1176             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1177             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1178
1179             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1180             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1181             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1182
1183             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1184             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1185             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1186             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1187
1188             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1189
1190             /* Inner loop uses 113 flops */
1191         }
1192
1193         /* End of innermost loop */
1194
1195         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1196                                               f+i_coord_offset,fshift+i_shift_offset);
1197
1198         /* Increment number of inner iterations */
1199         inneriter                  += j_index_end - j_index_start;
1200
1201         /* Outer loop uses 18 flops */
1202     }
1203
1204     /* Increment number of outer iterations */
1205     outeriter        += nri;
1206
1207     /* Update outer/inner flops */
1208
1209     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*113);
1210 }