src/gromacs/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecRF_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_ElecRF_VdwCSTab_GeomW3P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: ReactionField
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water3-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecRF_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     __m128           dummy_mask,cutoff_mask;
  93     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  94     __m128           one     = _mm_set1_ps(1.0);
  95     __m128           two     = _mm_set1_ps(2.0);
  96     x                = xx[0];
  97     f                = ff[0];
  98
  99     nri              = nlist->nri;
 100     iinr             = nlist->iinr;
 101     jindex           = nlist->jindex;
 102     jjnr             = nlist->jjnr;
 103     shiftidx         = nlist->shift;
 104     gid              = nlist->gid;
 105     shiftvec         = fr->shift_vec[0];
 106     fshift           = fr->fshift[0];
 107     facel            = _mm_set1_ps(fr->epsfac);
 108     charge           = mdatoms->chargeA;
 109     krf              = _mm_set1_ps(fr->ic->k_rf);
 110     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 111     crf              = _mm_set1_ps(fr->ic->c_rf);
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 122     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 123     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 124     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 125
 126     /* Avoid stupid compiler warnings */
 127     jnrA = jnrB = jnrC = jnrD = 0;
 128     j_coord_offsetA = 0;
 129     j_coord_offsetB = 0;
 130     j_coord_offsetC = 0;
 131     j_coord_offsetD = 0;
 132
 133     outeriter        = 0;
 134     inneriter        = 0;
 135
 136     for(iidx=0;iidx<4*DIM;iidx++)
 137     {
 138         scratch[iidx] = 0.0;
 139     }
 140
 141     /* Start outer loop over neighborlists */
 142     for(iidx=0; iidx<nri; iidx++)
 143     {
 144         /* Load shift vector for this list */
 145         i_shift_offset   = DIM*shiftidx[iidx];
 146
 147         /* Load limits for loop over neighbors */
 148         j_index_start    = jindex[iidx];
 149         j_index_end      = jindex[iidx+1];
 150
 151         /* Get outer coordinate index */
 152         inr              = iinr[iidx];
 153         i_coord_offset   = DIM*inr;
 154
 155         /* Load i particle coords and add shift vector */
 156         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 158
 159         fix0             = _mm_setzero_ps();
 160         fiy0             = _mm_setzero_ps();
 161         fiz0             = _mm_setzero_ps();
 162         fix1             = _mm_setzero_ps();
 163         fiy1             = _mm_setzero_ps();
 164         fiz1             = _mm_setzero_ps();
 165         fix2             = _mm_setzero_ps();
 166         fiy2             = _mm_setzero_ps();
 167         fiz2             = _mm_setzero_ps();
 168
 169         /* Reset potential sums */
 170         velecsum         = _mm_setzero_ps();
 171         vvdwsum          = _mm_setzero_ps();
 172
 173         /* Start inner kernel loop */
 174         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 175         {
 176
 177             /* Get j neighbor index, and coordinate index */
 178             jnrA             = jjnr[jidx];
 179             jnrB             = jjnr[jidx+1];
 180             jnrC             = jjnr[jidx+2];
 181             jnrD             = jjnr[jidx+3];
 182             j_coord_offsetA  = DIM*jnrA;
 183             j_coord_offsetB  = DIM*jnrB;
 184             j_coord_offsetC  = DIM*jnrC;
 185             j_coord_offsetD  = DIM*jnrD;
 186
 187             /* load j atom coordinates */
 188             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 189                                               x+j_coord_offsetC,x+j_coord_offsetD,
 190                                               &jx0,&jy0,&jz0);
 191
 192             /* Calculate displacement vector */
 193             dx00             = _mm_sub_ps(ix0,jx0);
 194             dy00             = _mm_sub_ps(iy0,jy0);
 195             dz00             = _mm_sub_ps(iz0,jz0);
 196             dx10             = _mm_sub_ps(ix1,jx0);
 197             dy10             = _mm_sub_ps(iy1,jy0);
 198             dz10             = _mm_sub_ps(iz1,jz0);
 199             dx20             = _mm_sub_ps(ix2,jx0);
 200             dy20             = _mm_sub_ps(iy2,jy0);
 201             dz20             = _mm_sub_ps(iz2,jz0);
 202
 203             /* Calculate squared distance and things based on it */
 204             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 205             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 206             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 207
 208             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 209             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 210             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 211
 212             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 213             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 214             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 215
 216             /* Load parameters for j particles */
 217             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 218                                                               charge+jnrC+0,charge+jnrD+0);
 219             vdwjidx0A        = 2*vdwtype[jnrA+0];
 220             vdwjidx0B        = 2*vdwtype[jnrB+0];
 221             vdwjidx0C        = 2*vdwtype[jnrC+0];
 222             vdwjidx0D        = 2*vdwtype[jnrD+0];
 223
 224             fjx0             = _mm_setzero_ps();
 225             fjy0             = _mm_setzero_ps();
 226             fjz0             = _mm_setzero_ps();
 227
 228             /**************************
 229              * CALCULATE INTERACTIONS *
 230              **************************/
 231
 232             r00              = _mm_mul_ps(rsq00,rinv00);
 233
 234             /* Compute parameters for interactions between i and j atoms */
 235             qq00             = _mm_mul_ps(iq0,jq0);
 236             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 237                                          vdwparam+vdwioffset0+vdwjidx0B,
 238                                          vdwparam+vdwioffset0+vdwjidx0C,
 239                                          vdwparam+vdwioffset0+vdwjidx0D,
 240                                          &c6_00,&c12_00);
 241
 242             /* Calculate table index by multiplying r with table scale and truncate to integer */
 243             rt               = _mm_mul_ps(r00,vftabscale);
 244             vfitab           = _mm_cvttps_epi32(rt);
 245 #ifdef __XOP__
 246             vfeps            = _mm_frcz_ps(rt);
 247 #else
 248             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 249 #endif
 250             twovfeps         = _mm_add_ps(vfeps,vfeps);
 251             vfitab           = _mm_slli_epi32(vfitab,3);
 252
 253             /* REACTION-FIELD ELECTROSTATICS */
 254             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 255             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 256
 257             /* CUBIC SPLINE TABLE DISPERSION */
 258             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 259             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 260             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 261             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 262             _MM_TRANSPOSE4_PS(Y,F,G,H);
 263             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 264             VV               = _mm_macc_ps(vfeps,Fp,Y);
 265             vvdw6            = _mm_mul_ps(c6_00,VV);
 266             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 267             fvdw6            = _mm_mul_ps(c6_00,FF);
 268
 269             /* CUBIC SPLINE TABLE REPULSION */
 270             vfitab           = _mm_add_epi32(vfitab,ifour);
 271             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 272             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 273             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 274             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 275             _MM_TRANSPOSE4_PS(Y,F,G,H);
 276             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 277             VV               = _mm_macc_ps(vfeps,Fp,Y);
 278             vvdw12           = _mm_mul_ps(c12_00,VV);
 279             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 280             fvdw12           = _mm_mul_ps(c12_00,FF);
 281             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 282             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 283
 284             /* Update potential sum for this i atom from the interaction with this j atom. */
 285             velecsum         = _mm_add_ps(velecsum,velec);
 286             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 287
 288             fscal            = _mm_add_ps(felec,fvdw);
 289
 290              /* Update vectorial force */
 291             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 292             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 293             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 294
 295             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 296             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 297             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 298
 299             /**************************
 300              * CALCULATE INTERACTIONS *
 301              **************************/
 302
 303             /* Compute parameters for interactions between i and j atoms */
 304             qq10             = _mm_mul_ps(iq1,jq0);
 305
 306             /* REACTION-FIELD ELECTROSTATICS */
 307             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 308             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 309
 310             /* Update potential sum for this i atom from the interaction with this j atom. */
 311             velecsum         = _mm_add_ps(velecsum,velec);
 312
 313             fscal            = felec;
 314
 315              /* Update vectorial force */
 316             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 317             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 318             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 319
 320             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 321             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 322             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 323
 324             /**************************
 325              * CALCULATE INTERACTIONS *
 326              **************************/
 327
 328             /* Compute parameters for interactions between i and j atoms */
 329             qq20             = _mm_mul_ps(iq2,jq0);
 330
 331             /* REACTION-FIELD ELECTROSTATICS */
 332             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 333             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 334
 335             /* Update potential sum for this i atom from the interaction with this j atom. */
 336             velecsum         = _mm_add_ps(velecsum,velec);
 337
 338             fscal            = felec;
 339
 340              /* Update vectorial force */
 341             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 342             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 343             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 344
 345             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 346             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 347             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 348
 349             fjptrA             = f+j_coord_offsetA;
 350             fjptrB             = f+j_coord_offsetB;
 351             fjptrC             = f+j_coord_offsetC;
 352             fjptrD             = f+j_coord_offsetD;
 353
 354             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 355
 356             /* Inner loop uses 140 flops */
 357         }
 358
 359         if(jidx<j_index_end)
 360         {
 361
 362             /* Get j neighbor index, and coordinate index */
 363             jnrlistA         = jjnr[jidx];
 364             jnrlistB         = jjnr[jidx+1];
 365             jnrlistC         = jjnr[jidx+2];
 366             jnrlistD         = jjnr[jidx+3];
 367             /* Sign of each element will be negative for non-real atoms.
 368              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 369              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 370              */
 371             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 372             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 373             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 374             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 375             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 376             j_coord_offsetA  = DIM*jnrA;
 377             j_coord_offsetB  = DIM*jnrB;
 378             j_coord_offsetC  = DIM*jnrC;
 379             j_coord_offsetD  = DIM*jnrD;
 380
 381             /* load j atom coordinates */
 382             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 383                                               x+j_coord_offsetC,x+j_coord_offsetD,
 384                                               &jx0,&jy0,&jz0);
 385
 386             /* Calculate displacement vector */
 387             dx00             = _mm_sub_ps(ix0,jx0);
 388             dy00             = _mm_sub_ps(iy0,jy0);
 389             dz00             = _mm_sub_ps(iz0,jz0);
 390             dx10             = _mm_sub_ps(ix1,jx0);
 391             dy10             = _mm_sub_ps(iy1,jy0);
 392             dz10             = _mm_sub_ps(iz1,jz0);
 393             dx20             = _mm_sub_ps(ix2,jx0);
 394             dy20             = _mm_sub_ps(iy2,jy0);
 395             dz20             = _mm_sub_ps(iz2,jz0);
 396
 397             /* Calculate squared distance and things based on it */
 398             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 399             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 400             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 401
 402             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 403             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 404             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 405
 406             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 407             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 408             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 409
 410             /* Load parameters for j particles */
 411             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 412                                                               charge+jnrC+0,charge+jnrD+0);
 413             vdwjidx0A        = 2*vdwtype[jnrA+0];
 414             vdwjidx0B        = 2*vdwtype[jnrB+0];
 415             vdwjidx0C        = 2*vdwtype[jnrC+0];
 416             vdwjidx0D        = 2*vdwtype[jnrD+0];
 417
 418             fjx0             = _mm_setzero_ps();
 419             fjy0             = _mm_setzero_ps();
 420             fjz0             = _mm_setzero_ps();
 421
 422             /**************************
 423              * CALCULATE INTERACTIONS *
 424              **************************/
 425
 426             r00              = _mm_mul_ps(rsq00,rinv00);
 427             r00              = _mm_andnot_ps(dummy_mask,r00);
 428
 429             /* Compute parameters for interactions between i and j atoms */
 430             qq00             = _mm_mul_ps(iq0,jq0);
 431             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 432                                          vdwparam+vdwioffset0+vdwjidx0B,
 433                                          vdwparam+vdwioffset0+vdwjidx0C,
 434                                          vdwparam+vdwioffset0+vdwjidx0D,
 435                                          &c6_00,&c12_00);
 436
 437             /* Calculate table index by multiplying r with table scale and truncate to integer */
 438             rt               = _mm_mul_ps(r00,vftabscale);
 439             vfitab           = _mm_cvttps_epi32(rt);
 440 #ifdef __XOP__
 441             vfeps            = _mm_frcz_ps(rt);
 442 #else
 443             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 444 #endif
 445             twovfeps         = _mm_add_ps(vfeps,vfeps);
 446             vfitab           = _mm_slli_epi32(vfitab,3);
 447
 448             /* REACTION-FIELD ELECTROSTATICS */
 449             velec            = _mm_mul_ps(qq00,_mm_sub_ps(_mm_macc_ps(krf,rsq00,rinv00),crf));
 450             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 451
 452             /* CUBIC SPLINE TABLE DISPERSION */
 453             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 454             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 455             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 456             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 457             _MM_TRANSPOSE4_PS(Y,F,G,H);
 458             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 459             VV               = _mm_macc_ps(vfeps,Fp,Y);
 460             vvdw6            = _mm_mul_ps(c6_00,VV);
 461             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 462             fvdw6            = _mm_mul_ps(c6_00,FF);
 463
 464             /* CUBIC SPLINE TABLE REPULSION */
 465             vfitab           = _mm_add_epi32(vfitab,ifour);
 466             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 467             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 468             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 469             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 470             _MM_TRANSPOSE4_PS(Y,F,G,H);
 471             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 472             VV               = _mm_macc_ps(vfeps,Fp,Y);
 473             vvdw12           = _mm_mul_ps(c12_00,VV);
 474             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 475             fvdw12           = _mm_mul_ps(c12_00,FF);
 476             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 477             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 478
 479             /* Update potential sum for this i atom from the interaction with this j atom. */
 480             velec            = _mm_andnot_ps(dummy_mask,velec);
 481             velecsum         = _mm_add_ps(velecsum,velec);
 482             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 483             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 484
 485             fscal            = _mm_add_ps(felec,fvdw);
 486
 487             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 488
 489              /* Update vectorial force */
 490             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 491             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 492             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 493
 494             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 495             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 496             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             /* Compute parameters for interactions between i and j atoms */
 503             qq10             = _mm_mul_ps(iq1,jq0);
 504
 505             /* REACTION-FIELD ELECTROSTATICS */
 506             velec            = _mm_mul_ps(qq10,_mm_sub_ps(_mm_macc_ps(krf,rsq10,rinv10),crf));
 507             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 508
 509             /* Update potential sum for this i atom from the interaction with this j atom. */
 510             velec            = _mm_andnot_ps(dummy_mask,velec);
 511             velecsum         = _mm_add_ps(velecsum,velec);
 512
 513             fscal            = felec;
 514
 515             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 516
 517              /* Update vectorial force */
 518             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 519             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 520             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 521
 522             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 523             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 524             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 525
 526             /**************************
 527              * CALCULATE INTERACTIONS *
 528              **************************/
 529
 530             /* Compute parameters for interactions between i and j atoms */
 531             qq20             = _mm_mul_ps(iq2,jq0);
 532
 533             /* REACTION-FIELD ELECTROSTATICS */
 534             velec            = _mm_mul_ps(qq20,_mm_sub_ps(_mm_macc_ps(krf,rsq20,rinv20),crf));
 535             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 536
 537             /* Update potential sum for this i atom from the interaction with this j atom. */
 538             velec            = _mm_andnot_ps(dummy_mask,velec);
 539             velecsum         = _mm_add_ps(velecsum,velec);
 540
 541             fscal            = felec;
 542
 543             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 544
 545              /* Update vectorial force */
 546             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 547             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 548             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 549
 550             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 551             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 552             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 553
 554             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 555             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 556             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 557             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 558
 559             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 560
 561             /* Inner loop uses 141 flops */
 562         }
 563
 564         /* End of innermost loop */
 565
 566         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 567                                               f+i_coord_offset,fshift+i_shift_offset);
 568
 569         ggid                        = gid[iidx];
 570         /* Update potential energies */
 571         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 572         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 573
 574         /* Increment number of inner iterations */
 575         inneriter                  += j_index_end - j_index_start;
 576
 577         /* Outer loop uses 20 flops */
 578     }
 579
 580     /* Increment number of outer iterations */
 581     outeriter        += nri;
 582
 583     /* Update outer/inner flops */
 584
 585     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*141);
 586 }
 587 /*
 588  * Gromacs nonbonded kernel:   nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 589  * Electrostatics interaction: ReactionField
 590  * VdW interaction:            CubicSplineTable
 591  * Geometry:                   Water3-Particle
 592  * Calculate force/pot:        Force
 593  */
 594 void
 595 nb_kernel_ElecRF_VdwCSTab_GeomW3P1_F_avx_128_fma_single
 596                     (t_nblist * gmx_restrict                nlist,
 597                      rvec * gmx_restrict                    xx,
 598                      rvec * gmx_restrict                    ff,
 599                      t_forcerec * gmx_restrict              fr,
 600                      t_mdatoms * gmx_restrict               mdatoms,
 601                      nb_kernel_data_t * gmx_restrict        kernel_data,
 602                      t_nrnb * gmx_restrict                  nrnb)
 603 {
 604     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 605      * just 0 for non-waters.
 606      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 607      * jnr indices corresponding to data put in the four positions in the SIMD register.
 608      */
 609     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 610     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 611     int              jnrA,jnrB,jnrC,jnrD;
 612     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 613     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 614     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 615     real             rcutoff_scalar;
 616     real             *shiftvec,*fshift,*x,*f;
 617     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 618     real             scratch[4*DIM];
 619     __m128           fscal,rcutoff,rcutoff2,jidxall;
 620     int              vdwioffset0;
 621     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 622     int              vdwioffset1;
 623     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 624     int              vdwioffset2;
 625     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 626     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 627     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 628     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 629     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 630     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 631     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 632     real             *charge;
 633     int              nvdwtype;
 634     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 635     int              *vdwtype;
 636     real             *vdwparam;
 637     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 638     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 639     __m128i          vfitab;
 640     __m128i          ifour       = _mm_set1_epi32(4);
 641     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 642     real             *vftab;
 643     __m128           dummy_mask,cutoff_mask;
 644     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 645     __m128           one     = _mm_set1_ps(1.0);
 646     __m128           two     = _mm_set1_ps(2.0);
 647     x                = xx[0];
 648     f                = ff[0];
 649
 650     nri              = nlist->nri;
 651     iinr             = nlist->iinr;
 652     jindex           = nlist->jindex;
 653     jjnr             = nlist->jjnr;
 654     shiftidx         = nlist->shift;
 655     gid              = nlist->gid;
 656     shiftvec         = fr->shift_vec[0];
 657     fshift           = fr->fshift[0];
 658     facel            = _mm_set1_ps(fr->epsfac);
 659     charge           = mdatoms->chargeA;
 660     krf              = _mm_set1_ps(fr->ic->k_rf);
 661     krf2             = _mm_set1_ps(fr->ic->k_rf*2.0);
 662     crf              = _mm_set1_ps(fr->ic->c_rf);
 663     nvdwtype         = fr->ntype;
 664     vdwparam         = fr->nbfp;
 665     vdwtype          = mdatoms->typeA;
 666
 667     vftab            = kernel_data->table_vdw->data;
 668     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 669
 670     /* Setup water-specific parameters */
 671     inr              = nlist->iinr[0];
 672     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 673     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 674     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 675     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 676
 677     /* Avoid stupid compiler warnings */
 678     jnrA = jnrB = jnrC = jnrD = 0;
 679     j_coord_offsetA = 0;
 680     j_coord_offsetB = 0;
 681     j_coord_offsetC = 0;
 682     j_coord_offsetD = 0;
 683
 684     outeriter        = 0;
 685     inneriter        = 0;
 686
 687     for(iidx=0;iidx<4*DIM;iidx++)
 688     {
 689         scratch[iidx] = 0.0;
 690     }
 691
 692     /* Start outer loop over neighborlists */
 693     for(iidx=0; iidx<nri; iidx++)
 694     {
 695         /* Load shift vector for this list */
 696         i_shift_offset   = DIM*shiftidx[iidx];
 697
 698         /* Load limits for loop over neighbors */
 699         j_index_start    = jindex[iidx];
 700         j_index_end      = jindex[iidx+1];
 701
 702         /* Get outer coordinate index */
 703         inr              = iinr[iidx];
 704         i_coord_offset   = DIM*inr;
 705
 706         /* Load i particle coords and add shift vector */
 707         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 708                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 709
 710         fix0             = _mm_setzero_ps();
 711         fiy0             = _mm_setzero_ps();
 712         fiz0             = _mm_setzero_ps();
 713         fix1             = _mm_setzero_ps();
 714         fiy1             = _mm_setzero_ps();
 715         fiz1             = _mm_setzero_ps();
 716         fix2             = _mm_setzero_ps();
 717         fiy2             = _mm_setzero_ps();
 718         fiz2             = _mm_setzero_ps();
 719
 720         /* Start inner kernel loop */
 721         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 722         {
 723
 724             /* Get j neighbor index, and coordinate index */
 725             jnrA             = jjnr[jidx];
 726             jnrB             = jjnr[jidx+1];
 727             jnrC             = jjnr[jidx+2];
 728             jnrD             = jjnr[jidx+3];
 729             j_coord_offsetA  = DIM*jnrA;
 730             j_coord_offsetB  = DIM*jnrB;
 731             j_coord_offsetC  = DIM*jnrC;
 732             j_coord_offsetD  = DIM*jnrD;
 733
 734             /* load j atom coordinates */
 735             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 736                                               x+j_coord_offsetC,x+j_coord_offsetD,
 737                                               &jx0,&jy0,&jz0);
 738
 739             /* Calculate displacement vector */
 740             dx00             = _mm_sub_ps(ix0,jx0);
 741             dy00             = _mm_sub_ps(iy0,jy0);
 742             dz00             = _mm_sub_ps(iz0,jz0);
 743             dx10             = _mm_sub_ps(ix1,jx0);
 744             dy10             = _mm_sub_ps(iy1,jy0);
 745             dz10             = _mm_sub_ps(iz1,jz0);
 746             dx20             = _mm_sub_ps(ix2,jx0);
 747             dy20             = _mm_sub_ps(iy2,jy0);
 748             dz20             = _mm_sub_ps(iz2,jz0);
 749
 750             /* Calculate squared distance and things based on it */
 751             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 752             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 753             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 754
 755             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 756             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 757             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 758
 759             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 760             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 761             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 762
 763             /* Load parameters for j particles */
 764             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 765                                                               charge+jnrC+0,charge+jnrD+0);
 766             vdwjidx0A        = 2*vdwtype[jnrA+0];
 767             vdwjidx0B        = 2*vdwtype[jnrB+0];
 768             vdwjidx0C        = 2*vdwtype[jnrC+0];
 769             vdwjidx0D        = 2*vdwtype[jnrD+0];
 770
 771             fjx0             = _mm_setzero_ps();
 772             fjy0             = _mm_setzero_ps();
 773             fjz0             = _mm_setzero_ps();
 774
 775             /**************************
 776              * CALCULATE INTERACTIONS *
 777              **************************/
 778
 779             r00              = _mm_mul_ps(rsq00,rinv00);
 780
 781             /* Compute parameters for interactions between i and j atoms */
 782             qq00             = _mm_mul_ps(iq0,jq0);
 783             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 784                                          vdwparam+vdwioffset0+vdwjidx0B,
 785                                          vdwparam+vdwioffset0+vdwjidx0C,
 786                                          vdwparam+vdwioffset0+vdwjidx0D,
 787                                          &c6_00,&c12_00);
 788
 789             /* Calculate table index by multiplying r with table scale and truncate to integer */
 790             rt               = _mm_mul_ps(r00,vftabscale);
 791             vfitab           = _mm_cvttps_epi32(rt);
 792 #ifdef __XOP__
 793             vfeps            = _mm_frcz_ps(rt);
 794 #else
 795             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 796 #endif
 797             twovfeps         = _mm_add_ps(vfeps,vfeps);
 798             vfitab           = _mm_slli_epi32(vfitab,3);
 799
 800             /* REACTION-FIELD ELECTROSTATICS */
 801             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 802
 803             /* CUBIC SPLINE TABLE DISPERSION */
 804             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 805             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 806             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 807             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 808             _MM_TRANSPOSE4_PS(Y,F,G,H);
 809             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 810             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 811             fvdw6            = _mm_mul_ps(c6_00,FF);
 812
 813             /* CUBIC SPLINE TABLE REPULSION */
 814             vfitab           = _mm_add_epi32(vfitab,ifour);
 815             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 816             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 817             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 818             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 819             _MM_TRANSPOSE4_PS(Y,F,G,H);
 820             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 821             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 822             fvdw12           = _mm_mul_ps(c12_00,FF);
 823             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 824
 825             fscal            = _mm_add_ps(felec,fvdw);
 826
 827              /* Update vectorial force */
 828             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 829             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 830             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 831
 832             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 833             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 834             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 835
 836             /**************************
 837              * CALCULATE INTERACTIONS *
 838              **************************/
 839
 840             /* Compute parameters for interactions between i and j atoms */
 841             qq10             = _mm_mul_ps(iq1,jq0);
 842
 843             /* REACTION-FIELD ELECTROSTATICS */
 844             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
 845
 846             fscal            = felec;
 847
 848              /* Update vectorial force */
 849             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 850             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 851             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 852
 853             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 854             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 855             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 856
 857             /**************************
 858              * CALCULATE INTERACTIONS *
 859              **************************/
 860
 861             /* Compute parameters for interactions between i and j atoms */
 862             qq20             = _mm_mul_ps(iq2,jq0);
 863
 864             /* REACTION-FIELD ELECTROSTATICS */
 865             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
 866
 867             fscal            = felec;
 868
 869              /* Update vectorial force */
 870             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 871             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 872             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 873
 874             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 875             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 876             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 877
 878             fjptrA             = f+j_coord_offsetA;
 879             fjptrB             = f+j_coord_offsetB;
 880             fjptrC             = f+j_coord_offsetC;
 881             fjptrD             = f+j_coord_offsetD;
 882
 883             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 884
 885             /* Inner loop uses 117 flops */
 886         }
 887
 888         if(jidx<j_index_end)
 889         {
 890
 891             /* Get j neighbor index, and coordinate index */
 892             jnrlistA         = jjnr[jidx];
 893             jnrlistB         = jjnr[jidx+1];
 894             jnrlistC         = jjnr[jidx+2];
 895             jnrlistD         = jjnr[jidx+3];
 896             /* Sign of each element will be negative for non-real atoms.
 897              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 898              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 899              */
 900             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 901             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 902             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 903             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 904             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 905             j_coord_offsetA  = DIM*jnrA;
 906             j_coord_offsetB  = DIM*jnrB;
 907             j_coord_offsetC  = DIM*jnrC;
 908             j_coord_offsetD  = DIM*jnrD;
 909
 910             /* load j atom coordinates */
 911             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 912                                               x+j_coord_offsetC,x+j_coord_offsetD,
 913                                               &jx0,&jy0,&jz0);
 914
 915             /* Calculate displacement vector */
 916             dx00             = _mm_sub_ps(ix0,jx0);
 917             dy00             = _mm_sub_ps(iy0,jy0);
 918             dz00             = _mm_sub_ps(iz0,jz0);
 919             dx10             = _mm_sub_ps(ix1,jx0);
 920             dy10             = _mm_sub_ps(iy1,jy0);
 921             dz10             = _mm_sub_ps(iz1,jz0);
 922             dx20             = _mm_sub_ps(ix2,jx0);
 923             dy20             = _mm_sub_ps(iy2,jy0);
 924             dz20             = _mm_sub_ps(iz2,jz0);
 925
 926             /* Calculate squared distance and things based on it */
 927             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 928             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 929             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 930
 931             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 932             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 933             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 934
 935             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 936             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 937             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 938
 939             /* Load parameters for j particles */
 940             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 941                                                               charge+jnrC+0,charge+jnrD+0);
 942             vdwjidx0A        = 2*vdwtype[jnrA+0];
 943             vdwjidx0B        = 2*vdwtype[jnrB+0];
 944             vdwjidx0C        = 2*vdwtype[jnrC+0];
 945             vdwjidx0D        = 2*vdwtype[jnrD+0];
 946
 947             fjx0             = _mm_setzero_ps();
 948             fjy0             = _mm_setzero_ps();
 949             fjz0             = _mm_setzero_ps();
 950
 951             /**************************
 952              * CALCULATE INTERACTIONS *
 953              **************************/
 954
 955             r00              = _mm_mul_ps(rsq00,rinv00);
 956             r00              = _mm_andnot_ps(dummy_mask,r00);
 957
 958             /* Compute parameters for interactions between i and j atoms */
 959             qq00             = _mm_mul_ps(iq0,jq0);
 960             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 961                                          vdwparam+vdwioffset0+vdwjidx0B,
 962                                          vdwparam+vdwioffset0+vdwjidx0C,
 963                                          vdwparam+vdwioffset0+vdwjidx0D,
 964                                          &c6_00,&c12_00);
 965
 966             /* Calculate table index by multiplying r with table scale and truncate to integer */
 967             rt               = _mm_mul_ps(r00,vftabscale);
 968             vfitab           = _mm_cvttps_epi32(rt);
 969 #ifdef __XOP__
 970             vfeps            = _mm_frcz_ps(rt);
 971 #else
 972             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 973 #endif
 974             twovfeps         = _mm_add_ps(vfeps,vfeps);
 975             vfitab           = _mm_slli_epi32(vfitab,3);
 976
 977             /* REACTION-FIELD ELECTROSTATICS */
 978             felec            = _mm_mul_ps(qq00,_mm_msub_ps(rinv00,rinvsq00,krf2));
 979
 980             /* CUBIC SPLINE TABLE DISPERSION */
 981             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 982             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 983             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 984             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 985             _MM_TRANSPOSE4_PS(Y,F,G,H);
 986             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 987             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 988             fvdw6            = _mm_mul_ps(c6_00,FF);
 989
 990             /* CUBIC SPLINE TABLE REPULSION */
 991             vfitab           = _mm_add_epi32(vfitab,ifour);
 992             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 993             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 994             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 995             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 996             _MM_TRANSPOSE4_PS(Y,F,G,H);
 997             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 998             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 999             fvdw12           = _mm_mul_ps(c12_00,FF);
1000             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1001
1002             fscal            = _mm_add_ps(felec,fvdw);
1003
1004             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1005
1006              /* Update vectorial force */
1007             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1008             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1009             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1010
1011             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1012             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1013             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1014
1015             /**************************
1016              * CALCULATE INTERACTIONS *
1017              **************************/
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq10             = _mm_mul_ps(iq1,jq0);
1021
1022             /* REACTION-FIELD ELECTROSTATICS */
1023             felec            = _mm_mul_ps(qq10,_mm_msub_ps(rinv10,rinvsq10,krf2));
1024
1025             fscal            = felec;
1026
1027             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1028
1029              /* Update vectorial force */
1030             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1031             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1032             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1033
1034             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1035             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1036             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1037
1038             /**************************
1039              * CALCULATE INTERACTIONS *
1040              **************************/
1041
1042             /* Compute parameters for interactions between i and j atoms */
1043             qq20             = _mm_mul_ps(iq2,jq0);
1044
1045             /* REACTION-FIELD ELECTROSTATICS */
1046             felec            = _mm_mul_ps(qq20,_mm_msub_ps(rinv20,rinvsq20,krf2));
1047
1048             fscal            = felec;
1049
1050             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1051
1052              /* Update vectorial force */
1053             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1054             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1055             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1056
1057             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1058             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1059             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1060
1061             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1062             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1063             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1064             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1065
1066             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1067
1068             /* Inner loop uses 118 flops */
1069         }
1070
1071         /* End of innermost loop */
1072
1073         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1074                                               f+i_coord_offset,fshift+i_shift_offset);
1075
1076         /* Increment number of inner iterations */
1077         inneriter                  += j_index_end - j_index_start;
1078
1079         /* Outer loop uses 18 flops */
1080     }
1081
1082     /* Increment number of outer iterations */
1083     outeriter        += nri;
1084
1085     /* Update outer/inner flops */
1086
1087     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*118);
1088 }