src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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_ElecCSTab_VdwCSTab_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            CubicSplineTable
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_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              vdwioffset3;
  76     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
  77     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  78     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  79     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  80     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  81     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  82     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
  83     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  84     real             *charge;
  85     int              nvdwtype;
  86     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  87     int              *vdwtype;
  88     real             *vdwparam;
  89     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
  90     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
  91     __m128i          vfitab;
  92     __m128i          ifour       = _mm_set1_epi32(4);
  93     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
  94     real             *vftab;
  95     __m128           dummy_mask,cutoff_mask;
  96     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
  97     __m128           one     = _mm_set1_ps(1.0);
  98     __m128           two     = _mm_set1_ps(2.0);
  99     x                = xx[0];
 100     f                = ff[0];
 101
 102     nri              = nlist->nri;
 103     iinr             = nlist->iinr;
 104     jindex           = nlist->jindex;
 105     jjnr             = nlist->jjnr;
 106     shiftidx         = nlist->shift;
 107     gid              = nlist->gid;
 108     shiftvec         = fr->shift_vec[0];
 109     fshift           = fr->fshift[0];
 110     facel            = _mm_set1_ps(fr->epsfac);
 111     charge           = mdatoms->chargeA;
 112     nvdwtype         = fr->ntype;
 113     vdwparam         = fr->nbfp;
 114     vdwtype          = mdatoms->typeA;
 115
 116     vftab            = kernel_data->table_elec_vdw->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 118
 119     /* Setup water-specific parameters */
 120     inr              = nlist->iinr[0];
 121     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 122     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 123     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 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_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 157                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 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         fix3             = _mm_setzero_ps();
 169         fiy3             = _mm_setzero_ps();
 170         fiz3             = _mm_setzero_ps();
 171
 172         /* Reset potential sums */
 173         velecsum         = _mm_setzero_ps();
 174         vvdwsum          = _mm_setzero_ps();
 175
 176         /* Start inner kernel loop */
 177         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 178         {
 179
 180             /* Get j neighbor index, and coordinate index */
 181             jnrA             = jjnr[jidx];
 182             jnrB             = jjnr[jidx+1];
 183             jnrC             = jjnr[jidx+2];
 184             jnrD             = jjnr[jidx+3];
 185             j_coord_offsetA  = DIM*jnrA;
 186             j_coord_offsetB  = DIM*jnrB;
 187             j_coord_offsetC  = DIM*jnrC;
 188             j_coord_offsetD  = DIM*jnrD;
 189
 190             /* load j atom coordinates */
 191             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 192                                               x+j_coord_offsetC,x+j_coord_offsetD,
 193                                               &jx0,&jy0,&jz0);
 194
 195             /* Calculate displacement vector */
 196             dx00             = _mm_sub_ps(ix0,jx0);
 197             dy00             = _mm_sub_ps(iy0,jy0);
 198             dz00             = _mm_sub_ps(iz0,jz0);
 199             dx10             = _mm_sub_ps(ix1,jx0);
 200             dy10             = _mm_sub_ps(iy1,jy0);
 201             dz10             = _mm_sub_ps(iz1,jz0);
 202             dx20             = _mm_sub_ps(ix2,jx0);
 203             dy20             = _mm_sub_ps(iy2,jy0);
 204             dz20             = _mm_sub_ps(iz2,jz0);
 205             dx30             = _mm_sub_ps(ix3,jx0);
 206             dy30             = _mm_sub_ps(iy3,jy0);
 207             dz30             = _mm_sub_ps(iz3,jz0);
 208
 209             /* Calculate squared distance and things based on it */
 210             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 211             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 212             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 213             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 214
 215             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 216             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 217             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 218             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 219
 220             /* Load parameters for j particles */
 221             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 222                                                               charge+jnrC+0,charge+jnrD+0);
 223             vdwjidx0A        = 2*vdwtype[jnrA+0];
 224             vdwjidx0B        = 2*vdwtype[jnrB+0];
 225             vdwjidx0C        = 2*vdwtype[jnrC+0];
 226             vdwjidx0D        = 2*vdwtype[jnrD+0];
 227
 228             fjx0             = _mm_setzero_ps();
 229             fjy0             = _mm_setzero_ps();
 230             fjz0             = _mm_setzero_ps();
 231
 232             /**************************
 233              * CALCULATE INTERACTIONS *
 234              **************************/
 235
 236             r00              = _mm_mul_ps(rsq00,rinv00);
 237
 238             /* Compute parameters for interactions between i and j atoms */
 239             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 240                                          vdwparam+vdwioffset0+vdwjidx0B,
 241                                          vdwparam+vdwioffset0+vdwjidx0C,
 242                                          vdwparam+vdwioffset0+vdwjidx0D,
 243                                          &c6_00,&c12_00);
 244
 245             /* Calculate table index by multiplying r with table scale and truncate to integer */
 246             rt               = _mm_mul_ps(r00,vftabscale);
 247             vfitab           = _mm_cvttps_epi32(rt);
 248 #ifdef __XOP__
 249             vfeps            = _mm_frcz_ps(rt);
 250 #else
 251             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 252 #endif
 253             twovfeps         = _mm_add_ps(vfeps,vfeps);
 254             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 255
 256             /* CUBIC SPLINE TABLE DISPERSION */
 257             vfitab           = _mm_add_epi32(vfitab,ifour);
 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             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 286
 287             fscal            = fvdw;
 288
 289              /* Update vectorial force */
 290             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 291             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 292             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 293
 294             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 295             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 296             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 297
 298             /**************************
 299              * CALCULATE INTERACTIONS *
 300              **************************/
 301
 302             r10              = _mm_mul_ps(rsq10,rinv10);
 303
 304             /* Compute parameters for interactions between i and j atoms */
 305             qq10             = _mm_mul_ps(iq1,jq0);
 306
 307             /* Calculate table index by multiplying r with table scale and truncate to integer */
 308             rt               = _mm_mul_ps(r10,vftabscale);
 309             vfitab           = _mm_cvttps_epi32(rt);
 310 #ifdef __XOP__
 311             vfeps            = _mm_frcz_ps(rt);
 312 #else
 313             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 314 #endif
 315             twovfeps         = _mm_add_ps(vfeps,vfeps);
 316             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 317
 318             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 319             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 320             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 321             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 322             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 323             _MM_TRANSPOSE4_PS(Y,F,G,H);
 324             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 325             VV               = _mm_macc_ps(vfeps,Fp,Y);
 326             velec            = _mm_mul_ps(qq10,VV);
 327             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 328             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 329
 330             /* Update potential sum for this i atom from the interaction with this j atom. */
 331             velecsum         = _mm_add_ps(velecsum,velec);
 332
 333             fscal            = felec;
 334
 335              /* Update vectorial force */
 336             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 337             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 338             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 339
 340             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 341             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 342             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 343
 344             /**************************
 345              * CALCULATE INTERACTIONS *
 346              **************************/
 347
 348             r20              = _mm_mul_ps(rsq20,rinv20);
 349
 350             /* Compute parameters for interactions between i and j atoms */
 351             qq20             = _mm_mul_ps(iq2,jq0);
 352
 353             /* Calculate table index by multiplying r with table scale and truncate to integer */
 354             rt               = _mm_mul_ps(r20,vftabscale);
 355             vfitab           = _mm_cvttps_epi32(rt);
 356 #ifdef __XOP__
 357             vfeps            = _mm_frcz_ps(rt);
 358 #else
 359             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 360 #endif
 361             twovfeps         = _mm_add_ps(vfeps,vfeps);
 362             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 363
 364             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 365             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 366             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 367             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 368             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 369             _MM_TRANSPOSE4_PS(Y,F,G,H);
 370             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 371             VV               = _mm_macc_ps(vfeps,Fp,Y);
 372             velec            = _mm_mul_ps(qq20,VV);
 373             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 374             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 375
 376             /* Update potential sum for this i atom from the interaction with this j atom. */
 377             velecsum         = _mm_add_ps(velecsum,velec);
 378
 379             fscal            = felec;
 380
 381              /* Update vectorial force */
 382             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 383             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 384             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 385
 386             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 387             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 388             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 389
 390             /**************************
 391              * CALCULATE INTERACTIONS *
 392              **************************/
 393
 394             r30              = _mm_mul_ps(rsq30,rinv30);
 395
 396             /* Compute parameters for interactions between i and j atoms */
 397             qq30             = _mm_mul_ps(iq3,jq0);
 398
 399             /* Calculate table index by multiplying r with table scale and truncate to integer */
 400             rt               = _mm_mul_ps(r30,vftabscale);
 401             vfitab           = _mm_cvttps_epi32(rt);
 402 #ifdef __XOP__
 403             vfeps            = _mm_frcz_ps(rt);
 404 #else
 405             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 406 #endif
 407             twovfeps         = _mm_add_ps(vfeps,vfeps);
 408             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 409
 410             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 411             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 412             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 413             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 414             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 415             _MM_TRANSPOSE4_PS(Y,F,G,H);
 416             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 417             VV               = _mm_macc_ps(vfeps,Fp,Y);
 418             velec            = _mm_mul_ps(qq30,VV);
 419             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 420             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 421
 422             /* Update potential sum for this i atom from the interaction with this j atom. */
 423             velecsum         = _mm_add_ps(velecsum,velec);
 424
 425             fscal            = felec;
 426
 427              /* Update vectorial force */
 428             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 429             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 430             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 431
 432             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 433             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 434             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 435
 436             fjptrA             = f+j_coord_offsetA;
 437             fjptrB             = f+j_coord_offsetB;
 438             fjptrC             = f+j_coord_offsetC;
 439             fjptrD             = f+j_coord_offsetD;
 440
 441             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 442
 443             /* Inner loop uses 197 flops */
 444         }
 445
 446         if(jidx<j_index_end)
 447         {
 448
 449             /* Get j neighbor index, and coordinate index */
 450             jnrlistA         = jjnr[jidx];
 451             jnrlistB         = jjnr[jidx+1];
 452             jnrlistC         = jjnr[jidx+2];
 453             jnrlistD         = jjnr[jidx+3];
 454             /* Sign of each element will be negative for non-real atoms.
 455              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 456              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 457              */
 458             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 459             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 460             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 461             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 462             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 463             j_coord_offsetA  = DIM*jnrA;
 464             j_coord_offsetB  = DIM*jnrB;
 465             j_coord_offsetC  = DIM*jnrC;
 466             j_coord_offsetD  = DIM*jnrD;
 467
 468             /* load j atom coordinates */
 469             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 470                                               x+j_coord_offsetC,x+j_coord_offsetD,
 471                                               &jx0,&jy0,&jz0);
 472
 473             /* Calculate displacement vector */
 474             dx00             = _mm_sub_ps(ix0,jx0);
 475             dy00             = _mm_sub_ps(iy0,jy0);
 476             dz00             = _mm_sub_ps(iz0,jz0);
 477             dx10             = _mm_sub_ps(ix1,jx0);
 478             dy10             = _mm_sub_ps(iy1,jy0);
 479             dz10             = _mm_sub_ps(iz1,jz0);
 480             dx20             = _mm_sub_ps(ix2,jx0);
 481             dy20             = _mm_sub_ps(iy2,jy0);
 482             dz20             = _mm_sub_ps(iz2,jz0);
 483             dx30             = _mm_sub_ps(ix3,jx0);
 484             dy30             = _mm_sub_ps(iy3,jy0);
 485             dz30             = _mm_sub_ps(iz3,jz0);
 486
 487             /* Calculate squared distance and things based on it */
 488             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 489             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 490             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 491             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 492
 493             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 494             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 495             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 496             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 497
 498             /* Load parameters for j particles */
 499             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 500                                                               charge+jnrC+0,charge+jnrD+0);
 501             vdwjidx0A        = 2*vdwtype[jnrA+0];
 502             vdwjidx0B        = 2*vdwtype[jnrB+0];
 503             vdwjidx0C        = 2*vdwtype[jnrC+0];
 504             vdwjidx0D        = 2*vdwtype[jnrD+0];
 505
 506             fjx0             = _mm_setzero_ps();
 507             fjy0             = _mm_setzero_ps();
 508             fjz0             = _mm_setzero_ps();
 509
 510             /**************************
 511              * CALCULATE INTERACTIONS *
 512              **************************/
 513
 514             r00              = _mm_mul_ps(rsq00,rinv00);
 515             r00              = _mm_andnot_ps(dummy_mask,r00);
 516
 517             /* Compute parameters for interactions between i and j atoms */
 518             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 519                                          vdwparam+vdwioffset0+vdwjidx0B,
 520                                          vdwparam+vdwioffset0+vdwjidx0C,
 521                                          vdwparam+vdwioffset0+vdwjidx0D,
 522                                          &c6_00,&c12_00);
 523
 524             /* Calculate table index by multiplying r with table scale and truncate to integer */
 525             rt               = _mm_mul_ps(r00,vftabscale);
 526             vfitab           = _mm_cvttps_epi32(rt);
 527 #ifdef __XOP__
 528             vfeps            = _mm_frcz_ps(rt);
 529 #else
 530             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 531 #endif
 532             twovfeps         = _mm_add_ps(vfeps,vfeps);
 533             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 534
 535             /* CUBIC SPLINE TABLE DISPERSION */
 536             vfitab           = _mm_add_epi32(vfitab,ifour);
 537             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 539             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 540             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 541             _MM_TRANSPOSE4_PS(Y,F,G,H);
 542             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 543             VV               = _mm_macc_ps(vfeps,Fp,Y);
 544             vvdw6            = _mm_mul_ps(c6_00,VV);
 545             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 546             fvdw6            = _mm_mul_ps(c6_00,FF);
 547
 548             /* CUBIC SPLINE TABLE REPULSION */
 549             vfitab           = _mm_add_epi32(vfitab,ifour);
 550             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 551             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 552             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 553             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 554             _MM_TRANSPOSE4_PS(Y,F,G,H);
 555             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 556             VV               = _mm_macc_ps(vfeps,Fp,Y);
 557             vvdw12           = _mm_mul_ps(c12_00,VV);
 558             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 559             fvdw12           = _mm_mul_ps(c12_00,FF);
 560             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 561             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 562
 563             /* Update potential sum for this i atom from the interaction with this j atom. */
 564             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 565             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 566
 567             fscal            = fvdw;
 568
 569             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 570
 571              /* Update vectorial force */
 572             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 573             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 574             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 575
 576             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 577             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 578             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 579
 580             /**************************
 581              * CALCULATE INTERACTIONS *
 582              **************************/
 583
 584             r10              = _mm_mul_ps(rsq10,rinv10);
 585             r10              = _mm_andnot_ps(dummy_mask,r10);
 586
 587             /* Compute parameters for interactions between i and j atoms */
 588             qq10             = _mm_mul_ps(iq1,jq0);
 589
 590             /* Calculate table index by multiplying r with table scale and truncate to integer */
 591             rt               = _mm_mul_ps(r10,vftabscale);
 592             vfitab           = _mm_cvttps_epi32(rt);
 593 #ifdef __XOP__
 594             vfeps            = _mm_frcz_ps(rt);
 595 #else
 596             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 597 #endif
 598             twovfeps         = _mm_add_ps(vfeps,vfeps);
 599             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 600
 601             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 602             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 603             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 604             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 605             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 606             _MM_TRANSPOSE4_PS(Y,F,G,H);
 607             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 608             VV               = _mm_macc_ps(vfeps,Fp,Y);
 609             velec            = _mm_mul_ps(qq10,VV);
 610             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 611             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 612
 613             /* Update potential sum for this i atom from the interaction with this j atom. */
 614             velec            = _mm_andnot_ps(dummy_mask,velec);
 615             velecsum         = _mm_add_ps(velecsum,velec);
 616
 617             fscal            = felec;
 618
 619             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 620
 621              /* Update vectorial force */
 622             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 623             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 624             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 625
 626             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 627             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 628             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 629
 630             /**************************
 631              * CALCULATE INTERACTIONS *
 632              **************************/
 633
 634             r20              = _mm_mul_ps(rsq20,rinv20);
 635             r20              = _mm_andnot_ps(dummy_mask,r20);
 636
 637             /* Compute parameters for interactions between i and j atoms */
 638             qq20             = _mm_mul_ps(iq2,jq0);
 639
 640             /* Calculate table index by multiplying r with table scale and truncate to integer */
 641             rt               = _mm_mul_ps(r20,vftabscale);
 642             vfitab           = _mm_cvttps_epi32(rt);
 643 #ifdef __XOP__
 644             vfeps            = _mm_frcz_ps(rt);
 645 #else
 646             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 647 #endif
 648             twovfeps         = _mm_add_ps(vfeps,vfeps);
 649             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 650
 651             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 652             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 653             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 654             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 655             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 656             _MM_TRANSPOSE4_PS(Y,F,G,H);
 657             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 658             VV               = _mm_macc_ps(vfeps,Fp,Y);
 659             velec            = _mm_mul_ps(qq20,VV);
 660             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 661             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 662
 663             /* Update potential sum for this i atom from the interaction with this j atom. */
 664             velec            = _mm_andnot_ps(dummy_mask,velec);
 665             velecsum         = _mm_add_ps(velecsum,velec);
 666
 667             fscal            = felec;
 668
 669             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 670
 671              /* Update vectorial force */
 672             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 673             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 674             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 675
 676             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 677             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 678             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 679
 680             /**************************
 681              * CALCULATE INTERACTIONS *
 682              **************************/
 683
 684             r30              = _mm_mul_ps(rsq30,rinv30);
 685             r30              = _mm_andnot_ps(dummy_mask,r30);
 686
 687             /* Compute parameters for interactions between i and j atoms */
 688             qq30             = _mm_mul_ps(iq3,jq0);
 689
 690             /* Calculate table index by multiplying r with table scale and truncate to integer */
 691             rt               = _mm_mul_ps(r30,vftabscale);
 692             vfitab           = _mm_cvttps_epi32(rt);
 693 #ifdef __XOP__
 694             vfeps            = _mm_frcz_ps(rt);
 695 #else
 696             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 697 #endif
 698             twovfeps         = _mm_add_ps(vfeps,vfeps);
 699             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 700
 701             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 702             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 703             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 704             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 705             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 706             _MM_TRANSPOSE4_PS(Y,F,G,H);
 707             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 708             VV               = _mm_macc_ps(vfeps,Fp,Y);
 709             velec            = _mm_mul_ps(qq30,VV);
 710             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 711             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 712
 713             /* Update potential sum for this i atom from the interaction with this j atom. */
 714             velec            = _mm_andnot_ps(dummy_mask,velec);
 715             velecsum         = _mm_add_ps(velecsum,velec);
 716
 717             fscal            = felec;
 718
 719             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 720
 721              /* Update vectorial force */
 722             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 723             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 724             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 725
 726             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 727             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 728             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 729
 730             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 731             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 732             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 733             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 734
 735             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 736
 737             /* Inner loop uses 201 flops */
 738         }
 739
 740         /* End of innermost loop */
 741
 742         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 743                                               f+i_coord_offset,fshift+i_shift_offset);
 744
 745         ggid                        = gid[iidx];
 746         /* Update potential energies */
 747         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 748         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 749
 750         /* Increment number of inner iterations */
 751         inneriter                  += j_index_end - j_index_start;
 752
 753         /* Outer loop uses 26 flops */
 754     }
 755
 756     /* Increment number of outer iterations */
 757     outeriter        += nri;
 758
 759     /* Update outer/inner flops */
 760
 761     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*201);
 762 }
 763 /*
 764  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 765  * Electrostatics interaction: CubicSplineTable
 766  * VdW interaction:            CubicSplineTable
 767  * Geometry:                   Water4-Particle
 768  * Calculate force/pot:        Force
 769  */
 770 void
 771 nb_kernel_ElecCSTab_VdwCSTab_GeomW4P1_F_avx_128_fma_single
 772                     (t_nblist * gmx_restrict                nlist,
 773                      rvec * gmx_restrict                    xx,
 774                      rvec * gmx_restrict                    ff,
 775                      t_forcerec * gmx_restrict              fr,
 776                      t_mdatoms * gmx_restrict               mdatoms,
 777                      nb_kernel_data_t * gmx_restrict        kernel_data,
 778                      t_nrnb * gmx_restrict                  nrnb)
 779 {
 780     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 781      * just 0 for non-waters.
 782      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 783      * jnr indices corresponding to data put in the four positions in the SIMD register.
 784      */
 785     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 786     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 787     int              jnrA,jnrB,jnrC,jnrD;
 788     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 789     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 790     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 791     real             rcutoff_scalar;
 792     real             *shiftvec,*fshift,*x,*f;
 793     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 794     real             scratch[4*DIM];
 795     __m128           fscal,rcutoff,rcutoff2,jidxall;
 796     int              vdwioffset0;
 797     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 798     int              vdwioffset1;
 799     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 800     int              vdwioffset2;
 801     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 802     int              vdwioffset3;
 803     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 804     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 805     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 806     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 807     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 808     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 809     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 810     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 811     real             *charge;
 812     int              nvdwtype;
 813     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 814     int              *vdwtype;
 815     real             *vdwparam;
 816     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 817     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 818     __m128i          vfitab;
 819     __m128i          ifour       = _mm_set1_epi32(4);
 820     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 821     real             *vftab;
 822     __m128           dummy_mask,cutoff_mask;
 823     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 824     __m128           one     = _mm_set1_ps(1.0);
 825     __m128           two     = _mm_set1_ps(2.0);
 826     x                = xx[0];
 827     f                = ff[0];
 828
 829     nri              = nlist->nri;
 830     iinr             = nlist->iinr;
 831     jindex           = nlist->jindex;
 832     jjnr             = nlist->jjnr;
 833     shiftidx         = nlist->shift;
 834     gid              = nlist->gid;
 835     shiftvec         = fr->shift_vec[0];
 836     fshift           = fr->fshift[0];
 837     facel            = _mm_set1_ps(fr->epsfac);
 838     charge           = mdatoms->chargeA;
 839     nvdwtype         = fr->ntype;
 840     vdwparam         = fr->nbfp;
 841     vdwtype          = mdatoms->typeA;
 842
 843     vftab            = kernel_data->table_elec_vdw->data;
 844     vftabscale       = _mm_set1_ps(kernel_data->table_elec_vdw->scale);
 845
 846     /* Setup water-specific parameters */
 847     inr              = nlist->iinr[0];
 848     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 849     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 850     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 851     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 852
 853     /* Avoid stupid compiler warnings */
 854     jnrA = jnrB = jnrC = jnrD = 0;
 855     j_coord_offsetA = 0;
 856     j_coord_offsetB = 0;
 857     j_coord_offsetC = 0;
 858     j_coord_offsetD = 0;
 859
 860     outeriter        = 0;
 861     inneriter        = 0;
 862
 863     for(iidx=0;iidx<4*DIM;iidx++)
 864     {
 865         scratch[iidx] = 0.0;
 866     }
 867
 868     /* Start outer loop over neighborlists */
 869     for(iidx=0; iidx<nri; iidx++)
 870     {
 871         /* Load shift vector for this list */
 872         i_shift_offset   = DIM*shiftidx[iidx];
 873
 874         /* Load limits for loop over neighbors */
 875         j_index_start    = jindex[iidx];
 876         j_index_end      = jindex[iidx+1];
 877
 878         /* Get outer coordinate index */
 879         inr              = iinr[iidx];
 880         i_coord_offset   = DIM*inr;
 881
 882         /* Load i particle coords and add shift vector */
 883         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 884                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 885
 886         fix0             = _mm_setzero_ps();
 887         fiy0             = _mm_setzero_ps();
 888         fiz0             = _mm_setzero_ps();
 889         fix1             = _mm_setzero_ps();
 890         fiy1             = _mm_setzero_ps();
 891         fiz1             = _mm_setzero_ps();
 892         fix2             = _mm_setzero_ps();
 893         fiy2             = _mm_setzero_ps();
 894         fiz2             = _mm_setzero_ps();
 895         fix3             = _mm_setzero_ps();
 896         fiy3             = _mm_setzero_ps();
 897         fiz3             = _mm_setzero_ps();
 898
 899         /* Start inner kernel loop */
 900         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 901         {
 902
 903             /* Get j neighbor index, and coordinate index */
 904             jnrA             = jjnr[jidx];
 905             jnrB             = jjnr[jidx+1];
 906             jnrC             = jjnr[jidx+2];
 907             jnrD             = jjnr[jidx+3];
 908             j_coord_offsetA  = DIM*jnrA;
 909             j_coord_offsetB  = DIM*jnrB;
 910             j_coord_offsetC  = DIM*jnrC;
 911             j_coord_offsetD  = DIM*jnrD;
 912
 913             /* load j atom coordinates */
 914             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 915                                               x+j_coord_offsetC,x+j_coord_offsetD,
 916                                               &jx0,&jy0,&jz0);
 917
 918             /* Calculate displacement vector */
 919             dx00             = _mm_sub_ps(ix0,jx0);
 920             dy00             = _mm_sub_ps(iy0,jy0);
 921             dz00             = _mm_sub_ps(iz0,jz0);
 922             dx10             = _mm_sub_ps(ix1,jx0);
 923             dy10             = _mm_sub_ps(iy1,jy0);
 924             dz10             = _mm_sub_ps(iz1,jz0);
 925             dx20             = _mm_sub_ps(ix2,jx0);
 926             dy20             = _mm_sub_ps(iy2,jy0);
 927             dz20             = _mm_sub_ps(iz2,jz0);
 928             dx30             = _mm_sub_ps(ix3,jx0);
 929             dy30             = _mm_sub_ps(iy3,jy0);
 930             dz30             = _mm_sub_ps(iz3,jz0);
 931
 932             /* Calculate squared distance and things based on it */
 933             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 934             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 935             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 936             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 937
 938             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 939             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 940             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 941             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 942
 943             /* Load parameters for j particles */
 944             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 945                                                               charge+jnrC+0,charge+jnrD+0);
 946             vdwjidx0A        = 2*vdwtype[jnrA+0];
 947             vdwjidx0B        = 2*vdwtype[jnrB+0];
 948             vdwjidx0C        = 2*vdwtype[jnrC+0];
 949             vdwjidx0D        = 2*vdwtype[jnrD+0];
 950
 951             fjx0             = _mm_setzero_ps();
 952             fjy0             = _mm_setzero_ps();
 953             fjz0             = _mm_setzero_ps();
 954
 955             /**************************
 956              * CALCULATE INTERACTIONS *
 957              **************************/
 958
 959             r00              = _mm_mul_ps(rsq00,rinv00);
 960
 961             /* Compute parameters for interactions between i and j atoms */
 962             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 963                                          vdwparam+vdwioffset0+vdwjidx0B,
 964                                          vdwparam+vdwioffset0+vdwjidx0C,
 965                                          vdwparam+vdwioffset0+vdwjidx0D,
 966                                          &c6_00,&c12_00);
 967
 968             /* Calculate table index by multiplying r with table scale and truncate to integer */
 969             rt               = _mm_mul_ps(r00,vftabscale);
 970             vfitab           = _mm_cvttps_epi32(rt);
 971 #ifdef __XOP__
 972             vfeps            = _mm_frcz_ps(rt);
 973 #else
 974             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 975 #endif
 976             twovfeps         = _mm_add_ps(vfeps,vfeps);
 977             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
 978
 979             /* CUBIC SPLINE TABLE DISPERSION */
 980             vfitab           = _mm_add_epi32(vfitab,ifour);
 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            = fvdw;
1003
1004              /* Update vectorial force */
1005             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1006             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1007             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1008
1009             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1010             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1011             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1012
1013             /**************************
1014              * CALCULATE INTERACTIONS *
1015              **************************/
1016
1017             r10              = _mm_mul_ps(rsq10,rinv10);
1018
1019             /* Compute parameters for interactions between i and j atoms */
1020             qq10             = _mm_mul_ps(iq1,jq0);
1021
1022             /* Calculate table index by multiplying r with table scale and truncate to integer */
1023             rt               = _mm_mul_ps(r10,vftabscale);
1024             vfitab           = _mm_cvttps_epi32(rt);
1025 #ifdef __XOP__
1026             vfeps            = _mm_frcz_ps(rt);
1027 #else
1028             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1029 #endif
1030             twovfeps         = _mm_add_ps(vfeps,vfeps);
1031             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1032
1033             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1034             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1035             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1036             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1037             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1038             _MM_TRANSPOSE4_PS(Y,F,G,H);
1039             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1040             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1041             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1042
1043             fscal            = felec;
1044
1045              /* Update vectorial force */
1046             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1047             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1048             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1049
1050             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1051             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1052             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1053
1054             /**************************
1055              * CALCULATE INTERACTIONS *
1056              **************************/
1057
1058             r20              = _mm_mul_ps(rsq20,rinv20);
1059
1060             /* Compute parameters for interactions between i and j atoms */
1061             qq20             = _mm_mul_ps(iq2,jq0);
1062
1063             /* Calculate table index by multiplying r with table scale and truncate to integer */
1064             rt               = _mm_mul_ps(r20,vftabscale);
1065             vfitab           = _mm_cvttps_epi32(rt);
1066 #ifdef __XOP__
1067             vfeps            = _mm_frcz_ps(rt);
1068 #else
1069             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1070 #endif
1071             twovfeps         = _mm_add_ps(vfeps,vfeps);
1072             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1073
1074             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1075             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1076             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1077             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1078             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1079             _MM_TRANSPOSE4_PS(Y,F,G,H);
1080             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1081             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1082             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1083
1084             fscal            = felec;
1085
1086              /* Update vectorial force */
1087             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1088             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1089             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1090
1091             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1092             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1093             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1094
1095             /**************************
1096              * CALCULATE INTERACTIONS *
1097              **************************/
1098
1099             r30              = _mm_mul_ps(rsq30,rinv30);
1100
1101             /* Compute parameters for interactions between i and j atoms */
1102             qq30             = _mm_mul_ps(iq3,jq0);
1103
1104             /* Calculate table index by multiplying r with table scale and truncate to integer */
1105             rt               = _mm_mul_ps(r30,vftabscale);
1106             vfitab           = _mm_cvttps_epi32(rt);
1107 #ifdef __XOP__
1108             vfeps            = _mm_frcz_ps(rt);
1109 #else
1110             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1111 #endif
1112             twovfeps         = _mm_add_ps(vfeps,vfeps);
1113             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1114
1115             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1116             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1117             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1118             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1119             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1120             _MM_TRANSPOSE4_PS(Y,F,G,H);
1121             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1122             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1123             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1124
1125             fscal            = felec;
1126
1127              /* Update vectorial force */
1128             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1129             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1130             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1131
1132             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1133             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1134             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1135
1136             fjptrA             = f+j_coord_offsetA;
1137             fjptrB             = f+j_coord_offsetB;
1138             fjptrC             = f+j_coord_offsetC;
1139             fjptrD             = f+j_coord_offsetD;
1140
1141             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1142
1143             /* Inner loop uses 177 flops */
1144         }
1145
1146         if(jidx<j_index_end)
1147         {
1148
1149             /* Get j neighbor index, and coordinate index */
1150             jnrlistA         = jjnr[jidx];
1151             jnrlistB         = jjnr[jidx+1];
1152             jnrlistC         = jjnr[jidx+2];
1153             jnrlistD         = jjnr[jidx+3];
1154             /* Sign of each element will be negative for non-real atoms.
1155              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1156              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1157              */
1158             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1159             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1160             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1161             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1162             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1163             j_coord_offsetA  = DIM*jnrA;
1164             j_coord_offsetB  = DIM*jnrB;
1165             j_coord_offsetC  = DIM*jnrC;
1166             j_coord_offsetD  = DIM*jnrD;
1167
1168             /* load j atom coordinates */
1169             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1170                                               x+j_coord_offsetC,x+j_coord_offsetD,
1171                                               &jx0,&jy0,&jz0);
1172
1173             /* Calculate displacement vector */
1174             dx00             = _mm_sub_ps(ix0,jx0);
1175             dy00             = _mm_sub_ps(iy0,jy0);
1176             dz00             = _mm_sub_ps(iz0,jz0);
1177             dx10             = _mm_sub_ps(ix1,jx0);
1178             dy10             = _mm_sub_ps(iy1,jy0);
1179             dz10             = _mm_sub_ps(iz1,jz0);
1180             dx20             = _mm_sub_ps(ix2,jx0);
1181             dy20             = _mm_sub_ps(iy2,jy0);
1182             dz20             = _mm_sub_ps(iz2,jz0);
1183             dx30             = _mm_sub_ps(ix3,jx0);
1184             dy30             = _mm_sub_ps(iy3,jy0);
1185             dz30             = _mm_sub_ps(iz3,jz0);
1186
1187             /* Calculate squared distance and things based on it */
1188             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1189             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1190             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1191             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1192
1193             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1194             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1195             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1196             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1197
1198             /* Load parameters for j particles */
1199             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1200                                                               charge+jnrC+0,charge+jnrD+0);
1201             vdwjidx0A        = 2*vdwtype[jnrA+0];
1202             vdwjidx0B        = 2*vdwtype[jnrB+0];
1203             vdwjidx0C        = 2*vdwtype[jnrC+0];
1204             vdwjidx0D        = 2*vdwtype[jnrD+0];
1205
1206             fjx0             = _mm_setzero_ps();
1207             fjy0             = _mm_setzero_ps();
1208             fjz0             = _mm_setzero_ps();
1209
1210             /**************************
1211              * CALCULATE INTERACTIONS *
1212              **************************/
1213
1214             r00              = _mm_mul_ps(rsq00,rinv00);
1215             r00              = _mm_andnot_ps(dummy_mask,r00);
1216
1217             /* Compute parameters for interactions between i and j atoms */
1218             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1219                                          vdwparam+vdwioffset0+vdwjidx0B,
1220                                          vdwparam+vdwioffset0+vdwjidx0C,
1221                                          vdwparam+vdwioffset0+vdwjidx0D,
1222                                          &c6_00,&c12_00);
1223
1224             /* Calculate table index by multiplying r with table scale and truncate to integer */
1225             rt               = _mm_mul_ps(r00,vftabscale);
1226             vfitab           = _mm_cvttps_epi32(rt);
1227 #ifdef __XOP__
1228             vfeps            = _mm_frcz_ps(rt);
1229 #else
1230             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1231 #endif
1232             twovfeps         = _mm_add_ps(vfeps,vfeps);
1233             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1234
1235             /* CUBIC SPLINE TABLE DISPERSION */
1236             vfitab           = _mm_add_epi32(vfitab,ifour);
1237             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1238             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1239             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1240             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1241             _MM_TRANSPOSE4_PS(Y,F,G,H);
1242             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1243             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1244             fvdw6            = _mm_mul_ps(c6_00,FF);
1245
1246             /* CUBIC SPLINE TABLE REPULSION */
1247             vfitab           = _mm_add_epi32(vfitab,ifour);
1248             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1249             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1250             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1251             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1252             _MM_TRANSPOSE4_PS(Y,F,G,H);
1253             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1254             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1255             fvdw12           = _mm_mul_ps(c12_00,FF);
1256             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1257
1258             fscal            = fvdw;
1259
1260             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1261
1262              /* Update vectorial force */
1263             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1264             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1265             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1266
1267             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1268             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1269             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1270
1271             /**************************
1272              * CALCULATE INTERACTIONS *
1273              **************************/
1274
1275             r10              = _mm_mul_ps(rsq10,rinv10);
1276             r10              = _mm_andnot_ps(dummy_mask,r10);
1277
1278             /* Compute parameters for interactions between i and j atoms */
1279             qq10             = _mm_mul_ps(iq1,jq0);
1280
1281             /* Calculate table index by multiplying r with table scale and truncate to integer */
1282             rt               = _mm_mul_ps(r10,vftabscale);
1283             vfitab           = _mm_cvttps_epi32(rt);
1284 #ifdef __XOP__
1285             vfeps            = _mm_frcz_ps(rt);
1286 #else
1287             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1288 #endif
1289             twovfeps         = _mm_add_ps(vfeps,vfeps);
1290             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1291
1292             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1293             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1294             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1295             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1296             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1297             _MM_TRANSPOSE4_PS(Y,F,G,H);
1298             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1299             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1300             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1301
1302             fscal            = felec;
1303
1304             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1305
1306              /* Update vectorial force */
1307             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1308             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1309             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1310
1311             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1312             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1313             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1314
1315             /**************************
1316              * CALCULATE INTERACTIONS *
1317              **************************/
1318
1319             r20              = _mm_mul_ps(rsq20,rinv20);
1320             r20              = _mm_andnot_ps(dummy_mask,r20);
1321
1322             /* Compute parameters for interactions between i and j atoms */
1323             qq20             = _mm_mul_ps(iq2,jq0);
1324
1325             /* Calculate table index by multiplying r with table scale and truncate to integer */
1326             rt               = _mm_mul_ps(r20,vftabscale);
1327             vfitab           = _mm_cvttps_epi32(rt);
1328 #ifdef __XOP__
1329             vfeps            = _mm_frcz_ps(rt);
1330 #else
1331             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1332 #endif
1333             twovfeps         = _mm_add_ps(vfeps,vfeps);
1334             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1335
1336             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1337             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1338             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1339             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1340             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1341             _MM_TRANSPOSE4_PS(Y,F,G,H);
1342             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1343             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1344             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1345
1346             fscal            = felec;
1347
1348             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1349
1350              /* Update vectorial force */
1351             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1352             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1353             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1354
1355             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1356             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1357             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1358
1359             /**************************
1360              * CALCULATE INTERACTIONS *
1361              **************************/
1362
1363             r30              = _mm_mul_ps(rsq30,rinv30);
1364             r30              = _mm_andnot_ps(dummy_mask,r30);
1365
1366             /* Compute parameters for interactions between i and j atoms */
1367             qq30             = _mm_mul_ps(iq3,jq0);
1368
1369             /* Calculate table index by multiplying r with table scale and truncate to integer */
1370             rt               = _mm_mul_ps(r30,vftabscale);
1371             vfitab           = _mm_cvttps_epi32(rt);
1372 #ifdef __XOP__
1373             vfeps            = _mm_frcz_ps(rt);
1374 #else
1375             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1376 #endif
1377             twovfeps         = _mm_add_ps(vfeps,vfeps);
1378             vfitab           = _mm_slli_epi32(_mm_add_epi32(vfitab,_mm_slli_epi32(vfitab,1)),2);
1379
1380             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1381             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1382             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1383             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1384             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1385             _MM_TRANSPOSE4_PS(Y,F,G,H);
1386             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1387             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1388             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1389
1390             fscal            = felec;
1391
1392             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1393
1394              /* Update vectorial force */
1395             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1396             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1397             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1398
1399             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1400             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1401             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1402
1403             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1404             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1405             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1406             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1407
1408             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1409
1410             /* Inner loop uses 181 flops */
1411         }
1412
1413         /* End of innermost loop */
1414
1415         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1416                                               f+i_coord_offset,fshift+i_shift_offset);
1417
1418         /* Increment number of inner iterations */
1419         inneriter                  += j_index_end - j_index_start;
1420
1421         /* Outer loop uses 24 flops */
1422     }
1423
1424     /* Increment number of outer iterations */
1425     outeriter        += nri;
1426
1427     /* Update outer/inner flops */
1428
1429     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*181);
1430 }