src/gmxlib/nonbonded/nb_kernel_avx_128_fma_single/nb_kernel_ElecCSTab_VdwLJ_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_VdwLJ_GeomW4P1_VF_avx_128_fma_single
  38  * Electrostatics interaction: CubicSplineTable
  39  * VdW interaction:            LennardJones
  40  * Geometry:                   Water4-Particle
  41  * Calculate force/pot:        PotentialAndForce
  42  */
  43 void
  44 nb_kernel_ElecCSTab_VdwLJ_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->data;
 117     vftabscale       = _mm_set1_ps(kernel_data->table_elec->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             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 216             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 217             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 218
 219             rinvsq00         = gmx_mm_inv_ps(rsq00);
 220
 221             /* Load parameters for j particles */
 222             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 223                                                               charge+jnrC+0,charge+jnrD+0);
 224             vdwjidx0A        = 2*vdwtype[jnrA+0];
 225             vdwjidx0B        = 2*vdwtype[jnrB+0];
 226             vdwjidx0C        = 2*vdwtype[jnrC+0];
 227             vdwjidx0D        = 2*vdwtype[jnrD+0];
 228
 229             fjx0             = _mm_setzero_ps();
 230             fjy0             = _mm_setzero_ps();
 231             fjz0             = _mm_setzero_ps();
 232
 233             /**************************
 234              * CALCULATE INTERACTIONS *
 235              **************************/
 236
 237             /* Compute parameters for interactions between i and j atoms */
 238             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 239                                          vdwparam+vdwioffset0+vdwjidx0B,
 240                                          vdwparam+vdwioffset0+vdwjidx0C,
 241                                          vdwparam+vdwioffset0+vdwjidx0D,
 242                                          &c6_00,&c12_00);
 243
 244             /* LENNARD-JONES DISPERSION/REPULSION */
 245
 246             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 247             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 248             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 249             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 250             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 251
 252             /* Update potential sum for this i atom from the interaction with this j atom. */
 253             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 254
 255             fscal            = fvdw;
 256
 257              /* Update vectorial force */
 258             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 259             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 260             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 261
 262             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 263             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 264             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 265
 266             /**************************
 267              * CALCULATE INTERACTIONS *
 268              **************************/
 269
 270             r10              = _mm_mul_ps(rsq10,rinv10);
 271
 272             /* Compute parameters for interactions between i and j atoms */
 273             qq10             = _mm_mul_ps(iq1,jq0);
 274
 275             /* Calculate table index by multiplying r with table scale and truncate to integer */
 276             rt               = _mm_mul_ps(r10,vftabscale);
 277             vfitab           = _mm_cvttps_epi32(rt);
 278 #ifdef __XOP__
 279             vfeps            = _mm_frcz_ps(rt);
 280 #else
 281             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 282 #endif
 283             twovfeps         = _mm_add_ps(vfeps,vfeps);
 284             vfitab           = _mm_slli_epi32(vfitab,2);
 285
 286             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 287             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 293             VV               = _mm_macc_ps(vfeps,Fp,Y);
 294             velec            = _mm_mul_ps(qq10,VV);
 295             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 296             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 297
 298             /* Update potential sum for this i atom from the interaction with this j atom. */
 299             velecsum         = _mm_add_ps(velecsum,velec);
 300
 301             fscal            = felec;
 302
 303              /* Update vectorial force */
 304             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 305             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 306             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 307
 308             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 309             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 310             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 311
 312             /**************************
 313              * CALCULATE INTERACTIONS *
 314              **************************/
 315
 316             r20              = _mm_mul_ps(rsq20,rinv20);
 317
 318             /* Compute parameters for interactions between i and j atoms */
 319             qq20             = _mm_mul_ps(iq2,jq0);
 320
 321             /* Calculate table index by multiplying r with table scale and truncate to integer */
 322             rt               = _mm_mul_ps(r20,vftabscale);
 323             vfitab           = _mm_cvttps_epi32(rt);
 324 #ifdef __XOP__
 325             vfeps            = _mm_frcz_ps(rt);
 326 #else
 327             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 328 #endif
 329             twovfeps         = _mm_add_ps(vfeps,vfeps);
 330             vfitab           = _mm_slli_epi32(vfitab,2);
 331
 332             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 333             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 334             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 335             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 336             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 337             _MM_TRANSPOSE4_PS(Y,F,G,H);
 338             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 339             VV               = _mm_macc_ps(vfeps,Fp,Y);
 340             velec            = _mm_mul_ps(qq20,VV);
 341             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 342             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 343
 344             /* Update potential sum for this i atom from the interaction with this j atom. */
 345             velecsum         = _mm_add_ps(velecsum,velec);
 346
 347             fscal            = felec;
 348
 349              /* Update vectorial force */
 350             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 351             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 352             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 353
 354             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 355             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 356             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 357
 358             /**************************
 359              * CALCULATE INTERACTIONS *
 360              **************************/
 361
 362             r30              = _mm_mul_ps(rsq30,rinv30);
 363
 364             /* Compute parameters for interactions between i and j atoms */
 365             qq30             = _mm_mul_ps(iq3,jq0);
 366
 367             /* Calculate table index by multiplying r with table scale and truncate to integer */
 368             rt               = _mm_mul_ps(r30,vftabscale);
 369             vfitab           = _mm_cvttps_epi32(rt);
 370 #ifdef __XOP__
 371             vfeps            = _mm_frcz_ps(rt);
 372 #else
 373             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 374 #endif
 375             twovfeps         = _mm_add_ps(vfeps,vfeps);
 376             vfitab           = _mm_slli_epi32(vfitab,2);
 377
 378             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 379             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 380             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 381             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 382             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 383             _MM_TRANSPOSE4_PS(Y,F,G,H);
 384             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 385             VV               = _mm_macc_ps(vfeps,Fp,Y);
 386             velec            = _mm_mul_ps(qq30,VV);
 387             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 388             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 389
 390             /* Update potential sum for this i atom from the interaction with this j atom. */
 391             velecsum         = _mm_add_ps(velecsum,velec);
 392
 393             fscal            = felec;
 394
 395              /* Update vectorial force */
 396             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 397             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 398             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 399
 400             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 401             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 402             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 403
 404             fjptrA             = f+j_coord_offsetA;
 405             fjptrB             = f+j_coord_offsetB;
 406             fjptrC             = f+j_coord_offsetC;
 407             fjptrD             = f+j_coord_offsetD;
 408
 409             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 410
 411             /* Inner loop uses 173 flops */
 412         }
 413
 414         if(jidx<j_index_end)
 415         {
 416
 417             /* Get j neighbor index, and coordinate index */
 418             jnrlistA         = jjnr[jidx];
 419             jnrlistB         = jjnr[jidx+1];
 420             jnrlistC         = jjnr[jidx+2];
 421             jnrlistD         = jjnr[jidx+3];
 422             /* Sign of each element will be negative for non-real atoms.
 423              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 424              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 425              */
 426             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 427             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 428             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 429             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 430             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 431             j_coord_offsetA  = DIM*jnrA;
 432             j_coord_offsetB  = DIM*jnrB;
 433             j_coord_offsetC  = DIM*jnrC;
 434             j_coord_offsetD  = DIM*jnrD;
 435
 436             /* load j atom coordinates */
 437             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 438                                               x+j_coord_offsetC,x+j_coord_offsetD,
 439                                               &jx0,&jy0,&jz0);
 440
 441             /* Calculate displacement vector */
 442             dx00             = _mm_sub_ps(ix0,jx0);
 443             dy00             = _mm_sub_ps(iy0,jy0);
 444             dz00             = _mm_sub_ps(iz0,jz0);
 445             dx10             = _mm_sub_ps(ix1,jx0);
 446             dy10             = _mm_sub_ps(iy1,jy0);
 447             dz10             = _mm_sub_ps(iz1,jz0);
 448             dx20             = _mm_sub_ps(ix2,jx0);
 449             dy20             = _mm_sub_ps(iy2,jy0);
 450             dz20             = _mm_sub_ps(iz2,jz0);
 451             dx30             = _mm_sub_ps(ix3,jx0);
 452             dy30             = _mm_sub_ps(iy3,jy0);
 453             dz30             = _mm_sub_ps(iz3,jz0);
 454
 455             /* Calculate squared distance and things based on it */
 456             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 457             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 458             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 459             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 460
 461             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 462             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 463             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 464
 465             rinvsq00         = gmx_mm_inv_ps(rsq00);
 466
 467             /* Load parameters for j particles */
 468             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 469                                                               charge+jnrC+0,charge+jnrD+0);
 470             vdwjidx0A        = 2*vdwtype[jnrA+0];
 471             vdwjidx0B        = 2*vdwtype[jnrB+0];
 472             vdwjidx0C        = 2*vdwtype[jnrC+0];
 473             vdwjidx0D        = 2*vdwtype[jnrD+0];
 474
 475             fjx0             = _mm_setzero_ps();
 476             fjy0             = _mm_setzero_ps();
 477             fjz0             = _mm_setzero_ps();
 478
 479             /**************************
 480              * CALCULATE INTERACTIONS *
 481              **************************/
 482
 483             /* Compute parameters for interactions between i and j atoms */
 484             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 485                                          vdwparam+vdwioffset0+vdwjidx0B,
 486                                          vdwparam+vdwioffset0+vdwjidx0C,
 487                                          vdwparam+vdwioffset0+vdwjidx0D,
 488                                          &c6_00,&c12_00);
 489
 490             /* LENNARD-JONES DISPERSION/REPULSION */
 491
 492             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 493             vvdw6            = _mm_mul_ps(c6_00,rinvsix);
 494             vvdw12           = _mm_mul_ps(c12_00,_mm_mul_ps(rinvsix,rinvsix));
 495             vvdw             = _mm_msub_ps(vvdw12,one_twelfth,_mm_mul_ps(vvdw6,one_sixth));
 496             fvdw             = _mm_mul_ps(_mm_sub_ps(vvdw12,vvdw6),rinvsq00);
 497
 498             /* Update potential sum for this i atom from the interaction with this j atom. */
 499             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 500             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 501
 502             fscal            = fvdw;
 503
 504             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 505
 506              /* Update vectorial force */
 507             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 508             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 509             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 510
 511             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 512             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 513             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 514
 515             /**************************
 516              * CALCULATE INTERACTIONS *
 517              **************************/
 518
 519             r10              = _mm_mul_ps(rsq10,rinv10);
 520             r10              = _mm_andnot_ps(dummy_mask,r10);
 521
 522             /* Compute parameters for interactions between i and j atoms */
 523             qq10             = _mm_mul_ps(iq1,jq0);
 524
 525             /* Calculate table index by multiplying r with table scale and truncate to integer */
 526             rt               = _mm_mul_ps(r10,vftabscale);
 527             vfitab           = _mm_cvttps_epi32(rt);
 528 #ifdef __XOP__
 529             vfeps            = _mm_frcz_ps(rt);
 530 #else
 531             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 532 #endif
 533             twovfeps         = _mm_add_ps(vfeps,vfeps);
 534             vfitab           = _mm_slli_epi32(vfitab,2);
 535
 536             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 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             velec            = _mm_mul_ps(qq10,VV);
 545             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 546             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 547
 548             /* Update potential sum for this i atom from the interaction with this j atom. */
 549             velec            = _mm_andnot_ps(dummy_mask,velec);
 550             velecsum         = _mm_add_ps(velecsum,velec);
 551
 552             fscal            = felec;
 553
 554             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 555
 556              /* Update vectorial force */
 557             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 558             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 559             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 560
 561             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 562             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 563             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 564
 565             /**************************
 566              * CALCULATE INTERACTIONS *
 567              **************************/
 568
 569             r20              = _mm_mul_ps(rsq20,rinv20);
 570             r20              = _mm_andnot_ps(dummy_mask,r20);
 571
 572             /* Compute parameters for interactions between i and j atoms */
 573             qq20             = _mm_mul_ps(iq2,jq0);
 574
 575             /* Calculate table index by multiplying r with table scale and truncate to integer */
 576             rt               = _mm_mul_ps(r20,vftabscale);
 577             vfitab           = _mm_cvttps_epi32(rt);
 578 #ifdef __XOP__
 579             vfeps            = _mm_frcz_ps(rt);
 580 #else
 581             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 582 #endif
 583             twovfeps         = _mm_add_ps(vfeps,vfeps);
 584             vfitab           = _mm_slli_epi32(vfitab,2);
 585
 586             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 587             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 588             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 589             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 590             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 591             _MM_TRANSPOSE4_PS(Y,F,G,H);
 592             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 593             VV               = _mm_macc_ps(vfeps,Fp,Y);
 594             velec            = _mm_mul_ps(qq20,VV);
 595             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 596             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 597
 598             /* Update potential sum for this i atom from the interaction with this j atom. */
 599             velec            = _mm_andnot_ps(dummy_mask,velec);
 600             velecsum         = _mm_add_ps(velecsum,velec);
 601
 602             fscal            = felec;
 603
 604             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 605
 606              /* Update vectorial force */
 607             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 608             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 609             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 610
 611             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 612             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 613             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 614
 615             /**************************
 616              * CALCULATE INTERACTIONS *
 617              **************************/
 618
 619             r30              = _mm_mul_ps(rsq30,rinv30);
 620             r30              = _mm_andnot_ps(dummy_mask,r30);
 621
 622             /* Compute parameters for interactions between i and j atoms */
 623             qq30             = _mm_mul_ps(iq3,jq0);
 624
 625             /* Calculate table index by multiplying r with table scale and truncate to integer */
 626             rt               = _mm_mul_ps(r30,vftabscale);
 627             vfitab           = _mm_cvttps_epi32(rt);
 628 #ifdef __XOP__
 629             vfeps            = _mm_frcz_ps(rt);
 630 #else
 631             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 632 #endif
 633             twovfeps         = _mm_add_ps(vfeps,vfeps);
 634             vfitab           = _mm_slli_epi32(vfitab,2);
 635
 636             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 637             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 638             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 639             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 640             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 641             _MM_TRANSPOSE4_PS(Y,F,G,H);
 642             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 643             VV               = _mm_macc_ps(vfeps,Fp,Y);
 644             velec            = _mm_mul_ps(qq30,VV);
 645             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 646             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
 647
 648             /* Update potential sum for this i atom from the interaction with this j atom. */
 649             velec            = _mm_andnot_ps(dummy_mask,velec);
 650             velecsum         = _mm_add_ps(velecsum,velec);
 651
 652             fscal            = felec;
 653
 654             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 655
 656              /* Update vectorial force */
 657             fix3             = _mm_macc_ps(dx30,fscal,fix3);
 658             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
 659             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
 660
 661             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
 662             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
 663             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
 664
 665             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 666             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 667             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 668             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 669
 670             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 671
 672             /* Inner loop uses 176 flops */
 673         }
 674
 675         /* End of innermost loop */
 676
 677         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
 678                                               f+i_coord_offset,fshift+i_shift_offset);
 679
 680         ggid                        = gid[iidx];
 681         /* Update potential energies */
 682         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 683         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 684
 685         /* Increment number of inner iterations */
 686         inneriter                  += j_index_end - j_index_start;
 687
 688         /* Outer loop uses 26 flops */
 689     }
 690
 691     /* Increment number of outer iterations */
 692     outeriter        += nri;
 693
 694     /* Update outer/inner flops */
 695
 696     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_VF,outeriter*26 + inneriter*176);
 697 }
 698 /*
 699  * Gromacs nonbonded kernel:   nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_single
 700  * Electrostatics interaction: CubicSplineTable
 701  * VdW interaction:            LennardJones
 702  * Geometry:                   Water4-Particle
 703  * Calculate force/pot:        Force
 704  */
 705 void
 706 nb_kernel_ElecCSTab_VdwLJ_GeomW4P1_F_avx_128_fma_single
 707                     (t_nblist * gmx_restrict                nlist,
 708                      rvec * gmx_restrict                    xx,
 709                      rvec * gmx_restrict                    ff,
 710                      t_forcerec * gmx_restrict              fr,
 711                      t_mdatoms * gmx_restrict               mdatoms,
 712                      nb_kernel_data_t * gmx_restrict        kernel_data,
 713                      t_nrnb * gmx_restrict                  nrnb)
 714 {
 715     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 716      * just 0 for non-waters.
 717      * Suffixes A,B,C,D refer to j loop unrolling done with AVX_128, e.g. for the four different
 718      * jnr indices corresponding to data put in the four positions in the SIMD register.
 719      */
 720     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 721     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 722     int              jnrA,jnrB,jnrC,jnrD;
 723     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 724     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 725     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 726     real             rcutoff_scalar;
 727     real             *shiftvec,*fshift,*x,*f;
 728     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 729     real             scratch[4*DIM];
 730     __m128           fscal,rcutoff,rcutoff2,jidxall;
 731     int              vdwioffset0;
 732     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 733     int              vdwioffset1;
 734     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 735     int              vdwioffset2;
 736     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 737     int              vdwioffset3;
 738     __m128           ix3,iy3,iz3,fix3,fiy3,fiz3,iq3,isai3;
 739     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 740     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 741     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 742     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 743     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 744     __m128           dx30,dy30,dz30,rsq30,rinv30,rinvsq30,r30,qq30,c6_30,c12_30;
 745     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 746     real             *charge;
 747     int              nvdwtype;
 748     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 749     int              *vdwtype;
 750     real             *vdwparam;
 751     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 752     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 753     __m128i          vfitab;
 754     __m128i          ifour       = _mm_set1_epi32(4);
 755     __m128           rt,vfeps,twovfeps,vftabscale,Y,F,G,H,Fp,VV,FF;
 756     real             *vftab;
 757     __m128           dummy_mask,cutoff_mask;
 758     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 759     __m128           one     = _mm_set1_ps(1.0);
 760     __m128           two     = _mm_set1_ps(2.0);
 761     x                = xx[0];
 762     f                = ff[0];
 763
 764     nri              = nlist->nri;
 765     iinr             = nlist->iinr;
 766     jindex           = nlist->jindex;
 767     jjnr             = nlist->jjnr;
 768     shiftidx         = nlist->shift;
 769     gid              = nlist->gid;
 770     shiftvec         = fr->shift_vec[0];
 771     fshift           = fr->fshift[0];
 772     facel            = _mm_set1_ps(fr->epsfac);
 773     charge           = mdatoms->chargeA;
 774     nvdwtype         = fr->ntype;
 775     vdwparam         = fr->nbfp;
 776     vdwtype          = mdatoms->typeA;
 777
 778     vftab            = kernel_data->table_elec->data;
 779     vftabscale       = _mm_set1_ps(kernel_data->table_elec->scale);
 780
 781     /* Setup water-specific parameters */
 782     inr              = nlist->iinr[0];
 783     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 784     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 785     iq3              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+3]));
 786     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 787
 788     /* Avoid stupid compiler warnings */
 789     jnrA = jnrB = jnrC = jnrD = 0;
 790     j_coord_offsetA = 0;
 791     j_coord_offsetB = 0;
 792     j_coord_offsetC = 0;
 793     j_coord_offsetD = 0;
 794
 795     outeriter        = 0;
 796     inneriter        = 0;
 797
 798     for(iidx=0;iidx<4*DIM;iidx++)
 799     {
 800         scratch[iidx] = 0.0;
 801     }
 802
 803     /* Start outer loop over neighborlists */
 804     for(iidx=0; iidx<nri; iidx++)
 805     {
 806         /* Load shift vector for this list */
 807         i_shift_offset   = DIM*shiftidx[iidx];
 808
 809         /* Load limits for loop over neighbors */
 810         j_index_start    = jindex[iidx];
 811         j_index_end      = jindex[iidx+1];
 812
 813         /* Get outer coordinate index */
 814         inr              = iinr[iidx];
 815         i_coord_offset   = DIM*inr;
 816
 817         /* Load i particle coords and add shift vector */
 818         gmx_mm_load_shift_and_4rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 819                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2,&ix3,&iy3,&iz3);
 820
 821         fix0             = _mm_setzero_ps();
 822         fiy0             = _mm_setzero_ps();
 823         fiz0             = _mm_setzero_ps();
 824         fix1             = _mm_setzero_ps();
 825         fiy1             = _mm_setzero_ps();
 826         fiz1             = _mm_setzero_ps();
 827         fix2             = _mm_setzero_ps();
 828         fiy2             = _mm_setzero_ps();
 829         fiz2             = _mm_setzero_ps();
 830         fix3             = _mm_setzero_ps();
 831         fiy3             = _mm_setzero_ps();
 832         fiz3             = _mm_setzero_ps();
 833
 834         /* Start inner kernel loop */
 835         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 836         {
 837
 838             /* Get j neighbor index, and coordinate index */
 839             jnrA             = jjnr[jidx];
 840             jnrB             = jjnr[jidx+1];
 841             jnrC             = jjnr[jidx+2];
 842             jnrD             = jjnr[jidx+3];
 843             j_coord_offsetA  = DIM*jnrA;
 844             j_coord_offsetB  = DIM*jnrB;
 845             j_coord_offsetC  = DIM*jnrC;
 846             j_coord_offsetD  = DIM*jnrD;
 847
 848             /* load j atom coordinates */
 849             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 850                                               x+j_coord_offsetC,x+j_coord_offsetD,
 851                                               &jx0,&jy0,&jz0);
 852
 853             /* Calculate displacement vector */
 854             dx00             = _mm_sub_ps(ix0,jx0);
 855             dy00             = _mm_sub_ps(iy0,jy0);
 856             dz00             = _mm_sub_ps(iz0,jz0);
 857             dx10             = _mm_sub_ps(ix1,jx0);
 858             dy10             = _mm_sub_ps(iy1,jy0);
 859             dz10             = _mm_sub_ps(iz1,jz0);
 860             dx20             = _mm_sub_ps(ix2,jx0);
 861             dy20             = _mm_sub_ps(iy2,jy0);
 862             dz20             = _mm_sub_ps(iz2,jz0);
 863             dx30             = _mm_sub_ps(ix3,jx0);
 864             dy30             = _mm_sub_ps(iy3,jy0);
 865             dz30             = _mm_sub_ps(iz3,jz0);
 866
 867             /* Calculate squared distance and things based on it */
 868             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 869             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 870             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 871             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
 872
 873             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 874             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 875             rinv30           = gmx_mm_invsqrt_ps(rsq30);
 876
 877             rinvsq00         = gmx_mm_inv_ps(rsq00);
 878
 879             /* Load parameters for j particles */
 880             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 881                                                               charge+jnrC+0,charge+jnrD+0);
 882             vdwjidx0A        = 2*vdwtype[jnrA+0];
 883             vdwjidx0B        = 2*vdwtype[jnrB+0];
 884             vdwjidx0C        = 2*vdwtype[jnrC+0];
 885             vdwjidx0D        = 2*vdwtype[jnrD+0];
 886
 887             fjx0             = _mm_setzero_ps();
 888             fjy0             = _mm_setzero_ps();
 889             fjz0             = _mm_setzero_ps();
 890
 891             /**************************
 892              * CALCULATE INTERACTIONS *
 893              **************************/
 894
 895             /* Compute parameters for interactions between i and j atoms */
 896             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 897                                          vdwparam+vdwioffset0+vdwjidx0B,
 898                                          vdwparam+vdwioffset0+vdwjidx0C,
 899                                          vdwparam+vdwioffset0+vdwjidx0D,
 900                                          &c6_00,&c12_00);
 901
 902             /* LENNARD-JONES DISPERSION/REPULSION */
 903
 904             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 905             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
 906
 907             fscal            = fvdw;
 908
 909              /* Update vectorial force */
 910             fix0             = _mm_macc_ps(dx00,fscal,fix0);
 911             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
 912             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
 913
 914             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
 915             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
 916             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
 917
 918             /**************************
 919              * CALCULATE INTERACTIONS *
 920              **************************/
 921
 922             r10              = _mm_mul_ps(rsq10,rinv10);
 923
 924             /* Compute parameters for interactions between i and j atoms */
 925             qq10             = _mm_mul_ps(iq1,jq0);
 926
 927             /* Calculate table index by multiplying r with table scale and truncate to integer */
 928             rt               = _mm_mul_ps(r10,vftabscale);
 929             vfitab           = _mm_cvttps_epi32(rt);
 930 #ifdef __XOP__
 931             vfeps            = _mm_frcz_ps(rt);
 932 #else
 933             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 934 #endif
 935             twovfeps         = _mm_add_ps(vfeps,vfeps);
 936             vfitab           = _mm_slli_epi32(vfitab,2);
 937
 938             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 939             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 940             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 941             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 942             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 943             _MM_TRANSPOSE4_PS(Y,F,G,H);
 944             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 945             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 946             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
 947
 948             fscal            = felec;
 949
 950              /* Update vectorial force */
 951             fix1             = _mm_macc_ps(dx10,fscal,fix1);
 952             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
 953             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
 954
 955             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
 956             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
 957             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
 958
 959             /**************************
 960              * CALCULATE INTERACTIONS *
 961              **************************/
 962
 963             r20              = _mm_mul_ps(rsq20,rinv20);
 964
 965             /* Compute parameters for interactions between i and j atoms */
 966             qq20             = _mm_mul_ps(iq2,jq0);
 967
 968             /* Calculate table index by multiplying r with table scale and truncate to integer */
 969             rt               = _mm_mul_ps(r20,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(vfitab,2);
 978
 979             /* CUBIC SPLINE TABLE ELECTROSTATICS */
 980             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
 981             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
 982             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
 983             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
 984             _MM_TRANSPOSE4_PS(Y,F,G,H);
 985             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
 986             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
 987             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
 988
 989             fscal            = felec;
 990
 991              /* Update vectorial force */
 992             fix2             = _mm_macc_ps(dx20,fscal,fix2);
 993             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
 994             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
 995
 996             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
 997             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
 998             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
 999
1000             /**************************
1001              * CALCULATE INTERACTIONS *
1002              **************************/
1003
1004             r30              = _mm_mul_ps(rsq30,rinv30);
1005
1006             /* Compute parameters for interactions between i and j atoms */
1007             qq30             = _mm_mul_ps(iq3,jq0);
1008
1009             /* Calculate table index by multiplying r with table scale and truncate to integer */
1010             rt               = _mm_mul_ps(r30,vftabscale);
1011             vfitab           = _mm_cvttps_epi32(rt);
1012 #ifdef __XOP__
1013             vfeps            = _mm_frcz_ps(rt);
1014 #else
1015             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1016 #endif
1017             twovfeps         = _mm_add_ps(vfeps,vfeps);
1018             vfitab           = _mm_slli_epi32(vfitab,2);
1019
1020             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1021             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1022             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1023             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1024             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1025             _MM_TRANSPOSE4_PS(Y,F,G,H);
1026             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1027             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1028             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1029
1030             fscal            = felec;
1031
1032              /* Update vectorial force */
1033             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1034             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1035             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1036
1037             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1038             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1039             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1040
1041             fjptrA             = f+j_coord_offsetA;
1042             fjptrB             = f+j_coord_offsetB;
1043             fjptrC             = f+j_coord_offsetC;
1044             fjptrD             = f+j_coord_offsetD;
1045
1046             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1047
1048             /* Inner loop uses 156 flops */
1049         }
1050
1051         if(jidx<j_index_end)
1052         {
1053
1054             /* Get j neighbor index, and coordinate index */
1055             jnrlistA         = jjnr[jidx];
1056             jnrlistB         = jjnr[jidx+1];
1057             jnrlistC         = jjnr[jidx+2];
1058             jnrlistD         = jjnr[jidx+3];
1059             /* Sign of each element will be negative for non-real atoms.
1060              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1061              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1062              */
1063             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1064             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1065             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1066             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1067             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1068             j_coord_offsetA  = DIM*jnrA;
1069             j_coord_offsetB  = DIM*jnrB;
1070             j_coord_offsetC  = DIM*jnrC;
1071             j_coord_offsetD  = DIM*jnrD;
1072
1073             /* load j atom coordinates */
1074             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1075                                               x+j_coord_offsetC,x+j_coord_offsetD,
1076                                               &jx0,&jy0,&jz0);
1077
1078             /* Calculate displacement vector */
1079             dx00             = _mm_sub_ps(ix0,jx0);
1080             dy00             = _mm_sub_ps(iy0,jy0);
1081             dz00             = _mm_sub_ps(iz0,jz0);
1082             dx10             = _mm_sub_ps(ix1,jx0);
1083             dy10             = _mm_sub_ps(iy1,jy0);
1084             dz10             = _mm_sub_ps(iz1,jz0);
1085             dx20             = _mm_sub_ps(ix2,jx0);
1086             dy20             = _mm_sub_ps(iy2,jy0);
1087             dz20             = _mm_sub_ps(iz2,jz0);
1088             dx30             = _mm_sub_ps(ix3,jx0);
1089             dy30             = _mm_sub_ps(iy3,jy0);
1090             dz30             = _mm_sub_ps(iz3,jz0);
1091
1092             /* Calculate squared distance and things based on it */
1093             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1094             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1095             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1096             rsq30            = gmx_mm_calc_rsq_ps(dx30,dy30,dz30);
1097
1098             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1099             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1100             rinv30           = gmx_mm_invsqrt_ps(rsq30);
1101
1102             rinvsq00         = gmx_mm_inv_ps(rsq00);
1103
1104             /* Load parameters for j particles */
1105             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1106                                                               charge+jnrC+0,charge+jnrD+0);
1107             vdwjidx0A        = 2*vdwtype[jnrA+0];
1108             vdwjidx0B        = 2*vdwtype[jnrB+0];
1109             vdwjidx0C        = 2*vdwtype[jnrC+0];
1110             vdwjidx0D        = 2*vdwtype[jnrD+0];
1111
1112             fjx0             = _mm_setzero_ps();
1113             fjy0             = _mm_setzero_ps();
1114             fjz0             = _mm_setzero_ps();
1115
1116             /**************************
1117              * CALCULATE INTERACTIONS *
1118              **************************/
1119
1120             /* Compute parameters for interactions between i and j atoms */
1121             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1122                                          vdwparam+vdwioffset0+vdwjidx0B,
1123                                          vdwparam+vdwioffset0+vdwjidx0C,
1124                                          vdwparam+vdwioffset0+vdwjidx0D,
1125                                          &c6_00,&c12_00);
1126
1127             /* LENNARD-JONES DISPERSION/REPULSION */
1128
1129             rinvsix          = _mm_mul_ps(_mm_mul_ps(rinvsq00,rinvsq00),rinvsq00);
1130             fvdw             = _mm_mul_ps(_mm_msub_ps(c12_00,rinvsix,c6_00),_mm_mul_ps(rinvsix,rinvsq00));
1131
1132             fscal            = fvdw;
1133
1134             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1135
1136              /* Update vectorial force */
1137             fix0             = _mm_macc_ps(dx00,fscal,fix0);
1138             fiy0             = _mm_macc_ps(dy00,fscal,fiy0);
1139             fiz0             = _mm_macc_ps(dz00,fscal,fiz0);
1140
1141             fjx0             = _mm_macc_ps(dx00,fscal,fjx0);
1142             fjy0             = _mm_macc_ps(dy00,fscal,fjy0);
1143             fjz0             = _mm_macc_ps(dz00,fscal,fjz0);
1144
1145             /**************************
1146              * CALCULATE INTERACTIONS *
1147              **************************/
1148
1149             r10              = _mm_mul_ps(rsq10,rinv10);
1150             r10              = _mm_andnot_ps(dummy_mask,r10);
1151
1152             /* Compute parameters for interactions between i and j atoms */
1153             qq10             = _mm_mul_ps(iq1,jq0);
1154
1155             /* Calculate table index by multiplying r with table scale and truncate to integer */
1156             rt               = _mm_mul_ps(r10,vftabscale);
1157             vfitab           = _mm_cvttps_epi32(rt);
1158 #ifdef __XOP__
1159             vfeps            = _mm_frcz_ps(rt);
1160 #else
1161             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1162 #endif
1163             twovfeps         = _mm_add_ps(vfeps,vfeps);
1164             vfitab           = _mm_slli_epi32(vfitab,2);
1165
1166             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1167             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1168             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1169             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1170             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1171             _MM_TRANSPOSE4_PS(Y,F,G,H);
1172             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1173             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1174             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq10,FF),_mm_mul_ps(vftabscale,rinv10)));
1175
1176             fscal            = felec;
1177
1178             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1179
1180              /* Update vectorial force */
1181             fix1             = _mm_macc_ps(dx10,fscal,fix1);
1182             fiy1             = _mm_macc_ps(dy10,fscal,fiy1);
1183             fiz1             = _mm_macc_ps(dz10,fscal,fiz1);
1184
1185             fjx0             = _mm_macc_ps(dx10,fscal,fjx0);
1186             fjy0             = _mm_macc_ps(dy10,fscal,fjy0);
1187             fjz0             = _mm_macc_ps(dz10,fscal,fjz0);
1188
1189             /**************************
1190              * CALCULATE INTERACTIONS *
1191              **************************/
1192
1193             r20              = _mm_mul_ps(rsq20,rinv20);
1194             r20              = _mm_andnot_ps(dummy_mask,r20);
1195
1196             /* Compute parameters for interactions between i and j atoms */
1197             qq20             = _mm_mul_ps(iq2,jq0);
1198
1199             /* Calculate table index by multiplying r with table scale and truncate to integer */
1200             rt               = _mm_mul_ps(r20,vftabscale);
1201             vfitab           = _mm_cvttps_epi32(rt);
1202 #ifdef __XOP__
1203             vfeps            = _mm_frcz_ps(rt);
1204 #else
1205             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1206 #endif
1207             twovfeps         = _mm_add_ps(vfeps,vfeps);
1208             vfitab           = _mm_slli_epi32(vfitab,2);
1209
1210             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1211             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1212             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1213             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1214             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1215             _MM_TRANSPOSE4_PS(Y,F,G,H);
1216             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1217             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1218             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq20,FF),_mm_mul_ps(vftabscale,rinv20)));
1219
1220             fscal            = felec;
1221
1222             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1223
1224              /* Update vectorial force */
1225             fix2             = _mm_macc_ps(dx20,fscal,fix2);
1226             fiy2             = _mm_macc_ps(dy20,fscal,fiy2);
1227             fiz2             = _mm_macc_ps(dz20,fscal,fiz2);
1228
1229             fjx0             = _mm_macc_ps(dx20,fscal,fjx0);
1230             fjy0             = _mm_macc_ps(dy20,fscal,fjy0);
1231             fjz0             = _mm_macc_ps(dz20,fscal,fjz0);
1232
1233             /**************************
1234              * CALCULATE INTERACTIONS *
1235              **************************/
1236
1237             r30              = _mm_mul_ps(rsq30,rinv30);
1238             r30              = _mm_andnot_ps(dummy_mask,r30);
1239
1240             /* Compute parameters for interactions between i and j atoms */
1241             qq30             = _mm_mul_ps(iq3,jq0);
1242
1243             /* Calculate table index by multiplying r with table scale and truncate to integer */
1244             rt               = _mm_mul_ps(r30,vftabscale);
1245             vfitab           = _mm_cvttps_epi32(rt);
1246 #ifdef __XOP__
1247             vfeps            = _mm_frcz_ps(rt);
1248 #else
1249             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1250 #endif
1251             twovfeps         = _mm_add_ps(vfeps,vfeps);
1252             vfitab           = _mm_slli_epi32(vfitab,2);
1253
1254             /* CUBIC SPLINE TABLE ELECTROSTATICS */
1255             Y                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,0) );
1256             F                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,1) );
1257             G                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,2) );
1258             H                = _mm_load_ps( vftab + _mm_extract_epi32(vfitab,3) );
1259             _MM_TRANSPOSE4_PS(Y,F,G,H);
1260             Fp               = _mm_macc_ps(vfeps,_mm_macc_ps(H,vfeps,G),F);
1261             FF               = _mm_macc_ps(vfeps,_mm_macc_ps(twovfeps,H,G),Fp);
1262             felec            = _mm_xor_ps(signbit,_mm_mul_ps(_mm_mul_ps(qq30,FF),_mm_mul_ps(vftabscale,rinv30)));
1263
1264             fscal            = felec;
1265
1266             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1267
1268              /* Update vectorial force */
1269             fix3             = _mm_macc_ps(dx30,fscal,fix3);
1270             fiy3             = _mm_macc_ps(dy30,fscal,fiy3);
1271             fiz3             = _mm_macc_ps(dz30,fscal,fiz3);
1272
1273             fjx0             = _mm_macc_ps(dx30,fscal,fjx0);
1274             fjy0             = _mm_macc_ps(dy30,fscal,fjy0);
1275             fjz0             = _mm_macc_ps(dz30,fscal,fjz0);
1276
1277             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1278             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1279             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1280             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1281
1282             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1283
1284             /* Inner loop uses 159 flops */
1285         }
1286
1287         /* End of innermost loop */
1288
1289         gmx_mm_update_iforce_4atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,fix3,fiy3,fiz3,
1290                                               f+i_coord_offset,fshift+i_shift_offset);
1291
1292         /* Increment number of inner iterations */
1293         inneriter                  += j_index_end - j_index_start;
1294
1295         /* Outer loop uses 24 flops */
1296     }
1297
1298     /* Increment number of outer iterations */
1299     outeriter        += nri;
1300
1301     /* Update outer/inner flops */
1302
1303     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W4_F,outeriter*24 + inneriter*159);
1304 }